1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. 6 * Copyright (c) 2008 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * Copyright (c) 2014 Kevin Lo 9 * All rights reserved. 10 * 11 * Portions of this software were developed by Robert N. M. Watson under 12 * contract to Juniper Networks, Inc. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 */ 38 39 #include <sys/cdefs.h> 40 #include "opt_inet.h" 41 #include "opt_inet6.h" 42 #include "opt_ipsec.h" 43 #include "opt_route.h" 44 #include "opt_rss.h" 45 46 #include <sys/param.h> 47 #include <sys/domain.h> 48 #include <sys/eventhandler.h> 49 #include <sys/jail.h> 50 #include <sys/kernel.h> 51 #include <sys/lock.h> 52 #include <sys/malloc.h> 53 #include <sys/mbuf.h> 54 #include <sys/priv.h> 55 #include <sys/proc.h> 56 #include <sys/protosw.h> 57 #include <sys/sdt.h> 58 #include <sys/signalvar.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/sx.h> 62 #include <sys/sysctl.h> 63 #include <sys/syslog.h> 64 #include <sys/systm.h> 65 66 #include <vm/uma.h> 67 68 #include <net/if.h> 69 #include <net/if_var.h> 70 #include <net/route.h> 71 #include <net/route/nhop.h> 72 #include <net/rss_config.h> 73 74 #include <netinet/in.h> 75 #include <netinet/in_kdtrace.h> 76 #include <netinet/in_fib.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 #include <netipsec/ipsec_support.h> 97 98 #include <machine/in_cksum.h> 99 100 #include <security/mac/mac_framework.h> 101 102 /* 103 * UDP and UDP-Lite protocols implementation. 104 * Per RFC 768, August, 1980. 105 * Per RFC 3828, July, 2004. 106 */ 107 108 VNET_DEFINE(int, udp_bind_all_fibs) = 1; 109 SYSCTL_INT(_net_inet_udp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN, 110 &VNET_NAME(udp_bind_all_fibs), 0, 111 "Bound sockets receive traffic from all FIBs"); 112 113 /* 114 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 115 * removes the only data integrity mechanism for packets and malformed 116 * packets that would otherwise be discarded due to bad checksums, and may 117 * cause problems (especially for NFS data blocks). 118 */ 119 VNET_DEFINE(int, udp_cksum) = 1; 120 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, 121 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 122 123 VNET_DEFINE(int, udp_log_in_vain) = 0; 124 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 125 &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets"); 126 127 VNET_DEFINE(int, udp_blackhole) = 0; 128 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 129 &VNET_NAME(udp_blackhole), 0, 130 "Do not send port unreachables for refused connects"); 131 VNET_DEFINE(bool, udp_blackhole_local) = false; 132 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 133 CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false, 134 "Enforce net.inet.udp.blackhole for locally originated packets"); 135 136 u_long udp_sendspace = 9216; /* really max datagram size */ 137 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 138 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 139 140 u_long udp_recvspace = 40 * (1024 + 141 #ifdef INET6 142 sizeof(struct sockaddr_in6) 143 #else 144 sizeof(struct sockaddr_in) 145 #endif 146 ); /* 40 1K datagrams */ 147 148 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 149 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 150 151 VNET_DEFINE(struct inpcbinfo, udbinfo); 152 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 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 #endif 169 170 INPCBSTORAGE_DEFINE(udpcbstor, udpcb, "udpinp", "udp_inpcb", "udp", "udphash"); 171 INPCBSTORAGE_DEFINE(udplitecbstor, udpcb, "udpliteinp", "udplite_inpcb", 172 "udplite", "udplitehash"); 173 174 static void 175 udp_vnet_init(void *arg __unused) 176 { 177 178 /* 179 * For now default to 2-tuple UDP hashing - until the fragment 180 * reassembly code can also update the flowid. 181 * 182 * Once we can calculate the flowid that way and re-establish 183 * a 4-tuple, flip this to 4-tuple. 184 */ 185 in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE); 186 /* Additional pcbinfo for UDP-Lite */ 187 in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE, 188 UDBHASHSIZE); 189 } 190 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 191 udp_vnet_init, NULL); 192 193 /* 194 * Kernel module interface for updating udpstat. The argument is an index 195 * into udpstat treated as an array of u_long. While this encodes the 196 * general layout of udpstat into the caller, it doesn't encode its location, 197 * so that future changes to add, for example, per-CPU stats support won't 198 * cause binary compatibility problems for kernel modules. 199 */ 200 void 201 kmod_udpstat_inc(int statnum) 202 { 203 204 counter_u64_add(VNET(udpstat)[statnum], 1); 205 } 206 207 #ifdef VIMAGE 208 static void 209 udp_destroy(void *unused __unused) 210 { 211 212 in_pcbinfo_destroy(&V_udbinfo); 213 in_pcbinfo_destroy(&V_ulitecbinfo); 214 } 215 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 216 #endif 217 218 #ifdef INET 219 /* 220 * Subroutine of udp_input(), which appends the provided mbuf chain to the 221 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 222 * contains the source address. If the socket ends up being an IPv6 socket, 223 * udp_append() will convert to a sockaddr_in6 before passing the address 224 * into the socket code. 225 * 226 * In the normal case udp_append() will return 'false', indicating that you 227 * must unlock the inpcb. However if a tunneling protocol is in place we 228 * increment the inpcb refcnt and unlock the inpcb, on return from the tunneling 229 * protocol we then decrement the reference count. If in_pcbrele_rlocked() 230 * returns 'true', indicating the inpcb is gone, we return that to the caller 231 * to tell them *not* to unlock the inpcb. In the case of multicast this will 232 * cause the distribution to stop (though most tunneling protocols known 233 * currently do *not* use multicast). 234 * 235 * The mbuf is always consumed. 