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 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 39 */ 40 41 #include <sys/cdefs.h> 42 __FBSDID("$FreeBSD$"); 43 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_ipsec.h" 47 #include "opt_route.h" 48 #include "opt_rss.h" 49 50 #include <sys/param.h> 51 #include <sys/domain.h> 52 #include <sys/eventhandler.h> 53 #include <sys/jail.h> 54 #include <sys/kernel.h> 55 #include <sys/lock.h> 56 #include <sys/malloc.h> 57 #include <sys/mbuf.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/protosw.h> 61 #include <sys/sdt.h> 62 #include <sys/signalvar.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/sx.h> 66 #include <sys/sysctl.h> 67 #include <sys/syslog.h> 68 #include <sys/systm.h> 69 70 #include <vm/uma.h> 71 72 #include <net/if.h> 73 #include <net/if_var.h> 74 #include <net/route.h> 75 #include <net/route/nhop.h> 76 #include <net/rss_config.h> 77 78 #include <netinet/in.h> 79 #include <netinet/in_kdtrace.h> 80 #include <netinet/in_fib.h> 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_systm.h> 83 #include <netinet/in_var.h> 84 #include <netinet/ip.h> 85 #ifdef INET6 86 #include <netinet/ip6.h> 87 #endif 88 #include <netinet/ip_icmp.h> 89 #include <netinet/icmp_var.h> 90 #include <netinet/ip_var.h> 91 #include <netinet/ip_options.h> 92 #ifdef INET6 93 #include <netinet6/ip6_var.h> 94 #endif 95 #include <netinet/udp.h> 96 #include <netinet/udp_var.h> 97 #include <netinet/udplite.h> 98 #include <netinet/in_rss.h> 99 100 #include <netipsec/ipsec_support.h> 101 102 #include <machine/in_cksum.h> 103 104 #include <security/mac/mac_framework.h> 105 106 /* 107 * UDP and UDP-Lite protocols implementation. 108 * Per RFC 768, August, 1980. 109 * Per RFC 3828, July, 2004. 110 */ 111 112 /* 113 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 114 * removes the only data integrity mechanism for packets and malformed 115 * packets that would otherwise be discarded due to bad checksums, and may 116 * cause problems (especially for NFS data blocks). 117 */ 118 VNET_DEFINE(int, udp_cksum) = 1; 119 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, 120 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 121 122 VNET_DEFINE(int, udp_log_in_vain) = 0; 123 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 124 &VNET_NAME(udp_log_in_vain), 0, "Log all incoming UDP packets"); 125 126 VNET_DEFINE(int, udp_blackhole) = 0; 127 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 128 &VNET_NAME(udp_blackhole), 0, 129 "Do not send port unreachables for refused connects"); 130 VNET_DEFINE(bool, udp_blackhole_local) = false; 131 SYSCTL_BOOL(_net_inet_udp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 132 CTLFLAG_RW, &VNET_NAME(udp_blackhole_local), false, 133 "Enforce net.inet.udp.blackhole for locally originated packets"); 134 135 u_long udp_sendspace = 9216; /* really max datagram size */ 136 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 137 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 138 139 u_long udp_recvspace = 40 * (1024 + 140 #ifdef INET6 141 sizeof(struct sockaddr_in6) 142 #else 143 sizeof(struct sockaddr_in) 144 #endif 145 ); /* 40 1K datagrams */ 146 147 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 148 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 149 150 VNET_DEFINE(struct inpcbinfo, udbinfo); 151 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 152 VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone); 153 #define V_udpcb_zone VNET(udpcb_zone) 154 155 #ifndef UDBHASHSIZE 156 #define UDBHASHSIZE 128 157 #endif 158 159 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 160 VNET_PCPUSTAT_SYSINIT(udpstat); 161 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 162 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 163 164 #ifdef VIMAGE 165 VNET_PCPUSTAT_SYSUNINIT(udpstat); 166 #endif /* VIMAGE */ 167 #ifdef INET 168 static void udp_detach(struct socket *so); 169 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 170 struct mbuf *, struct thread *, int); 171 #endif 172 173 INPCBSTORAGE_DEFINE(udpcbstor, "udpinp", "udp_inpcb", "udp", "udphash"); 174 INPCBSTORAGE_DEFINE(udplitecbstor, "udpliteinp", "udplite_inpcb", "udplite", 175 "udplitehash"); 176 177 static void 178 udp_vnet_init(void *arg __unused) 179 { 180 181 /* 182 * For now default to 2-tuple UDP hashing - until the fragment 183 * reassembly code can also update the flowid. 184 * 185 * Once we can calculate the flowid that way and re-establish 186 * a 4-tuple, flip this to 4-tuple. 187 */ 188 in_pcbinfo_init(&V_udbinfo, &udpcbstor, UDBHASHSIZE, UDBHASHSIZE); 189 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 190 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 191 uma_zone_set_max(V_udpcb_zone, maxsockets); 192 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 193 194 /* Additional pcbinfo for UDP-Lite */ 195 in_pcbinfo_init(&V_ulitecbinfo, &udplitecbstor, UDBHASHSIZE, 196 UDBHASHSIZE); 197 } 198 VNET_SYSINIT(udp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 199 udp_vnet_init, NULL); 200 201 /* 202 * Kernel module interface for updating udpstat. The argument is an index 203 * into udpstat treated as an array of u_long. While this encodes the 204 * general layout of udpstat into the caller, it doesn't encode its location, 205 * so that future changes to add, for example, per-CPU stats support won't 206 * cause binary compatibility problems for kernel modules. 