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 "opt_inet.h" 40 #include "opt_inet6.h" 41 #include "opt_ipsec.h" 42 #include "opt_rss.h" 43 44 #include <sys/param.h> 45 #include <sys/domain.h> 46 #include <sys/eventhandler.h> 47 #include <sys/jail.h> 48 #include <sys/kernel.h> 49 #include <sys/lock.h> 50 #include <sys/malloc.h> 51 #include <sys/mbuf.h> 52 #include <sys/priv.h> 53 #include <sys/proc.h> 54 #include <sys/protosw.h> 55 #include <sys/sdt.h> 56 #include <sys/signalvar.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/sx.h> 60 #include <sys/sysctl.h> 61 #include <sys/syslog.h> 62 #include <sys/systm.h> 63 64 #include <vm/uma.h> 65 66 #include <net/if.h> 67 #include <net/if_var.h> 68 #include <net/route.h> 69 #include <net/route/nhop.h> 70 #include <net/rss_config.h> 71 72 #include <netinet/in.h> 73 #include <netinet/in_kdtrace.h> 74 #include <netinet/in_fib.h> 75 #include <netinet/in_pcb.h> 76 #include <netinet/in_systm.h> 77 #include <netinet/in_var.h> 78 #include <netinet/ip.h> 79 #ifdef INET6 80 #include <netinet/ip6.h> 81 #endif 82 #include <netinet/ip_icmp.h> 83 #include <netinet/icmp_var.h> 84 #include <netinet/ip_var.h> 85 #include <netinet/ip_options.h> 86 #ifdef INET6 87 #include <netinet6/ip6_var.h> 88 #endif 89 #include <netinet/udp.h> 90 #include <netinet/udp_var.h> 91 #include <netinet/udplite.h> 92 #include <netinet/in_rss.h> 93 94 #include <netipsec/ipsec_support.h> 95 96 #include <machine/in_cksum.h> 97 98 #include <security/mac/mac_framework.h> 99 100 /* 101 * UDP and UDP-Lite protocols implementation. 102 * Per RFC 768, August, 1980. 103 * Per RFC 3828, July, 2004. 104 */ 105 106 VNET_DEFINE(int, udp_bind_all_fibs) = 1; 107 SYSCTL_INT(_net_inet_udp, OID_AUTO, bind_all_fibs, CTLFLAG_VNET | CTLFLAG_RDTUN, 108 &VNET_NAME(udp_bind_all_fibs), 0, 109 "Bound sockets receive traffic from all FIBs"); 110 111 /* 112 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 113 * removes the only data integrity mechanism for packets and malformed 114 * packets that would otherwise be discarded due to bad checksums, and may 115 * cause problems (especially for NFS data blocks). 116 */ 117 VNET_DEFINE(int, udp_cksum) = 1; 118 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, 119 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 120 121 VNET_DEFINE(int, udp_log_in_vain) = 0; 122 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 123 &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets"); 124 125 VNET_DEFINE(int, udp_blackhole) = 0; 126 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 127 &VNET_NAME(udp_blackhole), 0, 128 "Do not send port unreachables for refused connects"); 129 VNET_DEFINE(bool, udp_blackhole_local) = false; 130 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 131 CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false, 132 "Enforce net.inet.udp.blackhole for locally originated packets"); 133 134 u_long udp_sendspace = 9216; /* really max datagram size */ 135 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 136 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 137 138 u_long udp_recvspace = 40 * (1024 + 139 #ifdef INET6 140 sizeof(struct sockaddr_in6) 141 #else 142 sizeof(struct sockaddr_in) 143 #endif 144 ); /* 40 1K datagrams */ 145 146 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 147 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 148 149 VNET_DEFINE(struct inpcbinfo, udbinfo); 150 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 151 152 #ifndef UDBHASHSIZE 153 #define UDBHASHSIZE 128 154 #endif 155 156 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 157 VNET_PCPUSTAT_SYSINIT(udpstat); 158 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 159 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 160 161 #ifdef VIMAGE 162 VNET_PCPUSTAT_SYSUNINIT(udpstat); 163 #endif /* VIMAGE */ 164 #ifdef INET 165 static void udp_detach(struct socket *so); 166 #endif 167 168 INPCBSTORAGE_DEFINE(udpcbstor, udpcb, "udpinp", "udp_inpcb", "udphash"); 169 INPCBSTORAGE_DEFINE(udplitecbstor, udpcb, "udpliteinp", "udplite_inpcb", 170 "udplitehash"); 171 172 static void 173 udp_vnet_init(void *arg __unused) 174 { 175 176 /* 177 * For now default to 2-tuple UDP hashing - until the fragment 178 * reassembly code can also update the flowid. 179 * 180 * Once we can calculate the flowid that way and re-establish 181 * a 4-tuple, flip this to 4-tuple. 182 */ 183 in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE); 184 /* Additional pcbinfo for UDP-Lite */ 185 in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE, 186 UDBHASHSIZE); 187 } 188 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 189 udp_vnet_init, NULL); 190 191 /* 192 * Kernel module interface for updating udpstat. The argument is an index 193 * into udpstat treated as an array of u_long. While this encodes the 194 * general layout of udpstat into the caller, it doesn't encode its location, 195 * so that future changes to add, for example, per-CPU stats support won't 196 * cause binary compatibility problems for kernel modules. 197 */ 198 void 199 kmod_udpstat_inc(int statnum) 200 { 201 202 counter_u64_add(VNET(udpstat)[statnum], 1); 203 } 204 205 #ifdef VIMAGE 206 static void 207 udp_destroy(void *unused __unused) 208 { 209 210 in_pcbinfo_destroy(&V_udbinfo); 211 in_pcbinfo_destroy(&V_ulitecbinfo); 212 } 213 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 214 #endif 215 216 #ifdef INET 217 /* 218 * Subroutine of udp_input(), which appends the provided mbuf chain to the 219 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 220 * contains the source address. If the socket ends up being an IPv6 socket, 221 * udp_append() will convert to a sockaddr_in6 before passing the address 222 * into the socket code. 223 * 224 * In the normal case udp_append() will return 'false', indicating that you 225 * must unlock the inpcb. However if a tunneling protocol is in place we 226 * increment the inpcb refcnt and unlock the inpcb, on return from the tunneling 227 * protocol we then decrement the reference count. If in_pcbrele_rlocked() 228 * returns 'true', indicating the inpcb is gone, we return that to the caller 229 * to tell them *not* to unlock the inpcb. In the case of multicast this will 230 * cause the distribution to stop (though most tunneling protocols known 231 * currently do *not* use multicast). 