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 * 3. 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_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_ipsec.h" 45 #include "opt_rss.h" 46 47 #include <sys/param.h> 48 #include <sys/domain.h> 49 #include <sys/eventhandler.h> 50 #include <sys/jail.h> 51 #include <sys/kernel.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mbuf.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/protosw.h> 58 #include <sys/sdt.h> 59 #include <sys/signalvar.h> 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/sx.h> 63 #include <sys/sysctl.h> 64 #include <sys/syslog.h> 65 #include <sys/systm.h> 66 67 #include <vm/uma.h> 68 69 #include <net/if.h> 70 #include <net/if_var.h> 71 #include <net/route.h> 72 #include <net/rss_config.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 #include <netipsec/ipsec_support.h> 96 97 #include <machine/in_cksum.h> 98 99 #include <security/mac/mac_framework.h> 100 101 /* 102 * UDP and UDP-Lite protocols implementation. 103 * Per RFC 768, August, 1980. 104 * Per RFC 3828, July, 2004. 105 */ 106 107 /* 108 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 109 * removes the only data integrity mechanism for packets and malformed 110 * packets that would otherwise be discarded due to bad checksums, and may 111 * cause problems (especially for NFS data blocks). 112 */ 113 VNET_DEFINE(int, udp_cksum) = 1; 114 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, 115 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 116 117 int udp_log_in_vain = 0; 118 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 119 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 120 121 VNET_DEFINE(int, udp_blackhole) = 0; 122 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 123 &VNET_NAME(udp_blackhole), 0, 124 "Do not send port unreachables for refused connects"); 125 126 static VNET_DEFINE(int, udp_require_l2_bcast) = 0; 127 #define V_udp_require_l2_bcast VNET(udp_require_l2_bcast) 128 SYSCTL_INT(_net_inet_udp, OID_AUTO, require_l2_bcast, CTLFLAG_VNET | CTLFLAG_RW, 129 &VNET_NAME(udp_require_l2_bcast), 0, 130 "Only treat packets sent to an L2 broadcast address as broadcast packets"); 131 132 u_long udp_sendspace = 9216; /* really max datagram size */ 133 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 134 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 135 136 u_long udp_recvspace = 40 * (1024 + 137 #ifdef INET6 138 sizeof(struct sockaddr_in6) 139 #else 140 sizeof(struct sockaddr_in) 141 #endif 142 ); /* 40 1K datagrams */ 143 144 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 145 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 146 147 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 148 VNET_DEFINE(struct inpcbinfo, udbinfo); 149 VNET_DEFINE(struct inpcbhead, ulitecb); 150 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 151 static VNET_DEFINE(uma_zone_t, udpcb_zone); 152 #define V_udpcb_zone VNET(udpcb_zone) 153 154 #ifndef UDBHASHSIZE 155 #define UDBHASHSIZE 128 156 #endif 157 158 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 159 VNET_PCPUSTAT_SYSINIT(udpstat); 160 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 161 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 162 163 #ifdef VIMAGE 164 VNET_PCPUSTAT_SYSUNINIT(udpstat); 165 #endif /* VIMAGE */ 166 #ifdef INET 167 static void udp_detach(struct socket *so); 168 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 169 struct mbuf *, struct thread *); 170 #endif 171 172 static void 173 udp_zone_change(void *tag) 174 { 175 176 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 177 uma_zone_set_max(V_udpcb_zone, maxsockets); 178 } 179 180 static int 181 udp_inpcb_init(void *mem, int size, int flags) 182 { 183 struct inpcb *inp; 184 185 inp = mem; 186 INP_LOCK_INIT(inp, "inp", "udpinp"); 187 return (0); 188 } 189 190 static int 191 udplite_inpcb_init(void *mem, int size, int flags) 192 { 193 struct inpcb *inp; 194 195 inp = mem; 196 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 197 return (0); 198 } 199 200 void 201 udp_init(void) 202 { 203 204 /* 205 * For now default to 2-tuple UDP hashing - until the fragment 206 * reassembly code can also update the flowid. 207 * 208 * Once we can calculate the flowid that way and re-establish 209 * a 4-tuple, flip this to 4-tuple. 210 */ 211 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 212 "udp_inpcb", udp_inpcb_init, NULL, 0, 213 IPI_HASHFIELDS_2TUPLE); 214 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 215 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 216 uma_zone_set_max(V_udpcb_zone, maxsockets); 217 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 218 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 219 EVENTHANDLER_PRI_ANY); 220 } 221 222 void 223 udplite_init(void) 224 { 225 226 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 227 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL, 228 0, IPI_HASHFIELDS_2TUPLE); 229 } 230 231 /* 232 * Kernel module interface for updating udpstat. The argument is an index 233 * into udpstat treated as an array of u_long. While this encodes the 234 * general layout of udpstat into the caller, it doesn't encode its location, 235 * so that future changes to add, for example, per-CPU stats support won't 236 * cause binary compatibility problems for kernel modules. 