207 */ 208 void 209 kmod_udpstat_inc(int statnum) 210 { 211 212 counter_u64_add(VNET(udpstat)[statnum], 1); 213 } 214 215 int 216 udp_newudpcb(struct inpcb *inp) 217 { 218 struct udpcb *up; 219 220 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 221 if (up == NULL) 222 return (ENOBUFS); 223 inp->inp_ppcb = up; 224 return (0); 225 } 226 227 void 228 udp_discardcb(struct udpcb *up) 229 { 230 231 uma_zfree(V_udpcb_zone, up); 232 } 233 234 #ifdef VIMAGE 235 static void 236 udp_destroy(void *unused __unused) 237 { 238 239 in_pcbinfo_destroy(&V_udbinfo); 240 uma_zdestroy(V_udpcb_zone); 241 } 242 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 243 244 static void 245 udplite_destroy(void *unused __unused) 246 { 247 248 in_pcbinfo_destroy(&V_ulitecbinfo); 249 } 250 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 251 NULL); 252 #endif 253 254 #ifdef INET 255 /* 256 * Subroutine of udp_input(), which appends the provided mbuf chain to the 257 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 258 * contains the source address. If the socket ends up being an IPv6 socket, 259 * udp_append() will convert to a sockaddr_in6 before passing the address 260 * into the socket code. 261 * 262 * In the normal case udp_append() will return 0, indicating that you 263 * must unlock the inp. However if a tunneling protocol is in place we increment 264 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 265 * then decrement the reference count. If the inp_rele returns 1, indicating the 266 * inp is gone, we return that to the caller to tell them *not* to unlock 267 * the inp. In the case of multi-cast this will cause the distribution 268 * to stop (though most tunneling protocols known currently do *not* use 269 * multicast). 270 */ 271 static int 272 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 273 struct sockaddr_in *udp_in) 274 { 275 struct sockaddr *append_sa; 276 struct socket *so; 277 struct mbuf *tmpopts, *opts = NULL; 278 #ifdef INET6 279 struct sockaddr_in6 udp_in6; 280 #endif 281 struct udpcb *up; 282 bool filtered; 283 284 INP_LOCK_ASSERT(inp); 285 286 /* 287 * Engage the tunneling protocol. 288 */ 289 up = intoudpcb(inp); 290 if (up->u_tun_func != NULL) { 291 in_pcbref(inp); 292 INP_RUNLOCK(inp); 293 filtered = (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], 294 up->u_tun_ctx); 295 INP_RLOCK(inp); 296 if (filtered) 297 return (in_pcbrele_rlocked(inp)); 298 } 299 300 off += sizeof(struct udphdr); 301 302 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 303 /* Check AH/ESP integrity. */ 304 if (IPSEC_ENABLED(ipv4) && 305 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 306 m_freem(n); 307 return (0); 308 } 309 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 310 if (IPSEC_ENABLED(ipv4) && 311 UDPENCAP_INPUT(n, off, AF_INET) != 0) 312 return (0); /* Consumed. */ 313 } 314 #endif /* IPSEC */ 315 #ifdef MAC 316 if (mac_inpcb_check_deliver(inp, n) != 0) { 317 m_freem(n); 318 return (0); 319 } 320 #endif /* MAC */ 321 if (inp->inp_flags & INP_CONTROLOPTS || 322 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 323 #ifdef INET6 324 if (inp->inp_vflag & INP_IPV6) 325 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 326 else 327 #endif /* INET6 */ 328 ip_savecontrol(inp, &opts, ip, n); 329 } 330 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 331 tmpopts = sbcreatecontrol(&udp_in[1], 332 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP, 333 M_NOWAIT); 334 if (tmpopts) { 335 if (opts) { 336 tmpopts->m_next = opts; 337 opts = tmpopts; 338 } else 339 opts = tmpopts; 340 } 341 } 342 #ifdef INET6 343 if (inp->inp_vflag & INP_IPV6) { 344 bzero(&udp_in6, sizeof(udp_in6)); 345 udp_in6.sin6_len = sizeof(udp_in6); 346 udp_in6.sin6_family = AF_INET6; 347 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 348 append_sa = (struct sockaddr *)&udp_in6; 349 } else 350 #endif /* INET6 */ 351 append_sa = (struct sockaddr *)&udp_in[0]; 352 m_adj(n, off); 353 354 so = inp->inp_socket; 355 SOCKBUF_LOCK(&so->so_rcv); 356 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 357 soroverflow_locked(so); 358 m_freem(n); 359 if (opts) 360 m_freem(opts); 361 UDPSTAT_INC(udps_fullsock); 362 } else 363 sorwakeup_locked(so); 364 return (0); 365 } 366 367 static bool 368 udp_multi_match(const struct inpcb *inp, void *v) 369 { 370 struct ip *ip = v; 371 struct udphdr *uh = (struct udphdr *)(ip + 1); 372 373 if (inp->inp_lport != uh->uh_dport) 374 return (false); 375 #ifdef INET6 376 if ((inp->inp_vflag & INP_IPV4) == 0) 377 return (false); 378 #endif 379 if (inp->inp_laddr.s_addr != INADDR_ANY && 380 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 381 return (false); 382 if (inp->inp_faddr.s_addr != INADDR_ANY && 383 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 384 return (false); 385 if (inp->inp_fport != 0 && 386 inp->inp_fport != uh->uh_sport) 387 return (false); 388 389 return (true); 390 } 391 392 static int 393 udp_multi_input(struct mbuf *m, int proto, struct sockaddr_in *udp_in) 394 { 395 struct ip *ip = mtod(m, struct ip *); 396 struct inpcb_iterator inpi = INP_ITERATOR(udp_get_inpcbinfo(proto), 397 INPLOOKUP_RLOCKPCB, udp_multi_match, ip); 398 #ifdef KDTRACE_HOOKS 399 struct udphdr *uh = (struct udphdr *)(ip + 1); 400 #endif 401 struct inpcb *inp; 402 struct mbuf *n; 403 int appends = 0; 404 405 MPASS(ip->ip_hl == sizeof(struct ip) >> 2); 406 407 while ((inp = inp_next(&inpi)) != NULL) { 408 /* 409 * XXXRW: Because we weren't holding either the inpcb 410 * or the hash lock when we checked for a match 411 * before, we should probably recheck now that the 412 * inpcb lock is held. 413 */ 414 /* 415 * Handle socket delivery policy for any-source 416 * and source-specific multicast. [RFC3678] 417 */ 418 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 419 struct ip_moptions *imo; 420 struct sockaddr_in group; 421 int blocked; 422 423 imo = inp->inp_moptions; 424 if (imo == NULL) 425 continue; 426 bzero(&group, sizeof(struct sockaddr_in)); 427 group.sin_len = sizeof(struct sockaddr_in); 428 group.sin_family = AF_INET; 429 group.sin_addr = ip->ip_dst; 430 431 blocked = imo_multi_filter(imo, m->m_pkthdr.rcvif, 432 (struct sockaddr *)&group, 433 (struct sockaddr *)&udp_in[0]); 434 if (blocked != MCAST_PASS) { 435 if (blocked == MCAST_NOTGMEMBER) 436 IPSTAT_INC(ips_notmember); 437 if (blocked == MCAST_NOTSMEMBER || 438 blocked == MCAST_MUTED) 439 UDPSTAT_INC(udps_filtermcast); 440 continue; 441 } 442 } 443 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) { 444 if (proto == IPPROTO_UDPLITE) 445 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 446 else 447 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 448 if (udp_append(inp, ip, n, sizeof(struct ip), udp_in)) { 449 INP_RUNLOCK(inp); 450 break; 451 } else 452 appends++; 453 } 454 /* 455 * Don't look for additional matches if this one does 456 * not have either the SO_REUSEPORT or SO_REUSEADDR 457 * socket options set. This heuristic avoids 458 * searching through all pcbs in the common case of a 459 * non-shared port. It assumes that an application 460 * will never clear these options after setting them. 461 */ 462 if ((inp->inp_socket->so_options & 463 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) { 464 INP_RUNLOCK(inp); 465 break; 466 } 467 } 468 469 if (appends == 0) { 470 /* 471 * No matching pcb found; discard datagram. (No need 472 * to send an ICMP Port Unreachable for a broadcast 473 * or multicast datgram.) 