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("$FreeBSD$"); 41 42 #include "opt_ipfw.h" 43 #include "opt_inet.h" 44 #include "opt_inet6.h" 45 #include "opt_ipsec.h" 46 #include "opt_rss.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/if_var.h> 72 #include <net/route.h> 73 74 #include <netinet/in.h> 75 #include <netinet/in_kdtrace.h> 76 #include <netinet/in_pcb.h> 77 #include <netinet/in_systm.h> 78 #include <netinet/in_var.h> 79 #include <netinet/ip.h> 80 #ifdef INET6 81 #include <netinet/ip6.h> 82 #endif 83 #include <netinet/ip_icmp.h> 84 #include <netinet/icmp_var.h> 85 #include <netinet/ip_var.h> 86 #include <netinet/ip_options.h> 87 #ifdef INET6 88 #include <netinet6/ip6_var.h> 89 #endif 90 #include <netinet/udp.h> 91 #include <netinet/udp_var.h> 92 #include <netinet/udplite.h> 93 #include <netinet/in_rss.h> 94 95 #ifdef IPSEC 96 #include <netipsec/ipsec.h> 97 #include <netipsec/esp.h> 98 #endif 99 100 #include <machine/in_cksum.h> 101 102 #include <security/mac/mac_framework.h> 103 104 /* 105 * UDP and UDP-Lite protocols implementation. 106 * Per RFC 768, August, 1980. 107 * Per RFC 3828, July, 2004. 108 */ 109 110 /* 111 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 112 * removes the only data integrity mechanism for packets and malformed 113 * packets that would otherwise be discarded due to bad checksums, and may 114 * cause problems (especially for NFS data blocks). 115 */ 116 VNET_DEFINE(int, udp_cksum) = 1; 117 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, 118 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 119 120 int udp_log_in_vain = 0; 121 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 122 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 123 124 VNET_DEFINE(int, udp_blackhole) = 0; 125 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, 126 &VNET_NAME(udp_blackhole), 0, 127 "Do not send port unreachables for refused connects"); 128 129 u_long udp_sendspace = 9216; /* really max datagram size */ 130 /* 40 1K datagrams */ 131 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 132 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 133 134 u_long udp_recvspace = 40 * (1024 + 135 #ifdef INET6 136 sizeof(struct sockaddr_in6) 137 #else 138 sizeof(struct sockaddr_in) 139 #endif 140 ); 141 142 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 143 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 144 145 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 146 VNET_DEFINE(struct inpcbinfo, udbinfo); 147 VNET_DEFINE(struct inpcbhead, ulitecb); 148 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 149 static VNET_DEFINE(uma_zone_t, udpcb_zone); 150 #define V_udpcb_zone VNET(udpcb_zone) 151 152 #ifndef UDBHASHSIZE 153 #define UDBHASHSIZE 128 154 #endif 155 156 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 157 VNET_PCPUSTAT_SYSINIT(udpstat); 158 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 159 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 160 161 #ifdef VIMAGE 162 VNET_PCPUSTAT_SYSUNINIT(udpstat); 163 #endif /* VIMAGE */ 164 #ifdef INET 165 static void udp_detach(struct socket *so); 166 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 167 struct mbuf *, struct thread *); 168 #endif 169 170 #ifdef IPSEC 171 #ifdef IPSEC_NAT_T 172 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP) 173 #ifdef INET 174 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int); 175 #endif 176 #endif /* IPSEC_NAT_T */ 177 #endif /* IPSEC */ 178 179 static void 180 udp_zone_change(void *tag) 181 { 182 183 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 184 uma_zone_set_max(V_udpcb_zone, maxsockets); 185 } 186 187 static int 188 udp_inpcb_init(void *mem, int size, int flags) 189 { 190 struct inpcb *inp; 191 192 inp = mem; 193 INP_LOCK_INIT(inp, "inp", "udpinp"); 194 return (0); 195 } 196 197 static int 198 udplite_inpcb_init(void *mem, int size, int flags) 199 { 200 struct inpcb *inp; 201 202 inp = mem; 203 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 204 return (0); 205 } 206 207 void 208 udp_init(void) 209 { 210 211 /* 212 * For now default to 2-tuple UDP hashing - until the fragment 213 * reassembly code can also update the flowid. 214 * 215 * Once we can calculate the flowid that way and re-establish 216 * a 4-tuple, flip this to 4-tuple. 217 */ 218 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 219 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE, 220 IPI_HASHFIELDS_2TUPLE); 221 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 222 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 223 uma_zone_set_max(V_udpcb_zone, maxsockets); 224 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 225 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 226 EVENTHANDLER_PRI_ANY); 227 } 228 229 void 230 udplite_init(void) 231 { 232 233 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 234 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL, 235 UMA_ZONE_NOFREE, IPI_HASHFIELDS_2TUPLE); 236 } 237 238 /* 239 * Kernel module interface for updating udpstat. The argument is an index 240 * into udpstat treated as an array of u_long. While this encodes the 241 * general layout of udpstat into the caller, it doesn't encode its location, 242 * so that future changes to add, for example, per-CPU stats support won't 243 * cause binary compatibility problems for kernel modules. 244 */ 245 void 246 kmod_udpstat_inc(int statnum) 247 { 248 249 counter_u64_add(VNET(udpstat)[statnum], 1); 250 } 251 252 int 253 udp_newudpcb(struct inpcb *inp) 254 { 255 struct udpcb *up; 256 257 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 258 if (up == NULL) 259 return (ENOBUFS); 260 inp->inp_ppcb = up; 261 return (0); 262 } 263 264 void 265 udp_discardcb(struct udpcb *up) 266 { 267 268 uma_zfree(V_udpcb_zone, up); 269 } 270 271 #ifdef VIMAGE 272 void 273 udp_destroy(void) 274 { 275 276 in_pcbinfo_destroy(&V_udbinfo); 277 uma_zdestroy(V_udpcb_zone); 278 } 279 280 void 281 udplite_destroy(void) 282 { 283 284 in_pcbinfo_destroy(&V_ulitecbinfo); 285 } 286 #endif 287 288 #ifdef INET 289 /* 290 * Subroutine of udp_input(), which appends the provided mbuf chain to the 291 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 292 * contains the source address. If the socket ends up being an IPv6 socket, 293 * udp_append() will convert to a sockaddr_in6 before passing the address 294 * into the socket code. 295 */ 296 static void 297 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 298 struct sockaddr_in *udp_in) 299 { 300 struct sockaddr *append_sa; 301 struct socket *so; 302 struct mbuf *opts = 0; 303 #ifdef INET6 304 struct sockaddr_in6 udp_in6; 305 #endif 306 struct udpcb *up; 307 308 INP_LOCK_ASSERT(inp); 309 310 /* 311 * Engage the tunneling protocol. 