237 */ 238 void 239 kmod_udpstat_inc(int statnum) 240 { 241 242 counter_u64_add(VNET(udpstat)[statnum], 1); 243 } 244 245 int 246 udp_newudpcb(struct inpcb *inp) 247 { 248 struct udpcb *up; 249 250 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 251 if (up == NULL) 252 return (ENOBUFS); 253 inp->inp_ppcb = up; 254 return (0); 255 } 256 257 void 258 udp_discardcb(struct udpcb *up) 259 { 260 261 uma_zfree(V_udpcb_zone, up); 262 } 263 264 #ifdef VIMAGE 265 static void 266 udp_destroy(void *unused __unused) 267 { 268 269 in_pcbinfo_destroy(&V_udbinfo); 270 uma_zdestroy(V_udpcb_zone); 271 } 272 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 273 274 static void 275 udplite_destroy(void *unused __unused) 276 { 277 278 in_pcbinfo_destroy(&V_ulitecbinfo); 279 } 280 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 281 NULL); 282 #endif 283 284 #ifdef INET 285 /* 286 * Subroutine of udp_input(), which appends the provided mbuf chain to the 287 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 288 * contains the source address. If the socket ends up being an IPv6 socket, 289 * udp_append() will convert to a sockaddr_in6 before passing the address 290 * into the socket code. 291 * 292 * In the normal case udp_append() will return 0, indicating that you 293 * must unlock the inp. However if a tunneling protocol is in place we increment 294 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 295 * then decrement the reference count. If the inp_rele returns 1, indicating the 296 * inp is gone, we return that to the caller to tell them *not* to unlock 297 * the inp. In the case of multi-cast this will cause the distribution 298 * to stop (though most tunneling protocols known currently do *not* use 299 * multicast). 300 */ 301 static int 302 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 303 struct sockaddr_in *udp_in) 304 { 305 struct sockaddr *append_sa; 306 struct socket *so; 307 struct mbuf *opts = NULL; 308 #ifdef INET6 309 struct sockaddr_in6 udp_in6; 310 #endif 311 struct udpcb *up; 312 313 INP_LOCK_ASSERT(inp); 314 315 /* 316 * Engage the tunneling protocol. 317 */ 318 up = intoudpcb(inp); 319 if (up->u_tun_func != NULL) { 320 in_pcbref(inp); 321 INP_RUNLOCK(inp); 322 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in, 323 up->u_tun_ctx); 324 INP_RLOCK(inp); 325 return (in_pcbrele_rlocked(inp)); 326 } 327 328 off += sizeof(struct udphdr); 329 330 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 331 /* Check AH/ESP integrity. */ 332 if (IPSEC_ENABLED(ipv4) && 333 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 334 m_freem(n); 335 return (0); 336 } 337 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 338 if (IPSEC_ENABLED(ipv4) && 339 UDPENCAP_INPUT(n, off, AF_INET) != 0) 340 return (0); /* Consumed. */ 341 } 342 #endif /* IPSEC */ 343 #ifdef MAC 344 if (mac_inpcb_check_deliver(inp, n) != 0) { 345 m_freem(n); 346 return (0); 347 } 348 #endif /* MAC */ 349 if (inp->inp_flags & INP_CONTROLOPTS || 350 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 351 #ifdef INET6 352 if (inp->inp_vflag & INP_IPV6) 353 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 354 else 355 #endif /* INET6 */ 356 ip_savecontrol(inp, &opts, ip, n); 357 } 358 #ifdef INET6 359 if (inp->inp_vflag & INP_IPV6) { 360 bzero(&udp_in6, sizeof(udp_in6)); 361 udp_in6.sin6_len = sizeof(udp_in6); 362 udp_in6.sin6_family = AF_INET6; 363 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 364 append_sa = (struct sockaddr *)&udp_in6; 365 } else 366 #endif /* INET6 */ 367 append_sa = (struct sockaddr *)udp_in; 368 m_adj(n, off); 369 370 so = inp->inp_socket; 371 SOCKBUF_LOCK(&so->so_rcv); 372 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 373 SOCKBUF_UNLOCK(&so->so_rcv); 374 m_freem(n); 375 if (opts) 376 m_freem(opts); 377 UDPSTAT_INC(udps_fullsock); 378 } else 379 sorwakeup_locked(so); 380 return (0); 381 } 382 383 int 384 udp_input(struct mbuf **mp, int *offp, int proto) 385 { 386 struct ip *ip; 387 struct udphdr *uh; 388 struct ifnet *ifp; 389 struct inpcb *inp; 390 uint16_t len, ip_len; 391 struct inpcbinfo *pcbinfo; 392 struct ip save_ip; 393 struct sockaddr_in udp_in; 394 struct mbuf *m; 395 struct m_tag *fwd_tag; 396 int cscov_partial, iphlen; 397 398 m = *mp; 399 iphlen = *offp; 400 ifp = m->m_pkthdr.rcvif; 401 *mp = NULL; 402 UDPSTAT_INC(udps_ipackets); 403 404 /* 405 * Strip IP options, if any; should skip this, make available to 406 * user, and use on returned packets, but we don't yet have a way to 407 * check the checksum with options still present. 408 */ 409 if (iphlen > sizeof (struct ip)) { 410 ip_stripoptions(m); 411 iphlen = sizeof(struct ip); 412 } 413 414 /* 415 * Get IP and UDP header together in first mbuf. 416 */ 417 ip = mtod(m, struct ip *); 418 if (m->m_len < iphlen + sizeof(struct udphdr)) { 419 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 420 UDPSTAT_INC(udps_hdrops); 421 return (IPPROTO_DONE); 422 } 423 ip = mtod(m, struct ip *); 424 } 425 uh = (struct udphdr *)((caddr_t)ip + iphlen); 426 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 427 428 /* 429 * Destination port of 0 is illegal, based on RFC768. 430 */ 431 if (uh->uh_dport == 0) 432 goto badunlocked; 433 434 /* 435 * Construct sockaddr format source address. Stuff source address 436 * and datagram in user buffer. 437 */ 438 bzero(&udp_in, sizeof(udp_in)); 439 udp_in.sin_len = sizeof(udp_in); 440 udp_in.sin_family = AF_INET; 441 udp_in.sin_port = uh->uh_sport; 442 udp_in.sin_addr = ip->ip_src; 443 444 /* 445 * Make mbuf data length reflect UDP length. If not enough data to 446 * reflect UDP length, drop. 447 */ 448 len = ntohs((u_short)uh->uh_ulen); 449 ip_len = ntohs(ip->ip_len) - iphlen; 450 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 451 /* Zero means checksum over the complete packet. */ 452 if (len == 0) 453 len = ip_len; 454 cscov_partial = 0; 455 } 456 if (ip_len != len) { 457 if (len > ip_len || len < sizeof(struct udphdr)) { 458 UDPSTAT_INC(udps_badlen); 459 goto badunlocked; 460 } 461 if (proto == IPPROTO_UDP) 462 m_adj(m, len - ip_len); 463 } 464 465 /* 466 * Save a copy of the IP header in case we want restore it for 467 * sending an ICMP error message in response. 