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