232 * 233 * The mbuf is always consumed. 234 */ 235 static bool 236 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 237 struct sockaddr_in *udp_in) 238 { 239 struct sockaddr *append_sa; 240 struct socket *so; 241 struct mbuf *tmpopts, *opts = NULL; 242 #ifdef INET6 243 struct sockaddr_in6 udp_in6; 244 #endif 245 struct udpcb *up; 246 247 INP_LOCK_ASSERT(inp); 248 249 /* 250 * Engage the tunneling protocol. 251 */ 252 up = intoudpcb(inp); 253 if (up->u_tun_func != NULL) { 254 bool filtered; 255 256 in_pcbref(inp); 257 INP_RUNLOCK(inp); 258 filtered = (*up->u_tun_func)(n, off, inp, 259 (struct sockaddr *)&udp_in[0], up->u_tun_ctx); 260 INP_RLOCK(inp); 261 if (in_pcbrele_rlocked(inp)) { 262 if (!filtered) 263 m_freem(n); 264 return (true); 265 } 266 if (filtered) 267 return (false); 268 } 269 270 off += sizeof(struct udphdr); 271 272 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 273 /* Check AH/ESP integrity. */ 274 if (IPSEC_ENABLED(ipv4) && 275 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 276 m_freem(n); 277 return (false); 278 } 279 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 280 if (IPSEC_ENABLED(ipv4) && 281 UDPENCAP_INPUT(ipv4, n, off, AF_INET) != 0) 282 return (false); 283 } 284 #endif /* IPSEC */ 285 #ifdef MAC 286 if (mac_inpcb_check_deliver(inp, n) != 0) { 287 m_freem(n); 288 return (false); 289 } 290 #endif /* MAC */ 291 if (inp->inp_flags & INP_CONTROLOPTS || 292 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 293 #ifdef INET6 294 if (inp->inp_vflag & INP_IPV6) 295 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 296 else 297 #endif /* INET6 */ 298 ip_savecontrol(inp, &opts, ip, n); 299 } 300 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 301 tmpopts = sbcreatecontrol(&udp_in[1], 302 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP, 303 M_NOWAIT); 304 if (tmpopts) { 305 if (opts) { 306 tmpopts->m_next = opts; 307 opts = tmpopts; 308 } else 309 opts = tmpopts; 310 } 311 } 312 #ifdef INET6 313 if (inp->inp_vflag & INP_IPV6) { 314 bzero(&udp_in6, sizeof(udp_in6)); 315 udp_in6.sin6_len = sizeof(udp_in6); 316 udp_in6.sin6_family = AF_INET6; 317 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 318 append_sa = (struct sockaddr *)&udp_in6; 319 } else 320 #endif /* INET6 */ 321 append_sa = (struct sockaddr *)&udp_in[0]; 322 m_adj(n, off); 323 324 so = inp->inp_socket; 325 SOCKBUF_LOCK(&so->so_rcv); 326 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 327 soroverflow_locked(so); 328 m_freem(n); 329 if (opts) 330 m_freem(opts); 331 UDPSTAT_INC(udps_fullsock); 332 } else 333 sorwakeup_locked(so); 334 return (false); 335 } 336 337 static bool 338 udp_multi_match(const struct inpcb *inp, void *v) 339 { 340 struct ip *ip = v; 341 struct udphdr *uh = (struct udphdr *)(ip + 1); 342 343 if (inp->inp_lport != uh->uh_dport) 344 return (false); 345 #ifdef INET6 346 if ((inp->inp_vflag & INP_IPV4) == 0) 347 return (false); 348 #endif 349 if (inp->inp_laddr.s_addr != INADDR_ANY && 350 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 351 return (false); 352 if (inp->inp_faddr.s_addr != INADDR_ANY && 353 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 354 return (false); 355 if (inp->inp_fport != 0 && 356 inp->inp_fport != uh->uh_sport) 357 return (false); 358 359 return (true); 360 } 361 362 static int 363 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in) 364 { 365 struct ip *ip = mtod(m, struct ip *); 366 struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto), 367 INPLOOKUP_RLOCKPCB, udp_multi_match, ip); 368 #ifdef KDTRACE_HOOKS 369 struct udphdr *uh = (struct udphdr *)(ip + 1); 370 #endif 371 struct inpcb *inp; 372 struct mbuf *n; 373 int appends = 0, fib; 374 375 MPASS(ip->ip_hl == sizeof(struct ip) >> 2); 376 377 fib = M_GETFIB(m); 378 379 while ((inp = inp_next(&inpi)) != NULL) { 380 /* 381 * XXXRW: Because we weren't holding either the inpcb 382 * or the hash lock when we checked for a match 383 * before, we should probably recheck now that the 384 * inpcb lock is held. 385 */ 386 387 if (V_udp_bind_all_fibs == 0 && fib != inp->inp_inc.inc_fibnum) 388 /* 389 * Sockets bound to a specific FIB can only receive 390 * packets from that FIB. 391 */ 392 continue; 393 394 /* 395 * Handle socket delivery policy for any-source 396 * and source-specific multicast. [RFC3678] 397 */ 398 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 399 struct ip_moptions *imo; 400 struct sockaddr_in group; 401 int blocked; 402 403 imo = inp->inp_moptions; 404 if (imo == NULL) 405 continue; 406 bzero(&group, sizeof(struct sockaddr_in)); 407 group.sin_len = sizeof(struct sockaddr_in); 408 group.sin_family = AF_INET; 409 group.sin_addr = ip->ip_dst; 410 411 blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif, 412 (struct sockaddr *)&group, 413 (struct sockaddr *)&udp_in[0]); 414 if (blocked != MCAST_PASS) { 415 if (blocked == MCAST_NOTGMEMBER) 416 IPSTAT_INC(ips_notmember); 417 if (blocked == MCAST_NOTSMEMBER || 418 blocked == MCAST_MUTED) 419 UDPSTAT_INC(udps_filtermcast); 420 continue; 421 } 422 } 423 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) { 424 if (proto == IPPROTO_UDPLITE) 425 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 426 else 427 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 428 if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) { 429 break; 430 } else 431 appends++; 432 } 433 /* 434 * Don't look for additional matches if this one does 435 * not have either the SO_REUSEPORT or SO_REUSEADDR 436 * socket options set. This heuristic avoids 437 * searching through all pcbs in the common case of a 438 * non-shared port. It assumes that an application 439 * will never clear these options after setting them. 440 */ 441 if ((inp->inp_socket->so_options & 442 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) { 443 INP_RUNLOCK(inp); 444 break; 445 } 446 } 447 448 if (appends == 0) { 449 /* 450 * No matching pcb found; discard datagram. (No need 451 * to send an ICMP Port Unreachable for a broadcast 452 * or multicast datagram.) 