468 */ 469 if (!V_udp_blackhole) 470 save_ip = *ip; 471 else 472 memset(&save_ip, 0, sizeof(save_ip)); 473 474 /* 475 * Checksum extended UDP header and data. 476 */ 477 if (uh->uh_sum) { 478 u_short uh_sum; 479 480 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 481 !cscov_partial) { 482 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 483 uh_sum = m->m_pkthdr.csum_data; 484 else 485 uh_sum = in_pseudo(ip->ip_src.s_addr, 486 ip->ip_dst.s_addr, htonl((u_short)len + 487 m->m_pkthdr.csum_data + proto)); 488 uh_sum ^= 0xffff; 489 } else { 490 char b[9]; 491 492 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 493 bzero(((struct ipovly *)ip)->ih_x1, 9); 494 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 495 uh->uh_ulen : htons(ip_len); 496 uh_sum = in_cksum(m, len + sizeof (struct ip)); 497 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 498 } 499 if (uh_sum) { 500 UDPSTAT_INC(udps_badsum); 501 m_freem(m); 502 return (IPPROTO_DONE); 503 } 504 } else { 505 if (proto == IPPROTO_UDP) { 506 UDPSTAT_INC(udps_nosum); 507 } else { 508 /* UDPLite requires a checksum */ 509 /* XXX: What is the right UDPLite MIB counter here? */ 510 m_freem(m); 511 return (IPPROTO_DONE); 512 } 513 } 514 515 pcbinfo = udp_get_inpcbinfo(proto); 516 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 517 ((!V_udp_require_l2_bcast || m->m_flags & M_BCAST) && 518 in_broadcast(ip->ip_dst, ifp))) { 519 struct inpcb *last; 520 struct inpcbhead *pcblist; 521 struct ip_moptions *imo; 522 523 INP_INFO_RLOCK(pcbinfo); 524 pcblist = udp_get_pcblist(proto); 525 last = NULL; 526 LIST_FOREACH(inp, pcblist, inp_list) { 527 if (inp->inp_lport != uh->uh_dport) 528 continue; 529 #ifdef INET6 530 if ((inp->inp_vflag & INP_IPV4) == 0) 531 continue; 532 #endif 533 if (inp->inp_laddr.s_addr != INADDR_ANY && 534 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 535 continue; 536 if (inp->inp_faddr.s_addr != INADDR_ANY && 537 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 538 continue; 539 if (inp->inp_fport != 0 && 540 inp->inp_fport != uh->uh_sport) 541 continue; 542 543 INP_RLOCK(inp); 544 545 /* 546 * XXXRW: Because we weren't holding either the inpcb 547 * or the hash lock when we checked for a match 548 * before, we should probably recheck now that the 549 * inpcb lock is held. 550 */ 551 552 /* 553 * Handle socket delivery policy for any-source 554 * and source-specific multicast. [RFC3678] 555 */ 556 imo = inp->inp_moptions; 557 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 558 struct sockaddr_in group; 559 int blocked; 560 if (imo == NULL) { 561 INP_RUNLOCK(inp); 562 continue; 563 } 564 bzero(&group, sizeof(struct sockaddr_in)); 565 group.sin_len = sizeof(struct sockaddr_in); 566 group.sin_family = AF_INET; 567 group.sin_addr = ip->ip_dst; 568 569 blocked = imo_multi_filter(imo, ifp, 570 (struct sockaddr *)&group, 571 (struct sockaddr *)&udp_in); 572 if (blocked != MCAST_PASS) { 573 if (blocked == MCAST_NOTGMEMBER) 574 IPSTAT_INC(ips_notmember); 575 if (blocked == MCAST_NOTSMEMBER || 576 blocked == MCAST_MUTED) 577 UDPSTAT_INC(udps_filtermcast); 578 INP_RUNLOCK(inp); 579 continue; 580 } 581 } 582 if (last != NULL) { 583 struct mbuf *n; 584 585 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 586 NULL) { 587 UDP_PROBE(receive, NULL, last, ip, 588 last, uh); 589 if (udp_append(last, ip, n, iphlen, 590 &udp_in)) { 591 goto inp_lost; 592 } 593 } 594 INP_RUNLOCK(last); 595 } 596 last = inp; 597 /* 598 * Don't look for additional matches if this one does 599 * not have either the SO_REUSEPORT or SO_REUSEADDR 600 * socket options set. This heuristic avoids 601 * searching through all pcbs in the common case of a 602 * non-shared port. It assumes that an application 603 * will never clear these options after setting them. 604 */ 605 if ((last->inp_socket->so_options & 606 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 607 break; 608 } 609 610 if (last == NULL) { 611 /* 612 * No matching pcb found; discard datagram. (No need 613 * to send an ICMP Port Unreachable for a broadcast 614 * or multicast datgram.) 615 */ 616 UDPSTAT_INC(udps_noportbcast); 617 if (inp) 618 INP_RUNLOCK(inp); 619 INP_INFO_RUNLOCK(pcbinfo); 620 goto badunlocked; 621 } 622 UDP_PROBE(receive, NULL, last, ip, last, uh); 623 if (udp_append(last, ip, m, iphlen, &udp_in) == 0) 624 INP_RUNLOCK(last); 625 inp_lost: 626 INP_INFO_RUNLOCK(pcbinfo); 627 return (IPPROTO_DONE); 628 } 629 630 /* 631 * Locate pcb for datagram. 632 */ 633 634 /* 635 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 636 */ 637 if ((m->m_flags & M_IP_NEXTHOP) && 638 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 639 struct sockaddr_in *next_hop; 640 641 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 642 643 /* 644 * Transparently forwarded. Pretend to be the destination. 645 * Already got one like this? 646 */ 647 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 648 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 649 if (!inp) { 650 /* 651 * It's new. Try to find the ambushing socket. 652 * Because we've rewritten the destination address, 653 * any hardware-generated hash is ignored. 654 */ 655 inp = in_pcblookup(pcbinfo, ip->ip_src, 656 uh->uh_sport, next_hop->sin_addr, 657 next_hop->sin_port ? htons(next_hop->sin_port) : 658 uh->uh_dport, INPLOOKUP_WILDCARD | 659 INPLOOKUP_RLOCKPCB, ifp); 660 } 661 /* Remove the tag from the packet. We don't need it anymore. */ 662 m_tag_delete(m, fwd_tag); 663 m->m_flags &= ~M_IP_NEXTHOP; 664 } else 665 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 666 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 667 INPLOOKUP_RLOCKPCB, ifp, m); 668 if (inp == NULL) { 669 if (udp_log_in_vain) { 670 char src[INET_ADDRSTRLEN]; 671 char dst[INET_ADDRSTRLEN]; 672 673 log(LOG_INFO, 674 "Connection attempt to UDP %s:%d from %s:%d\n", 675 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 676 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 677 } 678 UDPSTAT_INC(udps_noport); 679 if (m->m_flags & (M_BCAST | M_MCAST)) { 680 UDPSTAT_INC(udps_noportbcast); 681 goto badunlocked; 682 } 683 if (V_udp_blackhole) 684 goto badunlocked; 685 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 686 goto badunlocked; 687 *ip = save_ip; 688 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 689 return (IPPROTO_DONE); 690 } 691 692 /* 693 * Check the minimum TTL for socket. 