236 */ 237 static bool 238 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 239 struct sockaddr_in *udp_in) 240 { 241 struct sockaddr *append_sa; 242 struct socket *so; 243 struct mbuf *tmpopts, *opts = NULL; 244 #ifdef INET6 245 struct sockaddr_in6 udp_in6; 246 #endif 247 struct udpcb *up; 248 249 INP_LOCK_ASSERT(inp); 250 251 /* 252 * Engage the tunneling protocol. 253 */ 254 up = intoudpcb(inp); 255 if (up->u_tun_func != NULL) { 256 bool filtered; 257 258 in_pcbref(inp); 259 INP_RUNLOCK(inp); 260 filtered = (*up->u_tun_func)(n, off, inp, 261 (struct sockaddr *)&udp_in[0], up->u_tun_ctx); 262 INP_RLOCK(inp); 263 if (in_pcbrele_rlocked(inp)) { 264 if (!filtered) 265 m_freem(n); 266 return (true); 267 } 268 if (filtered) 269 return (false); 270 } 271 272 off += sizeof(struct udphdr); 273 274 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 275 /* Check AH/ESP integrity. */ 276 if (IPSEC_ENABLED(ipv4) && 277 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 278 m_freem(n); 279 return (false); 280 } 281 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 282 if (IPSEC_ENABLED(ipv4) && 283 UDPENCAP_INPUT(ipv4, n, off, AF_INET) != 0) 284 return (false); 285 } 286 #endif /* IPSEC */ 287 #ifdef MAC 288 if (mac_inpcb_check_deliver(inp, n) != 0) { 289 m_freem(n); 290 return (false); 291 } 292 #endif /* MAC */ 293 if (inp->inp_flags & INP_CONTROLOPTS || 294 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 295 #ifdef INET6 296 if (inp->inp_vflag & INP_IPV6) 297 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 298 else 299 #endif /* INET6 */ 300 ip_savecontrol(inp, &opts, ip, n); 301 } 302 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 303 tmpopts = sbcreatecontrol(&udp_in[1], 304 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP, 305 M_NOWAIT); 306 if (tmpopts) { 307 if (opts) { 308 tmpopts->m_next = opts; 309 opts = tmpopts; 310 } else 311 opts = tmpopts; 312 } 313 } 314 #ifdef INET6 315 if (inp->inp_vflag & INP_IPV6) { 316 bzero(&udp_in6, sizeof(udp_in6)); 317 udp_in6.sin6_len = sizeof(udp_in6); 318 udp_in6.sin6_family = AF_INET6; 319 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 320 append_sa = (struct sockaddr *)&udp_in6; 321 } else 322 #endif /* INET6 */ 323 append_sa = (struct sockaddr *)&udp_in[0]; 324 m_adj(n, off); 325 326 so = inp->inp_socket; 327 SOCKBUF_LOCK(&so->so_rcv); 328 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 329 soroverflow_locked(so); 330 m_freem(n); 331 if (opts) 332 m_freem(opts); 333 UDPSTAT_INC(udps_fullsock); 334 } else 335 sorwakeup_locked(so); 336 return (false); 337 } 338 339 static bool 340 udp_multi_match(const struct inpcb *inp, void *v) 341 { 342 struct ip *ip = v; 343 struct udphdr *uh = (struct udphdr *)(ip + 1); 344 345 if (inp->inp_lport != uh->uh_dport) 346 return (false); 347 #ifdef INET6 348 if ((inp->inp_vflag & INP_IPV4) == 0) 349 return (false); 350 #endif 351 if (inp->inp_laddr.s_addr != INADDR_ANY && 352 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 353 return (false); 354 if (inp->inp_faddr.s_addr != INADDR_ANY && 355 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 356 return (false); 357 if (inp->inp_fport != 0 && 358 inp->inp_fport != uh->uh_sport) 359 return (false); 360 361 return (true); 362 } 363 364 static int 365 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in) 366 { 367 struct ip *ip = mtod(m, struct ip *); 368 struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto), 369 INPLOOKUP_RLOCKPCB, udp_multi_match, ip); 370 #ifdef KDTRACE_HOOKS 371 struct udphdr *uh = (struct udphdr *)(ip + 1); 372 #endif 373 struct inpcb *inp; 374 struct mbuf *n; 375 int appends = 0, fib; 376 377 MPASS(ip->ip_hl == sizeof(struct ip) >> 2); 378 379 fib = M_GETFIB(m); 380 381 while ((inp = inp_next(&inpi)) != NULL) { 382 /* 383 * XXXRW: Because we weren't holding either the inpcb 384 * or the hash lock when we checked for a match 385 * before, we should probably recheck now that the 386 * inpcb lock is held. 387 */ 388 389 if (V_udp_bind_all_fibs == 0 && fib != inp->inp_inc.inc_fibnum) 390 /* 391 * Sockets bound to a specific FIB can only receive 392 * packets from that FIB. 393 */ 394 continue; 395 396 /* 397 * Handle socket delivery policy for any-source 398 * and source-specific multicast. [RFC3678] 399 */ 400 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 401 struct ip_moptions *imo; 402 struct sockaddr_in group; 403 int blocked; 404 405 imo = inp->inp_moptions; 406 if (imo == NULL) 407 continue; 408 bzero(&group, sizeof(struct sockaddr_in)); 409 group.sin_len = sizeof(struct sockaddr_in); 410 group.sin_family = AF_INET; 411 group.sin_addr = ip->ip_dst; 412 413 blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif, 414 (struct sockaddr *)&group, 415 (struct sockaddr *)&udp_in[0]); 416 if (blocked != MCAST_PASS) { 417 if (blocked == MCAST_NOTGMEMBER) 418 IPSTAT_INC(ips_notmember); 419 if (blocked == MCAST_NOTSMEMBER || 420 blocked == MCAST_MUTED) 421 UDPSTAT_INC(udps_filtermcast); 422 continue; 423 } 424 } 425 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) { 426 if (proto == IPPROTO_UDPLITE) 427 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 428 else 429 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 430 if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) { 431 break; 432 } else 433 appends++; 434 } 435 /* 436 * Don't look for additional matches if this one does 437 * not have either the SO_REUSEPORT or SO_REUSEADDR 438 * socket options set. This heuristic avoids 439 * searching through all pcbs in the common case of a 440 * non-shared port. It assumes that an application 441 * will never clear these options after setting them. 442 */ 443 if ((inp->inp_socket->so_options & 444 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) { 445 INP_RUNLOCK(inp); 446 break; 447 } 448 } 449 450 if (appends == 0) { 451 /* 452 * No matching pcb found; discard datagram. (No need 453 * to send an ICMP Port Unreachable for a broadcast 454 * or multicast datagram.) 