453 */ 454 UDPSTAT_INC(udps_noport); 455 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 456 UDPSTAT_INC(udps_noportmcast); 457 else 458 UDPSTAT_INC(udps_noportbcast); 459 } 460 m_freem(m); 461 462 return (IPPROTO_DONE); 463 } 464 465 static int 466 udp_input(struct mbuf **mp, int *offp, int proto) 467 { 468 struct ip *ip; 469 struct udphdr *uh; 470 struct ifnet *ifp; 471 struct inpcb *inp; 472 uint16_t len, ip_len; 473 struct inpcbinfo *pcbinfo; 474 struct sockaddr_in udp_in[2]; 475 struct mbuf *m; 476 struct m_tag *fwd_tag; 477 int cscov_partial, iphlen, lookupflags; 478 479 m = *mp; 480 iphlen = *offp; 481 ifp = m->m_pkthdr.rcvif; 482 *mp = NULL; 483 UDPSTAT_INC(udps_ipackets); 484 485 /* 486 * Strip IP options, if any; should skip this, make available to 487 * user, and use on returned packets, but we don't yet have a way to 488 * check the checksum with options still present. 489 */ 490 if (iphlen > sizeof (struct ip)) { 491 ip_stripoptions(m); 492 iphlen = sizeof(struct ip); 493 } 494 495 /* 496 * Get IP and UDP header together in first mbuf. 497 */ 498 if (m->m_len < iphlen + sizeof(struct udphdr)) { 499 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 500 UDPSTAT_INC(udps_hdrops); 501 return (IPPROTO_DONE); 502 } 503 } 504 ip = mtod(m, struct ip *); 505 uh = (struct udphdr *)((caddr_t)ip + iphlen); 506 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 507 508 /* 509 * Destination port of 0 is illegal, based on RFC768. 510 */ 511 if (uh->uh_dport == 0) 512 goto badunlocked; 513 514 /* 515 * Construct sockaddr format source address. Stuff source address 516 * and datagram in user buffer. 517 */ 518 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 519 udp_in[0].sin_len = sizeof(struct sockaddr_in); 520 udp_in[0].sin_family = AF_INET; 521 udp_in[0].sin_port = uh->uh_sport; 522 udp_in[0].sin_addr = ip->ip_src; 523 udp_in[1].sin_len = sizeof(struct sockaddr_in); 524 udp_in[1].sin_family = AF_INET; 525 udp_in[1].sin_port = uh->uh_dport; 526 udp_in[1].sin_addr = ip->ip_dst; 527 528 /* 529 * Make mbuf data length reflect UDP length. If not enough data to 530 * reflect UDP length, drop. 531 */ 532 len = ntohs((u_short)uh->uh_ulen); 533 ip_len = ntohs(ip->ip_len) - iphlen; 534 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 535 /* Zero means checksum over the complete packet. */ 536 if (len == 0) 537 len = ip_len; 538 cscov_partial = 0; 539 } 540 if (ip_len != len) { 541 if (len > ip_len || len < sizeof(struct udphdr)) { 542 UDPSTAT_INC(udps_badlen); 543 goto badunlocked; 544 } 545 if (proto == IPPROTO_UDP) 546 m_adj(m, len - ip_len); 547 } 548 549 /* 550 * Checksum extended UDP header and data. 551 */ 552 if (uh->uh_sum) { 553 u_short uh_sum; 554 555 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 556 !cscov_partial) { 557 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 558 uh_sum = m->m_pkthdr.csum_data; 559 else 560 uh_sum = in_pseudo(ip->ip_src.s_addr, 561 ip->ip_dst.s_addr, htonl((u_short)len + 562 m->m_pkthdr.csum_data + proto)); 563 uh_sum ^= 0xffff; 564 } else if (m->m_pkthdr.csum_flags & CSUM_IP_UDP) { 565 /* 566 * Packet from local host (maybe from a VM). 567 * Checksum not required. 568 */ 569 uh_sum = 0; 570 } else { 571 char b[offsetof(struct ipovly, ih_src)]; 572 struct ipovly *ipov = (struct ipovly *)ip; 573 574 memcpy(b, ipov, sizeof(b)); 575 bzero(ipov, sizeof(ipov->ih_x1)); 576 ipov->ih_len = (proto == IPPROTO_UDP) ? 577 uh->uh_ulen : htons(ip_len); 578 uh_sum = in_cksum(m, len + sizeof (struct ip)); 579 memcpy(ipov, b, sizeof(b)); 580 } 581 if (uh_sum) { 582 UDPSTAT_INC(udps_badsum); 583 m_freem(m); 584 return (IPPROTO_DONE); 585 } 586 } else { 587 if (proto == IPPROTO_UDP) { 588 UDPSTAT_INC(udps_nosum); 589 } else { 590 /* UDPLite requires a checksum */ 591 /* XXX: What is the right UDPLite MIB counter here? */ 592 m_freem(m); 593 return (IPPROTO_DONE); 594 } 595 } 596 597 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 598 in_ifnet_broadcast(ip->ip_dst, ifp)) 599 return (udp_multi_input(m, proto, udp_in)); 600 601 pcbinfo = udp_get_inpcbinfo(proto); 602 603 /* 604 * Locate pcb for datagram. 605 */ 606 lookupflags = INPLOOKUP_RLOCKPCB | 607 (V_udp_bind_all_fibs ? 0 : INPLOOKUP_FIB); 608 609 /* 610 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 611 */ 612 if ((m->m_flags & M_IP_NEXTHOP) && 613 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 614 struct sockaddr_in *next_hop; 615 616 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 617 618 /* 619 * Transparently forwarded. Pretend to be the destination. 620 * Already got one like this? 621 */ 622 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 623 ip->ip_dst, uh->uh_dport, lookupflags, ifp, m); 624 if (!inp) { 625 /* 626 * It's new. Try to find the ambushing socket. 627 * Because we've rewritten the destination address, 628 * any hardware-generated hash is ignored. 629 */ 630 inp = in_pcblookup(pcbinfo, ip->ip_src, 631 uh->uh_sport, next_hop->sin_addr, 632 next_hop->sin_port ? htons(next_hop->sin_port) : 633 uh->uh_dport, INPLOOKUP_WILDCARD | lookupflags, 634 ifp); 635 } 636 /* Remove the tag from the packet. We don't need it anymore. */ 637 m_tag_delete(m, fwd_tag); 638 m->m_flags &= ~M_IP_NEXTHOP; 639 } else 640 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 641 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 642 lookupflags, ifp, m); 643 if (inp == NULL) { 644 if (V_udp_log_in_vain) { 645 char src[INET_ADDRSTRLEN]; 646 char dst[INET_ADDRSTRLEN]; 647 648 log(LOG_INFO, 649 "Connection attempt to UDP %s:%d from %s:%d\n", 650 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 651 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 652 } 653 if (proto == IPPROTO_UDPLITE) 654 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); 655 else 656 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 657 UDPSTAT_INC(udps_noport); 658 if (m->m_flags & M_MCAST) { 659 UDPSTAT_INC(udps_noportmcast); 660 goto badunlocked; 661 } 662 if (m->m_flags & M_BCAST) { 663 UDPSTAT_INC(udps_noportbcast); 664 goto badunlocked; 665 } 666 if (V_udp_blackhole && (V_udp_blackhole_local || 667 !in_localip(ip->ip_src))) 668 goto badunlocked; 669 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 670 goto badunlocked; 671 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 672 return (IPPROTO_DONE); 673 } 674 675 /* 676 * Check the minimum TTL for socket. 