694 */ 695 INP_RLOCK_ASSERT(inp); 696 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 697 INP_RUNLOCK(inp); 698 m_freem(m); 699 return (IPPROTO_DONE); 700 } 701 if (cscov_partial) { 702 struct udpcb *up; 703 704 up = intoudpcb(inp); 705 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 706 INP_RUNLOCK(inp); 707 m_freem(m); 708 return (IPPROTO_DONE); 709 } 710 } 711 712 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 713 if (udp_append(inp, ip, m, iphlen, &udp_in) == 0) 714 INP_RUNLOCK(inp); 715 return (IPPROTO_DONE); 716 717 badunlocked: 718 m_freem(m); 719 return (IPPROTO_DONE); 720 } 721 #endif /* INET */ 722 723 /* 724 * Notify a udp user of an asynchronous error; just wake up so that they can 725 * collect error status. 726 */ 727 struct inpcb * 728 udp_notify(struct inpcb *inp, int errno) 729 { 730 731 /* 732 * While udp_ctlinput() always calls udp_notify() with a read lock 733 * when invoking it directly, in_pcbnotifyall() currently uses write 734 * locks due to sharing code with TCP. For now, accept either a read 735 * or a write lock, but a read lock is sufficient. 736 */ 737 INP_LOCK_ASSERT(inp); 738 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 739 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 740 RTFREE(inp->inp_route.ro_rt); 741 inp->inp_route.ro_rt = (struct rtentry *)NULL; 742 } 743 744 inp->inp_socket->so_error = errno; 745 sorwakeup(inp->inp_socket); 746 sowwakeup(inp->inp_socket); 747 return (inp); 748 } 749 750 #ifdef INET 751 static void 752 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 753 struct inpcbinfo *pcbinfo) 754 { 755 struct ip *ip = vip; 756 struct udphdr *uh; 757 struct in_addr faddr; 758 struct inpcb *inp; 759 760 faddr = ((struct sockaddr_in *)sa)->sin_addr; 761 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 762 return; 763 764 if (PRC_IS_REDIRECT(cmd)) { 765 /* signal EHOSTDOWN, as it flushes the cached route */ 766 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 767 return; 768 } 769 770 /* 771 * Hostdead is ugly because it goes linearly through all PCBs. 772 * 773 * XXX: We never get this from ICMP, otherwise it makes an excellent 774 * DoS attack on machines with many connections. 775 */ 776 if (cmd == PRC_HOSTDEAD) 777 ip = NULL; 778 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 779 return; 780 if (ip != NULL) { 781 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 782 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 783 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 784 if (inp != NULL) { 785 INP_RLOCK_ASSERT(inp); 786 if (inp->inp_socket != NULL) { 787 udp_notify(inp, inetctlerrmap[cmd]); 788 } 789 INP_RUNLOCK(inp); 790 } else { 791 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 792 ip->ip_src, uh->uh_sport, 793 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 794 if (inp != NULL) { 795 struct udpcb *up; 796 797 up = intoudpcb(inp); 798 if (up->u_icmp_func != NULL) { 799 INP_RUNLOCK(inp); 800 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx); 801 } else { 802 INP_RUNLOCK(inp); 803 } 804 } 805 } 806 } else 807 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 808 udp_notify); 809 } 810 void 811 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 812 { 813 814 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 815 } 816 817 void 818 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 819 { 820 821 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 822 } 823 #endif /* INET */ 824 825 static int 826 udp_pcblist(SYSCTL_HANDLER_ARGS) 827 { 828 int error, i, n; 829 struct inpcb *inp, **inp_list; 830 inp_gen_t gencnt; 831 struct xinpgen xig; 832 833 /* 834 * The process of preparing the PCB list is too time-consuming and 835 * resource-intensive to repeat twice on every request. 836 */ 837 if (req->oldptr == 0) { 838 n = V_udbinfo.ipi_count; 839 n += imax(n / 8, 10); 840 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 841 return (0); 842 } 843 844 if (req->newptr != 0) 845 return (EPERM); 846 847 /* 848 * OK, now we're committed to doing something. 849 */ 850 INP_INFO_RLOCK(&V_udbinfo); 851 gencnt = V_udbinfo.ipi_gencnt; 852 n = V_udbinfo.ipi_count; 853 INP_INFO_RUNLOCK(&V_udbinfo); 854 855 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 856 + n * sizeof(struct xinpcb)); 857 if (error != 0) 858 return (error); 859 860 xig.xig_len = sizeof xig; 861 xig.xig_count = n; 862 xig.xig_gen = gencnt; 863 xig.xig_sogen = so_gencnt; 864 error = SYSCTL_OUT(req, &xig, sizeof xig); 865 if (error) 866 return (error); 867 868 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 869 if (inp_list == NULL) 870 return (ENOMEM); 871 872 INP_INFO_RLOCK(&V_udbinfo); 873 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 874 inp = LIST_NEXT(inp, inp_list)) { 875 INP_WLOCK(inp); 876 if (inp->inp_gencnt <= gencnt && 877 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 878 in_pcbref(inp); 879 inp_list[i++] = inp; 880 } 881 INP_WUNLOCK(inp); 882 } 883 INP_INFO_RUNLOCK(&V_udbinfo); 884 n = i; 885 886 error = 0; 887 for (i = 0; i < n; i++) { 888 inp = inp_list[i]; 889 INP_RLOCK(inp); 890 if (inp->inp_gencnt <= gencnt) { 891 struct xinpcb xi; 892 893 bzero(&xi, sizeof(xi)); 894 xi.xi_len = sizeof xi; 895 /* XXX should avoid extra copy */ 896 bcopy(inp, &xi.xi_inp, sizeof *inp); 897 if (inp->inp_socket) 898 sotoxsocket(inp->inp_socket, &xi.xi_socket); 899 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 900 INP_RUNLOCK(inp); 901 error = SYSCTL_OUT(req, &xi, sizeof xi); 902 } else 903 INP_RUNLOCK(inp); 904 } 905 INP_INFO_WLOCK(&V_udbinfo); 906 for (i = 0; i < n; i++) { 907 inp = inp_list[i]; 908 INP_RLOCK(inp); 909 if (!in_pcbrele_rlocked(inp)) 910 INP_RUNLOCK(inp); 911 } 912 INP_INFO_WUNLOCK(&V_udbinfo); 913 914 if (!error) { 915 /* 916 * Give the user an updated idea of our state. If the 917 * generation differs from what we told her before, she knows 918 * that something happened while we were processing this 919 * request, and it might be necessary to retry. 