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 inpcbhead, udb); /* from udp_var.h */ 151 VNET_DEFINE(struct inpcbinfo, udbinfo); 152 VNET_DEFINE(struct inpcbhead, ulitecb); 153 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 154 VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone); 155 #define V_udpcb_zone VNET(udpcb_zone) 156 157 #ifndef UDBHASHSIZE 158 #define UDBHASHSIZE 128 159 #endif 160 161 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 162 VNET_PCPUSTAT_SYSINIT(udpstat); 163 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 164 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 165 166 #ifdef VIMAGE 167 VNET_PCPUSTAT_SYSUNINIT(udpstat); 168 #endif /* VIMAGE */ 169 #ifdef INET 170 static void udp_detach(struct socket *so); 171 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 172 struct mbuf *, struct thread *, int); 173 #endif 174 175 static void 176 udp_zone_change(void *tag) 177 { 178 179 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 180 uma_zone_set_max(V_udpcb_zone, maxsockets); 181 } 182 183 static int 184 udp_inpcb_init(void *mem, int size, int flags) 185 { 186 struct inpcb *inp; 187 188 inp = mem; 189 INP_LOCK_INIT(inp, "inp", "udpinp"); 190 return (0); 191 } 192 193 static int 194 udplite_inpcb_init(void *mem, int size, int flags) 195 { 196 struct inpcb *inp; 197 198 inp = mem; 199 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 200 return (0); 201 } 202 203 void 204 udp_init(void) 205 { 206 207 /* 208 * For now default to 2-tuple UDP hashing - until the fragment 209 * reassembly code can also update the flowid. 210 * 211 * Once we can calculate the flowid that way and re-establish 212 * a 4-tuple, flip this to 4-tuple. 213 */ 214 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 215 "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE); 216 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 217 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 218 uma_zone_set_max(V_udpcb_zone, maxsockets); 219 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 220 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 221 EVENTHANDLER_PRI_ANY); 222 } 223 224 void 225 udplite_init(void) 226 { 227 228 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 229 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, 230 IPI_HASHFIELDS_2TUPLE); 231 } 232 233 /* 234 * Kernel module interface for updating udpstat. The argument is an index 235 * into udpstat treated as an array of u_long. While this encodes the 236 * general layout of udpstat into the caller, it doesn't encode its location, 237 * so that future changes to add, for example, per-CPU stats support won't 238 * cause binary compatibility problems for kernel modules. 239 */ 240 void 241 kmod_udpstat_inc(int statnum) 242 { 243 244 counter_u64_add(VNET(udpstat)[statnum], 1); 245 } 246 247 int 248 udp_newudpcb(struct inpcb *inp) 249 { 250 struct udpcb *up; 251 252 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 253 if (up == NULL) 254 return (ENOBUFS); 255 inp->inp_ppcb = up; 256 return (0); 257 } 258 259 void 260 udp_discardcb(struct udpcb *up) 261 { 262 263 uma_zfree(V_udpcb_zone, up); 264 } 265 266 #ifdef VIMAGE 267 static void 268 udp_destroy(void *unused __unused) 269 { 270 271 in_pcbinfo_destroy(&V_udbinfo); 272 uma_zdestroy(V_udpcb_zone); 273 } 274 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 275 276 static void 277 udplite_destroy(void *unused __unused) 278 { 279 280 in_pcbinfo_destroy(&V_ulitecbinfo); 281 } 282 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 283 NULL); 284 #endif 285 286 #ifdef INET 287 /* 288 * Subroutine of udp_input(), which appends the provided mbuf chain to the 289 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 