677 */ 678 INP_RLOCK_ASSERT(inp); 679 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 680 if (proto == IPPROTO_UDPLITE) 681 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 682 else 683 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 684 INP_RUNLOCK(inp); 685 m_freem(m); 686 return (IPPROTO_DONE); 687 } 688 if (cscov_partial) { 689 struct udpcb *up; 690 691 up = intoudpcb(inp); 692 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 693 INP_RUNLOCK(inp); 694 m_freem(m); 695 return (IPPROTO_DONE); 696 } 697 } 698 699 if (proto == IPPROTO_UDPLITE) 700 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 701 else 702 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 703 if (!udp_append(inp, ip, m, iphlen, udp_in)) 704 INP_RUNLOCK(inp); 705 return (IPPROTO_DONE); 706 707 badunlocked: 708 m_freem(m); 709 return (IPPROTO_DONE); 710 } 711 #endif /* INET */ 712 713 /* 714 * Notify a udp user of an asynchronous error; just wake up so that they can 715 * collect error status. 716 */ 717 struct inpcb * 718 udp_notify(struct inpcb *inp, int errno) 719 { 720 721 INP_WLOCK_ASSERT(inp); 722 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 723 errno == EHOSTDOWN) && inp->inp_route.ro_nh) { 724 NH_FREE(inp->inp_route.ro_nh); 725 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 726 } 727 728 inp->inp_socket->so_error = errno; 729 sorwakeup(inp->inp_socket); 730 sowwakeup(inp->inp_socket); 731 return (inp); 732 } 733 734 #ifdef INET 735 static void 736 udp_common_ctlinput(struct icmp *icmp, struct inpcbinfo *pcbinfo) 737 { 738 struct ip *ip = &icmp->icmp_ip; 739 struct udphdr *uh; 740 struct inpcb *inp; 741 742 if (icmp_errmap(icmp) == 0) 743 return; 744 745 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 746 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, ip->ip_src, 747 uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL); 748 if (inp != NULL) { 749 INP_WLOCK_ASSERT(inp); 750 if (inp->inp_socket != NULL) 751 udp_notify(inp, icmp_errmap(icmp)); 752 INP_WUNLOCK(inp); 753 } else { 754 inp = in_pcblookup(pcbinfo, ip->ip_dst, uh->uh_dport, 755 ip->ip_src, uh->uh_sport, 756 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 757 if (inp != NULL) { 758 struct udpcb *up; 759 udp_tun_icmp_t *func; 760 761 up = intoudpcb(inp); 762 func = up->u_icmp_func; 763 INP_RUNLOCK(inp); 764 if (func != NULL) 765 func(icmp); 766 } 767 } 768 } 769 770 static void 771 udp_ctlinput(struct icmp *icmp) 772 { 773 774 return (udp_common_ctlinput(icmp, &V_udbinfo)); 775 } 776 777 static void 778 udplite_ctlinput(struct icmp *icmp) 779 { 780 781 return (udp_common_ctlinput(icmp, &V_ulitecbinfo)); 782 } 783 #endif /* INET */ 784 785 static int 786 udp_pcblist(SYSCTL_HANDLER_ARGS) 787 { 788 struct inpcbinfo *pcbinfo = udp_get_inpcbinfo(arg2); 789 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 790 INPLOOKUP_RLOCKPCB); 791 struct xinpgen xig; 792 struct inpcb *inp; 793 int error; 794 795 if (req->newptr != 0) 796 return (EPERM); 797 798 if (req->oldptr == 0) { 799 int n; 800 801 n = pcbinfo->ipi_count; 802 n += imax(n / 8, 10); 803 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 804 return (0); 805 } 806 807 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 808 return (error); 809 810 bzero(&xig, sizeof(xig)); 811 xig.xig_len = sizeof xig; 812 xig.xig_count = pcbinfo->ipi_count; 813 xig.xig_gen = pcbinfo->ipi_gencnt; 814 xig.xig_sogen = so_gencnt; 815 error = SYSCTL_OUT(req, &xig, sizeof xig); 816 if (error) 817 return (error); 818 819 while ((inp = inp_next(&inpi)) != NULL) { 820 if (inp->inp_gencnt <= xig.xig_gen && 821 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 822 struct xinpcb xi; 823 824 in_pcbtoxinpcb(inp, &xi); 825 error = SYSCTL_OUT(req, &xi, sizeof xi); 826 if (error) { 827 INP_RUNLOCK(inp); 828 break; 829 } 830 } 831 } 832 833 if (!error) { 834 /* 835 * Give the user an updated idea of our state. If the 836 * generation differs from what we told her before, she knows 837 * that something happened while we were processing this 838 * request, and it might be necessary to retry. 839 */ 840 xig.xig_gen = pcbinfo->ipi_gencnt; 841 xig.xig_sogen = so_gencnt; 842 xig.xig_count = pcbinfo->ipi_count; 843 error = SYSCTL_OUT(req, &xig, sizeof xig); 844 } 845 846 return (error); 847 } 848 849 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 850 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, IPPROTO_UDP, 851 udp_pcblist, "S,xinpcb", 852 "List of active UDP sockets"); 853 854 SYSCTL_PROC(_net_inet_udplite, OID_AUTO, pcblist, 855 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, IPPROTO_UDPLITE, 856 udp_pcblist, "S,xinpcb", 857 "List of active UDP-Lite sockets"); 858 859 #ifdef INET 860 static int 861 udp_getcred(SYSCTL_HANDLER_ARGS) 862 { 863 struct xucred xuc; 864 struct sockaddr_in addrs[2]; 865 struct epoch_tracker et; 866 struct inpcb *inp; 867 int error; 868 869 if (req->newptr == NULL) 870 return (EINVAL); 871 error = priv_check(req->td, PRIV_NETINET_GETCRED); 872 if (error) 873 return (error); 874 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 875 if (error) 876 return (error); 877 NET_EPOCH_ENTER(et); 878 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 879 addrs[0].sin_addr, addrs[0].sin_port, 880 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 881 NET_EPOCH_EXIT(et); 882 if (inp != NULL) { 883 INP_RLOCK_ASSERT(inp); 884 if (inp->inp_socket == NULL) 885 error = ENOENT; 886 if (error == 0) 887 error = cr_canseeinpcb(req->td->td_ucred, inp); 888 if (error == 0) 889 cru2x(inp->inp_cred, &xuc); 890 INP_RUNLOCK(inp); 