920 */ 921 INP_INFO_RLOCK(&V_udbinfo); 922 xig.xig_gen = V_udbinfo.ipi_gencnt; 923 xig.xig_sogen = so_gencnt; 924 xig.xig_count = V_udbinfo.ipi_count; 925 INP_INFO_RUNLOCK(&V_udbinfo); 926 error = SYSCTL_OUT(req, &xig, sizeof xig); 927 } 928 free(inp_list, M_TEMP); 929 return (error); 930 } 931 932 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 933 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 934 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 935 936 #ifdef INET 937 static int 938 udp_getcred(SYSCTL_HANDLER_ARGS) 939 { 940 struct xucred xuc; 941 struct sockaddr_in addrs[2]; 942 struct inpcb *inp; 943 int error; 944 945 error = priv_check(req->td, PRIV_NETINET_GETCRED); 946 if (error) 947 return (error); 948 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 949 if (error) 950 return (error); 951 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 952 addrs[0].sin_addr, addrs[0].sin_port, 953 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 954 if (inp != NULL) { 955 INP_RLOCK_ASSERT(inp); 956 if (inp->inp_socket == NULL) 957 error = ENOENT; 958 if (error == 0) 959 error = cr_canseeinpcb(req->td->td_ucred, inp); 960 if (error == 0) 961 cru2x(inp->inp_cred, &xuc); 962 INP_RUNLOCK(inp); 963 } else 964 error = ENOENT; 965 if (error == 0) 966 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 967 return (error); 968 } 969 970 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 971 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 972 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 973 #endif /* INET */ 974 975 int 976 udp_ctloutput(struct socket *so, struct sockopt *sopt) 977 { 978 struct inpcb *inp; 979 struct udpcb *up; 980 int isudplite, error, optval; 981 982 error = 0; 983 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 984 inp = sotoinpcb(so); 985 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 986 INP_WLOCK(inp); 987 if (sopt->sopt_level != so->so_proto->pr_protocol) { 988 #ifdef INET6 989 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 990 INP_WUNLOCK(inp); 991 error = ip6_ctloutput(so, sopt); 992 } 993 #endif 994 #if defined(INET) && defined(INET6) 995 else 996 #endif 997 #ifdef INET 998 { 999 INP_WUNLOCK(inp); 1000 error = ip_ctloutput(so, sopt); 1001 } 1002 #endif 1003 return (error); 1004 } 1005 1006 switch (sopt->sopt_dir) { 1007 case SOPT_SET: 1008 switch (sopt->sopt_name) { 1009 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1010 #ifdef INET 1011 case UDP_ENCAP: 1012 if (!IPSEC_ENABLED(ipv4)) { 1013 INP_WUNLOCK(inp); 1014 return (ENOPROTOOPT); 1015 } 1016 error = UDPENCAP_PCBCTL(inp, sopt); 1017 break; 1018 #endif /* INET */ 1019 #endif /* IPSEC */ 1020 case UDPLITE_SEND_CSCOV: 1021 case UDPLITE_RECV_CSCOV: 1022 if (!isudplite) { 1023 INP_WUNLOCK(inp); 1024 error = ENOPROTOOPT; 1025 break; 1026 } 1027 INP_WUNLOCK(inp); 1028 error = sooptcopyin(sopt, &optval, sizeof(optval), 1029 sizeof(optval)); 1030 if (error != 0) 1031 break; 1032 inp = sotoinpcb(so); 1033 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1034 INP_WLOCK(inp); 1035 up = intoudpcb(inp); 1036 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1037 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1038 INP_WUNLOCK(inp); 1039 error = EINVAL; 1040 break; 1041 } 1042 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1043 up->u_txcslen = optval; 1044 else 1045 up->u_rxcslen = optval; 1046 INP_WUNLOCK(inp); 1047 break; 1048 default: 1049 INP_WUNLOCK(inp); 1050 error = ENOPROTOOPT; 1051 break; 1052 } 1053 break; 1054 case SOPT_GET: 1055 switch (sopt->sopt_name) { 1056 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1057 #ifdef INET 1058 case UDP_ENCAP: 1059 if (!IPSEC_ENABLED(ipv4)) { 1060 INP_WUNLOCK(inp); 1061 return (ENOPROTOOPT); 1062 } 1063 error = UDPENCAP_PCBCTL(inp, sopt); 1064 break; 1065 #endif /* INET */ 1066 #endif /* IPSEC */ 1067 case UDPLITE_SEND_CSCOV: 1068 case UDPLITE_RECV_CSCOV: 1069 if (!isudplite) { 1070 INP_WUNLOCK(inp); 1071 error = ENOPROTOOPT; 1072 break; 1073 } 1074 up = intoudpcb(inp); 1075 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1076 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1077 optval = up->u_txcslen; 1078 else 1079 optval = up->u_rxcslen; 1080 INP_WUNLOCK(inp); 1081 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1082 break; 1083 default: 1084 INP_WUNLOCK(inp); 1085 error = ENOPROTOOPT; 1086 break; 1087 } 1088 break; 1089 } 1090 return (error); 1091 } 1092 1093 #ifdef INET 1094 #define UH_WLOCKED 2 1095 #define UH_RLOCKED 1 1096 #define UH_UNLOCKED 0 1097 static int 1098 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1099 struct mbuf *control, struct thread *td) 1100 { 1101 struct udpiphdr *ui; 1102 int len = m->m_pkthdr.len; 1103 struct in_addr faddr, laddr; 1104 struct cmsghdr *cm; 1105 struct inpcbinfo *pcbinfo; 1106 struct sockaddr_in *sin, src; 1107 int cscov_partial = 0; 1108 int error = 0; 1109 int ipflags; 1110 u_short fport, lport; 1111 int unlock_udbinfo, unlock_inp; 1112 u_char tos; 1113 uint8_t pr; 1114 uint16_t cscov = 0; 1115 uint32_t flowid = 0; 1116 uint8_t flowtype = M_HASHTYPE_NONE; 1117 1118 /* 1119 * udp_output() may need to temporarily bind or connect the current 1120 * inpcb. As such, we don't know up front whether we will need the 1121 * pcbinfo lock or not. Do any work to decide what is needed up 1122 * front before acquiring any locks. 1123 */ 1124 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1125 if (control) 1126 m_freem(control); 1127 m_freem(m); 1128 return (EMSGSIZE); 1129 } 1130 1131 src.sin_family = 0; 1132 sin = (struct sockaddr_in *)addr; 1133 if (sin == NULL || 1134 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1135 INP_WLOCK(inp); 1136 unlock_inp = UH_WLOCKED; 1137 } else { 1138 INP_RLOCK(inp); 1139 unlock_inp = UH_RLOCKED; 1140 } 1141 tos = inp->inp_ip_tos; 1142 if (control != NULL) { 1143 /* 1144 * XXX: Currently, we assume all the optional information is 1145 * stored in a single mbuf. 1146 */ 1147 if (control->m_next) { 1148 if (unlock_inp == UH_WLOCKED) 1149 INP_WUNLOCK(inp); 1150 else 1151 INP_RUNLOCK(inp); 1152 m_freem(control); 1153 m_freem(m); 1154 return (EINVAL); 1155 } 1156 for (; control->m_len > 0; 1157 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1158 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1159 