290 * contains the source address. If the socket ends up being an IPv6 socket, 291 * udp_append() will convert to a sockaddr_in6 before passing the address 292 * into the socket code. 293 * 294 * In the normal case udp_append() will return 0, indicating that you 295 * must unlock the inp. However if a tunneling protocol is in place we increment 296 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 297 * then decrement the reference count. If the inp_rele returns 1, indicating the 298 * inp is gone, we return that to the caller to tell them *not* to unlock 299 * the inp. In the case of multi-cast this will cause the distribution 300 * to stop (though most tunneling protocols known currently do *not* use 301 * multicast). 302 */ 303 static int 304 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 305 struct sockaddr_in *udp_in) 306 { 307 struct sockaddr *append_sa; 308 struct socket *so; 309 struct mbuf *tmpopts, *opts = NULL; 310 #ifdef INET6 311 struct sockaddr_in6 udp_in6; 312 #endif 313 struct udpcb *up; 314 315 INP_LOCK_ASSERT(inp); 316 317 /* 318 * Engage the tunneling protocol. 319 */ 320 up = intoudpcb(inp); 321 if (up->u_tun_func != NULL) { 322 in_pcbref(inp); 323 INP_RUNLOCK(inp); 324 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], 325 up->u_tun_ctx); 326 INP_RLOCK(inp); 327 return (in_pcbrele_rlocked(inp)); 328 } 329 330 off += sizeof(struct udphdr); 331 332 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 333 /* Check AH/ESP integrity. */ 334 if (IPSEC_ENABLED(ipv4) && 335 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 336 m_freem(n); 337 return (0); 338 } 339 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 340 if (IPSEC_ENABLED(ipv4) && 341 UDPENCAP_INPUT(n, off, AF_INET) != 0) 342 return (0); /* Consumed. */ 343 } 344 #endif /* IPSEC */ 345 #ifdef MAC 346 if (mac_inpcb_check_deliver(inp, n) != 0) { 347 m_freem(n); 348 return (0); 349 } 350 #endif /* MAC */ 351 if (inp->inp_flags & INP_CONTROLOPTS || 352 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 353 #ifdef INET6 354 if (inp->inp_vflag & INP_IPV6) 355 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 356 else 357 #endif /* INET6 */ 358 ip_savecontrol(inp, &opts, ip, n); 359 } 360 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 361 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], 362 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); 363 if (tmpopts) { 364 if (opts) { 365 tmpopts->m_next = opts; 366 opts = tmpopts; 367 } else 368 opts = tmpopts; 369 } 370 } 371 #ifdef INET6 372 if (inp->inp_vflag & INP_IPV6) { 373 bzero(&udp_in6, sizeof(udp_in6)); 374 udp_in6.sin6_len = sizeof(udp_in6); 375 udp_in6.sin6_family = AF_INET6; 376 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 377 append_sa = (struct sockaddr *)&udp_in6; 378 } else 379 #endif /* INET6 */ 380 append_sa = (struct sockaddr *)&udp_in[0]; 381 m_adj(n, off); 382 383 so = inp->inp_socket; 384 SOCKBUF_LOCK(&so->so_rcv); 385 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 386 soroverflow_locked(so); 387 m_freem(n); 388 if (opts) 389 m_freem(opts); 390 UDPSTAT_INC(udps_fullsock); 391 } else 392 sorwakeup_locked(so); 393 return (0); 394 } 395 396 int 397 udp_input(struct mbuf **mp, int *offp, int proto) 398 { 399 struct ip *ip; 400 struct udphdr *uh; 401 struct ifnet *ifp; 402 struct inpcb *inp; 403 uint16_t len, ip_len; 404 struct inpcbinfo *pcbinfo; 405 struct sockaddr_in udp_in[2]; 406 struct mbuf *m; 407 struct m_tag *fwd_tag; 408 int cscov_partial, iphlen; 409 410 m = *mp; 411 iphlen = *offp; 412 ifp = m->m_pkthdr.rcvif; 413 *mp = NULL; 414 UDPSTAT_INC(udps_ipackets); 415 416 /* 417 * Strip IP options, if any; should skip this, make available to 418 * user, and use on returned packets, but we don't yet have a way to 419 * check the checksum with options still present. 