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 313558 2017-02-10 16:11:11Z vangyzen $"); 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, (struct sockaddr *)udp_in, 307 up->u_tun_ctx); 308 return; 309 } 310 311 off += sizeof(struct udphdr); 312 313#ifdef IPSEC 314 /* Check AH/ESP integrity. */ 315 if (ipsec4_in_reject(n, inp)) { 316 m_freem(n); 317 IPSECSTAT_INC(ips_in_polvio); 318 return; 319 } 320#ifdef IPSEC_NAT_T 321 up = intoudpcb(inp); 322 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 323 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */ 324 n = udp4_espdecap(inp, n, off); 325 if (n == NULL) /* Consumed. */ 326 return; 327 } 328#endif /* IPSEC_NAT_T */ 329#endif /* IPSEC */ 330#ifdef MAC 331 if (mac_inpcb_check_deliver(inp, n) != 0) { 332 m_freem(n); 333 return; 334 } 335#endif /* MAC */ 336 if (inp->inp_flags & INP_CONTROLOPTS || 337 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 338#ifdef INET6 339 if (inp->inp_vflag & INP_IPV6) 340 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 341 else 342#endif /* INET6 */ 343 ip_savecontrol(inp, &opts, ip, n); 344 } 345#ifdef INET6 346 if (inp->inp_vflag & INP_IPV6) { 347 bzero(&udp_in6, sizeof(udp_in6)); 348 udp_in6.sin6_len = sizeof(udp_in6); 349 udp_in6.sin6_family = AF_INET6; 350 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 351 append_sa = (struct sockaddr *)&udp_in6; 352 } else 353#endif /* INET6 */ 354 append_sa = (struct sockaddr *)udp_in; 355 m_adj(n, off); 356 357 so = inp->inp_socket; 358 SOCKBUF_LOCK(&so->so_rcv); 359 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 360 SOCKBUF_UNLOCK(&so->so_rcv); 361 m_freem(n); 362 if (opts) 363 m_freem(opts); 364 UDPSTAT_INC(udps_fullsock); 365 } else 366 sorwakeup_locked(so); 367} 368 369void 370udp_input(struct mbuf *m, int off) 371{ 372 int iphlen = off; 373 struct ip *ip; 374 struct udphdr *uh; 375 struct ifnet *ifp; 376 struct inpcb *inp; 377 uint16_t len, ip_len; 378 struct inpcbinfo *pcbinfo; 379 struct ip save_ip; 380 struct sockaddr_in udp_in; 381 struct m_tag *fwd_tag; 382 int cscov_partial; 383 uint8_t pr; 384 385 ifp = m->m_pkthdr.rcvif; 386 UDPSTAT_INC(udps_ipackets); 387 388 /* 389 * Strip IP options, if any; should skip this, make available to 390 * user, and use on returned packets, but we don't yet have a way to 391 * check the checksum with options still present. 392 */ 393 if (iphlen > sizeof (struct ip)) { 394 ip_stripoptions(m); 395 iphlen = sizeof(struct ip); 396 } 397 398 /* 399 * Get IP and UDP header together in first mbuf. 400 */ 401 ip = mtod(m, struct ip *); 402 if (m->m_len < iphlen + sizeof(struct udphdr)) { 403 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 404 UDPSTAT_INC(udps_hdrops); 405 return; 406 } 407 ip = mtod(m, struct ip *); 408 } 409 uh = (struct udphdr *)((caddr_t)ip + iphlen); 410 pr = ip->ip_p; 411 cscov_partial = (pr == IPPROTO_UDPLITE) ? 1 : 0; 412 413 /* 414 * Destination port of 0 is illegal, based on RFC768. 415 */ 416 if (uh->uh_dport == 0) 417 goto badunlocked; 418 419 /* 420 * Construct sockaddr format source address. Stuff source address 421 * and datagram in user buffer. 422 */ 423 bzero(&udp_in, sizeof(udp_in)); 424 udp_in.sin_len = sizeof(udp_in); 425 udp_in.sin_family = AF_INET; 426 udp_in.sin_port = uh->uh_sport; 427 udp_in.sin_addr = ip->ip_src; 428 429 /* 430 * Make mbuf data length reflect UDP length. If not enough data to 431 * reflect UDP length, drop. 432 */ 433 len = ntohs((u_short)uh->uh_ulen); 434 ip_len = ntohs(ip->ip_len) - iphlen; 435 if (pr == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 436 /* Zero means checksum over the complete packet. */ 437 if (len == 0) 438 len = ip_len; 439 cscov_partial = 0; 440 } 441 if (ip_len != len) { 442 if (len > ip_len || len < sizeof(struct udphdr)) { 443 UDPSTAT_INC(udps_badlen); 444 goto badunlocked; 445 } 446 if (pr == IPPROTO_UDP) 447 m_adj(m, len - ip_len); 448 } 449 450 /* 451 * Save a copy of the IP header in case we want restore it for 452 * sending an ICMP error message in response. 453 */ 454 if (!V_udp_blackhole) 455 save_ip = *ip; 456 else 457 memset(&save_ip, 0, sizeof(save_ip)); 458 459 /* 460 * Checksum extended UDP header and data. 461 */ 462 if (uh->uh_sum) { 463 u_short uh_sum; 464 465 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 466 !cscov_partial) { 467 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 468 uh_sum = m->m_pkthdr.csum_data; 469 else 470 uh_sum = in_pseudo(ip->ip_src.s_addr, 471 ip->ip_dst.s_addr, htonl((u_short)len + 472 m->m_pkthdr.csum_data + pr)); 473 uh_sum ^= 0xffff; 474 } else { 475 char b[9]; 476 477 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 478 bzero(((struct ipovly *)ip)->ih_x1, 9); 479 ((struct ipovly *)ip)->ih_len = (pr == IPPROTO_UDP) ? 