cm = mtod(control, struct cmsghdr *); 1160 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1161 || cm->cmsg_len > control->m_len) { 1162 error = EINVAL; 1163 break; 1164 } 1165 if (cm->cmsg_level != IPPROTO_IP) 1166 continue; 1167 1168 switch (cm->cmsg_type) { 1169 case IP_SENDSRCADDR: 1170 if (cm->cmsg_len != 1171 CMSG_LEN(sizeof(struct in_addr))) { 1172 error = EINVAL; 1173 break; 1174 } 1175 bzero(&src, sizeof(src)); 1176 src.sin_family = AF_INET; 1177 src.sin_len = sizeof(src); 1178 src.sin_port = inp->inp_lport; 1179 src.sin_addr = 1180 *(struct in_addr *)CMSG_DATA(cm); 1181 break; 1182 1183 case IP_TOS: 1184 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1185 error = EINVAL; 1186 break; 1187 } 1188 tos = *(u_char *)CMSG_DATA(cm); 1189 break; 1190 1191 case IP_FLOWID: 1192 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1193 error = EINVAL; 1194 break; 1195 } 1196 flowid = *(uint32_t *) CMSG_DATA(cm); 1197 break; 1198 1199 case IP_FLOWTYPE: 1200 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1201 error = EINVAL; 1202 break; 1203 } 1204 flowtype = *(uint32_t *) CMSG_DATA(cm); 1205 break; 1206 1207 #ifdef RSS 1208 case IP_RSSBUCKETID: 1209 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1210 error = EINVAL; 1211 break; 1212 } 1213 /* This is just a placeholder for now */ 1214 break; 1215 #endif /* RSS */ 1216 default: 1217 error = ENOPROTOOPT; 1218 break; 1219 } 1220 if (error) 1221 break; 1222 } 1223 m_freem(control); 1224 } 1225 if (error) { 1226 if (unlock_inp == UH_WLOCKED) 1227 INP_WUNLOCK(inp); 1228 else 1229 INP_RUNLOCK(inp); 1230 m_freem(m); 1231 return (error); 1232 } 1233 1234 /* 1235 * Depending on whether or not the application has bound or connected 1236 * the socket, we may have to do varying levels of work. The optimal 1237 * case is for a connected UDP socket, as a global lock isn't 1238 * required at all. 1239 * 1240 * In order to decide which we need, we require stability of the 1241 * inpcb binding, which we ensure by acquiring a read lock on the 1242 * inpcb. This doesn't strictly follow the lock order, so we play 1243 * the trylock and retry game; note that we may end up with more 1244 * conservative locks than required the second time around, so later 1245 * assertions have to accept that. Further analysis of the number of 1246 * misses under contention is required. 1247 * 1248 * XXXRW: Check that hash locking update here is correct. 1249 */ 1250 pr = inp->inp_socket->so_proto->pr_protocol; 1251 pcbinfo = udp_get_inpcbinfo(pr); 1252 sin = (struct sockaddr_in *)addr; 1253 if (sin != NULL && 1254 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1255 INP_HASH_WLOCK(pcbinfo); 1256 unlock_udbinfo = UH_WLOCKED; 1257 } else if ((sin != NULL && ( 1258 (sin->sin_addr.s_addr == INADDR_ANY) || 1259 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1260 (inp->inp_laddr.s_addr == INADDR_ANY) || 1261 (inp->inp_lport == 0))) || 1262 (src.sin_family == AF_INET)) { 1263 INP_HASH_RLOCK(pcbinfo); 1264 unlock_udbinfo = UH_RLOCKED; 1265 } else 1266 unlock_udbinfo = UH_UNLOCKED; 1267 1268 /* 1269 * If the IP_SENDSRCADDR control message was specified, override the 1270 * source address for this datagram. Its use is invalidated if the 1271 * address thus specified is incomplete or clobbers other inpcbs. 1272 */ 1273 laddr = inp->inp_laddr; 1274 lport = inp->inp_lport; 1275 if (src.sin_family == AF_INET) { 1276 INP_HASH_LOCK_ASSERT(pcbinfo); 1277 if ((lport == 0) || 1278 (laddr.s_addr == INADDR_ANY && 1279 src.sin_addr.s_addr == INADDR_ANY)) { 1280 error = EINVAL; 1281 goto release; 1282 } 1283 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1284 &laddr.s_addr, &lport, td->td_ucred); 1285 if (error) 1286 goto release; 1287 } 1288 1289 /* 1290 * If a UDP socket has been connected, then a local address/port will 1291 * have been selected and bound. 1292 * 1293 * If a UDP socket has not been connected to, then an explicit 1294 * destination address must be used, in which case a local 1295 * address/port may not have been selected and bound. 1296 */ 1297 if (sin != NULL) { 1298 INP_LOCK_ASSERT(inp); 1299 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1300 error = EISCONN; 1301 goto release; 1302 } 1303 1304 /* 1305 * Jail may rewrite the destination address, so let it do 1306 * that before we use it. 1307 */ 1308 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1309 if (error) 1310 goto release; 1311 1312 /* 1313 * If a local address or port hasn't yet been selected, or if 1314 * the destination address needs to be rewritten due to using 1315 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1316 * to do the heavy lifting. Once a port is selected, we 1317 * commit the binding back to the socket; we also commit the 1318 * binding of the address if in jail. 1319 * 1320 * If we already have a valid binding and we're not 1321 * requesting a destination address rewrite, use a fast path. 1322 */ 1323 if (inp->inp_laddr.s_addr == INADDR_ANY || 1324 inp->inp_lport == 0 || 1325 sin->sin_addr.s_addr == INADDR_ANY || 1326 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1327 INP_HASH_LOCK_ASSERT(pcbinfo); 1328 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1329 &lport, &faddr.s_addr, &fport, NULL, 1330 td->td_ucred); 1331 if (error) 1332 goto release; 1333 1334 /* 1335 * XXXRW: Why not commit the port if the address is 1336 * !INADDR_ANY? 1337 */ 1338 /* Commit the local port if newly assigned. */ 1339 if (inp->inp_laddr.s_addr == INADDR_ANY && 1340 inp->inp_lport == 0) { 1341 INP_WLOCK_ASSERT(inp); 1342 INP_HASH_WLOCK_ASSERT(pcbinfo); 1343 /* 1344 * Remember addr if jailed, to prevent 1345 * rebinding. 1346 */ 1347 if (prison_flag(td->td_ucred, PR_IP4)) 1348 inp->inp_laddr = laddr; 1349 inp->inp_lport = lport; 1350 if (in_pcbinshash(inp) != 0) { 1351 inp->inp_lport = 0; 1352 error = EAGAIN; 1353 goto release; 1354 } 1355 inp->inp_flags |= INP_ANONPORT; 1356 } 1357 } else { 1358 faddr = sin->sin_addr; 1359 fport = sin->sin_port; 1360 } 1361 } else { 1362 INP_LOCK_ASSERT(inp); 1363 faddr = inp->inp_faddr; 1364 fport = inp->inp_fport; 1365 if (faddr.s_addr == INADDR_ANY) { 1366 error = ENOTCONN; 1367 goto release; 1368 } 1369 } 1370 1371 /* 1372 * Calculate data length and get a mbuf for UDP, IP, and possible 1373 * link-layer headers. Immediate slide the data pointer back forward 1374 * since we won't use that space at this layer. 