420 */ 421 if (iphlen > sizeof (struct ip)) { 422 ip_stripoptions(m); 423 iphlen = sizeof(struct ip); 424 } 425 426 /* 427 * Get IP and UDP header together in first mbuf. 428 */ 429 if (m->m_len < iphlen + sizeof(struct udphdr)) { 430 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 431 UDPSTAT_INC(udps_hdrops); 432 return (IPPROTO_DONE); 433 } 434 } 435 ip = mtod(m, struct ip *); 436 uh = (struct udphdr *)((caddr_t)ip + iphlen); 437 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 438 439 /* 440 * Destination port of 0 is illegal, based on RFC768. 441 */ 442 if (uh->uh_dport == 0) 443 goto badunlocked; 444 445 /* 446 * Construct sockaddr format source address. Stuff source address 447 * and datagram in user buffer. 448 */ 449 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 450 udp_in[0].sin_len = sizeof(struct sockaddr_in); 451 udp_in[0].sin_family = AF_INET; 452 udp_in[0].sin_port = uh->uh_sport; 453 udp_in[0].sin_addr = ip->ip_src; 454 udp_in[1].sin_len = sizeof(struct sockaddr_in); 455 udp_in[1].sin_family = AF_INET; 456 udp_in[1].sin_port = uh->uh_dport; 457 udp_in[1].sin_addr = ip->ip_dst; 458 459 /* 460 * Make mbuf data length reflect UDP length. If not enough data to 461 * reflect UDP length, drop. 462 */ 463 len = ntohs((u_short)uh->uh_ulen); 464 ip_len = ntohs(ip->ip_len) - iphlen; 465 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 466 /* Zero means checksum over the complete packet. */ 467 if (len == 0) 468 len = ip_len; 469 cscov_partial = 0; 470 } 471 if (ip_len != len) { 472 if (len > ip_len || len < sizeof(struct udphdr)) { 473 UDPSTAT_INC(udps_badlen); 474 goto badunlocked; 475 } 476 if (proto == IPPROTO_UDP) 477 m_adj(m, len - ip_len); 478 } 479 480 /* 481 * Checksum extended UDP header and data. 482 */ 483 if (uh->uh_sum) { 484 u_short uh_sum; 485 486 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 487 !cscov_partial) { 488 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 489 uh_sum = m->m_pkthdr.csum_data; 490 else 491 uh_sum = in_pseudo(ip->ip_src.s_addr, 492 ip->ip_dst.s_addr, htonl((u_short)len + 493 m->m_pkthdr.csum_data + proto)); 494 uh_sum ^= 0xffff; 495 } else { 496 char b[offsetof(struct ipovly, ih_src)]; 497 struct ipovly *ipov = (struct ipovly *)ip; 498 499 bcopy(ipov, b, sizeof(b)); 500 bzero(ipov, sizeof(ipov->ih_x1)); 501 ipov->ih_len = (proto == IPPROTO_UDP) ? 502 uh->uh_ulen : htons(ip_len); 503 uh_sum = in_cksum(m, len + sizeof (struct ip)); 504 bcopy(b, ipov, sizeof(b)); 505 } 506 if (uh_sum) { 507 UDPSTAT_INC(udps_badsum); 508 m_freem(m); 509 return (IPPROTO_DONE); 510 } 511 } else { 512 if (proto == IPPROTO_UDP) { 513 UDPSTAT_INC(udps_nosum); 514 } else { 515 /* UDPLite requires a checksum */ 516 /* XXX: What is the right UDPLite MIB counter here? */ 517 m_freem(m); 518 return (IPPROTO_DONE); 519 } 520 } 521 522 pcbinfo = udp_get_inpcbinfo(proto); 523 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 524 in_broadcast(ip->ip_dst, ifp)) { 525 struct inpcb *last; 526 struct inpcbhead *pcblist; 527 528 NET_EPOCH_ASSERT(); 529 530 pcblist = udp_get_pcblist(proto); 531 last = NULL; 532 CK_LIST_FOREACH(inp, pcblist, inp_list) { 533 if (inp->inp_lport != uh->uh_dport) 534 continue; 535 #ifdef INET6 536 if ((inp->inp_vflag & INP_IPV4) == 0) 537 continue; 538 #endif 539 if (inp->inp_laddr.s_addr != INADDR_ANY && 540 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 541 continue; 542 if (inp->inp_faddr.s_addr != INADDR_ANY && 543 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 544 continue; 545 if (inp->inp_fport != 0 && 546 inp->inp_fport != uh->uh_sport) 547 continue; 548 549 INP_RLOCK(inp); 550 