312 */ 313 up = intoudpcb(inp); 314 if (up->u_tun_func != NULL) { 315 (*up->u_tun_func)(n, off, inp); 316 return; 317 } 318 319 if (n == NULL) 320 return; 321 322 off += sizeof(struct udphdr); 323 324 #ifdef IPSEC 325 /* Check AH/ESP integrity. */ 326 if (ipsec4_in_reject(n, inp)) { 327 m_freem(n); 328 IPSECSTAT_INC(ips_in_polvio); 329 return; 330 } 331 #ifdef IPSEC_NAT_T 332 up = intoudpcb(inp); 333 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 334 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */ 335 n = udp4_espdecap(inp, n, off); 336 if (n == NULL) /* Consumed. */ 337 return; 338 } 339 #endif /* IPSEC_NAT_T */ 340 #endif /* IPSEC */ 341 #ifdef MAC 342 if (mac_inpcb_check_deliver(inp, n) != 0) { 343 m_freem(n); 344 return; 345 } 346 #endif /* MAC */ 347 if (inp->inp_flags & INP_CONTROLOPTS || 348 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 349 #ifdef INET6 350 if (inp->inp_vflag & INP_IPV6) 351 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 352 else 353 #endif /* INET6 */ 354 ip_savecontrol(inp, &opts, ip, n); 355 } 356 #ifdef INET6 357 if (inp->inp_vflag & INP_IPV6) { 358 bzero(&udp_in6, sizeof(udp_in6)); 359 udp_in6.sin6_len = sizeof(udp_in6); 360 udp_in6.sin6_family = AF_INET6; 361 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 362 append_sa = (struct sockaddr *)&udp_in6; 363 } else 364 #endif /* INET6 */ 365 append_sa = (struct sockaddr *)udp_in; 366 m_adj(n, off); 367 368 so = inp->inp_socket; 369 SOCKBUF_LOCK(&so->so_rcv); 370 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 371 SOCKBUF_UNLOCK(&so->so_rcv); 372 m_freem(n); 373 if (opts) 374 m_freem(opts); 375 UDPSTAT_INC(udps_fullsock); 376 } else 377 sorwakeup_locked(so); 378 } 379 380 int 381 udp_input(struct mbuf **mp, int *offp, int proto) 382 { 383 struct ip *ip; 384 struct udphdr *uh; 385 struct ifnet *ifp; 386 struct inpcb *inp; 387 uint16_t len, ip_len; 388 struct inpcbinfo *pcbinfo; 389 struct ip save_ip; 390 struct sockaddr_in udp_in; 391 struct mbuf *m; 392 struct m_tag *fwd_tag; 393 int cscov_partial, iphlen; 394 395 m = *mp; 396 iphlen = *offp; 397 ifp = m->m_pkthdr.rcvif; 398 *mp = NULL; 399 UDPSTAT_INC(udps_ipackets); 400 401 /* 402 * Strip IP options, if any; should skip this, make available to 403 * user, and use on returned packets, but we don't yet have a way to 404 * check the checksum with options still present. 405 */ 406 if (iphlen > sizeof (struct ip)) { 407 ip_stripoptions(m); 408 iphlen = sizeof(struct ip); 409 } 410 411 /* 412 * Get IP and UDP header together in first mbuf. 413 */ 414 ip = mtod(m, struct ip *); 415 if (m->m_len < iphlen + sizeof(struct udphdr)) { 416 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 417 UDPSTAT_INC(udps_hdrops); 418 return (IPPROTO_DONE); 419 } 420 ip = mtod(m, struct ip *); 421 } 422 uh = (struct udphdr *)((caddr_t)ip + iphlen); 423 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 424 425 /* 426 * Destination port of 0 is illegal, based on RFC768. 427 */ 428 if (uh->uh_dport == 0) 429 goto badunlocked; 430 431 /* 432 * Construct sockaddr format source address. Stuff source address 433 * and datagram in user buffer. 434 */ 435 bzero(&udp_in, sizeof(udp_in)); 436 udp_in.sin_len = sizeof(udp_in); 437 udp_in.sin_family = AF_INET; 438 udp_in.sin_port = uh->uh_sport; 439 udp_in.sin_addr = ip->ip_src; 440 441 /* 442 * Make mbuf data length reflect UDP length. If not enough data to 443 * reflect UDP length, drop. 444 */ 445 len = ntohs((u_short)uh->uh_ulen); 446 ip_len = ntohs(ip->ip_len) - iphlen; 447 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 448 /* Zero means checksum over the complete packet. */ 449 if (len == 0) 450 len = ip_len; 451 cscov_partial = 0; 452 } 453 if (ip_len != len) { 454 if (len > ip_len || len < sizeof(struct udphdr)) { 455 UDPSTAT_INC(udps_badlen); 456 goto badunlocked; 457 } 458 if (proto == IPPROTO_UDP) 459 m_adj(m, len - ip_len); 460 } 461 462 /* 463 * Save a copy of the IP header in case we want restore it for 464 * sending an ICMP error message in response. 465 */ 466 if (!V_udp_blackhole) 467 save_ip = *ip; 468 else 469 memset(&save_ip, 0, sizeof(save_ip)); 470 471 /* 472 * Checksum extended UDP header and data. 473 */ 474 if (uh->uh_sum) { 475 u_short uh_sum; 476 477 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 478 !cscov_partial) { 479 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 480 uh_sum = m->m_pkthdr.csum_data; 481 else 482 uh_sum = in_pseudo(ip->ip_src.s_addr, 483 ip->ip_dst.s_addr, htonl((u_short)len + 484 m->m_pkthdr.csum_data + proto)); 485 uh_sum ^= 0xffff; 486 } else { 487 char b[9]; 488 489 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 490 bzero(((struct ipovly *)ip)->ih_x1, 9); 491 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 492 uh->uh_ulen : htons(ip_len); 493 uh_sum = in_cksum(m, len + sizeof (struct ip)); 494 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 495 } 496 if (uh_sum) { 497 UDPSTAT_INC(udps_badsum); 498 m_freem(m); 499 return (IPPROTO_DONE); 500 } 501 } else { 502 if (proto == IPPROTO_UDP) { 503 UDPSTAT_INC(udps_nosum); 504 } else { 505 /* UDPLite requires a checksum */ 506 /* XXX: What is the right UDPLite MIB counter here? */ 507 m_freem(m); 508 return (IPPROTO_DONE); 509 } 510 } 511 512 pcbinfo = get_inpcbinfo(proto); 513 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 514 in_broadcast(ip->ip_dst, ifp)) { 515 struct inpcb *last; 516 struct inpcbhead *pcblist; 517 struct ip_moptions *imo; 518 519 INP_INFO_RLOCK(pcbinfo); 520 pcblist = get_pcblist(proto); 521 last = NULL; 522 LIST_FOREACH(inp, pcblist, inp_list) { 523 if (inp->inp_lport != uh->uh_dport) 524 continue; 525 #ifdef INET6 526 if ((inp->inp_vflag & INP_IPV4) == 0) 527 continue; 528 #endif 529 if (inp->inp_laddr.s_addr != INADDR_ANY && 530 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 531 continue; 532 if (inp->inp_faddr.s_addr != INADDR_ANY && 533 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 534 continue; 535 if (inp->inp_fport != 0 && 536 inp->inp_fport != uh->uh_sport) 537 continue; 538 539 INP_RLOCK(inp); 540 541 /* 542 * XXXRW: Because we weren't holding either the inpcb 543 * or the hash lock when we checked for a match 544 * before, we should probably recheck now that the 545 * inpcb lock is held. 546 */ 547 548 /* 549 * Handle socket delivery policy for any-source 550 * and source-specific multicast. [RFC3678] 551 */ 552 imo = inp->inp_moptions; 553 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 554 struct sockaddr_in group; 555 int blocked; 556 if (imo == NULL) { 557 INP_RUNLOCK(inp); 558 continue; 559 } 560 bzero(&group, sizeof(struct sockaddr_in)); 561 group.sin_len = sizeof(struct sockaddr_in); 562 group.sin_family = AF_INET; 563 group.sin_addr = ip->ip_dst; 564 565 blocked = imo_multi_filter(imo, ifp, 566 (struct sockaddr *)&group, 567 (struct sockaddr *)&udp_in); 568 if (blocked != MCAST_PASS) { 569 if (blocked == MCAST_NOTGMEMBER) 570 IPSTAT_INC(ips_notmember); 571 if (blocked == MCAST_NOTSMEMBER || 572 blocked == MCAST_MUTED) 573 UDPSTAT_INC(udps_filtermcast); 574 INP_RUNLOCK(inp); 575 continue; 576 } 577 } 578 if (last != NULL) { 579 struct mbuf *n; 580 581 n = m_copy(m, 0, M_COPYALL); 582 udp_append(last, ip, n, iphlen, &udp_in); 583 INP_RUNLOCK(last); 584 } 585 last = inp; 586 /* 587 * Don't look for additional matches if this one does 588 * not have either the SO_REUSEPORT or SO_REUSEADDR 589 * socket options set. This heuristic avoids 590 * searching through all pcbs in the common case of a 591 * non-shared port. It assumes that an application 592 * will never clear these options after setting them. 593 */ 594 if ((last->inp_socket->so_options & 595 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 596 break; 597 } 598 599 if (last == NULL) { 600 /* 601 * No matching pcb found; discard datagram. (No need 602 * to send an ICMP Port Unreachable for a broadcast 603 * or multicast datgram.) 604 */ 605 UDPSTAT_INC(udps_noportbcast); 606 if (inp) 607 INP_RUNLOCK(inp); 608 INP_INFO_RUNLOCK(pcbinfo); 609 goto badunlocked; 610 } 611 udp_append(last, ip, m, iphlen, &udp_in); 612 INP_RUNLOCK(last); 613 INP_INFO_RUNLOCK(pcbinfo); 614 return (IPPROTO_DONE); 615 } 616 617 /* 618 * Locate pcb for datagram. 619 */ 620 621 /* 622 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 623 */ 624 if ((m->m_flags & M_IP_NEXTHOP) && 625 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 626 struct sockaddr_in *next_hop; 627 628 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 629 630 /* 631 * Transparently forwarded. Pretend to be the destination. 632 * Already got one like this? 633 */ 634 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 635 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 636 if (!inp) { 637 /* 638 * It's new. Try to find the ambushing socket. 639 * Because we've rewritten the destination address, 640 * any hardware-generated hash is ignored. 641 */ 642 inp = in_pcblookup(pcbinfo, ip->ip_src, 643 uh->uh_sport, next_hop->sin_addr, 644 next_hop->sin_port ? htons(next_hop->sin_port) : 645 uh->uh_dport, INPLOOKUP_WILDCARD | 646 INPLOOKUP_RLOCKPCB, ifp); 647 } 648 /* Remove the tag from the packet. We don't need it anymore. */ 649 m_tag_delete(m, fwd_tag); 650 m->m_flags &= ~M_IP_NEXTHOP; 651 } else 652 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 653 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 654 INPLOOKUP_RLOCKPCB, ifp, m); 655 if (inp == NULL) { 656 if (udp_log_in_vain) { 657 char buf[4*sizeof "123"]; 658 659 strcpy(buf, inet_ntoa(ip->ip_dst)); 660 log(LOG_INFO, 661 "Connection attempt to UDP %s:%d from %s:%d\n", 662 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 663 ntohs(uh->uh_sport)); 664 } 665 UDPSTAT_INC(udps_noport); 666 if (m->m_flags & (M_BCAST | M_MCAST)) { 667 UDPSTAT_INC(udps_noportbcast); 668 goto badunlocked; 669 } 670 if (V_udp_blackhole) 671 goto badunlocked; 672 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 673 goto badunlocked; 674 *ip = save_ip; 675 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 676 return (IPPROTO_DONE); 677 } 678 679 /* 680 * Check the minimum TTL for socket. 681 */ 682 INP_RLOCK_ASSERT(inp); 683 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 684 INP_RUNLOCK(inp); 685 m_freem(m); 686 return (IPPROTO_DONE); 687 } 688 if (cscov_partial) { 689 struct udpcb *up; 690 691 up = intoudpcb(inp); 692 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 693 INP_RUNLOCK(inp); 694 m_freem(m); 695 return (IPPROTO_DONE); 696 } 697 } 698 699 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 700 udp_append(inp, ip, m, iphlen, &udp_in); 701 INP_RUNLOCK(inp); 702 return (IPPROTO_DONE); 703 704 badunlocked: 705 m_freem(m); 706 return (IPPROTO_DONE); 707 } 708 #endif /* INET */ 709 710 /* 711 * Notify a udp user of an asynchronous error; just wake up so that they can 712 * collect error status. 713 */ 714 struct inpcb * 715 udp_notify(struct inpcb *inp, int errno) 716 { 717 718 /* 719 * While udp_ctlinput() always calls udp_notify() with a read lock 720 * when invoking it directly, in_pcbnotifyall() currently uses write 721 * locks due to sharing code with TCP. For now, accept either a read 722 * or a write lock, but a read lock is sufficient. 723 */ 724 INP_LOCK_ASSERT(inp); 725 726 inp->inp_socket->so_error = errno; 727 sorwakeup(inp->inp_socket); 728 sowwakeup(inp->inp_socket); 729 return (inp); 730 } 731 732 #ifdef INET 733 static void 734 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 735 struct inpcbinfo *pcbinfo) 736 { 737 struct ip *ip = vip; 738 struct udphdr *uh; 739 struct in_addr faddr; 740 struct inpcb *inp; 741 742 faddr = ((struct sockaddr_in *)sa)->sin_addr; 743 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 744 return; 745 746 /* 747 * Redirects don't need to be handled up here. 748 */ 749 if (PRC_IS_REDIRECT(cmd)) 750 return; 751 752 /* 753 * Hostdead is ugly because it goes linearly through all PCBs. 754 * 755 * XXX: We never get this from ICMP, otherwise it makes an excellent 756 * DoS attack on machines with many connections. 