1375 */ 1376 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1377 if (m == NULL) { 1378 error = ENOBUFS; 1379 goto release; 1380 } 1381 m->m_data += max_linkhdr; 1382 m->m_len -= max_linkhdr; 1383 m->m_pkthdr.len -= max_linkhdr; 1384 1385 /* 1386 * Fill in mbuf with extended UDP header and addresses and length put 1387 * into network format. 1388 */ 1389 ui = mtod(m, struct udpiphdr *); 1390 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1391 ui->ui_pr = pr; 1392 ui->ui_src = laddr; 1393 ui->ui_dst = faddr; 1394 ui->ui_sport = lport; 1395 ui->ui_dport = fport; 1396 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1397 if (pr == IPPROTO_UDPLITE) { 1398 struct udpcb *up; 1399 uint16_t plen; 1400 1401 up = intoudpcb(inp); 1402 cscov = up->u_txcslen; 1403 plen = (u_short)len + sizeof(struct udphdr); 1404 if (cscov >= plen) 1405 cscov = 0; 1406 ui->ui_len = htons(plen); 1407 ui->ui_ulen = htons(cscov); 1408 /* 1409 * For UDP-Lite, checksum coverage length of zero means 1410 * the entire UDPLite packet is covered by the checksum. 1411 */ 1412 cscov_partial = (cscov == 0) ? 0 : 1; 1413 } else 1414 ui->ui_v = IPVERSION << 4; 1415 1416 /* 1417 * Set the Don't Fragment bit in the IP header. 1418 */ 1419 if (inp->inp_flags & INP_DONTFRAG) { 1420 struct ip *ip; 1421 1422 ip = (struct ip *)&ui->ui_i; 1423 ip->ip_off |= htons(IP_DF); 1424 } 1425 1426 ipflags = 0; 1427 if (inp->inp_socket->so_options & SO_DONTROUTE) 1428 ipflags |= IP_ROUTETOIF; 1429 if (inp->inp_socket->so_options & SO_BROADCAST) 1430 ipflags |= IP_ALLOWBROADCAST; 1431 if (inp->inp_flags & INP_ONESBCAST) 1432 ipflags |= IP_SENDONES; 1433 1434 #ifdef MAC 1435 mac_inpcb_create_mbuf(inp, m); 1436 #endif 1437 1438 /* 1439 * Set up checksum and output datagram. 1440 */ 1441 ui->ui_sum = 0; 1442 if (pr == IPPROTO_UDPLITE) { 1443 if (inp->inp_flags & INP_ONESBCAST) 1444 faddr.s_addr = INADDR_BROADCAST; 1445 if (cscov_partial) { 1446 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1447 ui->ui_sum = 0xffff; 1448 } else { 1449 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1450 ui->ui_sum = 0xffff; 1451 } 1452 } else if (V_udp_cksum) { 1453 if (inp->inp_flags & INP_ONESBCAST) 1454 faddr.s_addr = INADDR_BROADCAST; 1455 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1456 htons((u_short)len + sizeof(struct udphdr) + pr)); 1457 m->m_pkthdr.csum_flags = CSUM_UDP; 1458 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1459 } 1460 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1461 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1462 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1463 UDPSTAT_INC(udps_opackets); 1464 1465 /* 1466 * Setup flowid / RSS information for outbound socket. 1467 * 1468 * Once the UDP code decides to set a flowid some other way, 1469 * this allows the flowid to be overridden by userland. 1470 */ 1471 if (flowtype != M_HASHTYPE_NONE) { 1472 m->m_pkthdr.flowid = flowid; 1473 M_HASHTYPE_SET(m, flowtype); 1474 #ifdef RSS 1475 } else { 1476 uint32_t hash_val, hash_type; 1477 /* 1478 * Calculate an appropriate RSS hash for UDP and 1479 * UDP Lite. 1480 * 1481 * The called function will take care of figuring out 1482 * whether a 2-tuple or 4-tuple hash is required based 1483 * on the currently configured scheme. 1484 * 1485 * Later later on connected socket values should be 1486 * cached in the inpcb and reused, rather than constantly 1487 * re-calculating it. 1488 * 1489 * UDP Lite is a different protocol number and will 1490 * likely end up being hashed as a 2-tuple until 1491 * RSS / NICs grow UDP Lite protocol awareness. 1492 */ 1493 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1494 pr, &hash_val, &hash_type) == 0) { 1495 m->m_pkthdr.flowid = hash_val; 1496 M_HASHTYPE_SET(m, hash_type); 1497 } 1498 #endif 1499 } 1500 1501 #ifdef RSS 1502 /* 1503 * Don't override with the inp cached flowid value. 1504 * 1505 * Depending upon the kind of send being done, the inp 1506 * flowid/flowtype values may actually not be appropriate 1507 * for this particular socket send. 1508 * 1509 * We should either leave the flowid at zero (which is what is 1510 * currently done) or set it to some software generated 1511 * hash value based on the packet contents. 1512 */ 1513 ipflags |= IP_NODEFAULTFLOWID; 1514 #endif /* RSS */ 1515 1516 if (unlock_udbinfo == UH_WLOCKED) 1517 INP_HASH_WUNLOCK(pcbinfo); 1518 else if (unlock_udbinfo == UH_RLOCKED) 1519 INP_HASH_RUNLOCK(pcbinfo); 1520 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1521 error = ip_output(m, inp->inp_options, 1522 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1523 inp->inp_moptions, inp); 1524 if (unlock_inp == UH_WLOCKED) 1525 INP_WUNLOCK(inp); 1526 else 1527 INP_RUNLOCK(inp); 1528 return (error); 1529 1530 release: 1531 if (unlock_udbinfo == UH_WLOCKED) { 1532 KASSERT(unlock_inp == UH_WLOCKED, 1533 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1534 INP_HASH_WUNLOCK(pcbinfo); 1535 INP_WUNLOCK(inp); 1536 } else if (unlock_udbinfo == UH_RLOCKED) { 1537 KASSERT(unlock_inp == UH_RLOCKED, 1538 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1539 INP_HASH_RUNLOCK(pcbinfo); 1540 INP_RUNLOCK(inp); 1541 } else if (unlock_inp == UH_WLOCKED) 1542 INP_WUNLOCK(inp); 1543 else 1544 INP_RUNLOCK(inp); 1545 m_freem(m); 1546 return (error); 1547 } 1548 1549 static void 1550 udp_abort(struct socket *so) 1551 { 1552 struct inpcb *inp; 1553 struct inpcbinfo *pcbinfo; 1554 1555 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1556 inp = sotoinpcb(so); 1557 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1558 INP_WLOCK(inp); 1559 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1560 INP_HASH_WLOCK(pcbinfo); 1561 in_pcbdisconnect(inp); 1562 