455 */ 456 UDPSTAT_INC(udps_noport); 457 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 458 UDPSTAT_INC(udps_noportmcast); 459 else 460 UDPSTAT_INC(udps_noportbcast); 461 } 462 m_freem(m); 463 464 return (IPPROTO_DONE); 465 } 466 467 static int 468 udp_input(struct mbuf **mp, int *offp, int proto) 469 { 470 struct ip *ip; 471 struct udphdr *uh; 472 struct ifnet *ifp; 473 struct inpcb *inp; 474 uint16_t len, ip_len; 475 struct inpcbinfo *pcbinfo; 476 struct sockaddr_in udp_in[2]; 477 struct mbuf *m; 478 struct m_tag *fwd_tag; 479 int cscov_partial, iphlen, lookupflags; 480 481 m = *mp; 482 iphlen = *offp; 483 ifp = m->m_pkthdr.rcvif; 484 *mp = NULL; 485 UDPSTAT_INC(udps_ipackets); 486 487 /* 488 * Strip IP options, if any; should skip this, make available to 489 * user, and use on returned packets, but we don't yet have a way to 490 * check the checksum with options still present. 491 */ 492 if (iphlen > sizeof (struct ip)) { 493 ip_stripoptions(m); 494 iphlen = sizeof(struct ip); 495 } 496 497 /* 498 * Get IP and UDP header together in first mbuf. 499 */ 500 if (m->m_len < iphlen + sizeof(struct udphdr)) { 501 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 502 UDPSTAT_INC(udps_hdrops); 503 return (IPPROTO_DONE); 504 } 505 } 506 ip = mtod(m, struct ip *); 507 uh = (struct udphdr *)((caddr_t)ip + iphlen); 508 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 509 510 /* 511 * Destination port of 0 is illegal, based on RFC768. 512 */ 513 if (uh->uh_dport == 0) 514 goto badunlocked; 515 516 /* 517 * Construct sockaddr format source address. Stuff source address 518 * and datagram in user buffer. 519 */ 520 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 521 udp_in[0].sin_len = sizeof(struct sockaddr_in); 522 udp_in[0].sin_family = AF_INET; 523 udp_in[0].sin_port = uh->uh_sport; 524 udp_in[0].sin_addr = ip->ip_src; 525 udp_in[1].sin_len = sizeof(struct sockaddr_in); 526 udp_in[1].sin_family = AF_INET; 527 udp_in[1].sin_port = uh->uh_dport; 528 udp_in[1].sin_addr = ip->ip_dst; 529 530 /* 531 * Make mbuf data length reflect UDP length. If not enough data to 532 * reflect UDP length, drop. 533 */ 534 len = ntohs((u_short)uh->uh_ulen); 535 ip_len = ntohs(ip->ip_len) - iphlen; 536 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 537 /* Zero means checksum over the complete packet. */ 538 if (len == 0) 539 len = ip_len; 540 cscov_partial = 0; 541 } 542 if (ip_len != len) { 543 if (len > ip_len || len < sizeof(struct udphdr)) { 544 UDPSTAT_INC(udps_badlen); 545 goto badunlocked; 546 } 547 if (proto == IPPROTO_UDP) 548 m_adj(m, len - ip_len); 549 } 550 551 /* 552 * Checksum extended UDP header and data. 553 */ 554 if (uh->uh_sum) { 555 u_short uh_sum; 556 557 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 558 !cscov_partial) { 559 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 560 uh_sum = m->m_pkthdr.csum_data; 561 else 562 uh_sum = in_pseudo(ip->ip_src.s_addr, 563 ip->ip_dst.s_addr, htonl((u_short)len + 564 m->m_pkthdr.csum_data + proto)); 565 uh_sum ^= 0xffff; 566 } else if (m->m_pkthdr.csum_flags & CSUM_IP_UDP) { 567 /* 568 * Packet from local host (maybe from a VM). 569 * Checksum not required. 570 */ 571 uh_sum = 0; 572 } else { 573 char b[offsetof(struct ipovly, ih_src)]; 574 struct ipovly *ipov = (struct ipovly *)ip; 575 576 memcpy(b, ipov, sizeof(b)); 577 bzero(ipov, sizeof(ipov->ih_x1)); 578 ipov->ih_len = (proto == IPPROTO_UDP) ? 579 uh->uh_ulen : htons(ip_len); 580 uh_sum = in_cksum(m, len + sizeof (struct ip)); 581 memcpy(ipov, b, sizeof(b)); 582 } 583 if (uh_sum) { 584 UDPSTAT_INC(udps_badsum); 585 m_freem(m); 586 return (IPPROTO_DONE); 587 } 588 } else { 589 if (proto == IPPROTO_UDP) { 590 UDPSTAT_INC(udps_nosum); 591 } else { 592 /* UDPLite requires a checksum */ 593 /* XXX: What is the right UDPLite MIB counter here? */ 594 m_freem(m); 595 return (IPPROTO_DONE); 596 } 597 } 598 599 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 600 in_ifnet_broadcast(ip->ip_dst, ifp)) 601 return (udp_multi_input(m, proto, udp_in)); 602 603 pcbinfo = udp_get_inpcbinfo(proto); 604 605 /* 606 * Locate pcb for datagram. 607 */ 608 lookupflags = INPLOOKUP_RLOCKPCB | 609 (V_udp_bind_all_fibs ? 0 : INPLOOKUP_FIB); 610 611 /* 612 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 613 */ 614 if ((m->m_flags & M_IP_NEXTHOP) && 615 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 616 struct sockaddr_in *next_hop; 617 618 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 619 620 /* 621 * Transparently forwarded. Pretend to be the destination. 622 * Already got one like this? 623 */ 624 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 625 ip->ip_dst, uh->uh_dport, lookupflags, ifp, m); 626 if (!inp) { 627 /* 628 * It's new. Try to find the ambushing socket. 629 * Because we've rewritten the destination address, 630 * any hardware-generated hash is ignored. 631 */ 632 inp = in_pcblookup(pcbinfo, ip->ip_src, 633 uh->uh_sport, next_hop->sin_addr, 634 next_hop->sin_port ? htons(next_hop->sin_port) : 635 uh->uh_dport, INPLOOKUP_WILDCARD | lookupflags, 636 ifp); 637 } 638 /* Remove the tag from the packet. We don't need it anymore. */ 639 m_tag_delete(m, fwd_tag); 640 m->m_flags &= ~M_IP_NEXTHOP; 641 } else 642 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 643 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 644 lookupflags, ifp, m); 645 if (inp == NULL) { 646 if (V_udp_log_in_vain) { 647 char src[INET_ADDRSTRLEN]; 648 char dst[INET_ADDRSTRLEN]; 649 650 log(LOG_INFO, 651 "Connection attempt to UDP %s:%d from %s:%d\n", 652 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 653 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 654 } 655 if (proto == IPPROTO_UDPLITE) 656 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); 657 else 658 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 659 UDPSTAT_INC(udps_noport); 660 if (m->m_flags & M_MCAST) { 661 UDPSTAT_INC(udps_noportmcast); 662 goto badunlocked; 663 } 664 if (m->m_flags & M_BCAST) { 665 UDPSTAT_INC(udps_noportbcast); 666 goto badunlocked; 667 } 668 if (V_udp_blackhole && (V_udp_blackhole_local || 669 !in_localip(ip->ip_src))) 670 goto badunlocked; 671 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 672 goto badunlocked; 673 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 674 return (IPPROTO_DONE); 675 } 676 677 /* 678 * Check the minimum TTL for socket. 