891 } else 892 error = ENOENT; 893 if (error == 0) 894 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 895 return (error); 896 } 897 898 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 899 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, 900 0, 0, udp_getcred, "S,xucred", 901 "Get the xucred of a UDP connection"); 902 #endif /* INET */ 903 904 int 905 udp_ctloutput(struct socket *so, struct sockopt *sopt) 906 { 907 struct inpcb *inp; 908 struct udpcb *up; 909 int isudplite, error, optval; 910 911 error = 0; 912 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 913 inp = sotoinpcb(so); 914 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 915 INP_WLOCK(inp); 916 if (sopt->sopt_level != so->so_proto->pr_protocol) { 917 #ifdef INET6 918 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 919 INP_WUNLOCK(inp); 920 error = ip6_ctloutput(so, sopt); 921 } 922 #endif 923 #if defined(INET) && defined(INET6) 924 else 925 #endif 926 #ifdef INET 927 { 928 INP_WUNLOCK(inp); 929 error = ip_ctloutput(so, sopt); 930 } 931 #endif 932 return (error); 933 } 934 935 switch (sopt->sopt_dir) { 936 case SOPT_SET: 937 switch (sopt->sopt_name) { 938 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 939 #if defined(INET) || defined(INET6) 940 case UDP_ENCAP: 941 #ifdef INET 942 if (INP_SOCKAF(so) == AF_INET) { 943 if (!IPSEC_ENABLED(ipv4)) { 944 INP_WUNLOCK(inp); 945 return (ENOPROTOOPT); 946 } 947 error = UDPENCAP_PCBCTL(ipv4, inp, sopt); 948 break; 949 } 950 #endif /* INET */ 951 #ifdef INET6 952 if (INP_SOCKAF(so) == AF_INET6) { 953 if (!IPSEC_ENABLED(ipv6)) { 954 INP_WUNLOCK(inp); 955 return (ENOPROTOOPT); 956 } 957 error = UDPENCAP_PCBCTL(ipv6, inp, sopt); 958 break; 959 } 960 #endif /* INET6 */ 961 INP_WUNLOCK(inp); 962 return (EINVAL); 963 #endif /* INET || INET6 */ 964 965 #endif /* IPSEC */ 966 case UDPLITE_SEND_CSCOV: 967 case UDPLITE_RECV_CSCOV: 968 if (!isudplite) { 969 INP_WUNLOCK(inp); 970 error = ENOPROTOOPT; 971 break; 972 } 973 INP_WUNLOCK(inp); 974 error = sooptcopyin(sopt, &optval, sizeof(optval), 975 sizeof(optval)); 976 if (error != 0) 977 break; 978 inp = sotoinpcb(so); 979 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 980 INP_WLOCK(inp); 981 up = intoudpcb(inp); 982 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 983 if ((optval != 0 && optval < 8) || (optval > 65535)) { 984 INP_WUNLOCK(inp); 985 error = EINVAL; 986 break; 987 } 988 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 989 up->u_txcslen = optval; 990 else 991 up->u_rxcslen = optval; 992 INP_WUNLOCK(inp); 993 break; 994 default: 995 INP_WUNLOCK(inp); 996 error = ENOPROTOOPT; 997 break; 998 } 999 break; 1000 case SOPT_GET: 1001 switch (sopt->sopt_name) { 1002 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1003 #if defined(INET) || defined(INET6) 1004 case UDP_ENCAP: 1005 #ifdef INET 1006 if (INP_SOCKAF(so) == AF_INET) { 1007 if (!IPSEC_ENABLED(ipv4)) { 1008 INP_WUNLOCK(inp); 1009 return (ENOPROTOOPT); 1010 } 1011 error = UDPENCAP_PCBCTL(ipv4, inp, sopt); 1012 break; 1013 } 1014 #endif /* INET */ 1015 #ifdef INET6 1016 if (INP_SOCKAF(so) == AF_INET6) { 1017 if (!IPSEC_ENABLED(ipv6)) { 1018 INP_WUNLOCK(inp); 1019 return (ENOPROTOOPT); 1020 } 1021 error = UDPENCAP_PCBCTL(ipv6, inp, sopt); 1022 break; 1023 } 1024 #endif /* INET6 */ 1025 INP_WUNLOCK(inp); 1026 return (EINVAL); 1027 #endif /* INET || INET6 */ 1028 1029 #endif /* IPSEC */ 1030 case UDPLITE_SEND_CSCOV: 1031 case UDPLITE_RECV_CSCOV: 1032 if (!isudplite) { 1033 INP_WUNLOCK(inp); 1034 error = ENOPROTOOPT; 1035 break; 1036 } 1037 up = intoudpcb(inp); 1038 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1039 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1040 optval = up->u_txcslen; 1041 else 1042 optval = up->u_rxcslen; 1043 INP_WUNLOCK(inp); 1044 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1045 break; 1046 default: 1047 INP_WUNLOCK(inp); 1048 error = ENOPROTOOPT; 1049 break; 1050 } 1051 break; 1052 } 1053 return (error); 1054 } 1055 1056 #ifdef INET 1057 #ifdef INET6 1058 /* The logic here is derived from ip6_setpktopt(). See comments there. */ 1059 static int 1060 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src, 1061 struct inpcb *inp, int flags) 1062 { 1063 struct ifnet *ifp; 1064 struct in6_pktinfo *pktinfo; 1065 struct in_addr ia; 1066 1067 NET_EPOCH_ASSERT(); 1068 1069 if ((flags & PRUS_IPV6) == 0) 1070 return (0); 1071 1072 if (cm->cmsg_level != IPPROTO_IPV6) 1073 return (0); 1074 1075 if (cm->cmsg_type != IPV6_2292PKTINFO && 1076 cm->cmsg_type != IPV6_PKTINFO) 1077 return (0); 1078 1079 if (cm->cmsg_len != 1080 CMSG_LEN(sizeof(struct in6_pktinfo))) 1081 return (EINVAL); 1082 1083 pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm); 1084 if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) && 1085 !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) 1086 return (EINVAL); 1087 1088 /* Validate the interface index if specified. */ 1089 if (pktinfo->ipi6_ifindex) { 1090 ifp = ifnet_byindex(pktinfo->ipi6_ifindex); 1091 if (ifp == NULL) 1092 return (ENXIO); 1093 } else 1094 ifp = NULL; 1095 if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { 1096 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1097 if (!in_ifhasaddr(ifp, ia)) 1098 return (EADDRNOTAVAIL); 1099 } 1100 1101 bzero(src, sizeof(*src)); 1102 src->sin_family = AF_INET; 1103 src->sin_len = sizeof(*src); 1104 src->sin_port = inp->inp_lport; 1105 src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1106 1107 return (0); 1108 } 1109 #endif /* INET6 */ 1110 1111 int 1112 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1113 struct mbuf *control, struct thread *td) 1114 { 1115 struct inpcb *inp; 1116 struct udpiphdr *ui; 1117 int len, error = 0; 1118 struct in_addr faddr, laddr; 1119 struct cmsghdr *cm; 1120 struct sockaddr_in *sin, src; 1121 struct epoch_tracker et; 1122 int cscov_partial = 0; 1123 int ipflags = 0; 1124 u_short fport, lport; 1125 u_char tos, vflagsav; 1126 uint8_t pr; 1127 uint16_t cscov = 0; 1128 uint32_t hash_val, hash_type, flowid = 0; 1129 uint8_t flowtype = M_HASHTYPE_NONE; 1130 bool use_cached_route; 1131 1132 inp = sotoinpcb(so); 1133 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1134 1135 if (addr != NULL) { 1136 if (addr->sa_family != AF_INET) 1137 error = EAFNOSUPPORT; 1138 else if (addr->sa_len != sizeof(struct sockaddr_in)) 1139 error = EINVAL; 1140 if (__predict_false(error != 0)) { 1141 m_freem(control); 1142 m_freem(m); 1143 return (error); 1144 } 1145 } 1146 1147 len = m->m_pkthdr.len; 1148 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1149 if (control) 1150 m_freem(control); 1151 m_freem(m); 1152 return (EMSGSIZE); 1153 } 1154 1155 src.sin_family = 0; 1156 sin = (struct sockaddr_in *)addr; 1157 1158 /* 1159 * udp_send() may need to bind the current inpcb. As such, we don't 1160 * know up front whether we will need the pcbinfo lock or not. Do any 1161 * work to decide what is needed up front before acquiring any locks. 