551 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 552 INP_RUNLOCK(inp); 553 continue; 554 } 555 556 /* 557 * XXXRW: Because we weren't holding either the inpcb 558 * or the hash lock when we checked for a match 559 * before, we should probably recheck now that the 560 * inpcb lock is held. 561 */ 562 563 /* 564 * Handle socket delivery policy for any-source 565 * and source-specific multicast. [RFC3678] 566 */ 567 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 568 struct ip_moptions *imo; 569 struct sockaddr_in group; 570 int blocked; 571 572 imo = inp->inp_moptions; 573 if (imo == NULL) { 574 INP_RUNLOCK(inp); 575 continue; 576 } 577 bzero(&group, sizeof(struct sockaddr_in)); 578 group.sin_len = sizeof(struct sockaddr_in); 579 group.sin_family = AF_INET; 580 group.sin_addr = ip->ip_dst; 581 582 blocked = imo_multi_filter(imo, ifp, 583 (struct sockaddr *)&group, 584 (struct sockaddr *)&udp_in[0]); 585 if (blocked != MCAST_PASS) { 586 if (blocked == MCAST_NOTGMEMBER) 587 IPSTAT_INC(ips_notmember); 588 if (blocked == MCAST_NOTSMEMBER || 589 blocked == MCAST_MUTED) 590 UDPSTAT_INC(udps_filtermcast); 591 INP_RUNLOCK(inp); 592 continue; 593 } 594 } 595 if (last != NULL) { 596 struct mbuf *n; 597 598 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 599 NULL) { 600 if (proto == IPPROTO_UDPLITE) 601 UDPLITE_PROBE(receive, NULL, last, ip, 602 last, uh); 603 else 604 UDP_PROBE(receive, NULL, last, ip, last, 605 uh); 606 if (udp_append(last, ip, n, iphlen, 607 udp_in)) { 608 INP_RUNLOCK(inp); 609 goto badunlocked; 610 } 611 } 612 /* Release PCB lock taken on previous pass. */ 613 INP_RUNLOCK(last); 614 } 615 last = inp; 616 /* 617 * Don't look for additional matches if this one does 618 * not have either the SO_REUSEPORT or SO_REUSEADDR 619 * socket options set. This heuristic avoids 620 * searching through all pcbs in the common case of a 621 * non-shared port. It assumes that an application 622 * will never clear these options after setting them. 623 */ 624 if ((last->inp_socket->so_options & 625 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) 626 break; 627 } 628 629 if (last == NULL) { 630 /* 631 * No matching pcb found; discard datagram. (No need 632 * to send an ICMP Port Unreachable for a broadcast 633 * or multicast datgram.) 634 */ 635 UDPSTAT_INC(udps_noport); 636 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 637 UDPSTAT_INC(udps_noportmcast); 638 else 639 UDPSTAT_INC(udps_noportbcast); 640 goto badunlocked; 641 } 642 if (proto == IPPROTO_UDPLITE) 643 UDPLITE_PROBE(receive, NULL, last, ip, last, uh); 644 else 645 UDP_PROBE(receive, NULL, last, ip, last, uh); 646 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 647 INP_RUNLOCK(last); 648 return (IPPROTO_DONE); 649 } 650 651 /* 652 * Locate pcb for datagram. 653 */ 654 655 /* 656 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 657 */ 658 if ((m->m_flags & M_IP_NEXTHOP) && 659 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 660 struct sockaddr_in *next_hop; 661 662 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 663 664 /* 665 * Transparently forwarded. Pretend to be the destination. 666 * Already got one like this? 667 */ 668 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 669 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 670 if (!inp) { 671 /* 672 * It's new. Try to find the ambushing socket. 673 * Because we've rewritten the destination address, 674 * any hardware-generated hash is ignored. 