480 uh->uh_ulen : htons(ip_len); 481 uh_sum = in_cksum(m, len + sizeof (struct ip)); 482 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 483 } 484 if (uh_sum) { 485 UDPSTAT_INC(udps_badsum); 486 m_freem(m); 487 return; 488 } 489 } else { 490 if (pr == IPPROTO_UDP) { 491 UDPSTAT_INC(udps_nosum); 492 } else { 493 /* UDPLite requires a checksum */ 494 /* XXX: What is the right UDPLite MIB counter here? */ 495 m_freem(m); 496 return; 497 } 498 } 499 500 pcbinfo = get_inpcbinfo(pr); 501 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 502 in_broadcast(ip->ip_dst, ifp)) { 503 struct inpcb *last; 504 struct inpcbhead *pcblist; 505 struct ip_moptions *imo; 506 507 INP_INFO_RLOCK(pcbinfo); 508 pcblist = get_pcblist(pr); 509 last = NULL; 510 LIST_FOREACH(inp, pcblist, inp_list) { 511 if (inp->inp_lport != uh->uh_dport) 512 continue; 513#ifdef INET6 514 if ((inp->inp_vflag & INP_IPV4) == 0) 515 continue; 516#endif 517 if (inp->inp_laddr.s_addr != INADDR_ANY && 518 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 519 continue; 520 if (inp->inp_faddr.s_addr != INADDR_ANY && 521 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 522 continue; 523 if (inp->inp_fport != 0 && 524 inp->inp_fport != uh->uh_sport) 525 continue; 526 527 INP_RLOCK(inp); 528 529 /* 530 * XXXRW: Because we weren't holding either the inpcb 531 * or the hash lock when we checked for a match 532 * before, we should probably recheck now that the 533 * inpcb lock is held. 534 */ 535 536 /* 537 * Handle socket delivery policy for any-source 538 * and source-specific multicast. [RFC3678] 539 */ 540 imo = inp->inp_moptions; 541 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 542 struct sockaddr_in group; 543 int blocked; 544 if (imo == NULL) { 545 INP_RUNLOCK(inp); 546 continue; 547 } 548 bzero(&group, sizeof(struct sockaddr_in)); 549 group.sin_len = sizeof(struct sockaddr_in); 550 group.sin_family = AF_INET; 551 group.sin_addr = ip->ip_dst; 552 553 blocked = imo_multi_filter(imo, ifp, 554 (struct sockaddr *)&group, 555 (struct sockaddr *)&udp_in); 556 if (blocked != MCAST_PASS) { 557 if (blocked == MCAST_NOTGMEMBER) 558 IPSTAT_INC(ips_notmember); 559 if (blocked == MCAST_NOTSMEMBER || 560 blocked == MCAST_MUTED) 561 UDPSTAT_INC(udps_filtermcast); 562 INP_RUNLOCK(inp); 563 continue; 564 } 565 } 566 if (last != NULL) { 567 struct mbuf *n; 568 569 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) { 570 UDP_PROBE(receive, NULL, last, ip, 571 last, uh); 572 udp_append(last, ip, n, iphlen, 573 &udp_in); 574 } 575 INP_RUNLOCK(last); 576 } 577 last = inp; 578 /* 579 * Don't look for additional matches if this one does 580 * not have either the SO_REUSEPORT or SO_REUSEADDR 581 * socket options set. This heuristic avoids 582 * searching through all pcbs in the common case of a 583 * non-shared port. It assumes that an application 584 * will never clear these options after setting them. 585 */ 586 if ((last->inp_socket->so_options & 587 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 588 break; 589 } 590 591 if (last == NULL) { 592 /* 593 * No matching pcb found; discard datagram. (No need 594 * to send an ICMP Port Unreachable for a broadcast 595 * or multicast datgram.) 596 */ 597 UDPSTAT_INC(udps_noportbcast); 598 if (inp) 599 INP_RUNLOCK(inp); 600 INP_INFO_RUNLOCK(pcbinfo); 601 goto badunlocked; 602 } 603 UDP_PROBE(receive, NULL, last, ip, last, uh); 604 udp_append(last, ip, m, iphlen, &udp_in); 605 INP_RUNLOCK(last); 606 INP_INFO_RUNLOCK(pcbinfo); 607 return; 608 } 609 610 /* 611 * Locate pcb for datagram. 612 */ 613 614 /* 615 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 616 */ 617 if ((m->m_flags & M_IP_NEXTHOP) && 618 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 619 struct sockaddr_in *next_hop; 620 621 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 622 623 /* 624 * Transparently forwarded. Pretend to be the destination. 625 * Already got one like this? 626 */ 627 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 628 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 629 if (!inp) { 630 /* 631 * It's new. Try to find the ambushing socket. 632 * Because we've rewritten the destination address, 633 * any hardware-generated hash is ignored. 634 */ 635 inp = in_pcblookup(pcbinfo, ip->ip_src, 636 uh->uh_sport, next_hop->sin_addr, 637 next_hop->sin_port ? htons(next_hop->sin_port) : 638 uh->uh_dport, INPLOOKUP_WILDCARD | 639 INPLOOKUP_RLOCKPCB, ifp); 640 } 641 /* Remove the tag from the packet. We don't need it anymore. */ 642 m_tag_delete(m, fwd_tag); 643 m->m_flags &= ~M_IP_NEXTHOP; 644 } else 645 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 646 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 647 INPLOOKUP_RLOCKPCB, ifp, m); 648 if (inp == NULL) { 649 if (udp_log_in_vain) { 650 char src[INET_ADDRSTRLEN]; 651 char dst[INET_ADDRSTRLEN]; 652 653 log(LOG_INFO, 654 "Connection attempt to UDP %s:%d from %s:%d\n", 655 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 656 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 657 } 658 UDPSTAT_INC(udps_noport); 659 if (m->m_flags & (M_BCAST | M_MCAST)) { 660 UDPSTAT_INC(udps_noportbcast); 661 goto badunlocked; 662 } 663 if (V_udp_blackhole) 664 goto badunlocked; 665 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 666 goto badunlocked; 667 *ip = save_ip; 668 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 669 return; 670 } 671 672 /* 673 * Check the minimum TTL for socket. 