679 */ 680 INP_RLOCK_ASSERT(inp); 681 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 682 if (proto == IPPROTO_UDPLITE) 683 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 684 else 685 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 686 INP_RUNLOCK(inp); 687 m_freem(m); 688 return (IPPROTO_DONE); 689 } 690 if (cscov_partial) { 691 struct udpcb *up; 692 693 up = intoudpcb(inp); 694 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 695 INP_RUNLOCK(inp); 696 m_freem(m); 697 return (IPPROTO_DONE); 698 } 699 } 700 701 if (proto == IPPROTO_UDPLITE) 702 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 703 else 704 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 705 if (!udp_append(inp, ip, m, iphlen, udp_in)) 706 INP_RUNLOCK(inp); 707 return (IPPROTO_DONE); 708 709 badunlocked: 710 m_freem(m); 711 return (IPPROTO_DONE); 712 } 713 #endif /* INET */ 714 715 /* 716 * Notify a udp user of an asynchronous error; just wake up so that they can 717 * collect error status. 718 */ 719 struct inpcb * 720 udp_notify(struct inpcb *inp, int errno) 721 { 722 723 INP_WLOCK_ASSERT(inp); 724 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 725 errno == EHOSTDOWN) && inp->inp_route.ro_nh) { 726 NH_FREE(inp->inp_route.ro_nh); 727 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 728 } 729 730 inp->inp_socket->so_error = errno; 731 sorwakeup(inp->inp_socket); 732 sowwakeup(inp->inp_socket); 733 return (inp); 734 } 735 736 #ifdef INET 737 static void 738 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo) 739 { 740 struct ip *ip = &icmp->icmp_ip; 741 struct udphdr *uh; 742 struct inpcb *inp; 743 744 if (icmp_errmap(icmp) == 0) 745 return; 746 747 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 748 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src, 749 uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL); 750 if (inp != NULL) { 751 INP_WLOCK_ASSERT(inp); 752 if (inp->inp_socket != NULL) 753 udp_notify(inp, icmp_errmap(icmp)); 754 INP_WUNLOCK(inp); 755 } else { 756 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, 757 ip->ip_src, uh->uh_sport, 758 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 759 if (inp != NULL) { 760 struct udpcb *up; 761 udp_tun_icmp_t *func; 762 763 up = intoudpcb(inp); 764 func = up->u_icmp_func; 765 INP_RUNLOCK(inp); 766 if (func != NULL) 767 func(icmp); 768 } 769 } 770 } 771 772 static void 773 udp_ctlinput(struct icmp *icmp) 774 { 775 776 return (udp_common_ctlinput(icmp, &V_udbinfo)); 777 } 778 779 static void 780 udplite_ctlinput(struct icmp *icmp) 781 { 782 783 return (udp_common_ctlinput(icmp, &V_ulitecbinfo)); 784 } 785 #endif /* INET */ 786 787 static int 788 udp_pcblist(SYSCTL_HANDLER_ARGS) 789 { 790 struct inpcbinfo *pcbinfo = udp_get_inpcbinfo(arg2); 791 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 792 INPLOOKUP_RLOCKPCB); 793 struct xinpgen xig; 794 struct inpcb *inp; 795 int error; 796 797 if (req->newptr != 0) 798 return (EPERM); 799 800 if (req->oldptr == 0) { 801 int n; 802 803 n = pcbinfo->ipi_count; 804 n += imax(n / 8, 10); 805 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 806 return (0); 807 } 808 809 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 810 return (error); 811 812 bzero(&xig, sizeof(xig)); 813 xig.xig_len = sizeof xig; 814 xig.xig_count = pcbinfo->ipi_count; 815 xig.xig_gen = pcbinfo->ipi_gencnt; 816 xig.xig_sogen = so_gencnt; 817 error = SYSCTL_OUT(req, &xig, sizeof xig); 818 if (error) 819 return (error); 820 821 while ((inp = inp_next(&inpi)) != NULL) { 822 if (inp->inp_gencnt <= xig.xig_gen && 823 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 824 struct xinpcb xi; 825 826 in_pcbtoxinpcb(inp, &xi); 827 error = SYSCTL_OUT(req, &xi, sizeof xi); 828 if (error) { 829 INP_RUNLOCK(inp); 830 break; 831 } 832 } 833 } 834 835 if (!error) { 836 /* 837 * Give the user an updated idea of our state. If the 838 * generation differs from what we told her before, she knows 839 * that something happened while we were processing this 840 * request, and it might be necessary to retry. 841 */ 842 xig.xig_gen = pcbinfo->ipi_gencnt; 843 xig.xig_sogen = so_gencnt; 844 xig.xig_count = pcbinfo->ipi_count; 845 error = SYSCTL_OUT(req, &xig, sizeof xig); 846 } 847 848 return (error); 849 } 850 851 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 852 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, IPPROTO_UDP, 853 udp_pcblist, "S,xinpcb", 854 "List of active UDP sockets"); 855 856 #ifdef INET 857 static int 858 udp_getcred(SYSCTL_HANDLER_ARGS) 859 { 860 struct xucred xuc; 861 struct sockaddr_in addrs[2]; 862 struct epoch_tracker et; 863 struct inpcb *inp; 864 int error; 865 866 if (req->newptr == NULL) 867 return (EINVAL); 868 error = priv_check(req->td, PRIV_NETINET_GETCRED); 869 if (error) 870 return (error); 871 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 872 if (error) 873 return (error); 874 NET_EPOCH_ENTER(et); 875 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 876 addrs[0].sin_addr, addrs[0].sin_port, 877 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 878 NET_EPOCH_EXIT(et); 879 if (inp != NULL) { 880 INP_RLOCK_ASSERT(inp); 881 if (inp->inp_socket == NULL) 882 error = ENOENT; 883 if (error == 0) 884 error = cr_canseeinpcb(req->td->td_ucred, inp); 885 if (error == 0) 886 cru2x(inp->inp_cred, &xuc); 887 INP_RUNLOCK(inp); 888 } else 889 error = ENOENT; 890 if (error == 0) 891 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 892 return (error); 893 } 894 895 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 896 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, 897 0, 0, udp_getcred, "S,xucred", 898 "Get the xucred of a UDP connection"); 899 #endif /* INET */ 900 901 int 902 udp_ctloutput(struct socket *so, struct sockopt *sopt) 903 { 904 