474 */ 475 UDPSTAT_INC(udps_noport); 476 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 477 UDPSTAT_INC(udps_noportmcast); 478 else 479 UDPSTAT_INC(udps_noportbcast); 480 } 481 m_freem(m); 482 483 return (IPPROTO_DONE); 484 } 485 486 static int 487 udp_input(struct mbuf **mp, int *offp, int proto) 488 { 489 struct ip *ip; 490 struct udphdr *uh; 491 struct ifnet *ifp; 492 struct inpcb *inp; 493 uint16_t len, ip_len; 494 struct inpcbinfo *pcbinfo; 495 struct sockaddr_in udp_in[2]; 496 struct mbuf *m; 497 struct m_tag *fwd_tag; 498 int cscov_partial, iphlen; 499 500 m = *mp; 501 iphlen = *offp; 502 ifp = m->m_pkthdr.rcvif; 503 *mp = NULL; 504 UDPSTAT_INC(udps_ipackets); 505 506 /* 507 * Strip IP options, if any; should skip this, make available to 508 * user, and use on returned packets, but we don't yet have a way to 509 * check the checksum with options still present. 510 */ 511 if (iphlen > sizeof (struct ip)) { 512 ip_stripoptions(m); 513 iphlen = sizeof(struct ip); 514 } 515 516 /* 517 * Get IP and UDP header together in first mbuf. 518 */ 519 if (m->m_len < iphlen + sizeof(struct udphdr)) { 520 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 521 UDPSTAT_INC(udps_hdrops); 522 return (IPPROTO_DONE); 523 } 524 } 525 ip = mtod(m, struct ip *); 526 uh = (struct udphdr *)((caddr_t)ip + iphlen); 527 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 528 529 /* 530 * Destination port of 0 is illegal, based on RFC768. 531 */ 532 if (uh->uh_dport == 0) 533 goto badunlocked; 534 535 /* 536 * Construct sockaddr format source address. Stuff source address 537 * and datagram in user buffer. 538 */ 539 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 540 udp_in[0].sin_len = sizeof(struct sockaddr_in); 541 udp_in[0].sin_family = AF_INET; 542 udp_in[0].sin_port = uh->uh_sport; 543 udp_in[0].sin_addr = ip->ip_src; 544 udp_in[1].sin_len = sizeof(struct sockaddr_in); 545 udp_in[1].sin_family = AF_INET; 546 udp_in[1].sin_port = uh->uh_dport; 547 udp_in[1].sin_addr = ip->ip_dst; 548 549 /* 550 * Make mbuf data length reflect UDP length. If not enough data to 551 * reflect UDP length, drop. 552 */ 553 len = ntohs((u_short)uh->uh_ulen); 554 ip_len = ntohs(ip->ip_len) - iphlen; 555 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 556 /* Zero means checksum over the complete packet. */ 557 if (len == 0) 558 len = ip_len; 559 cscov_partial = 0; 560 } 561 if (ip_len != len) { 562 if (len > ip_len || len < sizeof(struct udphdr)) { 563 UDPSTAT_INC(udps_badlen); 564 goto badunlocked; 565 } 566 if (proto == IPPROTO_UDP) 567 m_adj(m, len - ip_len); 568 } 569 570 /* 571 * Checksum extended UDP header and data. 572 */ 573 if (uh->uh_sum) { 574 u_short uh_sum; 575 576 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 577 !cscov_partial) { 578 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 579 uh_sum = m->m_pkthdr.csum_data; 580 else 581 uh_sum = in_pseudo(ip->ip_src.s_addr, 582 ip->ip_dst.s_addr, htonl((u_short)len + 583 m->m_pkthdr.csum_data + proto)); 584 uh_sum ^= 0xffff; 585 } else { 586 char b[offsetof(struct ipovly, ih_src)]; 587 struct ipovly *ipov = (struct ipovly *)ip; 588 589 bcopy(ipov, b, sizeof(b)); 590 bzero(ipov, sizeof(ipov->ih_x1)); 591 ipov->ih_len = (proto == IPPROTO_UDP) ? 592 uh->uh_ulen : htons(ip_len); 593 uh_sum = in_cksum(m, len + sizeof (struct ip)); 594 bcopy(b, ipov, sizeof(b)); 595 } 596 if (uh_sum) { 597 UDPSTAT_INC(udps_badsum); 598 m_freem(m); 599 return (IPPROTO_DONE); 600 } 601 } else { 602 if (proto == IPPROTO_UDP) { 603 UDPSTAT_INC(udps_nosum); 604 } else { 605 /* UDPLite requires a checksum */ 606 /* XXX: What is the right UDPLite MIB counter here? */ 607 m_freem(m); 608 return (IPPROTO_DONE); 609 } 610 } 611 612 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 613 in_broadcast(ip->ip_dst, ifp)) 614 return (udp_multi_input(m, proto, udp_in)); 615 616 pcbinfo = udp_get_inpcbinfo(proto); 617 618 /* 619 * Locate pcb for datagram. 620 * 621 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 622 */ 623 if ((m->m_flags & M_IP_NEXTHOP) && 624 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 625 struct sockaddr_in *next_hop; 626 627 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 628 629 /* 630 * Transparently forwarded. Pretend to be the destination. 631 * Already got one like this? 632 */ 633 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 634 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 635 if (!inp) { 636 /* 637 * It's new. Try to find the ambushing socket. 638 * Because we've rewritten the destination address, 639 * any hardware-generated hash is ignored. 640 */ 641 inp = in_pcblookup(pcbinfo, ip->ip_src, 642 uh->uh_sport, next_hop->sin_addr, 643 next_hop->sin_port ? htons(next_hop->sin_port) : 644 uh->uh_dport, INPLOOKUP_WILDCARD | 645 INPLOOKUP_RLOCKPCB, ifp); 646 } 647 /* Remove the tag from the packet. We don't need it anymore. */ 648 m_tag_delete(m, fwd_tag); 649 m->m_flags &= ~M_IP_NEXTHOP; 650 } else 651 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 652 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 653 INPLOOKUP_RLOCKPCB, ifp, m); 654 if (inp == NULL) { 655 if (V_udp_log_in_vain) { 656 char src[INET_ADDRSTRLEN]; 657 char dst[INET_ADDRSTRLEN]; 658 659 log(LOG_INFO, 660 "Connection attempt to UDP %s:%d from %s:%d\n", 661 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 662 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 663 } 664 if (proto == IPPROTO_UDPLITE) 665 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); 666 else 667 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 668 UDPSTAT_INC(udps_noport); 669 if (m->m_flags & (M_BCAST | M_MCAST)) { 670 UDPSTAT_INC(udps_noportbcast); 671 goto badunlocked; 672 } 673 if (V_udp_blackhole && (V_udp_blackhole_local || 674 !in_localip(ip->ip_src))) 675 goto badunlocked; 676 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 677 goto badunlocked; 678 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 679 return (IPPROTO_DONE); 680 } 681 682 /* 683 * Check the minimum TTL for socket. 