674 */ 675 INP_RLOCK_ASSERT(inp); 676 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 677 INP_RUNLOCK(inp); 678 m_freem(m); 679 return; 680 } 681 if (cscov_partial) { 682 struct udpcb *up; 683 684 up = intoudpcb(inp); 685 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 686 INP_RUNLOCK(inp); 687 m_freem(m); 688 return; 689 } 690 } 691 692 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 693 udp_append(inp, ip, m, iphlen, &udp_in); 694 INP_RUNLOCK(inp); 695 return; 696 697badunlocked: 698 m_freem(m); 699} 700#endif /* INET */ 701 702/* 703 * Notify a udp user of an asynchronous error; just wake up so that they can 704 * collect error status. 705 */ 706struct inpcb * 707udp_notify(struct inpcb *inp, int errno) 708{ 709 710 /* 711 * While udp_ctlinput() always calls udp_notify() with a read lock 712 * when invoking it directly, in_pcbnotifyall() currently uses write 713 * locks due to sharing code with TCP. For now, accept either a read 714 * or a write lock, but a read lock is sufficient. 715 */ 716 INP_LOCK_ASSERT(inp); 717 718 inp->inp_socket->so_error = errno; 719 sorwakeup(inp->inp_socket); 720 sowwakeup(inp->inp_socket); 721 return (inp); 722} 723 724#ifdef INET 725static void 726udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 727 struct inpcbinfo *pcbinfo) 728{ 729 struct ip *ip = vip; 730 struct udphdr *uh; 731 struct in_addr faddr; 732 struct inpcb *inp; 733 734 faddr = ((struct sockaddr_in *)sa)->sin_addr; 735 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 736 return; 737 738 /* 739 * Redirects don't need to be handled up here. 740 */ 741 if (PRC_IS_REDIRECT(cmd)) 742 return; 743 744 /* 745 * Hostdead is ugly because it goes linearly through all PCBs. 746 * 747 * XXX: We never get this from ICMP, otherwise it makes an excellent 748 * DoS attack on machines with many connections. 749 */ 750 if (cmd == PRC_HOSTDEAD) 751 ip = NULL; 752 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 753 return; 754 if (ip != NULL) { 755 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 756 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 757 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 758 if (inp != NULL) { 759 INP_RLOCK_ASSERT(inp); 760 if (inp->inp_socket != NULL) { 761 udp_notify(inp, inetctlerrmap[cmd]); 762 } 763 INP_RUNLOCK(inp); 764 } 765 } else 766 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 767 udp_notify); 768} 769void 770udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 771{ 772 773 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 774} 775 776void 777udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 778{ 779 780 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 781} 782#endif /* INET */ 783 784static int 785udp_pcblist(SYSCTL_HANDLER_ARGS) 786{ 787 int error, i, n; 788 struct inpcb *inp, **inp_list; 789 inp_gen_t gencnt; 790 struct xinpgen xig; 791 792 /* 793 * The process of preparing the PCB list is too time-consuming and 794 * resource-intensive to repeat twice on every request. 795 */ 796 if (req->oldptr == 0) { 797 n = V_udbinfo.ipi_count; 798 n += imax(n / 8, 10); 799 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 800 return (0); 801 } 802 803 if (req->newptr != 0) 804 return (EPERM); 805 806 /* 807 * OK, now we're committed to doing something. 808 */ 809 INP_INFO_RLOCK(&V_udbinfo); 810 gencnt = V_udbinfo.ipi_gencnt; 811 n = V_udbinfo.ipi_count; 812 INP_INFO_RUNLOCK(&V_udbinfo); 813 814 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 815 + n * sizeof(struct xinpcb)); 816 if (error != 0) 817 return (error); 818 819 xig.xig_len = sizeof xig; 820 xig.xig_count = n; 821 xig.xig_gen = gencnt; 822 xig.xig_sogen = so_gencnt; 823 error = SYSCTL_OUT(req, &xig, sizeof xig); 824 if (error) 825 return (error); 826 827 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 828 if (inp_list == 0) 829 return (ENOMEM); 830 831 INP_INFO_RLOCK(&V_udbinfo); 832 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 833 inp = LIST_NEXT(inp, inp_list)) { 834 INP_WLOCK(inp); 835 if (inp->inp_gencnt <= gencnt && 836 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 837 in_pcbref(inp); 838 inp_list[i++] = inp; 839 } 840 INP_WUNLOCK(inp); 841 } 842 INP_INFO_RUNLOCK(&V_udbinfo); 843 n = i; 844 845 error = 0; 846 for (i = 0; i < n; i++) { 847 inp = inp_list[i]; 848 INP_RLOCK(inp); 849 if (inp->inp_gencnt <= gencnt) { 850 struct xinpcb xi; 851 852 bzero(&xi, sizeof(xi)); 853 xi.xi_len = sizeof xi; 854 /* XXX should avoid extra copy */ 855 bcopy(inp, &xi.xi_inp, sizeof *inp); 856 if (inp->inp_socket) 857 sotoxsocket(inp->inp_socket, &xi.xi_socket); 858 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 859 INP_RUNLOCK(inp); 860 error = SYSCTL_OUT(req, &xi, sizeof xi); 861 } else 862 INP_RUNLOCK(inp); 863 } 864 INP_INFO_WLOCK(&V_udbinfo); 865 for (i = 0; i < n; i++) { 866 inp = inp_list[i]; 867 INP_RLOCK(inp); 868 if (!in_pcbrele_rlocked(inp)) 869 INP_RUNLOCK(inp); 870 } 871 INP_INFO_WUNLOCK(&V_udbinfo); 872 873 if (!error) { 874 /* 875 * Give the user an updated idea of our state. If the 876 * generation differs from what we told her before, she knows 877 * that something happened while we were processing this 878 * request, and it might be necessary to retry. 