1162 * 1163 * We will need network epoch in either case, to safely lookup into 1164 * pcb hash. 1165 */ 1166 use_cached_route = sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0); 1167 if (use_cached_route || (flags & PRUS_IPV6) != 0) 1168 INP_WLOCK(inp); 1169 else 1170 INP_RLOCK(inp); 1171 NET_EPOCH_ENTER(et); 1172 #ifdef INET6 1173 if ((flags & PRUS_IPV6) != 0) { 1174 if ((inp->in6p_outputopts != NULL) && 1175 (inp->in6p_outputopts->ip6po_tclass != -1)) 1176 tos = (u_char)inp->in6p_outputopts->ip6po_tclass; 1177 else 1178 tos = 0; 1179 } else { 1180 tos = inp->inp_ip_tos; 1181 } 1182 #else 1183 tos = inp->inp_ip_tos; 1184 #endif 1185 if (control != NULL) { 1186 /* 1187 * XXX: Currently, we assume all the optional information is 1188 * stored in a single mbuf. 1189 */ 1190 if (control->m_next) { 1191 m_freem(control); 1192 error = EINVAL; 1193 goto release; 1194 } 1195 for (; control->m_len > 0; 1196 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1197 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1198 cm = mtod(control, struct cmsghdr *); 1199 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1200 || cm->cmsg_len > control->m_len) { 1201 error = EINVAL; 1202 break; 1203 } 1204 #ifdef INET6 1205 error = udp_v4mapped_pktinfo(cm, &src, inp, flags); 1206 if (error != 0) 1207 break; 1208 if (((flags & PRUS_IPV6) != 0) && 1209 (cm->cmsg_level == IPPROTO_IPV6) && 1210 (cm->cmsg_type == IPV6_TCLASS)) { 1211 int tclass; 1212 1213 if (cm->cmsg_len != CMSG_LEN(sizeof(int))) { 1214 error = EINVAL; 1215 break; 1216 } 1217 tclass = *(int *)CMSG_DATA(cm); 1218 if (tclass < -1 || tclass > 255) { 1219 error = EINVAL; 1220 break; 1221 } 1222 if (tclass != -1) 1223 tos = (u_char)tclass; 1224 } 1225 #endif 1226 if (cm->cmsg_level != IPPROTO_IP) 1227 continue; 1228 1229 switch (cm->cmsg_type) { 1230 case IP_SENDSRCADDR: 1231 if (cm->cmsg_len != 1232 CMSG_LEN(sizeof(struct in_addr))) { 1233 error = EINVAL; 1234 break; 1235 } 1236 bzero(&src, sizeof(src)); 1237 src.sin_family = AF_INET; 1238 src.sin_len = sizeof(src); 1239 src.sin_port = inp->inp_lport; 1240 src.sin_addr = 1241 *(struct in_addr *)CMSG_DATA(cm); 1242 break; 1243 1244 case IP_TOS: 1245 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1246 error = EINVAL; 1247 break; 1248 } 1249 tos = *(u_char *)CMSG_DATA(cm); 1250 break; 1251 1252 case IP_FLOWID: 1253 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1254 error = EINVAL; 1255 break; 1256 } 1257 flowid = *(uint32_t *) CMSG_DATA(cm); 1258 break; 1259 1260 case IP_FLOWTYPE: 1261 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1262 error = EINVAL; 1263 break; 1264 } 1265 flowtype = *(uint32_t *) CMSG_DATA(cm); 1266 break; 1267 1268 #ifdef RSS 1269 case IP_RSSBUCKETID: 1270 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1271 error = EINVAL; 1272 break; 1273 } 1274 /* This is just a placeholder for now */ 1275 break; 1276 #endif /* RSS */ 1277 default: 1278 error = ENOPROTOOPT; 1279 break; 1280 } 1281 if (error) 1282 break; 1283 } 1284 m_freem(control); 1285 control = NULL; 1286 } 1287 if (error) 1288 goto release; 1289 1290 pr = inp->inp_socket->so_proto->pr_protocol; 1291 1292 /* 1293 * If the IP_SENDSRCADDR control message was specified, override the 1294 * source address for this datagram. Its use is invalidated if the 1295 * address thus specified is incomplete or clobbers other inpcbs. 1296 */ 1297 laddr = inp->inp_laddr; 1298 lport = inp->inp_lport; 1299 if (src.sin_family == AF_INET) { 1300 if ((lport == 0) || 1301 (laddr.s_addr == INADDR_ANY && 1302 src.sin_addr.s_addr == INADDR_ANY)) { 1303 error = EINVAL; 1304 goto release; 1305 } 1306 if ((flags & PRUS_IPV6) != 0) { 1307 vflagsav = inp->inp_vflag; 1308 inp->inp_vflag |= INP_IPV4; 1309 inp->inp_vflag &= ~INP_IPV6; 1310 } 1311 error = in_pcbbind_setup(inp, &src, &laddr.s_addr, &lport, 1312 V_udp_bind_all_fibs ? 0 : INPBIND_FIB, td->td_ucred); 1313 if ((flags & PRUS_IPV6) != 0) 1314 inp->inp_vflag = vflagsav; 1315 if (error) 1316 goto release; 1317 } 1318 1319 /* 1320 * If a UDP socket has been connected, then a local address/port will 1321 * have been selected and bound. 1322 * 1323 * If a UDP socket has not been connected to, then an explicit 1324 * destination address must be used, in which case a local 1325 * address/port may not have been selected and bound. 1326 */ 1327 if (sin != NULL) { 1328 INP_LOCK_ASSERT(inp); 1329 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1330 error = EISCONN; 1331 goto release; 1332 } 1333 1334 /* 1335 * Jail may rewrite the destination address, so let it do 1336 * that before we use it. 1337 */ 1338 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1339 if (error) 1340 goto release; 1341 /* 1342 * sendto(2) on unconnected UDP socket results in implicit 1343 * binding to INADDR_ANY and anonymous port. This has two 1344 * side effects: 1345 * 1) after first sendto(2) the socket will receive datagrams 1346 * destined to the selected port. 1347 * 2) subsequent sendto(2) calls will use the same source port. 1348 */ 1349 if (inp->inp_lport == 0) { 1350 struct sockaddr_in wild = { 1351 .sin_family = AF_INET, 1352 .sin_len = sizeof(struct sockaddr_in), 1353 }; 1354 1355 error = in_pcbbind(inp, &wild, V_udp_bind_all_fibs ? 