struct inpcb *inp; 905 struct udpcb *up; 906 int isudplite, error, optval; 907 908 error = 0; 909 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 910 inp = sotoinpcb(so); 911 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 912 INP_WLOCK(inp); 913 if (sopt->sopt_level != so->so_proto->pr_protocol) { 914 #ifdef INET6 915 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 916 INP_WUNLOCK(inp); 917 error = ip6_ctloutput(so, sopt); 918 } 919 #endif 920 #if defined(INET) && defined(INET6) 921 else 922 #endif 923 #ifdef INET 924 { 925 INP_WUNLOCK(inp); 926 error = ip_ctloutput(so, sopt); 927 } 928 #endif 929 return (error); 930 } 931 932 switch (sopt->sopt_dir) { 933 case SOPT_SET: 934 switch (sopt->sopt_name) { 935 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 936 #if defined(INET) || defined(INET6) 937 case UDP_ENCAP: 938 #ifdef INET 939 if (INP_SOCKAF(so) == AF_INET) { 940 if (!IPSEC_ENABLED(ipv4)) { 941 INP_WUNLOCK(inp); 942 return (ENOPROTOOPT); 943 } 944 error = UDPENCAP_PCBCTL(ipv4, inp, sopt); 945 break; 946 } 947 #endif /* INET */ 948 #ifdef INET6 949 if (INP_SOCKAF(so) == AF_INET6) { 950 if (!IPSEC_ENABLED(ipv6)) { 951 INP_WUNLOCK(inp); 952 return (ENOPROTOOPT); 953 } 954 error = UDPENCAP_PCBCTL(ipv6, inp, sopt); 955 break; 956 } 957 #endif /* INET6 */ 958 INP_WUNLOCK(inp); 959 return (EINVAL); 960 #endif /* INET || INET6 */ 961 962 #endif /* IPSEC */ 963 case UDPLITE_SEND_CSCOV: 964 case UDPLITE_RECV_CSCOV: 965 if (!isudplite) { 966 INP_WUNLOCK(inp); 967 error = ENOPROTOOPT; 968 break; 969 } 970 INP_WUNLOCK(inp); 971 error = sooptcopyin(sopt, &optval, sizeof(optval), 972 sizeof(optval)); 973 if (error != 0) 974 break; 975 inp = sotoinpcb(so); 976 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 977 INP_WLOCK(inp); 978 up = intoudpcb(inp); 979 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 980 if ((optval != 0 && optval < 8) || (optval > 65535)) { 981 INP_WUNLOCK(inp); 982 error = EINVAL; 983 break; 984 } 985 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 986 up->u_txcslen = optval; 987 else 988 up->u_rxcslen = optval; 989 INP_WUNLOCK(inp); 990 break; 991 default: 992 INP_WUNLOCK(inp); 993 error = ENOPROTOOPT; 994 break; 995 } 996 break; 997 case SOPT_GET: 998 switch (sopt->sopt_name) { 999 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1000 #if defined(INET) || defined(INET6) 1001 case UDP_ENCAP: 1002 #ifdef INET 1003 if (INP_SOCKAF(so) == AF_INET) { 1004 if (!IPSEC_ENABLED(ipv4)) { 1005 INP_WUNLOCK(inp); 1006 return (ENOPROTOOPT); 1007 } 1008 error = UDPENCAP_PCBCTL(ipv4, inp, sopt); 1009 break; 1010 } 1011 #endif /* INET */ 1012 #ifdef INET6 1013 if (INP_SOCKAF(so) == AF_INET6) { 1014 if (!IPSEC_ENABLED(ipv6)) { 1015 INP_WUNLOCK(inp); 1016 return (ENOPROTOOPT); 1017 } 1018 error = UDPENCAP_PCBCTL(ipv6, inp, sopt); 1019 break; 1020 } 1021 #endif /* INET6 */ 1022 INP_WUNLOCK(inp); 1023 return (EINVAL); 1024 #endif /* INET || INET6 */ 1025 1026 #endif /* IPSEC */ 1027 case UDPLITE_SEND_CSCOV: 1028 case UDPLITE_RECV_CSCOV: 1029 if (!isudplite) { 1030 INP_WUNLOCK(inp); 1031 error = ENOPROTOOPT; 1032 break; 1033 } 1034 up = intoudpcb(inp); 1035 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1036 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1037 optval = up->u_txcslen; 1038 else 1039 optval = up->u_rxcslen; 1040 INP_WUNLOCK(inp); 1041 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1042 break; 1043 default: 1044 INP_WUNLOCK(inp); 1045 error = ENOPROTOOPT; 1046 break; 1047 } 1048 break; 1049 } 1050 return (error); 1051 } 1052 1053 #ifdef INET 1054 #ifdef INET6 1055 /* The logic here is derived from ip6_setpktopt(). See comments there. */ 1056 static int 1057 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src, 1058 struct inpcb *inp, int flags) 1059 { 1060 struct ifnet *ifp; 1061 struct in6_pktinfo *pktinfo; 1062 struct in_addr ia; 1063 1064 NET_EPOCH_ASSERT(); 1065 1066 if ((flags & PRUS_IPV6) == 0) 1067 return (0); 1068 1069 if (cm->cmsg_level != IPPROTO_IPV6) 1070 return (0); 1071 1072 if (cm->cmsg_type != IPV6_2292PKTINFO && 1073 cm->cmsg_type != IPV6_PKTINFO) 1074 return (0); 1075 1076 if (cm->cmsg_len != 1077 CMSG_LEN(sizeof(struct in6_pktinfo))) 1078 return (EINVAL); 1079 1080 pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm); 1081 if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) && 1082 !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) 1083 return (EINVAL); 1084 1085 /* Validate the interface index if specified. */ 1086 if (pktinfo->ipi6_ifindex) { 1087 ifp = ifnet_byindex(pktinfo->ipi6_ifindex); 1088 if (ifp == NULL) 1089 return (ENXIO); 1090 } else 1091 ifp = NULL; 1092 if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { 1093 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1094 if (!in_ifhasaddr(ifp, ia)) 1095 return (EADDRNOTAVAIL); 1096 } 1097 1098 bzero(src, sizeof(*src)); 1099 src->sin_family = AF_INET; 1100 src->sin_len = sizeof(*src); 1101 src->sin_port = inp->inp_lport; 1102 src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1103 1104 return (0); 1105 } 1106 #endif /* INET6 */ 1107 1108 int 1109 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1110 struct mbuf *control, struct thread *td) 1111 { 1112 struct inpcb *inp; 1113 struct udpiphdr *ui; 1114 int len, error = 0; 1115 struct in_addr faddr, laddr; 1116 struct cmsghdr *cm; 1117 struct inpcbinfo *pcbinfo; 1118 struct sockaddr_in *sin, src; 1119 struct epoch_tracker et; 1120 int cscov_partial = 0; 1121 int ipflags = 0; 1122 u_short fport, lport; 1123 u_char tos, vflagsav; 1124 uint8_t pr; 1125 uint16_t cscov = 0; 1126 uint32_t flowid = 0; 1127 uint8_t flowtype = M_HASHTYPE_NONE; 1128 bool use_cached_route; 1129 1130 inp = sotoinpcb(so); 1131 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1132 1133 if (addr != NULL) { 1134 if (addr->sa_family != AF_INET) 1135 error = EAFNOSUPPORT; 1136 else if (addr->sa_len != sizeof(struct sockaddr_in)) 1137 error = EINVAL; 1138 if (__predict_false(error != 0)) { 1139 m_freem(control); 1140 m_freem(m); 1141 return (error); 1142 } 1143 } 1144 1145 len = m->m_pkthdr.len; 1146 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1147 if (control) 1148 m_freem(control); 1149 m_freem(m); 1150 return (EMSGSIZE); 1151 } 1152 1153 src.sin_family = 0; 1154 sin = (struct sockaddr_in *)addr; 1155 1156 /* 1157 * udp_send() may need to bind the current inpcb. As such, we don't 1158 * know up front whether we will need the pcbinfo lock or not. Do any 1159 * work to decide what is needed up front before acquiring any locks. 