684 */ 685 INP_RLOCK_ASSERT(inp); 686 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 687 if (proto == IPPROTO_UDPLITE) 688 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 689 else 690 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 691 INP_RUNLOCK(inp); 692 m_freem(m); 693 return (IPPROTO_DONE); 694 } 695 if (cscov_partial) { 696 struct udpcb *up; 697 698 up = intoudpcb(inp); 699 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 700 INP_RUNLOCK(inp); 701 m_freem(m); 702 return (IPPROTO_DONE); 703 } 704 } 705 706 if (proto == IPPROTO_UDPLITE) 707 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 708 else 709 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 710 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 711 INP_RUNLOCK(inp); 712 return (IPPROTO_DONE); 713 714 badunlocked: 715 m_freem(m); 716 return (IPPROTO_DONE); 717 } 718 #endif /* INET */ 719 720 /* 721 * Notify a udp user of an asynchronous error; just wake up so that they can 722 * collect error status. 723 */ 724 struct inpcb * 725 udp_notify(struct inpcb *inp, int errno) 726 { 727 728 INP_WLOCK_ASSERT(inp); 729 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 730 errno == EHOSTDOWN) && inp->inp_route.ro_nh) { 731 NH_FREE(inp->inp_route.ro_nh); 732 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 733 } 734 735 inp->inp_socket->so_error = errno; 736 sorwakeup(inp->inp_socket); 737 sowwakeup(inp->inp_socket); 738 return (inp); 739 } 740 741 #ifdef INET 742 static void 743 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 744 struct inpcbinfo *pcbinfo) 745 { 746 struct ip *ip = vip; 747 struct udphdr *uh; 748 struct in_addr faddr; 749 struct inpcb *inp; 750 751 faddr = ((struct sockaddr_in *)sa)->sin_addr; 752 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 753 return; 754 755 if (PRC_IS_REDIRECT(cmd)) { 756 /* signal EHOSTDOWN, as it flushes the cached route */ 757 in_pcbnotifyall(pcbinfo, faddr, EHOSTDOWN, udp_notify); 758 return; 759 } 760 761 /* 762 * Hostdead is ugly because it goes linearly through all PCBs. 763 * 764 * XXX: We never get this from ICMP, otherwise it makes an excellent 765 * DoS attack on machines with many connections. 766 */ 767 if (cmd == PRC_HOSTDEAD) 768 ip = NULL; 769 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 770 return; 771 if (ip != NULL) { 772 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 773 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 774 ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL); 775 if (inp != NULL) { 776 INP_WLOCK_ASSERT(inp); 777 if (inp->inp_socket != NULL) { 778 udp_notify(inp, inetctlerrmap[cmd]); 779 } 780 INP_WUNLOCK(inp); 781 } else { 782 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 783 ip->ip_src, uh->uh_sport, 784 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 785 if (inp != NULL) { 786 struct udpcb *up; 787 void *ctx; 788 udp_tun_icmp_t func; 789 790 up = intoudpcb(inp); 791 ctx = up->u_tun_ctx; 792 func = up->u_icmp_func; 793 INP_RUNLOCK(inp); 794 if (func != NULL) 795 (*func)(cmd, sa, vip, ctx); 796 } 797 } 798 } else 799 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 800 udp_notify); 801 } 802 803 static void 804 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 805 { 806 807 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 808 } 809 810 static void 811 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 812 { 813 814 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 815 } 816 #endif /* INET */ 817 818 static int 819 udp_pcblist(SYSCTL_HANDLER_ARGS) 820 { 821 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_udbinfo, 822 INPLOOKUP_RLOCKPCB); 823 struct xinpgen xig; 824 struct inpcb *inp; 825 int error; 826 827 if (req->newptr != 0) 828 return (EPERM); 829 830 if (req->oldptr == 0) { 831 int n; 832 833 n = V_udbinfo.ipi_count; 834 n += imax(n / 8, 10); 835 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 836 return (0); 837 } 838 839 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 840 return (error); 841 842 bzero(&xig, sizeof(xig)); 843 xig.xig_len = sizeof xig; 844 xig.xig_count = V_udbinfo.ipi_count; 845 xig.xig_gen = V_udbinfo.ipi_gencnt; 846 xig.xig_sogen = so_gencnt; 847 error = SYSCTL_OUT(req, &xig, sizeof xig); 848 if (error) 849 return (error); 850 851 while ((inp = inp_next(&inpi)) != NULL) { 852 if (inp->inp_gencnt <= xig.xig_gen && 853 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 854 struct xinpcb xi; 855 856 in_pcbtoxinpcb(inp, &xi); 857 error = SYSCTL_OUT(req, &xi, sizeof xi); 858 if (error) { 859 INP_RUNLOCK(inp); 860 break; 861 } 862 } 863 } 864 865 if (!error) { 866 /* 867 * Give the user an updated idea of our state. If the 868 * generation differs from what we told her before, she knows 869 * that something happened while we were processing this 870 * request, and it might be necessary to retry. 