757 */ 758 if (cmd == PRC_HOSTDEAD) 759 ip = NULL; 760 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 761 return; 762 if (ip != NULL) { 763 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 764 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 765 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 766 if (inp != NULL) { 767 INP_RLOCK_ASSERT(inp); 768 if (inp->inp_socket != NULL) { 769 udp_notify(inp, inetctlerrmap[cmd]); 770 } 771 INP_RUNLOCK(inp); 772 } 773 } else 774 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 775 udp_notify); 776 } 777 void 778 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 779 { 780 781 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 782 } 783 784 void 785 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 786 { 787 788 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 789 } 790 #endif /* INET */ 791 792 static int 793 udp_pcblist(SYSCTL_HANDLER_ARGS) 794 { 795 int error, i, n; 796 struct inpcb *inp, **inp_list; 797 inp_gen_t gencnt; 798 struct xinpgen xig; 799 800 /* 801 * The process of preparing the PCB list is too time-consuming and 802 * resource-intensive to repeat twice on every request. 803 */ 804 if (req->oldptr == 0) { 805 n = V_udbinfo.ipi_count; 806 n += imax(n / 8, 10); 807 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 808 return (0); 809 } 810 811 if (req->newptr != 0) 812 return (EPERM); 813 814 /* 815 * OK, now we're committed to doing something. 816 */ 817 INP_INFO_RLOCK(&V_udbinfo); 818 gencnt = V_udbinfo.ipi_gencnt; 819 n = V_udbinfo.ipi_count; 820 INP_INFO_RUNLOCK(&V_udbinfo); 821 822 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 823 + n * sizeof(struct xinpcb)); 824 if (error != 0) 825 return (error); 826 827 xig.xig_len = sizeof xig; 828 xig.xig_count = n; 829 xig.xig_gen = gencnt; 830 xig.xig_sogen = so_gencnt; 831 error = SYSCTL_OUT(req, &xig, sizeof xig); 832 if (error) 833 return (error); 834 835 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 836 if (inp_list == 0) 837 return (ENOMEM); 838 839 INP_INFO_RLOCK(&V_udbinfo); 840 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 841 inp = LIST_NEXT(inp, inp_list)) { 842 INP_WLOCK(inp); 843 if (inp->inp_gencnt <= gencnt && 844 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 845 in_pcbref(inp); 846 inp_list[i++] = inp; 847 } 848 INP_WUNLOCK(inp); 849 } 850 INP_INFO_RUNLOCK(&V_udbinfo); 851 n = i; 852 853 error = 0; 854 for (i = 0; i < n; i++) { 855 inp = inp_list[i]; 856 INP_RLOCK(inp); 857 if (inp->inp_gencnt <= gencnt) { 858 struct xinpcb xi; 859 860 bzero(&xi, sizeof(xi)); 861 xi.xi_len = sizeof xi; 862 /* XXX should avoid extra copy */ 863 bcopy(inp, &xi.xi_inp, sizeof *inp); 864 if (inp->inp_socket) 865 sotoxsocket(inp->inp_socket, &xi.xi_socket); 866 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 867 INP_RUNLOCK(inp); 868 error = SYSCTL_OUT(req, &xi, sizeof xi); 869 } else 870 INP_RUNLOCK(inp); 871 } 872 INP_INFO_WLOCK(&V_udbinfo); 873 for (i = 0; i < n; i++) { 874 inp = inp_list[i]; 875 INP_RLOCK(inp); 876 if (!in_pcbrele_rlocked(inp)) 877 INP_RUNLOCK(inp); 878 } 879 INP_INFO_WUNLOCK(&V_udbinfo); 880 881 if (!error) { 882 /* 883 * Give the user an updated idea of our state. If the 884 * generation differs from what we told her before, she knows 885 * that something happened while we were processing this 886 * request, and it might be necessary to retry. 887 */ 888 INP_INFO_RLOCK(&V_udbinfo); 889 xig.xig_gen = V_udbinfo.ipi_gencnt; 890 xig.xig_sogen = so_gencnt; 891 xig.xig_count = V_udbinfo.ipi_count; 892 INP_INFO_RUNLOCK(&V_udbinfo); 893 error = SYSCTL_OUT(req, &xig, sizeof xig); 894 } 895 free(inp_list, M_TEMP); 896 return (error); 897 } 898 899 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 900 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 901 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 902 903 #ifdef INET 904 static int 905 udp_getcred(SYSCTL_HANDLER_ARGS) 906 { 907 struct xucred xuc; 908 struct sockaddr_in addrs[2]; 909 struct inpcb *inp; 910 int error; 911 912 error = priv_check(req->td, PRIV_NETINET_GETCRED); 913 if (error) 914 return (error); 915 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 916 if (error) 917 return (error); 918 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 919 addrs[0].sin_addr, addrs[0].sin_port, 920 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 921 if (inp != NULL) { 922 INP_RLOCK_ASSERT(inp); 923 if (inp->inp_socket == NULL) 924 error = ENOENT; 925 if (error == 0) 926 error = cr_canseeinpcb(req->td->td_ucred, inp); 927 if (error == 0) 928 cru2x(inp->inp_cred, &xuc); 929 INP_RUNLOCK(inp); 930 } else 931 error = ENOENT; 932 if (error == 0) 933 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 934 return (error); 935 } 936 937 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 938 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 939 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 940 #endif /* INET */ 941 942 int 943 udp_ctloutput(struct socket *so, struct sockopt *sopt) 944 { 945 struct inpcb *inp; 946 struct udpcb *up; 947 int isudplite, error, optval; 948 949 error = 0; 950 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 951 inp = sotoinpcb(so); 952 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 953 INP_WLOCK(inp); 954 if (sopt->sopt_level != so->so_proto->pr_protocol) { 955 #ifdef INET6 956 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 957 INP_WUNLOCK(inp); 958 error = ip6_ctloutput(so, sopt); 959 } 960 #endif 961 #if defined(INET) && defined(INET6) 962 else 963 #endif 964 #ifdef INET 965 { 966 INP_WUNLOCK(inp); 967 error = ip_ctloutput(so, sopt); 968 } 969 #endif 970 return (error); 971 } 972 973 switch (sopt->sopt_dir) { 974 case SOPT_SET: 975 switch (sopt->sopt_name) { 976 case UDP_ENCAP: 977 INP_WUNLOCK(inp); 978 error = sooptcopyin(sopt, &optval, sizeof optval, 979 sizeof optval); 980 if (error) 981 break; 982 inp = sotoinpcb(so); 983 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 984 INP_WLOCK(inp); 985 #ifdef IPSEC_NAT_T 986 up = intoudpcb(inp); 987 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 988 #endif 989 switch (optval) { 990 case 0: 991 /* Clear all UDP encap. */ 992 #ifdef IPSEC_NAT_T 993 up->u_flags &= ~UF_ESPINUDP_ALL; 994 #endif 995 break; 996 #ifdef IPSEC_NAT_T 997 case UDP_ENCAP_ESPINUDP: 998 case UDP_ENCAP_ESPINUDP_NON_IKE: 999 up->u_flags &= ~UF_ESPINUDP_ALL; 1000 if (optval == UDP_ENCAP_ESPINUDP) 1001 up->u_flags |= UF_ESPINUDP; 1002 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 1003 up->u_flags |= UF_ESPINUDP_NON_IKE; 1004 break; 1005 #endif 1006 default: 1007 error = EINVAL; 1008 break; 1009 } 1010 INP_WUNLOCK(inp); 