1160 * 1161 * We will need network epoch in either case, to safely lookup into 1162 * pcb hash. 1163 */ 1164 use_cached_route = sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0); 1165 if (use_cached_route || (flags & PRUS_IPV6) != 0) 1166 INP_WLOCK(inp); 1167 else 1168 INP_RLOCK(inp); 1169 NET_EPOCH_ENTER(et); 1170 tos = inp->inp_ip_tos; 1171 if (control != NULL) { 1172 /* 1173 * XXX: Currently, we assume all the optional information is 1174 * stored in a single mbuf. 1175 */ 1176 if (control->m_next) { 1177 m_freem(control); 1178 error = EINVAL; 1179 goto release; 1180 } 1181 for (; control->m_len > 0; 1182 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1183 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1184 cm = mtod(control, struct cmsghdr *); 1185 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1186 || cm->cmsg_len > control->m_len) { 1187 error = EINVAL; 1188 break; 1189 } 1190 #ifdef INET6 1191 error = udp_v4mapped_pktinfo(cm, &src, inp, flags); 1192 if (error != 0) 1193 break; 1194 #endif 1195 if (cm->cmsg_level != IPPROTO_IP) 1196 continue; 1197 1198 switch (cm->cmsg_type) { 1199 case IP_SENDSRCADDR: 1200 if (cm->cmsg_len != 1201 CMSG_LEN(sizeof(struct in_addr))) { 1202 error = EINVAL; 1203 break; 1204 } 1205 bzero(&src, sizeof(src)); 1206 src.sin_family = AF_INET; 1207 src.sin_len = sizeof(src); 1208 src.sin_port = inp->inp_lport; 1209 src.sin_addr = 1210 *(struct in_addr *)CMSG_DATA(cm); 1211 break; 1212 1213 case IP_TOS: 1214 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1215 error = EINVAL; 1216 break; 1217 } 1218 tos = *(u_char *)CMSG_DATA(cm); 1219 break; 1220 1221 case IP_FLOWID: 1222 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1223 error = EINVAL; 1224 break; 1225 } 1226 flowid = *(uint32_t *) CMSG_DATA(cm); 1227 break; 1228 1229 case IP_FLOWTYPE: 1230 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1231 error = EINVAL; 1232 break; 1233 } 1234 flowtype = *(uint32_t *) CMSG_DATA(cm); 1235 break; 1236 1237 #ifdef RSS 1238 case IP_RSSBUCKETID: 1239 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1240 error = EINVAL; 1241 break; 1242 } 1243 /* This is just a placeholder for now */ 1244 break; 1245 #endif /* RSS */ 1246 default: 1247 error = ENOPROTOOPT; 1248 break; 1249 } 1250 if (error) 1251 break; 1252 } 1253 m_freem(control); 1254 control = NULL; 1255 } 1256 if (error) 1257 goto release; 1258 1259 pr = inp->inp_socket->so_proto->pr_protocol; 1260 pcbinfo = udp_get_inpcbinfo(pr); 1261 1262 /* 1263 * If the IP_SENDSRCADDR control message was specified, override the 1264 * source address for this datagram. Its use is invalidated if the 1265 * address thus specified is incomplete or clobbers other inpcbs. 1266 */ 1267 laddr = inp->inp_laddr; 1268 lport = inp->inp_lport; 1269 if (src.sin_family == AF_INET) { 1270 if ((lport == 0) || 1271 (laddr.s_addr == INADDR_ANY && 1272 src.sin_addr.s_addr == INADDR_ANY)) { 1273 error = EINVAL; 1274 goto release; 1275 } 1276 if ((flags & PRUS_IPV6) != 0) { 1277 vflagsav = inp->inp_vflag; 1278 inp->inp_vflag |= INP_IPV4; 1279 inp->inp_vflag &= ~INP_IPV6; 1280 } 1281 INP_HASH_WLOCK(pcbinfo); 1282 error = in_pcbbind_setup(inp, &src, &laddr.s_addr, &lport, 1283 V_udp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred); 1284 INP_HASH_WUNLOCK(pcbinfo); 1285 if ((flags & PRUS_IPV6) != 0) 1286 inp->inp_vflag = vflagsav; 1287 if (error) 1288 goto release; 1289 } 1290 1291 /* 1292 * If a UDP socket has been connected, then a local address/port will 1293 * have been selected and bound. 1294 * 1295 * If a UDP socket has not been connected to, then an explicit 1296 * destination address must be used, in which case a local 1297 * address/port may not have been selected and bound. 1298 */ 1299 if (sin != NULL) { 1300 INP_LOCK_ASSERT(inp); 1301 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1302 error = EISCONN; 1303 goto release; 1304 } 1305 1306 /* 1307 * Jail may rewrite the destination address, so let it do 1308 * that before we use it. 1309 */ 1310 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1311 if (error) 1312 goto release; 1313 /* 1314 * sendto(2) on unconnected UDP socket results in implicit 1315 * binding to INADDR_ANY and anonymous port. This has two 1316 * side effects: 1317 * 1) after first sendto(2) the socket will receive datagrams 1318 * destined to the selected port. 1319 * 2) subsequent sendto(2) calls will use the same source port. 1320 */ 1321 if (inp->inp_lport == 0) { 1322 struct sockaddr_in wild = { 1323 .sin_family = AF_INET, 1324 .sin_len = sizeof(struct sockaddr_in), 1325 }; 1326 1327 INP_HASH_WLOCK(pcbinfo); 1328 error = in_pcbbind(inp, &wild, V_udp_bind_all_fibs ? 1329 0 : INPBIND_FIB, td->td_ucred); 1330 INP_HASH_WUNLOCK(pcbinfo); 1331 if (error) 1332 goto release; 1333 lport = inp->inp_lport; 1334 laddr = inp->inp_laddr; 1335 } 1336 if (laddr.s_addr == INADDR_ANY) { 1337 error = in_pcbladdr(inp, &sin->sin_addr, &laddr, 1338 td->td_ucred); 1339 if (error) 1340 goto release; 1341 } 1342 faddr = sin->sin_addr; 1343 fport = sin->sin_port; 1344 } else { 1345 INP_LOCK_ASSERT(inp); 1346 faddr = inp->inp_faddr; 1347 fport = inp->inp_fport; 1348 if (faddr.s_addr == INADDR_ANY) { 1349 error = ENOTCONN; 1350 goto release; 1351 } 1352 } 1353 1354 /* 1355 * Calculate data length and get a mbuf for UDP, IP, and possible 1356 * link-layer headers. Immediate slide the data pointer back forward 1357 * since we won't use that space at this layer. 