675 */ 676 inp = in_pcblookup(pcbinfo, ip->ip_src, 677 uh->uh_sport, next_hop->sin_addr, 678 next_hop->sin_port ? htons(next_hop->sin_port) : 679 uh->uh_dport, INPLOOKUP_WILDCARD | 680 INPLOOKUP_RLOCKPCB, ifp); 681 } 682 /* Remove the tag from the packet. We don't need it anymore. */ 683 m_tag_delete(m, fwd_tag); 684 m->m_flags &= ~M_IP_NEXTHOP; 685 } else 686 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 687 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 688 INPLOOKUP_RLOCKPCB, ifp, m); 689 if (inp == NULL) { 690 if (V_udp_log_in_vain) { 691 char src[INET_ADDRSTRLEN]; 692 char dst[INET_ADDRSTRLEN]; 693 694 log(LOG_INFO, 695 "Connection attempt to UDP %s:%d from %s:%d\n", 696 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 697 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 698 } 699 if (proto == IPPROTO_UDPLITE) 700 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); 701 else 702 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 703 UDPSTAT_INC(udps_noport); 704 if (m->m_flags & (M_BCAST | M_MCAST)) { 705 UDPSTAT_INC(udps_noportbcast); 706 goto badunlocked; 707 } 708 if (V_udp_blackhole && (V_udp_blackhole_local || 709 !in_localip(ip->ip_src))) 710 goto badunlocked; 711 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 712 goto badunlocked; 713 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 714 return (IPPROTO_DONE); 715 } 716 717 /* 718 * Check the minimum TTL for socket. 719 */ 720 INP_RLOCK_ASSERT(inp); 721 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 722 if (proto == IPPROTO_UDPLITE) 723 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 724 else 725 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 726 INP_RUNLOCK(inp); 727 m_freem(m); 728 return (IPPROTO_DONE); 729 } 730 if (cscov_partial) { 731 struct udpcb *up; 732 733 up = intoudpcb(inp); 734 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 735 INP_RUNLOCK(inp); 736 m_freem(m); 737 return (IPPROTO_DONE); 738 } 739 } 740 741 if (proto == IPPROTO_UDPLITE) 742 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 743 else 744 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 745 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 746 INP_RUNLOCK(inp); 747 return (IPPROTO_DONE); 748 749 badunlocked: 750 m_freem(m); 751 return (IPPROTO_DONE); 752 } 753 #endif /* INET */ 754 755 /* 756 * Notify a udp user of an asynchronous error; just wake up so that they can 757 * collect error status. 758 */ 759 struct inpcb * 760 udp_notify(struct inpcb *inp, int errno) 761 { 762 763 INP_WLOCK_ASSERT(inp); 764 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 765 errno == EHOSTDOWN) && inp->inp_route.ro_nh) { 766 NH_FREE(inp->inp_route.ro_nh); 767 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 768 } 769 770 inp->inp_socket->so_error = errno; 771 sorwakeup(inp->inp_socket); 772 sowwakeup(inp->inp_socket); 773 return (inp); 774 } 775 776 #ifdef INET 777 static void 778 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 779 struct inpcbinfo *pcbinfo) 780 { 781 struct ip *ip = vip; 782 struct udphdr *uh; 783 struct in_addr faddr; 784 struct inpcb *inp; 785 786 faddr = ((struct sockaddr_in *)sa)->sin_addr; 787 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 788 return; 789 790 if (PRC_IS_REDIRECT(cmd)) { 791 /* signal EHOSTDOWN, as it flushes the cached route */ 792 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 793 return; 794 } 795 796 /* 797 * Hostdead is ugly because it goes linearly through all PCBs. 798 * 799 * XXX: We never get this from ICMP, otherwise it makes an excellent 800 * DoS attack on machines with many connections. 