879 */ 880 INP_INFO_RLOCK(&V_udbinfo); 881 xig.xig_gen = V_udbinfo.ipi_gencnt; 882 xig.xig_sogen = so_gencnt; 883 xig.xig_count = V_udbinfo.ipi_count; 884 INP_INFO_RUNLOCK(&V_udbinfo); 885 error = SYSCTL_OUT(req, &xig, sizeof xig); 886 } 887 free(inp_list, M_TEMP); 888 return (error); 889} 890 891SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 892 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 893 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 894 895#ifdef INET 896static int 897udp_getcred(SYSCTL_HANDLER_ARGS) 898{ 899 struct xucred xuc; 900 struct sockaddr_in addrs[2]; 901 struct inpcb *inp; 902 int error; 903 904 error = priv_check(req->td, PRIV_NETINET_GETCRED); 905 if (error) 906 return (error); 907 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 908 if (error) 909 return (error); 910 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 911 addrs[0].sin_addr, addrs[0].sin_port, 912 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 913 if (inp != NULL) { 914 INP_RLOCK_ASSERT(inp); 915 if (inp->inp_socket == NULL) 916 error = ENOENT; 917 if (error == 0) 918 error = cr_canseeinpcb(req->td->td_ucred, inp); 919 if (error == 0) 920 cru2x(inp->inp_cred, &xuc); 921 INP_RUNLOCK(inp); 922 } else 923 error = ENOENT; 924 if (error == 0) 925 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 926 return (error); 927} 928 929SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 930 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 931 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 932#endif /* INET */ 933 934int 935udp_ctloutput(struct socket *so, struct sockopt *sopt) 936{ 937 struct inpcb *inp; 938 struct udpcb *up; 939 int isudplite, error, optval; 940 941 error = 0; 942 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 943 inp = sotoinpcb(so); 944 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 945 INP_WLOCK(inp); 946 if (sopt->sopt_level != so->so_proto->pr_protocol) { 947#ifdef INET6 948 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 949 INP_WUNLOCK(inp); 950 error = ip6_ctloutput(so, sopt); 951 } 952#endif 953#if defined(INET) && defined(INET6) 954 else 955#endif 956#ifdef INET 957 { 958 INP_WUNLOCK(inp); 959 error = ip_ctloutput(so, sopt); 960 } 961#endif 962 return (error); 963 } 964 965 switch (sopt->sopt_dir) { 966 case SOPT_SET: 967 switch (sopt->sopt_name) { 968 case UDP_ENCAP: 969 INP_WUNLOCK(inp); 970 error = sooptcopyin(sopt, &optval, sizeof optval, 971 sizeof optval); 972 if (error) 973 break; 974 inp = sotoinpcb(so); 975 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 976 INP_WLOCK(inp); 977#ifdef IPSEC_NAT_T 978 up = intoudpcb(inp); 979 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 980#endif 981 switch (optval) { 982 case 0: 983 /* Clear all UDP encap. */ 984#ifdef IPSEC_NAT_T 985 up->u_flags &= ~UF_ESPINUDP_ALL; 986#endif 987 break; 988#ifdef IPSEC_NAT_T 989 case UDP_ENCAP_ESPINUDP: 990 case UDP_ENCAP_ESPINUDP_NON_IKE: 991 up->u_flags &= ~UF_ESPINUDP_ALL; 992 if (optval == UDP_ENCAP_ESPINUDP) 993 up->u_flags |= UF_ESPINUDP; 994 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 995 up->u_flags |= UF_ESPINUDP_NON_IKE; 996 break; 997#endif 998 default: 999 error = EINVAL; 1000 break; 1001 } 1002 INP_WUNLOCK(inp); 1003 break; 1004 case UDPLITE_SEND_CSCOV: 1005 case UDPLITE_RECV_CSCOV: 1006 if (!isudplite) { 1007 INP_WUNLOCK(inp); 1008 error = ENOPROTOOPT; 1009 break; 1010 } 1011 INP_WUNLOCK(inp); 1012 error = sooptcopyin(sopt, &optval, sizeof(optval), 1013 sizeof(optval)); 1014 if (error != 0) 1015 break; 1016 inp = sotoinpcb(so); 1017 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1018 INP_WLOCK(inp); 1019 up = intoudpcb(inp); 1020 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1021 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1022 INP_WUNLOCK(inp); 1023 error = EINVAL; 1024 break; 1025 } 1026 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1027 up->u_txcslen = optval; 1028 else 1029 up->u_rxcslen = optval; 1030 INP_WUNLOCK(inp); 1031 break; 1032 default: 1033 INP_WUNLOCK(inp); 1034 error = ENOPROTOOPT; 1035 break; 1036 } 1037 break; 1038 case SOPT_GET: 1039 switch (sopt->sopt_name) { 1040#ifdef IPSEC_NAT_T 1041 case UDP_ENCAP: 1042 up = intoudpcb(inp); 1043 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1044 optval = up->u_flags & UF_ESPINUDP_ALL; 1045 INP_WUNLOCK(inp); 1046 error = sooptcopyout(sopt, &optval, sizeof optval); 1047 break; 1048#endif 1049 case UDPLITE_SEND_CSCOV: 1050 case UDPLITE_RECV_CSCOV: 1051 if (!isudplite) { 1052 INP_WUNLOCK(inp); 1053 error = ENOPROTOOPT; 1054 break; 1055 } 1056 up = intoudpcb(inp); 1057 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1058 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1059 optval = up->u_txcslen; 1060 else 1061 optval = up->u_rxcslen; 1062 INP_WUNLOCK(inp); 1063 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1064 break; 1065 default: 1066 INP_WUNLOCK(inp); 1067 error = ENOPROTOOPT; 1068 break; 1069 } 1070 break; 1071 } 1072 return (error); 1073} 1074 1075#ifdef INET 1076#define UH_WLOCKED 2 1077#define UH_RLOCKED 1 1078#define UH_UNLOCKED 0 1079static int 1080udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1081 struct mbuf *control, struct thread *td) 1082{ 1083 struct udpiphdr *ui; 1084 int len = m->m_pkthdr.len; 1085 struct in_addr faddr, laddr; 1086 struct cmsghdr *cm; 1087 struct inpcbinfo *pcbinfo; 1088 struct sockaddr_in *sin, src; 1089 int cscov_partial = 0; 1090 int error = 0; 1091 int ipflags; 1092 u_short fport, lport; 1093 int unlock_udbinfo; 1094 u_char tos; 1095 uint8_t pr; 1096 uint16_t cscov = 0; 1097 1098 /* 1099 * udp_output() may need to temporarily bind or connect the current 1100 * inpcb. As such, we don't know up front whether we will need the 1101 * pcbinfo lock or not. Do any work to decide what is needed up 1102 * front before acquiring any locks. 