1358 */ 1359 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1360 if (m == NULL) { 1361 error = ENOBUFS; 1362 goto release; 1363 } 1364 m->m_data += max_linkhdr; 1365 m->m_len -= max_linkhdr; 1366 m->m_pkthdr.len -= max_linkhdr; 1367 1368 /* 1369 * Fill in mbuf with extended UDP header and addresses and length put 1370 * into network format. 1371 */ 1372 ui = mtod(m, struct udpiphdr *); 1373 /* 1374 * Filling only those fields of udpiphdr that participate in the 1375 * checksum calculation. The rest must be zeroed and will be filled 1376 * later. 1377 */ 1378 bzero(ui->ui_x1, sizeof(ui->ui_x1)); 1379 ui->ui_pr = pr; 1380 ui->ui_src = laddr; 1381 ui->ui_dst = faddr; 1382 ui->ui_sport = lport; 1383 ui->ui_dport = fport; 1384 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1385 if (pr == IPPROTO_UDPLITE) { 1386 struct udpcb *up; 1387 uint16_t plen; 1388 1389 up = intoudpcb(inp); 1390 cscov = up->u_txcslen; 1391 plen = (u_short)len + sizeof(struct udphdr); 1392 if (cscov >= plen) 1393 cscov = 0; 1394 ui->ui_len = htons(plen); 1395 ui->ui_ulen = htons(cscov); 1396 /* 1397 * For UDP-Lite, checksum coverage length of zero means 1398 * the entire UDPLite packet is covered by the checksum. 1399 */ 1400 cscov_partial = (cscov == 0) ? 0 : 1; 1401 } 1402 1403 if (inp->inp_socket->so_options & SO_DONTROUTE) 1404 ipflags |= IP_ROUTETOIF; 1405 if (inp->inp_socket->so_options & SO_BROADCAST) 1406 ipflags |= IP_ALLOWBROADCAST; 1407 if (inp->inp_flags & INP_ONESBCAST) 1408 ipflags |= IP_SENDONES; 1409 1410 #ifdef MAC 1411 mac_inpcb_create_mbuf(inp, m); 1412 #endif 1413 1414 /* 1415 * Set up checksum and output datagram. 1416 */ 1417 ui->ui_sum = 0; 1418 if (pr == IPPROTO_UDPLITE) { 1419 if (inp->inp_flags & INP_ONESBCAST) 1420 faddr.s_addr = INADDR_BROADCAST; 1421 if (cscov_partial) { 1422 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1423 ui->ui_sum = 0xffff; 1424 } else { 1425 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1426 ui->ui_sum = 0xffff; 1427 } 1428 } else if (V_udp_cksum) { 1429 if (inp->inp_flags & INP_ONESBCAST) 1430 faddr.s_addr = INADDR_BROADCAST; 1431 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1432 htons((u_short)len + sizeof(struct udphdr) + pr)); 1433 m->m_pkthdr.csum_flags = CSUM_UDP; 1434 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1435 } 1436 /* 1437 * After finishing the checksum computation, fill the remaining fields 1438 * of udpiphdr. 1439 */ 1440 ((struct ip *)ui)->ip_v = IPVERSION; 1441 ((struct ip *)ui)->ip_tos = tos; 1442 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1443 if (inp->inp_flags & INP_DONTFRAG) 1444 ((struct ip *)ui)->ip_off |= htons(IP_DF); 1445 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; 1446 UDPSTAT_INC(udps_opackets); 1447 1448 /* 1449 * Setup flowid / RSS information for outbound socket. 1450 * 1451 * Once the UDP code decides to set a flowid some other way, 1452 * this allows the flowid to be overridden by userland. 1453 */ 1454 if (flowtype != M_HASHTYPE_NONE) { 1455 m->m_pkthdr.flowid = flowid; 1456 M_HASHTYPE_SET(m, flowtype); 1457 } 1458 #if defined(ROUTE_MPATH) || defined(RSS) 1459 else if (CALC_FLOWID_OUTBOUND_SENDTO) { 1460 uint32_t hash_val, hash_type; 1461 1462 hash_val = fib4_calc_packet_hash(laddr, faddr, 1463 lport, fport, pr, &hash_type); 1464 m->m_pkthdr.flowid = hash_val; 1465 M_HASHTYPE_SET(m, hash_type); 1466 } 1467 1468 /* 1469 * Don't override with the inp cached flowid value. 1470 * 1471 * Depending upon the kind of send being done, the inp 1472 * flowid/flowtype values may actually not be appropriate 1473 * for this particular socket send. 1474 * 1475 * We should either leave the flowid at zero (which is what is 1476 * currently done) or set it to some software generated 1477 * hash value based on the packet contents. 1478 */ 1479 ipflags |= IP_NODEFAULTFLOWID; 1480 #endif /* RSS */ 1481 1482 if (pr == IPPROTO_UDPLITE) 1483 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1484 else 1485 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1486 error = ip_output(m, inp->inp_options, 1487 use_cached_route ? &inp->inp_route : NULL, ipflags, 1488 inp->inp_moptions, inp); 1489 INP_UNLOCK(inp); 1490 NET_EPOCH_EXIT(et); 1491 return (error); 1492 1493 release: 1494 INP_UNLOCK(inp); 1495 NET_EPOCH_EXIT(et); 1496 m_freem(m); 1497 return (error); 1498 } 1499 1500 void 1501 udp_abort(struct socket *so) 1502 { 1503 struct inpcb *inp; 1504 struct inpcbinfo *pcbinfo; 1505 1506 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1507 inp = sotoinpcb(so); 1508 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1509 INP_WLOCK(inp); 1510 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1511 INP_HASH_WLOCK(pcbinfo); 1512 in_pcbdisconnect(inp); 1513 INP_HASH_WUNLOCK(pcbinfo); 1514 soisdisconnected(so); 1515 } 1516 INP_WUNLOCK(inp); 1517 } 1518 1519 static int 1520 udp_attach(struct socket *so, int proto, struct thread *td) 1521 { 1522 static uint32_t udp_flowid; 1523 struct inpcbinfo *pcbinfo; 1524 struct inpcb *inp; 1525 struct udpcb *up; 1526 int error; 1527 1528 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1529 inp = sotoinpcb(so); 1530 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1531 error = soreserve(so, udp_sendspace, udp_recvspace); 1532 if (error) 1533 return (error); 1534 error = in_pcballoc(so, pcbinfo); 1535 if (error) 1536 return (error); 1537 1538 inp = sotoinpcb(so); 1539 inp->inp_ip_ttl = V_ip_defttl; 1540 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1541 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1542 up = intoudpcb(inp); 1543 bzero(&up->u_start_zero, u_zero_size); 1544 INP_WUNLOCK(inp); 1545 1546 return (0); 1547 } 1548 #endif /* INET */ 1549 1550 int 1551 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1552 { 1553 struct inpcb *inp; 1554 struct udpcb *up; 1555 1556 KASSERT(so->so_type == SOCK_DGRAM, 1557 ("udp_set_kernel_tunneling: !dgram")); 1558 inp = sotoinpcb(so); 1559 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1560 INP_WLOCK(inp); 1561 up = intoudpcb(inp); 1562 if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) || 1563 (up->u_icmp_func != NULL))) { 1564 INP_WUNLOCK(inp); 1565 return (EBUSY); 1566 } 1567 up->u_tun_func = f; 1568 up->u_icmp_func = i; 1569 up->u_tun_ctx = ctx; 1570 INP_WUNLOCK(inp); 1571 return (0); 1572 } 1573 1574 #ifdef INET 1575 static int 1576 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1577 { 1578 struct inpcb *inp; 1579 struct inpcbinfo *pcbinfo; 1580 struct sockaddr_in *sinp; 1581 int error; 1582 1583 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1584 inp = sotoinpcb(so); 1585 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1586 1587 sinp = (struct sockaddr_in *)nam; 1588 if (nam->sa_family != AF_INET) { 1589 /* 1590 * Preserve compatibility with old programs. 