871 */ 872 xig.xig_gen = V_udbinfo.ipi_gencnt; 873 xig.xig_sogen = so_gencnt; 874 xig.xig_count = V_udbinfo.ipi_count; 875 error = SYSCTL_OUT(req, &xig, sizeof xig); 876 } 877 878 return (error); 879 } 880 881 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 882 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 883 udp_pcblist, "S,xinpcb", 884 "List of active UDP sockets"); 885 886 #ifdef INET 887 static int 888 udp_getcred(SYSCTL_HANDLER_ARGS) 889 { 890 struct xucred xuc; 891 struct sockaddr_in addrs[2]; 892 struct epoch_tracker et; 893 struct inpcb *inp; 894 int error; 895 896 error = priv_check(req->td, PRIV_NETINET_GETCRED); 897 if (error) 898 return (error); 899 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 900 if (error) 901 return (error); 902 NET_EPOCH_ENTER(et); 903 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 904 addrs[0].sin_addr, addrs[0].sin_port, 905 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 906 NET_EPOCH_EXIT(et); 907 if (inp != NULL) { 908 INP_RLOCK_ASSERT(inp); 909 if (inp->inp_socket == NULL) 910 error = ENOENT; 911 if (error == 0) 912 error = cr_canseeinpcb(req->td->td_ucred, inp); 913 if (error == 0) 914 cru2x(inp->inp_cred, &xuc); 915 INP_RUNLOCK(inp); 916 } else 917 error = ENOENT; 918 if (error == 0) 919 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 920 return (error); 921 } 922 923 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 924 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, 925 0, 0, udp_getcred, "S,xucred", 926 "Get the xucred of a UDP connection"); 927 #endif /* INET */ 928 929 int 930 udp_ctloutput(struct socket *so, struct sockopt *sopt) 931 { 932 struct inpcb *inp; 933 struct udpcb *up; 934 int isudplite, error, optval; 935 936 error = 0; 937 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 938 inp = sotoinpcb(so); 939 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 940 INP_WLOCK(inp); 941 if (sopt->sopt_level != so->so_proto->pr_protocol) { 942 #ifdef INET6 943 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 944 INP_WUNLOCK(inp); 945 error = ip6_ctloutput(so, sopt); 946 } 947 #endif 948 #if defined(INET) && defined(INET6) 949 else 950 #endif 951 #ifdef INET 952 { 953 INP_WUNLOCK(inp); 954 error = ip_ctloutput(so, sopt); 955 } 956 #endif 957 return (error); 958 } 959 960 switch (sopt->sopt_dir) { 961 case SOPT_SET: 962 switch (sopt->sopt_name) { 963 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 964 #ifdef INET 965 case UDP_ENCAP: 966 if (!IPSEC_ENABLED(ipv4)) { 967 INP_WUNLOCK(inp); 968 return (ENOPROTOOPT); 969 } 970 error = UDPENCAP_PCBCTL(inp, sopt); 971 break; 972 #endif /* INET */ 973 #endif /* IPSEC */ 974 case UDPLITE_SEND_CSCOV: 975 case UDPLITE_RECV_CSCOV: 976 if (!isudplite) { 977 INP_WUNLOCK(inp); 978 error = ENOPROTOOPT; 979 break; 980 } 981 INP_WUNLOCK(inp); 982 error = sooptcopyin(sopt, &optval, sizeof(optval), 983 sizeof(optval)); 984 if (error != 0) 985 break; 986 inp = sotoinpcb(so); 987 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 988 INP_WLOCK(inp); 989 up = intoudpcb(inp); 990 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 991 if ((optval != 0 && optval < 8) || (optval > 65535)) { 992 INP_WUNLOCK(inp); 993 error = EINVAL; 994 break; 995 } 996 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 997 up->u_txcslen = optval; 998 else 999 up->u_rxcslen = optval; 1000 INP_WUNLOCK(inp); 1001 break; 1002 default: 1003 INP_WUNLOCK(inp); 1004 error = ENOPROTOOPT; 1005 break; 1006 } 1007 break; 1008 case SOPT_GET: 1009 switch (sopt->sopt_name) { 1010 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1011 #ifdef INET 1012 case UDP_ENCAP: 1013 if (!IPSEC_ENABLED(ipv4)) { 1014 INP_WUNLOCK(inp); 1015 return (ENOPROTOOPT); 1016 } 1017 error = UDPENCAP_PCBCTL(inp, sopt); 1018 break; 1019 #endif /* INET */ 1020 #endif /* IPSEC */ 1021 case UDPLITE_SEND_CSCOV: 1022 case UDPLITE_RECV_CSCOV: 1023 if (!isudplite) { 1024 INP_WUNLOCK(inp); 1025 error = ENOPROTOOPT; 1026 break; 1027 } 1028 up = intoudpcb(inp); 1029 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1030 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1031 optval = up->u_txcslen; 1032 else 1033 optval = up->u_rxcslen; 1034 INP_WUNLOCK(inp); 1035 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1036 break; 1037 default: 1038 INP_WUNLOCK(inp); 1039 error = ENOPROTOOPT; 1040 break; 1041 } 1042 break; 1043 } 1044 return (error); 1045 } 1046 1047 #ifdef INET 1048 #ifdef INET6 1049 /* The logic here is derived from ip6_setpktopt(). See comments there. */ 1050 static int 1051 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src, 1052 struct inpcb *inp, int flags) 1053 { 1054 struct ifnet *ifp; 1055 struct in6_pktinfo *pktinfo; 1056 struct in_addr ia; 1057 1058 if ((flags & PRUS_IPV6) == 0) 1059 return (0); 1060 1061 if (cm->cmsg_level != IPPROTO_IPV6) 1062 return (0); 1063 1064 if (cm->cmsg_type != IPV6_2292PKTINFO && 1065 cm->cmsg_type != IPV6_PKTINFO) 1066 return (0); 1067 1068 if (cm->cmsg_len != 1069 CMSG_LEN(sizeof(struct in6_pktinfo))) 1070 return (EINVAL); 1071 1072 pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm); 1073 if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) && 1074 !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) 1075 return (EINVAL); 1076 1077 /* Validate the interface index if specified. */ 1078 if (pktinfo->ipi6_ifindex) { 1079 struct epoch_tracker et; 1080 1081 NET_EPOCH_ENTER(et); 1082 ifp = ifnet_byindex(pktinfo->ipi6_ifindex); 1083 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */ 1084 if (ifp == NULL) 1085 return (ENXIO); 1086 } else 1087 ifp = NULL; 1088 if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { 1089 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1090 if (in_ifhasaddr(ifp, ia) == 0) 1091 return (EADDRNOTAVAIL); 1092 } 1093 1094 bzero(src, sizeof(*src)); 1095 src->sin_family = AF_INET; 1096 src->sin_len = sizeof(*src); 1097 src->sin_port = inp->inp_lport; 1098 src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1099 1100 return (0); 1101 } 1102 #endif 1103 1104 static int 1105 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1106 struct mbuf *control, struct thread *td, int flags) 1107 { 1108 struct udpiphdr *ui; 1109 int len = m->m_pkthdr.len; 1110 struct in_addr faddr, laddr; 1111 struct cmsghdr *cm; 1112 struct inpcbinfo *pcbinfo; 1113 struct sockaddr_in *sin, src; 1114 struct epoch_tracker et; 1115 int cscov_partial = 0; 1116 int error = 0; 1117 int ipflags = 0; 1118 u_short fport, lport; 1119 u_char tos; 1120 uint8_t pr; 1121 uint16_t cscov = 0; 1122 uint32_t flowid = 0; 1123 uint8_t flowtype = M_HASHTYPE_NONE; 1124 1125 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1126 if (control) 1127 m_freem(control); 1128 m_freem(m); 1129 return (EMSGSIZE); 1130 } 1131 1132 src.sin_family = 0; 1133 sin = (struct sockaddr_in *)addr; 1134 1135 /* 1136 * udp_output() may need to temporarily bind or connect the current 1137 * inpcb. As such, we don't know up front whether we will need the 1138 * pcbinfo lock or not. Do any work to decide what is needed up 1139 * front before acquiring any locks. 