801 */ 802 if (cmd == PRC_HOSTDEAD) 803 ip = NULL; 804 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 805 return; 806 if (ip != NULL) { 807 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 808 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 809 ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL); 810 if (inp != NULL) { 811 INP_WLOCK_ASSERT(inp); 812 if (inp->inp_socket != NULL) { 813 udp_notify(inp, inetctlerrmap[cmd]); 814 } 815 INP_WUNLOCK(inp); 816 } else { 817 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 818 ip->ip_src, uh->uh_sport, 819 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 820 if (inp != NULL) { 821 struct udpcb *up; 822 void *ctx; 823 udp_tun_icmp_t func; 824 825 up = intoudpcb(inp); 826 ctx = up->u_tun_ctx; 827 func = up->u_icmp_func; 828 INP_RUNLOCK(inp); 829 if (func != NULL) 830 (*func)(cmd, sa, vip, ctx); 831 } 832 } 833 } else 834 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 835 udp_notify); 836 } 837 void 838 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 839 { 840 841 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 842 } 843 844 void 845 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 846 { 847 848 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 849 } 850 #endif /* INET */ 851 852 static int 853 udp_pcblist(SYSCTL_HANDLER_ARGS) 854 { 855 struct xinpgen xig; 856 struct epoch_tracker et; 857 struct inpcb *inp; 858 int error; 859 860 if (req->newptr != 0) 861 return (EPERM); 862 863 if (req->oldptr == 0) { 864 int n; 865 866 n = V_udbinfo.ipi_count; 867 n += imax(n / 8, 10); 868 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 869 return (0); 870 } 871 872 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 873 return (error); 874 875 bzero(&xig, sizeof(xig)); 876 xig.xig_len = sizeof xig; 877 xig.xig_count = V_udbinfo.ipi_count; 878 xig.xig_gen = V_udbinfo.ipi_gencnt; 879 xig.xig_sogen = so_gencnt; 880 error = SYSCTL_OUT(req, &xig, sizeof xig); 881 if (error) 882 return (error); 883 884 NET_EPOCH_ENTER(et); 885 for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead); 886 inp != NULL; 887 inp = CK_LIST_NEXT(inp, inp_list)) { 888 INP_RLOCK(inp); 889 if (inp->inp_gencnt <= xig.xig_gen && 890 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 891 struct xinpcb xi; 892 893 in_pcbtoxinpcb(inp, &xi); 894 INP_RUNLOCK(inp); 895 error = SYSCTL_OUT(req, &xi, sizeof xi); 896 if (error) 897 break; 898 } else 899 INP_RUNLOCK(inp); 900 } 901 NET_EPOCH_EXIT(et); 902 903 if (!error) { 904 /* 905 * Give the user an updated idea of our state. If the 906 * generation differs from what we told her before, she knows 907 * that something happened while we were processing this 908 * request, and it might be necessary to retry. 909 */ 910 xig.xig_gen = V_udbinfo.ipi_gencnt; 911 xig.xig_sogen = so_gencnt; 912 xig.xig_count = V_udbinfo.ipi_count; 913 error = SYSCTL_OUT(req, &xig, sizeof xig); 914 } 915 916 return (error); 917 } 918 919 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 920 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 921 udp_pcblist, "S,xinpcb", 922 "List of active UDP sockets"); 923 924 #ifdef INET 925 static int 926 udp_getcred(SYSCTL_HANDLER_ARGS) 927 { 928 struct xucred xuc; 929 struct sockaddr_in addrs[2]; 930 struct epoch_tracker et; 931 struct inpcb *inp; 932 int error; 933 934 error = priv_check(req->td, PRIV_NETINET_GETCRED); 935 if (error) 936 return (error); 937 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 938 if (error) 939 return (error); 940 NET_EPOCH_ENTER(et); 941 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 942 addrs[0].sin_addr, addrs[0].sin_port, 943 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 944 NET_EPOCH_EXIT(et); 945 if (inp != NULL) { 946 INP_RLOCK_ASSERT(inp); 947 if (inp->inp_socket == NULL) 948 error = ENOENT; 949 if (error == 0) 950 error = cr_canseeinpcb(req->td->td_ucred, inp); 951 if (error == 0) 952 cru2x(inp->inp_cred, &xuc); 953 INP_RUNLOCK(inp); 954 } else 955 error = ENOENT; 956 if (error == 0) 957 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 958 return (error); 