1011 break; 1012 case UDPLITE_SEND_CSCOV: 1013 case UDPLITE_RECV_CSCOV: 1014 if (!isudplite) { 1015 INP_WUNLOCK(inp); 1016 error = ENOPROTOOPT; 1017 break; 1018 } 1019 INP_WUNLOCK(inp); 1020 error = sooptcopyin(sopt, &optval, sizeof(optval), 1021 sizeof(optval)); 1022 if (error != 0) 1023 break; 1024 inp = sotoinpcb(so); 1025 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1026 INP_WLOCK(inp); 1027 up = intoudpcb(inp); 1028 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1029 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1030 INP_WUNLOCK(inp); 1031 error = EINVAL; 1032 break; 1033 } 1034 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1035 up->u_txcslen = optval; 1036 else 1037 up->u_rxcslen = optval; 1038 INP_WUNLOCK(inp); 1039 break; 1040 default: 1041 INP_WUNLOCK(inp); 1042 error = ENOPROTOOPT; 1043 break; 1044 } 1045 break; 1046 case SOPT_GET: 1047 switch (sopt->sopt_name) { 1048 #ifdef IPSEC_NAT_T 1049 case UDP_ENCAP: 1050 up = intoudpcb(inp); 1051 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1052 optval = up->u_flags & UF_ESPINUDP_ALL; 1053 INP_WUNLOCK(inp); 1054 error = sooptcopyout(sopt, &optval, sizeof optval); 1055 break; 1056 #endif 1057 case UDPLITE_SEND_CSCOV: 1058 case UDPLITE_RECV_CSCOV: 1059 if (!isudplite) { 1060 INP_WUNLOCK(inp); 1061 error = ENOPROTOOPT; 1062 break; 1063 } 1064 up = intoudpcb(inp); 1065 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1066 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1067 optval = up->u_txcslen; 1068 else 1069 optval = up->u_rxcslen; 1070 INP_WUNLOCK(inp); 1071 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1072 break; 1073 default: 1074 INP_WUNLOCK(inp); 1075 error = ENOPROTOOPT; 1076 break; 1077 } 1078 break; 1079 } 1080 return (error); 1081 } 1082 1083 #ifdef INET 1084 #define UH_WLOCKED 2 1085 #define UH_RLOCKED 1 1086 #define UH_UNLOCKED 0 1087 static int 1088 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1089 struct mbuf *control, struct thread *td) 1090 { 1091 struct udpiphdr *ui; 1092 int len = m->m_pkthdr.len; 1093 struct in_addr faddr, laddr; 1094 struct cmsghdr *cm; 1095 struct inpcbinfo *pcbinfo; 1096 struct sockaddr_in *sin, src; 1097 int cscov_partial = 0; 1098 int error = 0; 1099 int ipflags; 1100 u_short fport, lport; 1101 int unlock_udbinfo; 1102 u_char tos; 1103 uint8_t pr; 1104 uint16_t cscov = 0; 1105 uint32_t flowid = 0; 1106 int flowid_type = 0; 1107 int use_flowid = 0; 1108 1109 /* 1110 * udp_output() may need to temporarily bind or connect the current 1111 * inpcb. As such, we don't know up front whether we will need the 1112 * pcbinfo lock or not. Do any work to decide what is needed up 1113 * front before acquiring any locks. 1114 */ 1115 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1116 if (control) 1117 m_freem(control); 1118 m_freem(m); 1119 return (EMSGSIZE); 1120 } 1121 1122 src.sin_family = 0; 1123 INP_RLOCK(inp); 1124 tos = inp->inp_ip_tos; 1125 if (control != NULL) { 1126 /* 1127 * XXX: Currently, we assume all the optional information is 1128 * stored in a single mbuf. 1129 */ 1130 if (control->m_next) { 1131 INP_RUNLOCK(inp); 1132 m_freem(control); 1133 m_freem(m); 1134 return (EINVAL); 1135 } 1136 for (; control->m_len > 0; 1137 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1138 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1139 cm = mtod(control, struct cmsghdr *); 1140 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1141 || cm->cmsg_len > control->m_len) { 1142 error = EINVAL; 1143 break; 1144 } 1145 if (cm->cmsg_level != IPPROTO_IP) 1146 continue; 1147 1148 switch (cm->cmsg_type) { 1149 case IP_SENDSRCADDR: 1150 if (cm->cmsg_len != 1151 CMSG_LEN(sizeof(struct in_addr))) { 1152 error = EINVAL; 1153 break; 1154 } 1155 bzero(&src, sizeof(src)); 1156 src.sin_family = AF_INET; 1157 src.sin_len = sizeof(src); 1158 src.sin_port = inp->inp_lport; 1159 src.sin_addr = 1160 *(struct in_addr *)CMSG_DATA(cm); 1161 break; 1162 1163 case IP_TOS: 1164 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1165 error = EINVAL; 1166 break; 1167 } 1168 tos = *(u_char *)CMSG_DATA(cm); 1169 break; 1170 1171 case IP_FLOWID: 1172 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1173 error = EINVAL; 1174 break; 1175 } 1176 flowid = *(uint32_t *) CMSG_DATA(cm); 1177 break; 1178 1179 case IP_FLOWTYPE: 1180 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1181 error = EINVAL; 1182 break; 1183 } 1184 flowid_type = *(uint32_t *) CMSG_DATA(cm); 1185 use_flowid = 1; 1186 break; 1187 1188 #ifdef RSS 1189 case IP_RSSBUCKETID: 1190 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1191 error = EINVAL; 1192 break; 1193 } 1194 /* This is just a placeholder for now */ 1195 break; 1196 #endif /* RSS */ 1197 default: 1198 error = ENOPROTOOPT; 1199 break; 1200 } 1201 if (error) 1202 break; 1203 } 1204 m_freem(control); 1205 } 1206 if (error) { 1207 INP_RUNLOCK(inp); 1208 m_freem(m); 1209 return (error); 1210 } 1211 1212 /* 1213 * Depending on whether or not the application has bound or connected 1214 * the socket, we may have to do varying levels of work. The optimal 1215 * case is for a connected UDP socket, as a global lock isn't 1216 * required at all. 1217 * 1218 * In order to decide which we need, we require stability of the 1219 * inpcb binding, which we ensure by acquiring a read lock on the 1220 * inpcb. This doesn't strictly follow the lock order, so we play 1221 * the trylock and retry game; note that we may end up with more 1222 * conservative locks than required the second time around, so later 1223 * assertions have to accept that. Further analysis of the number of 1224 * misses under contention is required. 1225 * 1226 * XXXRW: Check that hash locking update here is correct. 1227 */ 1228 pr = inp->inp_socket->so_proto->pr_protocol; 1229 pcbinfo = get_inpcbinfo(pr); 1230 sin = (struct sockaddr_in *)addr; 1231 if (sin != NULL && 1232 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1233 INP_RUNLOCK(inp); 1234 INP_WLOCK(inp); 1235 INP_HASH_WLOCK(pcbinfo); 1236 unlock_udbinfo = UH_WLOCKED; 1237 } else if ((sin != NULL && ( 1238 (sin->sin_addr.s_addr == INADDR_ANY) || 1239 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1240 (inp->inp_laddr.s_addr == INADDR_ANY) || 1241 (inp->inp_lport == 0))) || 1242 (src.sin_family == AF_INET)) { 1243 INP_HASH_RLOCK(pcbinfo); 1244 unlock_udbinfo = UH_RLOCKED; 1245 } else 1246 unlock_udbinfo = UH_UNLOCKED; 1247 1248 /* 1249 * If the IP_SENDSRCADDR control message was specified, override the 1250 * source address for this datagram. Its use is invalidated if the 1251 * address thus specified is incomplete or clobbers other inpcbs. 