inp->inp_laddr.s_addr = INADDR_ANY; 1563 INP_HASH_WUNLOCK(pcbinfo); 1564 soisdisconnected(so); 1565 } 1566 INP_WUNLOCK(inp); 1567 } 1568 1569 static int 1570 udp_attach(struct socket *so, int proto, struct thread *td) 1571 { 1572 struct inpcb *inp; 1573 struct inpcbinfo *pcbinfo; 1574 int error; 1575 1576 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1577 inp = sotoinpcb(so); 1578 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1579 error = soreserve(so, udp_sendspace, udp_recvspace); 1580 if (error) 1581 return (error); 1582 INP_INFO_WLOCK(pcbinfo); 1583 error = in_pcballoc(so, pcbinfo); 1584 if (error) { 1585 INP_INFO_WUNLOCK(pcbinfo); 1586 return (error); 1587 } 1588 1589 inp = sotoinpcb(so); 1590 inp->inp_vflag |= INP_IPV4; 1591 inp->inp_ip_ttl = V_ip_defttl; 1592 1593 error = udp_newudpcb(inp); 1594 if (error) { 1595 in_pcbdetach(inp); 1596 in_pcbfree(inp); 1597 INP_INFO_WUNLOCK(pcbinfo); 1598 return (error); 1599 } 1600 1601 INP_WUNLOCK(inp); 1602 INP_INFO_WUNLOCK(pcbinfo); 1603 return (0); 1604 } 1605 #endif /* INET */ 1606 1607 int 1608 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1609 { 1610 struct inpcb *inp; 1611 struct udpcb *up; 1612 1613 KASSERT(so->so_type == SOCK_DGRAM, 1614 ("udp_set_kernel_tunneling: !dgram")); 1615 inp = sotoinpcb(so); 1616 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1617 INP_WLOCK(inp); 1618 up = intoudpcb(inp); 1619 if ((up->u_tun_func != NULL) || 1620 (up->u_icmp_func != NULL)) { 1621 INP_WUNLOCK(inp); 1622 return (EBUSY); 1623 } 1624 up->u_tun_func = f; 1625 up->u_icmp_func = i; 1626 up->u_tun_ctx = ctx; 1627 INP_WUNLOCK(inp); 1628 return (0); 1629 } 1630 1631 #ifdef INET 1632 static int 1633 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1634 { 1635 struct inpcb *inp; 1636 struct inpcbinfo *pcbinfo; 1637 int error; 1638 1639 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1640 inp = sotoinpcb(so); 1641 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1642 INP_WLOCK(inp); 1643 INP_HASH_WLOCK(pcbinfo); 1644 error = in_pcbbind(inp, nam, td->td_ucred); 1645 INP_HASH_WUNLOCK(pcbinfo); 1646 INP_WUNLOCK(inp); 1647 return (error); 1648 } 1649 1650 static void 1651 udp_close(struct socket *so) 1652 { 1653 struct inpcb *inp; 1654 struct inpcbinfo *pcbinfo; 1655 1656 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1657 inp = sotoinpcb(so); 1658 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1659 INP_WLOCK(inp); 1660 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1661 INP_HASH_WLOCK(pcbinfo); 1662 in_pcbdisconnect(inp); 1663 inp->inp_laddr.s_addr = INADDR_ANY; 1664 INP_HASH_WUNLOCK(pcbinfo); 1665 soisdisconnected(so); 1666 } 1667 INP_WUNLOCK(inp); 1668 } 1669 1670 static int 1671 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1672 { 1673 struct inpcb *inp; 1674 struct inpcbinfo *pcbinfo; 1675 struct sockaddr_in *sin; 1676 int error; 1677 1678 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1679 inp = sotoinpcb(so); 1680 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1681 INP_WLOCK(inp); 1682 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1683 INP_WUNLOCK(inp); 1684 return (EISCONN); 1685 } 1686 sin = (struct sockaddr_in *)nam; 1687 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1688 if (error != 0) { 1689 INP_WUNLOCK(inp); 1690 return (error); 1691 } 1692 INP_HASH_WLOCK(pcbinfo); 1693 error = in_pcbconnect(inp, nam, td->td_ucred); 1694 INP_HASH_WUNLOCK(pcbinfo); 1695 if (error == 0) 1696 soisconnected(so); 1697 INP_WUNLOCK(inp); 1698 return (error); 1699 } 1700 1701 static void 1702 udp_detach(struct socket *so) 1703 { 1704 struct inpcb *inp; 1705 struct inpcbinfo *pcbinfo; 1706 struct udpcb *up; 1707 1708 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1709 inp = sotoinpcb(so); 1710 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1711 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1712 ("udp_detach: not disconnected")); 1713 INP_INFO_WLOCK(pcbinfo); 1714 INP_WLOCK(inp); 1715 up = intoudpcb(inp); 1716 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1717 inp->inp_ppcb = NULL; 1718 in_pcbdetach(inp); 1719 in_pcbfree(inp); 1720 INP_INFO_WUNLOCK(pcbinfo); 1721 udp_discardcb(up); 1722 } 1723 1724 static int 1725 udp_disconnect(struct socket *so) 1726 { 1727 struct inpcb *inp; 1728 struct inpcbinfo *pcbinfo; 1729 1730 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1731 inp = sotoinpcb(so); 1732 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1733 INP_WLOCK(inp); 1734 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1735 INP_WUNLOCK(inp); 1736 return (ENOTCONN); 1737 } 1738 INP_HASH_WLOCK(pcbinfo); 1739 in_pcbdisconnect(inp); 1740 inp->inp_laddr.s_addr = INADDR_ANY; 1741 INP_HASH_WUNLOCK(pcbinfo); 1742 SOCK_LOCK(so); 1743 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1744 SOCK_UNLOCK(so); 1745 INP_WUNLOCK(inp); 1746 return (0); 1747 } 1748 1749 static int 1750 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1751 struct mbuf *control, struct thread *td) 1752 { 1753 struct inpcb *inp; 1754 1755 inp = sotoinpcb(so); 1756 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1757 return (udp_output(inp, m, addr, control, td)); 1758 } 1759 #endif /* INET */ 1760 1761 int 1762 udp_shutdown(struct socket *so) 1763 { 1764 struct inpcb *inp; 1765 1766 inp = sotoinpcb(so); 1767 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1768 INP_WLOCK(inp); 1769 socantsendmore(so); 1770 INP_WUNLOCK(inp); 1771 return (0); 1772 } 1773 1774 #ifdef INET 1775 struct pr_usrreqs udp_usrreqs = { 1776 .pru_abort = udp_abort, 1777 .pru_attach = udp_attach, 1778 .pru_bind = udp_bind, 1779 .pru_connect = udp_connect, 1780 .pru_control = in_control, 1781 .pru_detach = udp_detach, 1782 .pru_disconnect = udp_disconnect, 1783 .pru_peeraddr = in_getpeeraddr, 1784 .pru_send = udp_send, 1785 .pru_soreceive = soreceive_dgram, 1786 .pru_sosend = sosend_dgram, 1787 .pru_shutdown = udp_shutdown, 1788 .pru_sockaddr = in_getsockaddr, 1789 .pru_sosetlabel = in_pcbsosetlabel, 1790 .pru_close = udp_close, 1791 }; 1792 #endif /* INET */ 1793