1140 * 1141 * We will need network epoch in either case, to safely lookup into 1142 * pcb hash. 1143 */ 1144 if (sin == NULL || 1145 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) 1146 INP_WLOCK(inp); 1147 else 1148 INP_RLOCK(inp); 1149 NET_EPOCH_ENTER(et); 1150 tos = inp->inp_ip_tos; 1151 if (control != NULL) { 1152 /* 1153 * XXX: Currently, we assume all the optional information is 1154 * stored in a single mbuf. 1155 */ 1156 if (control->m_next) { 1157 m_freem(control); 1158 error = EINVAL; 1159 goto release; 1160 } 1161 for (; control->m_len > 0; 1162 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1163 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1164 cm = mtod(control, struct cmsghdr *); 1165 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1166 || cm->cmsg_len > control->m_len) { 1167 error = EINVAL; 1168 break; 1169 } 1170 #ifdef INET6 1171 error = udp_v4mapped_pktinfo(cm, &src, inp, flags); 1172 if (error != 0) 1173 break; 1174 #endif 1175 if (cm->cmsg_level != IPPROTO_IP) 1176 continue; 1177 1178 switch (cm->cmsg_type) { 1179 case IP_SENDSRCADDR: 1180 if (cm->cmsg_len != 1181 CMSG_LEN(sizeof(struct in_addr))) { 1182 error = EINVAL; 1183 break; 1184 } 1185 bzero(&src, sizeof(src)); 1186 src.sin_family = AF_INET; 1187 src.sin_len = sizeof(src); 1188 src.sin_port = inp->inp_lport; 1189 src.sin_addr = 1190 *(struct in_addr *)CMSG_DATA(cm); 1191 break; 1192 1193 case IP_TOS: 1194 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1195 error = EINVAL; 1196 break; 1197 } 1198 tos = *(u_char *)CMSG_DATA(cm); 1199 break; 1200 1201 case IP_FLOWID: 1202 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1203 error = EINVAL; 1204 break; 1205 } 1206 flowid = *(uint32_t *) CMSG_DATA(cm); 1207 break; 1208 1209 case IP_FLOWTYPE: 1210 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1211 error = EINVAL; 1212 break; 1213 } 1214 flowtype = *(uint32_t *) CMSG_DATA(cm); 1215 break; 1216 1217 #ifdef RSS 1218 case IP_RSSBUCKETID: 1219 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1220 error = EINVAL; 1221 break; 1222 } 1223 /* This is just a placeholder for now */ 1224 break; 1225 #endif /* RSS */ 1226 default: 1227 error = ENOPROTOOPT; 1228 break; 1229 } 1230 if (error) 1231 break; 1232 } 1233 m_freem(control); 1234 control = NULL; 1235 } 1236 if (error) 1237 goto release; 1238 1239 pr = inp->inp_socket->so_proto->pr_protocol; 1240 pcbinfo = udp_get_inpcbinfo(pr); 1241 1242 /* 1243 * If the IP_SENDSRCADDR control message was specified, override the 1244 * source address for this datagram. Its use is invalidated if the 1245 * address thus specified is incomplete or clobbers other inpcbs. 1246 */ 1247 laddr = inp->inp_laddr; 1248 lport = inp->inp_lport; 1249 if (src.sin_family == AF_INET) { 1250 if ((lport == 0) || 1251 (laddr.s_addr == INADDR_ANY && 1252 src.sin_addr.s_addr == INADDR_ANY)) { 1253 error = EINVAL; 1254 goto release; 1255 } 1256 INP_HASH_WLOCK(pcbinfo); 1257 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1258 &laddr.s_addr, &lport, td->td_ucred); 1259 INP_HASH_WUNLOCK(pcbinfo); 1260 if (error) 1261 goto release; 1262 } 1263 1264 /* 1265 * If a UDP socket has been connected, then a local address/port will 1266 * have been selected and bound. 1267 * 1268 * If a UDP socket has not been connected to, then an explicit 1269 * destination address must be used, in which case a local 1270 * address/port may not have been selected and bound. 1271 */ 1272 if (sin != NULL) { 1273 INP_LOCK_ASSERT(inp); 1274 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1275 error = EISCONN; 1276 goto release; 1277 } 1278 1279 /* 1280 * Jail may rewrite the destination address, so let it do 1281 * that before we use it. 1282 */ 1283 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1284 if (error) 1285 goto release; 1286 1287 /* 1288 * If a local address or port hasn't yet been selected, or if 1289 * the destination address needs to be rewritten due to using 1290 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1291 * to do the heavy lifting. Once a port is selected, we 1292 * commit the binding back to the socket; we also commit the 1293 * binding of the address if in jail. 1294 * 1295 * If we already have a valid binding and we're not 1296 * requesting a destination address rewrite, use a fast path. 1297 */ 1298 if (inp->inp_laddr.s_addr == INADDR_ANY || 1299 inp->inp_lport == 0 || 1300 sin->sin_addr.s_addr == INADDR_ANY || 1301 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1302 INP_HASH_WLOCK(pcbinfo); 1303 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1304 &lport, &faddr.s_addr, &fport, NULL, 1305 td->td_ucred); 1306 if (error) { 1307 INP_HASH_WUNLOCK(pcbinfo); 1308 goto release; 1309 } 1310 1311 /* 1312 * XXXRW: Why not commit the port if the address is 1313 * !INADDR_ANY? 1314 */ 1315 /* Commit the local port if newly assigned. */ 1316 if (inp->inp_laddr.s_addr == INADDR_ANY && 1317 inp->inp_lport == 0) { 1318 INP_WLOCK_ASSERT(inp); 1319 /* 1320 * Remember addr if jailed, to prevent 1321 * rebinding. 1322 */ 1323 if (prison_flag(td->td_ucred, PR_IP4)) 1324 inp->inp_laddr = laddr; 1325 inp->inp_lport = lport; 1326 error = in_pcbinshash(inp); 1327 INP_HASH_WUNLOCK(pcbinfo); 1328 if (error != 0) { 1329 inp->inp_lport = 0; 1330 error = EAGAIN; 1331 goto release; 1332 } 1333 inp->inp_flags |= INP_ANONPORT; 1334 } else 1335 INP_HASH_WUNLOCK(pcbinfo); 1336 } else { 1337 faddr = sin->sin_addr; 1338 fport = sin->sin_port; 1339 } 1340 } else { 1341 INP_LOCK_ASSERT(inp); 1342 faddr = inp->inp_faddr; 1343 fport = inp->inp_fport; 1344 if (faddr.s_addr == INADDR_ANY) { 1345 error = ENOTCONN; 1346 goto release; 1347 } 1348 } 1349 1350 /* 1351 * Calculate data length and get a mbuf for UDP, IP, and possible 1352 * link-layer headers. Immediate slide the data pointer back forward 1353 * since we won't use that space at this layer. 