1252 */ 1253 laddr = inp->inp_laddr; 1254 lport = inp->inp_lport; 1255 if (src.sin_family == AF_INET) { 1256 INP_HASH_LOCK_ASSERT(pcbinfo); 1257 if ((lport == 0) || 1258 (laddr.s_addr == INADDR_ANY && 1259 src.sin_addr.s_addr == INADDR_ANY)) { 1260 error = EINVAL; 1261 goto release; 1262 } 1263 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1264 &laddr.s_addr, &lport, td->td_ucred); 1265 if (error) 1266 goto release; 1267 } 1268 1269 /* 1270 * If a UDP socket has been connected, then a local address/port will 1271 * have been selected and bound. 1272 * 1273 * If a UDP socket has not been connected to, then an explicit 1274 * destination address must be used, in which case a local 1275 * address/port may not have been selected and bound. 1276 */ 1277 if (sin != NULL) { 1278 INP_LOCK_ASSERT(inp); 1279 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1280 error = EISCONN; 1281 goto release; 1282 } 1283 1284 /* 1285 * Jail may rewrite the destination address, so let it do 1286 * that before we use it. 1287 */ 1288 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1289 if (error) 1290 goto release; 1291 1292 /* 1293 * If a local address or port hasn't yet been selected, or if 1294 * the destination address needs to be rewritten due to using 1295 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1296 * to do the heavy lifting. Once a port is selected, we 1297 * commit the binding back to the socket; we also commit the 1298 * binding of the address if in jail. 1299 * 1300 * If we already have a valid binding and we're not 1301 * requesting a destination address rewrite, use a fast path. 1302 */ 1303 if (inp->inp_laddr.s_addr == INADDR_ANY || 1304 inp->inp_lport == 0 || 1305 sin->sin_addr.s_addr == INADDR_ANY || 1306 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1307 INP_HASH_LOCK_ASSERT(pcbinfo); 1308 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1309 &lport, &faddr.s_addr, &fport, NULL, 1310 td->td_ucred); 1311 if (error) 1312 goto release; 1313 1314 /* 1315 * XXXRW: Why not commit the port if the address is 1316 * !INADDR_ANY? 1317 */ 1318 /* Commit the local port if newly assigned. */ 1319 if (inp->inp_laddr.s_addr == INADDR_ANY && 1320 inp->inp_lport == 0) { 1321 INP_WLOCK_ASSERT(inp); 1322 INP_HASH_WLOCK_ASSERT(pcbinfo); 1323 /* 1324 * Remember addr if jailed, to prevent 1325 * rebinding. 1326 */ 1327 if (prison_flag(td->td_ucred, PR_IP4)) 1328 inp->inp_laddr = laddr; 1329 inp->inp_lport = lport; 1330 if (in_pcbinshash(inp) != 0) { 1331 inp->inp_lport = 0; 1332 error = EAGAIN; 1333 goto release; 1334 } 1335 inp->inp_flags |= INP_ANONPORT; 1336 } 1337 } else { 1338 faddr = sin->sin_addr; 1339 fport = sin->sin_port; 1340 } 1341 } else { 1342 INP_LOCK_ASSERT(inp); 1343 faddr = inp->inp_faddr; 1344 fport = inp->inp_fport; 1345 if (faddr.s_addr == INADDR_ANY) { 1346 error = ENOTCONN; 1347 goto release; 1348 } 1349 } 1350 1351 /* 1352 * Calculate data length and get a mbuf for UDP, IP, and possible 1353 * link-layer headers. Immediate slide the data pointer back forward 1354 * since we won't use that space at this layer. 1355 */ 1356 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1357 if (m == NULL) { 1358 error = ENOBUFS; 1359 goto release; 1360 } 1361 m->m_data += max_linkhdr; 1362 m->m_len -= max_linkhdr; 1363 m->m_pkthdr.len -= max_linkhdr; 1364 1365 /* 1366 * Fill in mbuf with extended UDP header and addresses and length put 1367 * into network format. 1368 */ 1369 ui = mtod(m, struct udpiphdr *); 1370 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1371 ui->ui_pr = pr; 1372 ui->ui_src = laddr; 1373 ui->ui_dst = faddr; 1374 ui->ui_sport = lport; 1375 ui->ui_dport = fport; 1376 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1377 if (pr == IPPROTO_UDPLITE) { 1378 struct udpcb *up; 1379 uint16_t plen; 1380 1381 up = intoudpcb(inp); 1382 cscov = up->u_txcslen; 1383 plen = (u_short)len + sizeof(struct udphdr); 1384 if (cscov >= plen) 1385 cscov = 0; 1386 ui->ui_len = htons(plen); 1387 ui->ui_ulen = htons(cscov); 1388 /* 1389 * For UDP-Lite, checksum coverage length of zero means 1390 * the entire UDPLite packet is covered by the checksum. 1391 */ 1392 cscov_partial = (cscov == 0) ? 0 : 1; 1393 } else 1394 ui->ui_v = IPVERSION << 4; 1395 1396 /* 1397 * Set the Don't Fragment bit in the IP header. 1398 */ 1399 if (inp->inp_flags & INP_DONTFRAG) { 1400 struct ip *ip; 1401 1402 ip = (struct ip *)&ui->ui_i; 1403 ip->ip_off |= htons(IP_DF); 1404 } 1405 1406 ipflags = 0; 1407 if (inp->inp_socket->so_options & SO_DONTROUTE) 1408 ipflags |= IP_ROUTETOIF; 1409 if (inp->inp_socket->so_options & SO_BROADCAST) 1410 ipflags |= IP_ALLOWBROADCAST; 1411 if (inp->inp_flags & INP_ONESBCAST) 1412 ipflags |= IP_SENDONES; 1413 1414 #ifdef MAC 1415 mac_inpcb_create_mbuf(inp, m); 1416 #endif 1417 1418 /* 1419 * Set up checksum and output datagram. 1420 */ 1421 ui->ui_sum = 0; 1422 if (pr == IPPROTO_UDPLITE) { 1423 if (inp->inp_flags & INP_ONESBCAST) 1424 faddr.s_addr = INADDR_BROADCAST; 1425 if (cscov_partial) { 1426 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1427 ui->ui_sum = 0xffff; 1428 } else { 1429 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1430 ui->ui_sum = 0xffff; 1431 } 1432 } else if (V_udp_cksum) { 1433 if (inp->inp_flags & INP_ONESBCAST) 1434 faddr.s_addr = INADDR_BROADCAST; 1435 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1436 htons((u_short)len + sizeof(struct udphdr) + pr)); 1437 m->m_pkthdr.csum_flags = CSUM_UDP; 1438 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1439 } 1440 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1441 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1442 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1443 UDPSTAT_INC(udps_opackets); 1444 1445 /* 1446 * Setup flowid / RSS information for outbound socket. 1447 * 1448 * Once the UDP code decides to set a flowid some other way, 1449 * this allows the flowid to be overridden by userland. 