1103 */ 1104 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1105 if (control) 1106 m_freem(control); 1107 m_freem(m); 1108 return (EMSGSIZE); 1109 } 1110 1111 src.sin_family = 0; 1112 INP_RLOCK(inp); 1113 tos = inp->inp_ip_tos; 1114 if (control != NULL) { 1115 /* 1116 * XXX: Currently, we assume all the optional information is 1117 * stored in a single mbuf. 1118 */ 1119 if (control->m_next) { 1120 INP_RUNLOCK(inp); 1121 m_freem(control); 1122 m_freem(m); 1123 return (EINVAL); 1124 } 1125 for (; control->m_len > 0; 1126 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1127 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1128 cm = mtod(control, struct cmsghdr *); 1129 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1130 || cm->cmsg_len > control->m_len) { 1131 error = EINVAL; 1132 break; 1133 } 1134 if (cm->cmsg_level != IPPROTO_IP) 1135 continue; 1136 1137 switch (cm->cmsg_type) { 1138 case IP_SENDSRCADDR: 1139 if (cm->cmsg_len != 1140 CMSG_LEN(sizeof(struct in_addr))) { 1141 error = EINVAL; 1142 break; 1143 } 1144 bzero(&src, sizeof(src)); 1145 src.sin_family = AF_INET; 1146 src.sin_len = sizeof(src); 1147 src.sin_port = inp->inp_lport; 1148 src.sin_addr = 1149 *(struct in_addr *)CMSG_DATA(cm); 1150 break; 1151 1152 case IP_TOS: 1153 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1154 error = EINVAL; 1155 break; 1156 } 1157 tos = *(u_char *)CMSG_DATA(cm); 1158 break; 1159 1160 default: 1161 error = ENOPROTOOPT; 1162 break; 1163 } 1164 if (error) 1165 break; 1166 } 1167 m_freem(control); 1168 } 1169 if (error) { 1170 INP_RUNLOCK(inp); 1171 m_freem(m); 1172 return (error); 1173 } 1174 1175 /* 1176 * Depending on whether or not the application has bound or connected 1177 * the socket, we may have to do varying levels of work. The optimal 1178 * case is for a connected UDP socket, as a global lock isn't 1179 * required at all. 1180 * 1181 * In order to decide which we need, we require stability of the 1182 * inpcb binding, which we ensure by acquiring a read lock on the 1183 * inpcb. This doesn't strictly follow the lock order, so we play 1184 * the trylock and retry game; note that we may end up with more 1185 * conservative locks than required the second time around, so later 1186 * assertions have to accept that. Further analysis of the number of 1187 * misses under contention is required. 1188 * 1189 * XXXRW: Check that hash locking update here is correct. 1190 */ 1191 pr = inp->inp_socket->so_proto->pr_protocol; 1192 pcbinfo = get_inpcbinfo(pr); 1193 sin = (struct sockaddr_in *)addr; 1194 if (sin != NULL && 1195 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1196 INP_RUNLOCK(inp); 1197 INP_WLOCK(inp); 1198 INP_HASH_WLOCK(pcbinfo); 1199 unlock_udbinfo = UH_WLOCKED; 1200 } else if ((sin != NULL && ( 1201 (sin->sin_addr.s_addr == INADDR_ANY) || 1202 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1203 (inp->inp_laddr.s_addr == INADDR_ANY) || 1204 (inp->inp_lport == 0))) || 1205 (src.sin_family == AF_INET)) { 1206 INP_HASH_RLOCK(pcbinfo); 1207 unlock_udbinfo = UH_RLOCKED; 1208 } else 1209 unlock_udbinfo = UH_UNLOCKED; 1210 1211 /* 1212 * If the IP_SENDSRCADDR control message was specified, override the 1213 * source address for this datagram. Its use is invalidated if the 1214 * address thus specified is incomplete or clobbers other inpcbs. 1215 */ 1216 laddr = inp->inp_laddr; 1217 lport = inp->inp_lport; 1218 if (src.sin_family == AF_INET) { 1219 INP_HASH_LOCK_ASSERT(pcbinfo); 1220 if ((lport == 0) || 1221 (laddr.s_addr == INADDR_ANY && 1222 src.sin_addr.s_addr == INADDR_ANY)) { 1223 error = EINVAL; 1224 goto release; 1225 } 1226 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1227 &laddr.s_addr, &lport, td->td_ucred); 1228 if (error) 1229 goto release; 1230 } 1231 1232 /* 1233 * If a UDP socket has been connected, then a local address/port will 1234 * have been selected and bound. 1235 * 1236 * If a UDP socket has not been connected to, then an explicit 1237 * destination address must be used, in which case a local 1238 * address/port may not have been selected and bound. 