959 } 960 961 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 962 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE, 963 0, 0, udp_getcred, "S,xucred", 964 "Get the xucred of a UDP connection"); 965 #endif /* INET */ 966 967 int 968 udp_ctloutput(struct socket *so, struct sockopt *sopt) 969 { 970 struct inpcb *inp; 971 struct udpcb *up; 972 int isudplite, error, optval; 973 974 error = 0; 975 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 976 inp = sotoinpcb(so); 977 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 978 INP_WLOCK(inp); 979 if (sopt->sopt_level != so->so_proto->pr_protocol) { 980 #ifdef INET6 981 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 982 INP_WUNLOCK(inp); 983 error = ip6_ctloutput(so, sopt); 984 } 985 #endif 986 #if defined(INET) && defined(INET6) 987 else 988 #endif 989 #ifdef INET 990 { 991 INP_WUNLOCK(inp); 992 error = ip_ctloutput(so, sopt); 993 } 994 #endif 995 return (error); 996 } 997 998 switch (sopt->sopt_dir) { 999 case SOPT_SET: 1000 switch (sopt->sopt_name) { 1001 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1002 #ifdef INET 1003 case UDP_ENCAP: 1004 if (!IPSEC_ENABLED(ipv4)) { 1005 INP_WUNLOCK(inp); 1006 return (ENOPROTOOPT); 1007 } 1008 error = UDPENCAP_PCBCTL(inp, sopt); 1009 break; 1010 #endif /* INET */ 1011 #endif /* IPSEC */ 1012 case UDPLITE_SEND_CSCOV: 1013 case UDPLITE_RECV_CSCOV: 1014 if (!isudplite) { 1015 INP_WUNLOCK(inp); 1016 error = ENOPROTOOPT; 1017 break; 1018 } 1019 INP_WUNLOCK(inp); 1020 error = sooptcopyin(sopt, &optval, sizeof(optval), 1021 sizeof(optval)); 1022 if (error != 0) 1023 break; 1024 inp = sotoinpcb(so); 1025 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1026 INP_WLOCK(inp); 1027 up = intoudpcb(inp); 1028 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1029 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1030 INP_WUNLOCK(inp); 1031 error = EINVAL; 1032 break; 1033 } 1034 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1035 up->u_txcslen = optval; 1036 else 1037 up->u_rxcslen = optval; 1038 INP_WUNLOCK(inp); 1039 break; 1040 default: 1041 INP_WUNLOCK(inp); 1042 error = ENOPROTOOPT; 1043 break; 1044 } 1045 break; 1046 case SOPT_GET: 1047 switch (sopt->sopt_name) { 1048 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1049 #ifdef INET 1050 case UDP_ENCAP: 1051 if (!IPSEC_ENABLED(ipv4)) { 1052 INP_WUNLOCK(inp); 1053 return (ENOPROTOOPT); 1054 } 1055 error = UDPENCAP_PCBCTL(inp, sopt); 1056 break; 1057 #endif /* INET */ 1058 #endif /* IPSEC */ 1059 case UDPLITE_SEND_CSCOV: 1060 case UDPLITE_RECV_CSCOV: 1061 if (!isudplite) { 1062 INP_WUNLOCK(inp); 1063 error = ENOPROTOOPT; 1064 break; 1065 } 1066 up = intoudpcb(inp); 1067 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1068 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1069 optval = up->u_txcslen; 1070 else 1071 optval = up->u_rxcslen; 1072 INP_WUNLOCK(inp); 1073 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1074 break; 1075 default: 1076 INP_WUNLOCK(inp); 1077 error = ENOPROTOOPT; 1078 break; 1079 } 1080 break; 1081 } 1082 return (error); 1083 } 1084 1085 #ifdef INET 1086 #ifdef INET6 1087 /* The logic here is derived from ip6_setpktopt(). See comments there. */ 1088 static int 1089 udp_v4mapped_pktinfo(struct cmsghdr *cm, struct sockaddr_in * src, 1090 struct inpcb *inp, int flags) 1091 { 1092 struct ifnet *ifp; 1093 struct in6_pktinfo *pktinfo; 1094 struct in_addr ia; 1095 1096 if ((flags & PRUS_IPV6) == 0) 1097 return (0); 1098 1099 if (cm->cmsg_level != IPPROTO_IPV6) 1100 return (0); 1101 1102 if (cm->cmsg_type != IPV6_2292PKTINFO && 1103 cm->cmsg_type != IPV6_PKTINFO) 1104 return (0); 1105 1106 if (cm->cmsg_len != 1107 CMSG_LEN(sizeof(struct in6_pktinfo))) 1108 return (EINVAL); 1109 1110 pktinfo = (struct in6_pktinfo *)CMSG_DATA(cm); 1111 if (!IN6_IS_ADDR_V4MAPPED(&pktinfo->ipi6_addr) && 1112 !