1450 */ 1451 if (use_flowid) { 1452 m->m_flags |= M_FLOWID; 1453 m->m_pkthdr.flowid = flowid; 1454 M_HASHTYPE_SET(m, flowid_type); 1455 #ifdef RSS 1456 } else { 1457 uint32_t hash_val, hash_type; 1458 /* 1459 * Calculate an appropriate RSS hash for UDP and 1460 * UDP Lite. 1461 * 1462 * The called function will take care of figuring out 1463 * whether a 2-tuple or 4-tuple hash is required based 1464 * on the currently configured scheme. 1465 * 1466 * Later later on connected socket values should be 1467 * cached in the inpcb and reused, rather than constantly 1468 * re-calculating it. 1469 * 1470 * UDP Lite is a different protocol number and will 1471 * likely end up being hashed as a 2-tuple until 1472 * RSS / NICs grow UDP Lite protocol awareness. 1473 */ 1474 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1475 pr, &hash_val, &hash_type) == 0) { 1476 m->m_pkthdr.flowid = hash_val; 1477 m->m_flags |= M_FLOWID; 1478 M_HASHTYPE_SET(m, hash_type); 1479 } 1480 #endif 1481 } 1482 1483 #ifdef RSS 1484 /* 1485 * Don't override with the inp cached flowid value. 1486 * 1487 * Depending upon the kind of send being done, the inp 1488 * flowid/flowtype values may actually not be appropriate 1489 * for this particular socket send. 1490 * 1491 * We should either leave the flowid at zero (which is what is 1492 * currently done) or set it to some software generated 1493 * hash value based on the packet contents. 1494 */ 1495 ipflags |= IP_NODEFAULTFLOWID; 1496 #endif /* RSS */ 1497 1498 if (unlock_udbinfo == UH_WLOCKED) 1499 INP_HASH_WUNLOCK(pcbinfo); 1500 else if (unlock_udbinfo == UH_RLOCKED) 1501 INP_HASH_RUNLOCK(pcbinfo); 1502 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1503 error = ip_output(m, inp->inp_options, NULL, ipflags, 1504 inp->inp_moptions, inp); 1505 if (unlock_udbinfo == UH_WLOCKED) 1506 INP_WUNLOCK(inp); 1507 else 1508 INP_RUNLOCK(inp); 1509 return (error); 1510 1511 release: 1512 if (unlock_udbinfo == UH_WLOCKED) { 1513 INP_HASH_WUNLOCK(pcbinfo); 1514 INP_WUNLOCK(inp); 1515 } else if (unlock_udbinfo == UH_RLOCKED) { 1516 INP_HASH_RUNLOCK(pcbinfo); 1517 INP_RUNLOCK(inp); 1518 } else 1519 INP_RUNLOCK(inp); 1520 m_freem(m); 1521 return (error); 1522 } 1523 1524 1525 #if defined(IPSEC) && defined(IPSEC_NAT_T) 1526 /* 1527 * Potentially decap ESP in UDP frame. Check for an ESP header 1528 * and optional marker; if present, strip the UDP header and 1529 * push the result through IPSec. 1530 * 1531 * Returns mbuf to be processed (potentially re-allocated) or 1532 * NULL if consumed and/or processed. 1533 */ 1534 static struct mbuf * 1535 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1536 { 1537 size_t minlen, payload, skip, iphlen; 1538 caddr_t data; 1539 struct udpcb *up; 1540 struct m_tag *tag; 1541 struct udphdr *udphdr; 1542 struct ip *ip; 1543 1544 INP_RLOCK_ASSERT(inp); 1545 1546 /* 1547 * Pull up data so the longest case is contiguous: 1548 * IP/UDP hdr + non ESP marker + ESP hdr. 1549 */ 1550 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1551 if (minlen > m->m_pkthdr.len) 1552 minlen = m->m_pkthdr.len; 1553 if ((m = m_pullup(m, minlen)) == NULL) { 1554 IPSECSTAT_INC(ips_in_inval); 1555 return (NULL); /* Bypass caller processing. */ 1556 } 1557 data = mtod(m, caddr_t); /* Points to ip header. */ 1558 payload = m->m_len - off; /* Size of payload. */ 1559 1560 if (payload == 1 && data[off] == '\xff') 1561 return (m); /* NB: keepalive packet, no decap. */ 1562 1563 up = intoudpcb(inp); 1564 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1565 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1566 ("u_flags 0x%x", up->u_flags)); 1567 1568 /* 1569 * Check that the payload is large enough to hold an 1570 * ESP header and compute the amount of data to remove. 1571 * 1572 * NB: the caller has already done a pullup for us. 1573 * XXX can we assume alignment and eliminate bcopys? 1574 */ 1575 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1576 /* 1577 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1578 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1579 * possible AH mode non-IKE marker+non-ESP marker 1580 * from draft-ietf-ipsec-udp-encaps-00.txt. 1581 */ 1582 uint64_t marker; 1583 1584 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1585 return (m); /* NB: no decap. */ 1586 bcopy(data + off, &marker, sizeof(uint64_t)); 1587 if (marker != 0) /* Non-IKE marker. */ 1588 return (m); /* NB: no decap. */ 1589 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1590 } else { 1591 uint32_t spi; 1592 1593 if (payload <= sizeof(struct esp)) { 1594 IPSECSTAT_INC(ips_in_inval); 1595 m_freem(m); 1596 return (NULL); /* Discard. */ 1597 } 1598 bcopy(data + off, &spi, sizeof(uint32_t)); 1599 if (spi == 0) /* Non-ESP marker. */ 1600 return (m); /* NB: no decap. */ 1601 skip = sizeof(struct udphdr); 1602 } 1603 1604 /* 1605 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1606 * the UDP ports. This is required if we want to select 1607 * the right SPD for multiple hosts behind same NAT. 1608 * 1609 * NB: ports are maintained in network byte order everywhere 1610 * in the NAT-T code. 1611 */ 1612 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1613 2 * sizeof(uint16_t), M_NOWAIT); 1614 if (tag == NULL) { 1615 IPSECSTAT_INC(ips_in_nomem); 1616 m_freem(m); 1617 return (NULL); /* Discard. */ 1618 } 1619 iphlen = off - sizeof(struct udphdr); 1620 udphdr = (struct udphdr *)(data + iphlen); 1621 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1622 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1623 m_tag_prepend(m, tag); 1624 1625 /* 1626 * Remove the UDP header (and possibly the non ESP marker) 1627 * IP header length is iphlen 1628 * Before: 1629 * <--- off ---> 1630 * +----+------+-----+ 1631 * | IP | UDP | ESP | 1632 * +----+------+-----+ 1633 * <-skip-> 1634 * After: 1635 * +----+-----+ 1636 * | IP | ESP | 1637 * +----+-----+ 1638 * <-skip-> 1639 */ 1640 ovbcopy(data, data + skip, iphlen); 1641 m_adj(m, skip); 1642 1643 ip = mtod(m, struct ip *); 1644 ip->ip_len = htons(ntohs(ip->ip_len) - skip); 1645 ip->ip_p = IPPROTO_ESP; 1646 1647 /* 1648 * We cannot yet update the cksums so clear any 1649 * h/w cksum flags as they are no longer valid. 