1356 0 : INPBIND_FIB, td->td_ucred); 1357 if (error) 1358 goto release; 1359 lport = inp->inp_lport; 1360 laddr = inp->inp_laddr; 1361 } 1362 if (laddr.s_addr == INADDR_ANY) { 1363 error = in_pcbladdr(inp, &sin->sin_addr, &laddr, 1364 td->td_ucred); 1365 if (error) 1366 goto release; 1367 } 1368 faddr = sin->sin_addr; 1369 fport = sin->sin_port; 1370 } else { 1371 INP_LOCK_ASSERT(inp); 1372 faddr = inp->inp_faddr; 1373 fport = inp->inp_fport; 1374 if (faddr.s_addr == INADDR_ANY) { 1375 error = ENOTCONN; 1376 goto release; 1377 } 1378 } 1379 1380 /* 1381 * Calculate data length and get a mbuf for UDP, IP, and possible 1382 * link-layer headers. Immediate slide the data pointer back forward 1383 * since we won't use that space at this layer. 1384 */ 1385 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1386 if (m == NULL) { 1387 error = ENOBUFS; 1388 goto release; 1389 } 1390 m->m_data += max_linkhdr; 1391 m->m_len -= max_linkhdr; 1392 m->m_pkthdr.len -= max_linkhdr; 1393 1394 /* 1395 * Fill in mbuf with extended UDP header and addresses and length put 1396 * into network format. 1397 */ 1398 ui = mtod(m, struct udpiphdr *); 1399 /* 1400 * Filling only those fields of udpiphdr that participate in the 1401 * checksum calculation. The rest must be zeroed and will be filled 1402 * later. 1403 */ 1404 bzero(ui->ui_x1, sizeof(ui->ui_x1)); 1405 ui->ui_pr = pr; 1406 ui->ui_src = laddr; 1407 ui->ui_dst = faddr; 1408 ui->ui_sport = lport; 1409 ui->ui_dport = fport; 1410 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1411 if (pr == IPPROTO_UDPLITE) { 1412 struct udpcb *up; 1413 uint16_t plen; 1414 1415 up = intoudpcb(inp); 1416 cscov = up->u_txcslen; 1417 plen = (u_short)len + sizeof(struct udphdr); 1418 if (cscov >= plen) 1419 cscov = 0; 1420 ui->ui_len = htons(plen); 1421 ui->ui_ulen = htons(cscov); 1422 /* 1423 * For UDP-Lite, checksum coverage length of zero means 1424 * the entire UDPLite packet is covered by the checksum. 1425 */ 1426 cscov_partial = (cscov == 0) ? 0 : 1; 1427 } 1428 1429 if (inp->inp_socket->so_options & SO_DONTROUTE) 1430 ipflags |= IP_ROUTETOIF; 1431 if (inp->inp_socket->so_options & SO_BROADCAST) 1432 ipflags |= IP_ALLOWBROADCAST; 1433 if (inp->inp_flags & INP_ONESBCAST) 1434 ipflags |= IP_SENDONES; 1435 1436 #ifdef MAC 1437 mac_inpcb_create_mbuf(inp, m); 1438 #endif 1439 1440 /* 1441 * Set up checksum and output datagram. 1442 */ 1443 ui->ui_sum = 0; 1444 if (pr == IPPROTO_UDPLITE) { 1445 if (inp->inp_flags & INP_ONESBCAST) 1446 faddr.s_addr = INADDR_BROADCAST; 1447 if (cscov_partial) { 1448 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1449 ui->ui_sum = 0xffff; 1450 } else { 1451 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1452 ui->ui_sum = 0xffff; 1453 } 1454 } else if (V_udp_cksum) { 1455 if (inp->inp_flags & INP_ONESBCAST) 1456 faddr.s_addr = INADDR_BROADCAST; 1457 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1458 htons((u_short)len + sizeof(struct udphdr) + pr)); 1459 m->m_pkthdr.csum_flags = CSUM_UDP; 1460 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1461 } 1462 /* 1463 * After finishing the checksum computation, fill the remaining fields 1464 * of udpiphdr. 1465 */ 1466 ((struct ip *)ui)->ip_v = IPVERSION; 1467 ((struct ip *)ui)->ip_tos = tos; 1468 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1469 if (inp->inp_flags & INP_DONTFRAG) 1470 ((struct ip *)ui)->ip_off |= htons(IP_DF); 1471 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; 1472 UDPSTAT_INC(udps_opackets); 1473 1474 /* 1475 * Setup flowid / RSS information for outbound socket. 1476 * 1477 * Once the UDP code decides to set a flowid some other way, 1478 * this allows the flowid to be overridden by userland. 1479 */ 1480 if (flowtype != M_HASHTYPE_NONE) { 1481 m->m_pkthdr.flowid = flowid; 1482 M_HASHTYPE_SET(m, flowtype); 1483 } else if (CALC_FLOWID_OUTBOUND_SENDTO) { 1484 hash_val = fib4_calc_packet_hash(laddr, faddr, 1485 lport, fport, pr, &hash_type); 1486 m->m_pkthdr.flowid = hash_val; 1487 M_HASHTYPE_SET(m, hash_type); 1488 } 1489 1490 /* 1491 * Don't override with the inp cached flowid value. 1492 * 1493 * Depending upon the kind of send being done, the inp 1494 * flowid/flowtype values may actually not be appropriate 1495 * for this particular socket send. 1496 * 1497 * We should either leave the flowid at zero (which is what is 1498 * currently done) or set it to some software generated 1499 * hash value based on the packet contents. 1500 */ 1501 ipflags |= IP_NODEFAULTFLOWID; 1502 1503 if (pr == IPPROTO_UDPLITE) 1504 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1505 else 1506 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1507 error = ip_output(m, inp->inp_options, 1508 use_cached_route ? &inp->inp_route : NULL, ipflags, 1509 inp->inp_moptions, inp); 1510 INP_UNLOCK(inp); 1511 NET_EPOCH_EXIT(et); 1512 return (error); 1513 1514 release: 1515 INP_UNLOCK(inp); 1516 NET_EPOCH_EXIT(et); 1517 m_freem(m); 1518 return (error); 1519 } 1520 1521 void 1522 udp_abort(struct socket *so) 1523 { 1524 struct inpcb *inp; 1525 1526 inp = sotoinpcb(so); 1527 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1528 INP_WLOCK(inp); 1529 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1530 in_pcbdisconnect(inp); 1531 inp->inp_laddr.s_addr = INADDR_ANY; 1532 soisdisconnected(so); 1533 } 1534 INP_WUNLOCK(inp); 1535 } 1536 1537 static int 1538 udp_attach(struct socket *so, int proto, struct thread *td) 1539 { 1540 static uint32_t udp_flowid; 1541 struct inpcbinfo *pcbinfo; 1542 struct inpcb *inp; 1543 struct udpcb *up; 1544 int error; 1545 1546 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1547 inp = sotoinpcb(so); 1548 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1549 error = soreserve(so, udp_sendspace, udp_recvspace); 1550 if (error) 1551 return (error); 1552 error = in_pcballoc(so, pcbinfo); 1553 if (error) 1554 return (error); 1555 1556 inp = sotoinpcb(so); 1557 inp->inp_ip_ttl = V_ip_defttl; 1558 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1559 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1560 up = intoudpcb(inp); 1561 bzero(&up->u_start_zero, u_zero_size); 1562 INP_WUNLOCK(inp); 1563 1564 return (0); 1565 } 1566 #endif /* INET */ 1567 1568 int 1569 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1570 { 1571 struct inpcb *inp; 1572 struct udpcb *up; 1573 1574 KASSERT(so->so_type == SOCK_DGRAM, 1575 ("udp_set_kernel_tunneling: !dgram")); 1576 inp = sotoinpcb(so); 1577 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1578 INP_WLOCK(inp); 1579 up = intoudpcb(inp); 1580 if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) || 1581 (up->u_icmp_func != NULL))) { 1582 INP_WUNLOCK(inp); 1583 return (EBUSY); 1584 } 1585 up->u_tun_func = f; 1586 up->u_icmp_func = i; 1587 up->u_tun_ctx = ctx; 1588 INP_WUNLOCK(inp); 1589 return (0); 1590 } 1591 1592 #ifdef INET 1593 static int 1594 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1595 { 1596 struct inpcb *inp; 1597 struct sockaddr_in *sinp; 1598 int error; 1599 1600 inp = sotoinpcb(so); 1601 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1602 1603 sinp = (struct sockaddr_in *)nam; 1604 if (nam->sa_family != AF_INET) { 1605 /* 1606 * Preserve compatibility with old programs. 1607 */ 1608 if (nam->sa_family != AF_UNSPEC || 1609 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 1610 sinp->sin_addr.s_addr != INADDR_ANY) 1611 return (EAFNOSUPPORT); 1612 nam->sa_family = AF_INET; 1613 } 1614 if (nam->sa_len != sizeof(struct sockaddr_in)) 1615 return (EINVAL); 1616 1617 INP_WLOCK(inp); 1618 error = in_pcbbind(inp, sinp, V_udp_bind_all_fibs ? 0 : INPBIND_FIB, 1619 td->td_ucred); 1620 INP_WUNLOCK(inp); 1621 return (error); 1622 } 1623 1624 static void 1625 udp_close(struct socket *so) 1626 { 1627 struct inpcb *inp; 1628 1629 inp = sotoinpcb(so); 1630 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1631 INP_WLOCK(inp); 1632 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1633 in_pcbdisconnect(inp); 1634 inp->inp_laddr.s_addr = INADDR_ANY; 1635 soisdisconnected(so); 1636 } 1637 INP_WUNLOCK(inp); 1638 } 1639 1640 static int 1641 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1642 { 1643 struct epoch_tracker et; 1644 struct inpcb *inp; 1645 struct sockaddr_in *sin; 1646 int error; 1647 1648 inp = sotoinpcb(so); 1649 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1650 1651 sin = (struct sockaddr_in *)nam; 1652 if (sin->sin_family != AF_INET) 1653 return (EAFNOSUPPORT); 1654 if (sin->sin_len != sizeof(*sin)) 1655 return (EINVAL); 1656 1657 INP_WLOCK(inp); 1658 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1659 INP_WUNLOCK(inp); 1660 return (EISCONN); 1661 } 1662 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1663 if (error != 0) { 1664 INP_WUNLOCK(inp); 1665 return (error); 1666 } 1667 NET_EPOCH_ENTER(et); 1668 error = in_pcbconnect(inp, sin, td->td_ucred); 1669 NET_EPOCH_EXIT(et); 1670 if (error == 0) 1671 soisconnected(so); 1672 INP_WUNLOCK(inp); 1673 return (error); 1674 } 1675 1676 static void 1677 udp_detach(struct socket *so) 1678 { 1679 struct inpcb *inp; 1680 1681 inp = sotoinpcb(so); 1682 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1683 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1684 ("udp_detach: not disconnected")); 1685 INP_WLOCK(inp); 1686 in_pcbfree(inp); 1687 } 1688 1689 int 1690 udp_disconnect(struct socket *so) 1691 { 1692 struct inpcb *inp; 1693 1694 inp = sotoinpcb(so); 1695 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1696 INP_WLOCK(inp); 1697 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1698 INP_WUNLOCK(inp); 1699 return (ENOTCONN); 1700 } 1701 in_pcbdisconnect(inp); 1702 inp->inp_laddr.s_addr = INADDR_ANY; 1703 SOCK_LOCK(so); 1704 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1705 SOCK_UNLOCK(so); 1706 INP_WUNLOCK(inp); 1707 return (0); 1708 } 1709 #endif /* INET */ 1710 1711 int 1712 udp_shutdown(struct socket *so, enum shutdown_how how) 1713 { 1714 int error; 1715 1716 SOCK_LOCK(so); 1717 if (!(so->so_state & SS_ISCONNECTED)) 1718 /* 1719 * POSIX mandates us to just return ENOTCONN when shutdown(2) is 1720 * invoked on a datagram sockets, however historically we would 1721 * actually tear socket down. This is known to be leveraged by 1722 * some applications to unblock process waiting in recv(2) by 1723 * other process that it shares that socket with. Try to meet 1724 * both backward-compatibility and POSIX requirements by forcing 1725 * ENOTCONN but still flushing buffers and performing wakeup(9). 1726 * 1727 * XXXGL: it remains unknown what applications expect this 1728 * behavior and is this isolated to unix/dgram or inet/dgram or 1729 * both. See: D10351, D3039. 1730 */ 1731 error = ENOTCONN; 1732 else 1733 error = 0; 1734 SOCK_UNLOCK(so); 1735 1736 switch (how) { 1737 case SHUT_RD: 1738 sorflush(so); 1739 break; 1740 case SHUT_RDWR: 1741 sorflush(so); 1742 /* FALLTHROUGH */ 1743 case SHUT_WR: 1744 socantsendmore(so); 1745 } 1746 1747 return (error); 1748 } 1749 1750 #ifdef INET 1751 #define UDP_PROTOSW \ 1752 .pr_type = SOCK_DGRAM, \ 1753 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH, \ 1754 .pr_ctloutput = udp_ctloutput, \ 1755 .pr_abort = udp_abort, \ 1756 .pr_attach = udp_attach, \ 1757 .pr_bind = udp_bind, \ 1758 .pr_connect = udp_connect, \ 1759 .pr_control = in_control, \ 1760 .pr_detach = udp_detach, \ 1761 .pr_disconnect = udp_disconnect, \ 1762 .pr_peeraddr = in_getpeeraddr, \ 1763 .pr_send = udp_send, \ 1764 .pr_soreceive = soreceive_dgram, \ 1765 .pr_sosend = sosend_dgram, \ 1766 .pr_shutdown = udp_shutdown, \ 1767 .pr_sockaddr = in_getsockaddr, \ 1768 .pr_sosetlabel = in_pcbsosetlabel, \ 1769 .pr_close = udp_close 1770 1771 struct protosw udp_protosw = { 1772 .pr_protocol = IPPROTO_UDP, 1773 UDP_PROTOSW 1774 }; 1775 1776 struct protosw udplite_protosw = { 1777 .pr_protocol = IPPROTO_UDPLITE, 1778 UDP_PROTOSW 1779 }; 1780 1781 static void 1782 udp_init(void *arg __unused) 1783 { 1784 1785 IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput); 1786 IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput); 1787 } 1788 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL); 1789 #endif /* INET */ 1790