1591 */ 1592 if (nam->sa_family != AF_UNSPEC || 1593 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 1594 sinp->sin_addr.s_addr != INADDR_ANY) 1595 return (EAFNOSUPPORT); 1596 nam->sa_family = AF_INET; 1597 } 1598 if (nam->sa_len != sizeof(struct sockaddr_in)) 1599 return (EINVAL); 1600 1601 INP_WLOCK(inp); 1602 INP_HASH_WLOCK(pcbinfo); 1603 error = in_pcbbind(inp, sinp, V_udp_bind_all_fibs ? 0 : INPBIND_FIB, 1604 td->td_ucred); 1605 INP_HASH_WUNLOCK(pcbinfo); 1606 INP_WUNLOCK(inp); 1607 return (error); 1608 } 1609 1610 static void 1611 udp_close(struct socket *so) 1612 { 1613 struct inpcb *inp; 1614 struct inpcbinfo *pcbinfo; 1615 1616 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1617 inp = sotoinpcb(so); 1618 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1619 INP_WLOCK(inp); 1620 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1621 INP_HASH_WLOCK(pcbinfo); 1622 in_pcbdisconnect(inp); 1623 INP_HASH_WUNLOCK(pcbinfo); 1624 soisdisconnected(so); 1625 } 1626 INP_WUNLOCK(inp); 1627 } 1628 1629 static int 1630 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1631 { 1632 struct epoch_tracker et; 1633 struct inpcb *inp; 1634 struct inpcbinfo *pcbinfo; 1635 struct sockaddr_in *sin; 1636 int error; 1637 1638 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1639 inp = sotoinpcb(so); 1640 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1641 1642 sin = (struct sockaddr_in *)nam; 1643 if (sin->sin_family != AF_INET) 1644 return (EAFNOSUPPORT); 1645 if (sin->sin_len != sizeof(*sin)) 1646 return (EINVAL); 1647 1648 INP_WLOCK(inp); 1649 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1650 INP_WUNLOCK(inp); 1651 return (EISCONN); 1652 } 1653 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1654 if (error != 0) { 1655 INP_WUNLOCK(inp); 1656 return (error); 1657 } 1658 NET_EPOCH_ENTER(et); 1659 INP_HASH_WLOCK(pcbinfo); 1660 error = in_pcbconnect(inp, sin, td->td_ucred); 1661 INP_HASH_WUNLOCK(pcbinfo); 1662 NET_EPOCH_EXIT(et); 1663 if (error == 0) 1664 soisconnected(so); 1665 INP_WUNLOCK(inp); 1666 return (error); 1667 } 1668 1669 static void 1670 udp_detach(struct socket *so) 1671 { 1672 struct inpcb *inp; 1673 1674 inp = sotoinpcb(so); 1675 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1676 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1677 ("udp_detach: not disconnected")); 1678 INP_WLOCK(inp); 1679 in_pcbfree(inp); 1680 } 1681 1682 int 1683 udp_disconnect(struct socket *so) 1684 { 1685 struct inpcb *inp; 1686 struct inpcbinfo *pcbinfo; 1687 1688 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1689 inp = sotoinpcb(so); 1690 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1691 INP_WLOCK(inp); 1692 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1693 INP_WUNLOCK(inp); 1694 return (ENOTCONN); 1695 } 1696 INP_HASH_WLOCK(pcbinfo); 1697 in_pcbdisconnect(inp); 1698 INP_HASH_WUNLOCK(pcbinfo); 1699 SOCK_LOCK(so); 1700 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1701 SOCK_UNLOCK(so); 1702 INP_WUNLOCK(inp); 1703 return (0); 1704 } 1705 #endif /* INET */ 1706 1707 int 1708 udp_shutdown(struct socket *so, enum shutdown_how how) 1709 { 1710 int error; 1711 1712 SOCK_LOCK(so); 1713 if (!(so->so_state & SS_ISCONNECTED)) 1714 /* 1715 * POSIX mandates us to just return ENOTCONN when shutdown(2) is 1716 * invoked on a datagram sockets, however historically we would 1717 * actually tear socket down. This is known to be leveraged by 1718 * some applications to unblock process waiting in recv(2) by 1719 * other process that it shares that socket with. Try to meet 1720 * both backward-compatibility and POSIX requirements by forcing 1721 * ENOTCONN but still flushing buffers and performing wakeup(9). 1722 * 1723 * XXXGL: it remains unknown what applications expect this 1724 * behavior and is this isolated to unix/dgram or inet/dgram or 1725 * both. See: D10351, D3039. 1726 */ 1727 error = ENOTCONN; 1728 else 1729 error = 0; 1730 SOCK_UNLOCK(so); 1731 1732 switch (how) { 1733 case SHUT_RD: 1734 sorflush(so); 1735 break; 1736 case SHUT_RDWR: 1737 sorflush(so); 1738 /* FALLTHROUGH */ 1739 case SHUT_WR: 1740 socantsendmore(so); 1741 } 1742 1743 return (error); 1744 } 1745 1746 #ifdef INET 1747 #define UDP_PROTOSW \ 1748 .pr_type = SOCK_DGRAM, \ 1749 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH, \ 1750 .pr_ctloutput = udp_ctloutput, \ 1751 .pr_abort = udp_abort, \ 1752 .pr_attach = udp_attach, \ 1753 .pr_bind = udp_bind, \ 1754 .pr_connect = udp_connect, \ 1755 .pr_control = in_control, \ 1756 .pr_detach = udp_detach, \ 1757 .pr_disconnect = udp_disconnect, \ 1758 .pr_peeraddr = in_getpeeraddr, \ 1759 .pr_send = udp_send, \ 1760 .pr_soreceive = soreceive_dgram, \ 1761 .pr_sosend = sosend_dgram, \ 1762 .pr_shutdown = udp_shutdown, \ 1763 .pr_sockaddr = in_getsockaddr, \ 1764 .pr_sosetlabel = in_pcbsosetlabel, \ 1765 .pr_close = udp_close 1766 1767 struct protosw udp_protosw = { 1768 .pr_protocol = IPPROTO_UDP, 1769 UDP_PROTOSW 1770 }; 1771 1772 struct protosw udplite_protosw = { 1773 .pr_protocol = IPPROTO_UDPLITE, 1774 UDP_PROTOSW 1775 }; 1776 1777 static void 1778 udp_init(void *arg __unused) 1779 { 1780 1781 IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput); 1782 IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput); 1783 } 1784 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL); 1785 #endif /* INET */ 1786