1239 */ 1240 if (sin != NULL) { 1241 INP_LOCK_ASSERT(inp); 1242 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1243 error = EISCONN; 1244 goto release; 1245 } 1246 1247 /* 1248 * Jail may rewrite the destination address, so let it do 1249 * that before we use it. 1250 */ 1251 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1252 if (error) 1253 goto release; 1254 1255 /* 1256 * If a local address or port hasn't yet been selected, or if 1257 * the destination address needs to be rewritten due to using 1258 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1259 * to do the heavy lifting. Once a port is selected, we 1260 * commit the binding back to the socket; we also commit the 1261 * binding of the address if in jail. 1262 * 1263 * If we already have a valid binding and we're not 1264 * requesting a destination address rewrite, use a fast path. 1265 */ 1266 if (inp->inp_laddr.s_addr == INADDR_ANY || 1267 inp->inp_lport == 0 || 1268 sin->sin_addr.s_addr == INADDR_ANY || 1269 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1270 INP_HASH_LOCK_ASSERT(pcbinfo); 1271 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1272 &lport, &faddr.s_addr, &fport, NULL, 1273 td->td_ucred); 1274 if (error) 1275 goto release; 1276 1277 /* 1278 * XXXRW: Why not commit the port if the address is 1279 * !INADDR_ANY? 1280 */ 1281 /* Commit the local port if newly assigned. */ 1282 if (inp->inp_laddr.s_addr == INADDR_ANY && 1283 inp->inp_lport == 0) { 1284 INP_WLOCK_ASSERT(inp); 1285 INP_HASH_WLOCK_ASSERT(pcbinfo); 1286 /* 1287 * Remember addr if jailed, to prevent 1288 * rebinding. 1289 */ 1290 if (prison_flag(td->td_ucred, PR_IP4)) 1291 inp->inp_laddr = laddr; 1292 inp->inp_lport = lport; 1293 if (in_pcbinshash(inp) != 0) { 1294 inp->inp_lport = 0; 1295 error = EAGAIN; 1296 goto release; 1297 } 1298 inp->inp_flags |= INP_ANONPORT; 1299 } 1300 } else { 1301 faddr = sin->sin_addr; 1302 fport = sin->sin_port; 1303 } 1304 } else { 1305 INP_LOCK_ASSERT(inp); 1306 faddr = inp->inp_faddr; 1307 fport = inp->inp_fport; 1308 if (faddr.s_addr == INADDR_ANY) { 1309 error = ENOTCONN; 1310 goto release; 1311 } 1312 } 1313 1314 /* 1315 * Calculate data length and get a mbuf for UDP, IP, and possible 1316 * link-layer headers. Immediate slide the data pointer back forward 1317 * since we won't use that space at this layer. 1318 */ 1319 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1320 if (m == NULL) { 1321 error = ENOBUFS; 1322 goto release; 1323 } 1324 m->m_data += max_linkhdr; 1325 m->m_len -= max_linkhdr; 1326 m->m_pkthdr.len -= max_linkhdr; 1327 1328 /* 1329 * Fill in mbuf with extended UDP header and addresses and length put 1330 * into network format. 1331 */ 1332 ui = mtod(m, struct udpiphdr *); 1333 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1334 ui->ui_pr = pr; 1335 ui->ui_src = laddr; 1336 ui->ui_dst = faddr; 1337 ui->ui_sport = lport; 1338 ui->ui_dport = fport; 1339 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1340 if (pr == IPPROTO_UDPLITE) { 1341 struct udpcb *up; 1342 uint16_t plen; 1343 1344 up = intoudpcb(inp); 1345 cscov = up->u_txcslen; 1346 plen = (u_short)len + sizeof(struct udphdr); 1347 if (cscov >= plen) 1348 cscov = 0; 1349 ui->ui_len = htons(plen); 1350 ui->ui_ulen = htons(cscov); 1351 /* 1352 * For UDP-Lite, checksum coverage length of zero means 1353 * the entire UDPLite packet is covered by the checksum. 1354 */ 1355 cscov_partial = (cscov == 0) ? 0 : 1; 1356 } else 1357 ui->ui_v = IPVERSION << 4; 1358 1359 /* 1360 * Set the Don't Fragment bit in the IP header. 1361 */ 1362 if (inp->inp_flags & INP_DONTFRAG) { 1363 struct ip *ip; 1364 1365 ip = (struct ip *)&ui->ui_i; 1366 ip->ip_off |= htons(IP_DF); 1367 } 1368 1369 ipflags = 0; 1370 if (inp->inp_socket->so_options & SO_DONTROUTE) 1371 ipflags |= IP_ROUTETOIF; 1372 if (inp->inp_socket->so_options & SO_BROADCAST) 1373 ipflags |= IP_ALLOWBROADCAST; 1374 if (inp->inp_flags & INP_ONESBCAST) 1375 ipflags |= IP_SENDONES; 1376 1377#ifdef MAC 1378 mac_inpcb_create_mbuf(inp, m); 1379#endif 1380 1381 /* 1382 * Set up checksum and output datagram. 1383 */ 1384 ui->ui_sum = 0; 1385 if (pr == IPPROTO_UDPLITE) { 1386 if (inp->inp_flags & INP_ONESBCAST) 1387 faddr.s_addr = INADDR_BROADCAST; 1388 if (cscov_partial) { 1389 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1390 ui->ui_sum = 0xffff; 1391 } else { 1392 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1393 ui->ui_sum = 0xffff; 1394 } 1395 } else if (V_udp_cksum) { 1396 if (inp->inp_flags & INP_ONESBCAST) 1397 faddr.s_addr = INADDR_BROADCAST; 1398 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1399 htons((u_short)len + sizeof(struct udphdr) + pr)); 1400 m->m_pkthdr.csum_flags = CSUM_UDP; 1401 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1402 } 1403 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1404 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1405 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1406 UDPSTAT_INC(udps_opackets); 1407 1408 if (unlock_udbinfo == UH_WLOCKED) 1409 INP_HASH_WUNLOCK(pcbinfo); 1410 else if (unlock_udbinfo == UH_RLOCKED) 1411 INP_HASH_RUNLOCK(pcbinfo); 1412 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1413 error = ip_output(m, inp->inp_options, NULL, ipflags, 1414 inp->inp_moptions, inp); 1415 if (unlock_udbinfo == UH_WLOCKED) 1416 INP_WUNLOCK(inp); 1417 else 1418 INP_RUNLOCK(inp); 1419 return (error); 1420 1421release: 1422 if (unlock_udbinfo == UH_WLOCKED) { 1423 INP_HASH_WUNLOCK(pcbinfo); 1424 INP_WUNLOCK(inp); 1425 } else if (unlock_udbinfo == UH_RLOCKED) { 1426 INP_HASH_RUNLOCK(pcbinfo); 1427 INP_RUNLOCK(inp); 1428 } else 1429 INP_RUNLOCK(inp); 1430 m_freem(m); 1431 return (error); 1432} 1433 1434 1435#if defined(IPSEC) && defined(IPSEC_NAT_T) 1436/* 1437 * Potentially decap ESP in UDP frame. Check for an ESP header 1438 * and optional marker; if present, strip the UDP header and 1439 * push the result through IPSec. 