1354 */ 1355 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1356 if (m == NULL) { 1357 error = ENOBUFS; 1358 goto release; 1359 } 1360 m->m_data += max_linkhdr; 1361 m->m_len -= max_linkhdr; 1362 m->m_pkthdr.len -= max_linkhdr; 1363 1364 /* 1365 * Fill in mbuf with extended UDP header and addresses and length put 1366 * into network format. 1367 */ 1368 ui = mtod(m, struct udpiphdr *); 1369 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1370 ui->ui_v = IPVERSION << 4; 1371 ui->ui_pr = pr; 1372 ui->ui_src = laddr; 1373 ui->ui_dst = faddr; 1374 ui->ui_sport = lport; 1375 ui->ui_dport = fport; 1376 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1377 if (pr == IPPROTO_UDPLITE) { 1378 struct udpcb *up; 1379 uint16_t plen; 1380 1381 up = intoudpcb(inp); 1382 cscov = up->u_txcslen; 1383 plen = (u_short)len + sizeof(struct udphdr); 1384 if (cscov >= plen) 1385 cscov = 0; 1386 ui->ui_len = htons(plen); 1387 ui->ui_ulen = htons(cscov); 1388 /* 1389 * For UDP-Lite, checksum coverage length of zero means 1390 * the entire UDPLite packet is covered by the checksum. 1391 */ 1392 cscov_partial = (cscov == 0) ? 0 : 1; 1393 } 1394 1395 /* 1396 * Set the Don't Fragment bit in the IP header. 1397 */ 1398 if (inp->inp_flags & INP_DONTFRAG) { 1399 struct ip *ip; 1400 1401 ip = (struct ip *)&ui->ui_i; 1402 ip->ip_off |= htons(IP_DF); 1403 } 1404 1405 if (inp->inp_socket->so_options & SO_DONTROUTE) 1406 ipflags |= IP_ROUTETOIF; 1407 if (inp->inp_socket->so_options & SO_BROADCAST) 1408 ipflags |= IP_ALLOWBROADCAST; 1409 if (inp->inp_flags & INP_ONESBCAST) 1410 ipflags |= IP_SENDONES; 1411 1412 #ifdef MAC 1413 mac_inpcb_create_mbuf(inp, m); 1414 #endif 1415 1416 /* 1417 * Set up checksum and output datagram. 1418 */ 1419 ui->ui_sum = 0; 1420 if (pr == IPPROTO_UDPLITE) { 1421 if (inp->inp_flags & INP_ONESBCAST) 1422 faddr.s_addr = INADDR_BROADCAST; 1423 if (cscov_partial) { 1424 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1425 ui->ui_sum = 0xffff; 1426 } else { 1427 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1428 ui->ui_sum = 0xffff; 1429 } 1430 } else if (V_udp_cksum) { 1431 if (inp->inp_flags & INP_ONESBCAST) 1432 faddr.s_addr = INADDR_BROADCAST; 1433 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1434 htons((u_short)len + sizeof(struct udphdr) + pr)); 1435 m->m_pkthdr.csum_flags = CSUM_UDP; 1436 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1437 } 1438 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1439 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1440 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1441 UDPSTAT_INC(udps_opackets); 1442 1443 /* 1444 * Setup flowid / RSS information for outbound socket. 1445 * 1446 * Once the UDP code decides to set a flowid some other way, 1447 * this allows the flowid to be overridden by userland. 1448 */ 1449 if (flowtype != M_HASHTYPE_NONE) { 1450 m->m_pkthdr.flowid = flowid; 1451 M_HASHTYPE_SET(m, flowtype); 1452 } 1453 #if defined(ROUTE_MPATH) || defined(RSS) 1454 else if (CALC_FLOWID_OUTBOUND_SENDTO) { 1455 uint32_t hash_val, hash_type; 1456 1457 hash_val = fib4_calc_packet_hash(laddr, faddr, 1458 lport, fport, pr, &hash_type); 1459 m->m_pkthdr.flowid = hash_val; 1460 M_HASHTYPE_SET(m, hash_type); 1461 } 1462 1463 /* 1464 * Don't override with the inp cached flowid value. 1465 * 1466 * Depending upon the kind of send being done, the inp 1467 * flowid/flowtype values may actually not be appropriate 1468 * for this particular socket send. 1469 * 1470 * We should either leave the flowid at zero (which is what is 1471 * currently done) or set it to some software generated 1472 * hash value based on the packet contents. 1473 */ 1474 ipflags |= IP_NODEFAULTFLOWID; 1475 #endif /* RSS */ 1476 1477 if (pr == IPPROTO_UDPLITE) 1478 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1479 else 1480 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1481 error = ip_output(m, inp->inp_options, 1482 INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags, 1483 inp->inp_moptions, inp); 1484 INP_UNLOCK(inp); 1485 NET_EPOCH_EXIT(et); 1486 return (error); 1487 1488 release: 1489 INP_UNLOCK(inp); 1490 NET_EPOCH_EXIT(et); 1491 m_freem(m); 1492 return (error); 1493 } 1494 1495 pr_abort_t udp_abort; /* shared with udp6_usrreq.c */ 1496 void 1497 udp_abort(struct socket *so) 1498 { 1499 struct inpcb *inp; 1500 struct inpcbinfo *pcbinfo; 1501 1502 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1503 inp = sotoinpcb(so); 1504 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1505 INP_WLOCK(inp); 1506 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1507 INP_HASH_WLOCK(pcbinfo); 1508 in_pcbdisconnect(inp); 1509 inp->inp_laddr.s_addr = INADDR_ANY; 1510 INP_HASH_WUNLOCK(pcbinfo); 1511 soisdisconnected(so); 1512 } 1513 INP_WUNLOCK(inp); 1514 } 1515 1516 static int 1517 udp_attach(struct socket *so, int proto, struct thread *td) 1518 { 1519 static uint32_t udp_flowid; 1520 struct inpcb *inp; 1521 struct inpcbinfo *pcbinfo; 1522 int error; 1523 1524 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1525 inp = sotoinpcb(so); 1526 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1527 error = soreserve(so, udp_sendspace, udp_recvspace); 1528 if (error) 1529 return (error); 1530 error = in_pcballoc(so, pcbinfo); 1531 if (error) 1532 return (error); 1533 1534 inp = sotoinpcb(so); 1535 inp->inp_ip_ttl = V_ip_defttl; 1536 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1537 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1538 1539 error = udp_newudpcb(inp); 1540 if (error) { 1541 in_pcbdetach(inp); 1542 in_pcbfree(inp); 1543 return (error); 1544 } 1545 INP_WUNLOCK(inp); 1546 1547 return (0); 1548 } 1549 #endif /* INET */ 1550 1551 int 1552 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1553 { 1554 