1650 */ 1651 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1652 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1653 1654 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1655 return (NULL); /* NB: consumed, bypass processing. */ 1656 } 1657 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1658 1659 static void 1660 udp_abort(struct socket *so) 1661 { 1662 struct inpcb *inp; 1663 struct inpcbinfo *pcbinfo; 1664 1665 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1666 inp = sotoinpcb(so); 1667 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1668 INP_WLOCK(inp); 1669 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1670 INP_HASH_WLOCK(pcbinfo); 1671 in_pcbdisconnect(inp); 1672 inp->inp_laddr.s_addr = INADDR_ANY; 1673 INP_HASH_WUNLOCK(pcbinfo); 1674 soisdisconnected(so); 1675 } 1676 INP_WUNLOCK(inp); 1677 } 1678 1679 static int 1680 udp_attach(struct socket *so, int proto, struct thread *td) 1681 { 1682 struct inpcb *inp; 1683 struct inpcbinfo *pcbinfo; 1684 int error; 1685 1686 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1687 inp = sotoinpcb(so); 1688 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1689 error = soreserve(so, udp_sendspace, udp_recvspace); 1690 if (error) 1691 return (error); 1692 INP_INFO_WLOCK(pcbinfo); 1693 error = in_pcballoc(so, pcbinfo); 1694 if (error) { 1695 INP_INFO_WUNLOCK(pcbinfo); 1696 return (error); 1697 } 1698 1699 inp = sotoinpcb(so); 1700 inp->inp_vflag |= INP_IPV4; 1701 inp->inp_ip_ttl = V_ip_defttl; 1702 1703 error = udp_newudpcb(inp); 1704 if (error) { 1705 in_pcbdetach(inp); 1706 in_pcbfree(inp); 1707 INP_INFO_WUNLOCK(pcbinfo); 1708 return (error); 1709 } 1710 1711 INP_WUNLOCK(inp); 1712 INP_INFO_WUNLOCK(pcbinfo); 1713 return (0); 1714 } 1715 #endif /* INET */ 1716 1717 int 1718 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f) 1719 { 1720 struct inpcb *inp; 1721 struct udpcb *up; 1722 1723 KASSERT(so->so_type == SOCK_DGRAM, 1724 ("udp_set_kernel_tunneling: !dgram")); 1725 inp = sotoinpcb(so); 1726 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1727 INP_WLOCK(inp); 1728 up = intoudpcb(inp); 1729 if (up->u_tun_func != NULL) { 1730 INP_WUNLOCK(inp); 1731 return (EBUSY); 1732 } 1733 up->u_tun_func = f; 1734 INP_WUNLOCK(inp); 1735 return (0); 1736 } 1737 1738 #ifdef INET 1739 static int 1740 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1741 { 1742 struct inpcb *inp; 1743 struct inpcbinfo *pcbinfo; 1744 int error; 1745 1746 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1747 inp = sotoinpcb(so); 1748 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1749 INP_WLOCK(inp); 1750 INP_HASH_WLOCK(pcbinfo); 1751 error = in_pcbbind(inp, nam, td->td_ucred); 1752 INP_HASH_WUNLOCK(pcbinfo); 1753 INP_WUNLOCK(inp); 1754 return (error); 1755 } 1756 1757 static void 1758 udp_close(struct socket *so) 1759 { 1760 struct inpcb *inp; 1761 struct inpcbinfo *pcbinfo; 1762 1763 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1764 inp = sotoinpcb(so); 1765 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1766 INP_WLOCK(inp); 1767 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1768 INP_HASH_WLOCK(pcbinfo); 1769 in_pcbdisconnect(inp); 1770 inp->inp_laddr.s_addr = INADDR_ANY; 1771 INP_HASH_WUNLOCK(pcbinfo); 1772 soisdisconnected(so); 1773 } 1774 INP_WUNLOCK(inp); 1775 } 1776 1777 static int 1778 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1779 { 1780 struct inpcb *inp; 1781 struct inpcbinfo *pcbinfo; 1782 struct sockaddr_in *sin; 1783 int error; 1784 1785 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1786 inp = sotoinpcb(so); 1787 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1788 INP_WLOCK(inp); 1789 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1790 INP_WUNLOCK(inp); 1791 return (EISCONN); 1792 } 1793 sin = (struct sockaddr_in *)nam; 1794 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1795 if (error != 0) { 1796 INP_WUNLOCK(inp); 1797 return (error); 1798 } 1799 INP_HASH_WLOCK(pcbinfo); 1800 error = in_pcbconnect(inp, nam, td->td_ucred); 1801 INP_HASH_WUNLOCK(pcbinfo); 1802 if (error == 0) 1803 soisconnected(so); 1804 INP_WUNLOCK(inp); 1805 return (error); 1806 } 1807 1808 static void 1809 udp_detach(struct socket *so) 1810 { 1811 struct inpcb *inp; 1812 struct inpcbinfo *pcbinfo; 1813 struct udpcb *up; 1814 1815 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1816 inp = sotoinpcb(so); 1817 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1818 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1819 ("udp_detach: not disconnected")); 1820 INP_INFO_WLOCK(pcbinfo); 1821 INP_WLOCK(inp); 1822 up = intoudpcb(inp); 1823 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1824 inp->inp_ppcb = NULL; 1825 in_pcbdetach(inp); 1826 in_pcbfree(inp); 1827 INP_INFO_WUNLOCK(pcbinfo); 1828 udp_discardcb(up); 1829 } 1830 1831 static int 1832 udp_disconnect(struct socket *so) 1833 { 1834 struct inpcb *inp; 1835 struct inpcbinfo *pcbinfo; 1836 1837 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1838 inp = sotoinpcb(so); 1839 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1840 INP_WLOCK(inp); 1841 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1842 INP_WUNLOCK(inp); 1843 return (ENOTCONN); 1844 } 1845 INP_HASH_WLOCK(pcbinfo); 1846 in_pcbdisconnect(inp); 1847 inp->inp_laddr.s_addr = INADDR_ANY; 1848 INP_HASH_WUNLOCK(pcbinfo); 1849 SOCK_LOCK(so); 1850 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1851 SOCK_UNLOCK(so); 1852 INP_WUNLOCK(inp); 1853 return (0); 1854 } 1855 1856 static int 1857 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1858 struct mbuf *control, struct thread *td) 1859 { 1860 struct inpcb *inp; 1861 1862 inp = sotoinpcb(so); 1863 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1864 return (udp_output(inp, m, addr, control, td)); 1865 } 1866 #endif /* INET */ 1867 1868 int 1869 udp_shutdown(struct socket *so) 1870 { 1871 struct inpcb *inp; 1872 1873 inp = sotoinpcb(so); 1874 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1875 INP_WLOCK(inp); 1876 socantsendmore(so); 1877 INP_WUNLOCK(inp); 1878 return (0); 1879 } 1880 1881 #ifdef INET 1882 struct pr_usrreqs udp_usrreqs = { 1883 .pru_abort = udp_abort, 1884 .pru_attach = udp_attach, 1885 .pru_bind = udp_bind, 1886 .pru_connect = udp_connect, 1887 .pru_control = in_control, 1888 .pru_detach = udp_detach, 1889 .pru_disconnect = udp_disconnect, 1890 .pru_peeraddr = in_getpeeraddr, 1891 .pru_send = udp_send, 1892 .pru_soreceive = soreceive_dgram, 1893 .pru_sosend = sosend_dgram, 1894 .pru_shutdown = udp_shutdown, 1895 .pru_sockaddr = in_getsockaddr, 1896 .pru_sosetlabel = in_pcbsosetlabel, 1897 .pru_close = udp_close, 1898 }; 1899 #endif /* INET */ 1900