1440 * 1441 * Returns mbuf to be processed (potentially re-allocated) or 1442 * NULL if consumed and/or processed. 1443 */ 1444static struct mbuf * 1445udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1446{ 1447 size_t minlen, payload, skip, iphlen; 1448 caddr_t data; 1449 struct udpcb *up; 1450 struct m_tag *tag; 1451 struct udphdr *udphdr; 1452 struct ip *ip; 1453 1454 INP_RLOCK_ASSERT(inp); 1455 1456 /* 1457 * Pull up data so the longest case is contiguous: 1458 * IP/UDP hdr + non ESP marker + ESP hdr. 1459 */ 1460 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1461 if (minlen > m->m_pkthdr.len) 1462 minlen = m->m_pkthdr.len; 1463 if ((m = m_pullup(m, minlen)) == NULL) { 1464 IPSECSTAT_INC(ips_in_inval); 1465 return (NULL); /* Bypass caller processing. */ 1466 } 1467 data = mtod(m, caddr_t); /* Points to ip header. */ 1468 payload = m->m_len - off; /* Size of payload. */ 1469 1470 if (payload == 1 && data[off] == '\xff') 1471 return (m); /* NB: keepalive packet, no decap. */ 1472 1473 up = intoudpcb(inp); 1474 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1475 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1476 ("u_flags 0x%x", up->u_flags)); 1477 1478 /* 1479 * Check that the payload is large enough to hold an 1480 * ESP header and compute the amount of data to remove. 1481 * 1482 * NB: the caller has already done a pullup for us. 1483 * XXX can we assume alignment and eliminate bcopys? 1484 */ 1485 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1486 /* 1487 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1488 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1489 * possible AH mode non-IKE marker+non-ESP marker 1490 * from draft-ietf-ipsec-udp-encaps-00.txt. 1491 */ 1492 uint64_t marker; 1493 1494 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1495 return (m); /* NB: no decap. */ 1496 bcopy(data + off, &marker, sizeof(uint64_t)); 1497 if (marker != 0) /* Non-IKE marker. */ 1498 return (m); /* NB: no decap. */ 1499 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1500 } else { 1501 uint32_t spi; 1502 1503 if (payload <= sizeof(struct esp)) { 1504 IPSECSTAT_INC(ips_in_inval); 1505 m_freem(m); 1506 return (NULL); /* Discard. */ 1507 } 1508 bcopy(data + off, &spi, sizeof(uint32_t)); 1509 if (spi == 0) /* Non-ESP marker. */ 1510 return (m); /* NB: no decap. */ 1511 skip = sizeof(struct udphdr); 1512 } 1513 1514 /* 1515 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1516 * the UDP ports. This is required if we want to select 1517 * the right SPD for multiple hosts behind same NAT. 1518 * 1519 * NB: ports are maintained in network byte order everywhere 1520 * in the NAT-T code. 1521 */ 1522 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1523 2 * sizeof(uint16_t), M_NOWAIT); 1524 if (tag == NULL) { 1525 IPSECSTAT_INC(ips_in_nomem); 1526 m_freem(m); 1527 return (NULL); /* Discard. */ 1528 } 1529 iphlen = off - sizeof(struct udphdr); 1530 udphdr = (struct udphdr *)(data + iphlen); 1531 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1532 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1533 m_tag_prepend(m, tag); 1534 1535 /* 1536 * Remove the UDP header (and possibly the non ESP marker) 1537 * IP header length is iphlen 1538 * Before: 1539 * <--- off ---> 1540 * +----+------+-----+ 1541 * | IP | UDP | ESP | 1542 * +----+------+-----+ 1543 * <-skip-> 1544 * After: 1545 * +----+-----+ 1546 * | IP | ESP | 1547 * +----+-----+ 1548 * <-skip-> 1549 */ 1550 ovbcopy(data, data + skip, iphlen); 1551 m_adj(m, skip); 1552 1553 ip = mtod(m, struct ip *); 1554 ip->ip_len = htons(ntohs(ip->ip_len) - skip); 1555 ip->ip_p = IPPROTO_ESP; 1556 1557 /* 1558 * We cannot yet update the cksums so clear any 1559 * h/w cksum flags as they are no longer valid. 1560 */ 1561 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1562 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1563 1564 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1565 return (NULL); /* NB: consumed, bypass processing. */ 1566} 1567#endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1568 1569static void 1570udp_abort(struct socket *so) 1571{ 1572 struct inpcb *inp; 1573 struct inpcbinfo *pcbinfo; 1574 1575 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1576 inp = sotoinpcb(so); 1577 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1578 INP_WLOCK(inp); 1579 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1580 INP_HASH_WLOCK(pcbinfo); 1581 in_pcbdisconnect(inp); 1582 inp->inp_laddr.s_addr = INADDR_ANY; 1583 INP_HASH_WUNLOCK(pcbinfo); 1584 soisdisconnected(so); 1585 } 1586 INP_WUNLOCK(inp); 1587} 1588 1589static int 1590udp_attach(struct socket *so, int proto, struct thread *td) 1591{ 1592 struct inpcb *inp; 1593 struct inpcbinfo *pcbinfo; 1594 int error; 1595 1596 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1597 inp = sotoinpcb(so); 1598 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1599 error = soreserve(so, udp_sendspace, udp_recvspace); 1600 if (error) 1601 return (error); 1602 INP_INFO_WLOCK(pcbinfo); 1603 error = in_pcballoc(so, pcbinfo); 1604 if (error) { 1605 INP_INFO_WUNLOCK(pcbinfo); 1606 return (error); 1607 } 1608 1609 inp = sotoinpcb(so); 1610 inp->inp_vflag |= INP_IPV4; 1611 