struct inpcb *inp; 1555 struct udpcb *up; 1556 1557 KASSERT(so->so_type == SOCK_DGRAM, 1558 ("udp_set_kernel_tunneling: !dgram")); 1559 inp = sotoinpcb(so); 1560 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1561 INP_WLOCK(inp); 1562 up = intoudpcb(inp); 1563 if ((f != NULL || i != NULL) && ((up->u_tun_func != NULL) || 1564 (up->u_icmp_func != NULL))) { 1565 INP_WUNLOCK(inp); 1566 return (EBUSY); 1567 } 1568 up->u_tun_func = f; 1569 up->u_icmp_func = i; 1570 up->u_tun_ctx = ctx; 1571 INP_WUNLOCK(inp); 1572 return (0); 1573 } 1574 1575 #ifdef INET 1576 static int 1577 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1578 { 1579 struct inpcb *inp; 1580 struct inpcbinfo *pcbinfo; 1581 struct sockaddr_in *sinp; 1582 int error; 1583 1584 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1585 inp = sotoinpcb(so); 1586 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1587 1588 sinp = (struct sockaddr_in *)nam; 1589 if (nam->sa_family != AF_INET) { 1590 /* 1591 * Preserve compatibility with old programs. 1592 */ 1593 if (nam->sa_family != AF_UNSPEC || 1594 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 1595 sinp->sin_addr.s_addr != INADDR_ANY) 1596 return (EAFNOSUPPORT); 1597 nam->sa_family = AF_INET; 1598 } 1599 if (nam->sa_len != sizeof(struct sockaddr_in)) 1600 return (EINVAL); 1601 1602 INP_WLOCK(inp); 1603 INP_HASH_WLOCK(pcbinfo); 1604 error = in_pcbbind(inp, nam, 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->inp_laddr.s_addr = INADDR_ANY; 1624 INP_HASH_WUNLOCK(pcbinfo); 1625 soisdisconnected(so); 1626 } 1627 INP_WUNLOCK(inp); 1628 } 1629 1630 static int 1631 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1632 { 1633 struct epoch_tracker et; 1634 struct inpcb *inp; 1635 struct inpcbinfo *pcbinfo; 1636 struct sockaddr_in *sin; 1637 int error; 1638 1639 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1640 inp = sotoinpcb(so); 1641 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1642 1643 sin = (struct sockaddr_in *)nam; 1644 if (sin->sin_family != AF_INET) 1645 return (EAFNOSUPPORT); 1646 if (sin->sin_len != sizeof(*sin)) 1647 return (EINVAL); 1648 1649 INP_WLOCK(inp); 1650 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1651 INP_WUNLOCK(inp); 1652 return (EISCONN); 1653 } 1654 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1655 if (error != 0) { 1656 INP_WUNLOCK(inp); 1657 return (error); 1658 } 1659 NET_EPOCH_ENTER(et); 1660 INP_HASH_WLOCK(pcbinfo); 1661 error = in_pcbconnect(inp, nam, td->td_ucred, true); 1662 INP_HASH_WUNLOCK(pcbinfo); 1663 NET_EPOCH_EXIT(et); 1664 if (error == 0) 1665 soisconnected(so); 1666 INP_WUNLOCK(inp); 1667 return (error); 1668 } 1669 1670 static void 1671 udp_detach(struct socket *so) 1672 { 1673 struct inpcb *inp; 1674 struct udpcb *up; 1675 1676 inp = sotoinpcb(so); 1677 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1678 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1679 ("udp_detach: not disconnected")); 1680 INP_WLOCK(inp); 1681 up = intoudpcb(inp); 1682 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1683 inp->inp_ppcb = NULL; 1684 in_pcbdetach(inp); 1685 in_pcbfree(inp); 1686 udp_discardcb(up); 1687 } 1688 1689 pr_disconnect_t udp_disconnect; /* shared with udp6_usrreq.c */ 1690 int 1691 udp_disconnect(struct socket *so) 1692 { 1693 struct inpcb *inp; 1694 struct inpcbinfo *pcbinfo; 1695 1696 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1697 inp = sotoinpcb(so); 1698 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1699 INP_WLOCK(inp); 1700 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1701 INP_WUNLOCK(inp); 1702 return (ENOTCONN); 1703 } 1704 INP_HASH_WLOCK(pcbinfo); 1705 in_pcbdisconnect(inp); 1706 inp->inp_laddr.s_addr = INADDR_ANY; 1707 INP_HASH_WUNLOCK(pcbinfo); 1708 SOCK_LOCK(so); 1709 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1710 SOCK_UNLOCK(so); 1711 INP_WUNLOCK(inp); 1712 return (0); 1713 } 1714 1715 pr_send_t udp_send; /* shared with udp6_usrreq.c */ 1716 int 1717 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1718 struct mbuf *control, struct thread *td) 1719 { 1720 struct inpcb *inp; 1721 int error; 1722 1723 inp = sotoinpcb(so); 1724 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1725 1726 if (addr != NULL) { 1727 error = 0; 1728 if (addr->sa_family != AF_INET) 1729 error = EAFNOSUPPORT; 1730 else if (addr->sa_len != sizeof(struct sockaddr_in)) 1731 error = EINVAL; 1732 if (__predict_false(error != 0)) { 1733 m_freem(control); 1734 m_freem(m); 1735 return (error); 1736 } 1737 } 1738 return (udp_output(inp, m, addr, control, td, flags)); 1739 } 1740 #endif /* INET */ 1741 1742 int 1743 udp_shutdown(struct socket *so) 1744 { 1745 struct inpcb *inp; 1746 1747 inp = sotoinpcb(so); 1748 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1749 INP_WLOCK(inp); 1750 socantsendmore(so); 1751 INP_WUNLOCK(inp); 1752 return (0); 1753 } 1754 1755 #ifdef INET 1756 #define UDP_PROTOSW \ 1757 .pr_type = SOCK_DGRAM, \ 1758 .pr_flags = PR_ATOMIC | PR_ADDR | PR_CAPATTACH, \ 1759 .pr_ctloutput = udp_ctloutput, \ 1760 .pr_abort = udp_abort, \ 1761 .pr_attach = udp_attach, \ 1762 .pr_bind = udp_bind, \ 1763 .pr_connect = udp_connect, \ 1764 .pr_control = in_control, \ 1765 .pr_detach = udp_detach, \ 1766 .pr_disconnect = udp_disconnect, \ 1767 .pr_peeraddr = in_getpeeraddr, \ 1768 .pr_send = udp_send, \ 1769 .pr_soreceive = soreceive_dgram, \ 1770 .pr_sosend = sosend_dgram, \ 1771 .pr_shutdown = udp_shutdown, \ 1772 .pr_sockaddr = in_getsockaddr, \ 1773 .pr_sosetlabel = in_pcbsosetlabel, \ 1774 .pr_close = udp_close 1775 1776 struct protosw udp_protosw = { 1777 .pr_protocol = IPPROTO_UDP, 1778 UDP_PROTOSW 1779 }; 1780 1781 struct protosw udplite_protosw = { 1782 .pr_protocol = IPPROTO_UDPLITE, 1783 UDP_PROTOSW 1784 }; 1785 1786 static void 1787 udp_init(void *arg __unused) 1788 { 1789 1790 IPPROTO_REGISTER(IPPROTO_UDP, udp_input, udp_ctlinput); 1791 IPPROTO_REGISTER(IPPROTO_UDPLITE, udp_input, udplite_ctlinput); 1792 } 1793 SYSINIT(udp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, udp_init, NULL); 1794 #endif /* INET */ 1795