inp->inp_ip_ttl = V_ip_defttl; 1612 1613 error = udp_newudpcb(inp); 1614 if (error) { 1615 in_pcbdetach(inp); 1616 in_pcbfree(inp); 1617 INP_INFO_WUNLOCK(pcbinfo); 1618 return (error); 1619 } 1620 1621 INP_WUNLOCK(inp); 1622 INP_INFO_WUNLOCK(pcbinfo); 1623 return (0); 1624} 1625#endif /* INET */ 1626 1627int 1628udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, void *ctx) 1629{ 1630 struct inpcb *inp; 1631 struct udpcb *up; 1632 1633 KASSERT(so->so_type == SOCK_DGRAM, 1634 ("udp_set_kernel_tunneling: !dgram")); 1635 inp = sotoinpcb(so); 1636 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1637 INP_WLOCK(inp); 1638 up = intoudpcb(inp); 1639 if (up->u_tun_func != NULL) { 1640 INP_WUNLOCK(inp); 1641 return (EBUSY); 1642 } 1643 up->u_tun_func = f; 1644 up->u_tun_ctx = ctx; 1645 INP_WUNLOCK(inp); 1646 return (0); 1647} 1648 1649#ifdef INET 1650static int 1651udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1652{ 1653 struct inpcb *inp; 1654 struct inpcbinfo *pcbinfo; 1655 int error; 1656 1657 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1658 inp = sotoinpcb(so); 1659 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1660 INP_WLOCK(inp); 1661 INP_HASH_WLOCK(pcbinfo); 1662 error = in_pcbbind(inp, nam, td->td_ucred); 1663 INP_HASH_WUNLOCK(pcbinfo); 1664 INP_WUNLOCK(inp); 1665 return (error); 1666} 1667 1668static void 1669udp_close(struct socket *so) 1670{ 1671 struct inpcb *inp; 1672 struct inpcbinfo *pcbinfo; 1673 1674 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1675 inp = sotoinpcb(so); 1676 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1677 INP_WLOCK(inp); 1678 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1679 INP_HASH_WLOCK(pcbinfo); 1680 in_pcbdisconnect(inp); 1681 inp->inp_laddr.s_addr = INADDR_ANY; 1682 INP_HASH_WUNLOCK(pcbinfo); 1683 soisdisconnected(so); 1684 } 1685 INP_WUNLOCK(inp); 1686} 1687 1688static int 1689udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1690{ 1691 struct inpcb *inp; 1692 struct inpcbinfo *pcbinfo; 1693 struct sockaddr_in *sin; 1694 int error; 1695 1696 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1697 inp = sotoinpcb(so); 1698 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1699 INP_WLOCK(inp); 1700 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1701 INP_WUNLOCK(inp); 1702 return (EISCONN); 1703 } 1704 sin = (struct sockaddr_in *)nam; 1705 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1706 if (error != 0) { 1707 INP_WUNLOCK(inp); 1708 return (error); 1709 } 1710 INP_HASH_WLOCK(pcbinfo); 1711 error = in_pcbconnect(inp, nam, td->td_ucred); 1712 INP_HASH_WUNLOCK(pcbinfo); 1713 if (error == 0) 1714 soisconnected(so); 1715 INP_WUNLOCK(inp); 1716 return (error); 1717} 1718 1719static void 1720udp_detach(struct socket *so) 1721{ 1722 struct inpcb *inp; 1723 struct inpcbinfo *pcbinfo; 1724 struct udpcb *up; 1725 1726 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1727 inp = sotoinpcb(so); 1728 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1729 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1730 ("udp_detach: not disconnected")); 1731 INP_INFO_WLOCK(pcbinfo); 1732 INP_WLOCK(inp); 1733 up = intoudpcb(inp); 1734 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1735 inp->inp_ppcb = NULL; 1736 in_pcbdetach(inp); 1737 in_pcbfree(inp); 1738 INP_INFO_WUNLOCK(pcbinfo); 1739 udp_discardcb(up); 1740} 1741 1742static int 1743udp_disconnect(struct socket *so) 1744{ 1745 struct inpcb *inp; 1746 struct inpcbinfo *pcbinfo; 1747 1748 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1749 inp = sotoinpcb(so); 1750 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1751 INP_WLOCK(inp); 1752 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1753 INP_WUNLOCK(inp); 1754 return (ENOTCONN); 1755 } 1756 INP_HASH_WLOCK(pcbinfo); 1757 in_pcbdisconnect(inp); 1758 inp->inp_laddr.s_addr = INADDR_ANY; 1759 INP_HASH_WUNLOCK(pcbinfo); 1760 SOCK_LOCK(so); 1761 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1762 SOCK_UNLOCK(so); 1763 INP_WUNLOCK(inp); 1764 return (0); 1765} 1766 1767static int 1768udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1769 struct mbuf *control, struct thread *td) 1770{ 1771 struct inpcb *inp; 1772 1773 inp = sotoinpcb(so); 1774 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1775 return (udp_output(inp, m, addr, control, td)); 1776} 1777#endif /* INET */ 1778 1779int 1780udp_shutdown(struct socket *so) 1781{ 1782 struct inpcb *inp; 1783 1784 inp = sotoinpcb(so); 1785 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1786 INP_WLOCK(inp); 1787 socantsendmore(so); 1788 INP_WUNLOCK(inp); 1789 return (0); 1790} 1791 1792#ifdef INET 1793struct pr_usrreqs udp_usrreqs = { 1794 .pru_abort = udp_abort, 1795 .pru_attach = udp_attach, 1796 .pru_bind = udp_bind, 1797 .pru_connect = udp_connect, 1798 .pru_control = in_control, 1799 .pru_detach = udp_detach, 1800 .pru_disconnect = udp_disconnect, 1801 .pru_peeraddr = in_getpeeraddr, 1802 .pru_send = udp_send, 1803 .pru_soreceive = soreceive_dgram, 1804 .pru_sosend = sosend_dgram, 1805 .pru_shutdown = udp_shutdown, 1806 .pru_sockaddr = in_getsockaddr, 1807 .pru_sosetlabel = in_pcbsosetlabel, 1808 .pru_close = udp_close, 1809}; 1810#endif /* INET */ 1811