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) 1113 return (EINVAL); 1114 1115 /* Validate the interface index if specified. */ 1116 if (pktinfo->ipi6_ifindex > V_if_index) 1117 return (ENXIO); 1118 1119 ifp = NULL; 1120 if (pktinfo->ipi6_ifindex) { 1121 ifp = ifnet_byindex(pktinfo->ipi6_ifindex); 1122 if (ifp == NULL) 1123 return (ENXIO); 1124 } 1125 if (ifp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&pktinfo->ipi6_addr)) { 1126 ia.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1127 if (in_ifhasaddr(ifp, ia) == 0) 1128 return (EADDRNOTAVAIL); 1129 } 1130 1131 bzero(src, sizeof(*src)); 1132 src->sin_family = AF_INET; 1133 src->sin_len = sizeof(*src); 1134 src->sin_port = inp->inp_lport; 1135 src->sin_addr.s_addr = pktinfo->ipi6_addr.s6_addr32[3]; 1136 1137 return (0); 1138 } 1139 #endif 1140 1141 static int 1142 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1143 struct mbuf *control, struct thread *td, int flags) 1144 { 1145 struct udpiphdr *ui; 1146 int len = m->m_pkthdr.len; 1147 struct in_addr faddr, laddr; 1148 struct cmsghdr *cm; 1149 struct inpcbinfo *pcbinfo; 1150 struct sockaddr_in *sin, src; 1151 struct epoch_tracker et; 1152 int cscov_partial = 0; 1153 int error = 0; 1154 int ipflags = 0; 1155 u_short fport, lport; 1156 u_char tos; 1157 uint8_t pr; 1158 uint16_t cscov = 0; 1159 uint32_t flowid = 0; 1160 uint8_t flowtype = M_HASHTYPE_NONE; 1161 1162 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1163 if (control) 1164 m_freem(control); 1165 m_freem(m); 1166 return (EMSGSIZE); 1167 } 1168 1169 src.sin_family = 0; 1170 sin = (struct sockaddr_in *)addr; 1171 1172 /* 1173 * udp_output() may need to temporarily bind or connect the current 1174 * inpcb. As such, we don't know up front whether we will need the 1175 * pcbinfo lock or not. Do any work to decide what is needed up 1176 * front before acquiring any locks. 1177 * 1178 * We will need network epoch in either case, to safely lookup into 1179 * pcb hash. 1180 */ 1181 if (sin == NULL || 1182 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) 1183 INP_WLOCK(inp); 1184 else 1185 INP_RLOCK(inp); 1186 NET_EPOCH_ENTER(et); 1187 tos = inp->inp_ip_tos; 1188 if (control != NULL) { 1189 /* 1190 * XXX: Currently, we assume all the optional information is 1191 * stored in a single mbuf. 1192 */ 1193 if (control->m_next) { 1194 m_freem(control); 1195 error = EINVAL; 1196 goto release; 1197 } 1198 for (; control->m_len > 0; 1199 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1200 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1201 cm = mtod(control, struct cmsghdr *); 1202 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1203 || cm->cmsg_len > control->m_len) { 1204 error = EINVAL; 1205 break; 1206 } 1207 #ifdef INET6 1208 error = udp_v4mapped_pktinfo(cm, &src, inp, flags); 1209 if (error != 0) 1210 break; 1211 #endif 1212 if (cm->cmsg_level != IPPROTO_IP) 1213 continue; 1214 1215 switch (cm->cmsg_type) { 1216 case IP_SENDSRCADDR: 1217 if (cm->cmsg_len != 1218 CMSG_LEN(sizeof(struct in_addr))) { 1219 error = EINVAL; 1220 break; 1221 } 1222 bzero(&src, sizeof(src)); 1223 src.sin_family = AF_INET; 1224 src.sin_len = sizeof(src); 1225 src.sin_port = inp->inp_lport; 1226 src.sin_addr = 1227 *(struct in_addr *)CMSG_DATA(cm); 1228 break; 1229 1230 case IP_TOS: 1231 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1232 error = EINVAL; 1233 break; 1234 } 1235 tos = *(u_char *)CMSG_DATA(cm); 1236 break; 1237 1238 case IP_FLOWID: 1239 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1240 error = EINVAL; 1241 break; 1242 } 1243 flowid = *(uint32_t *) CMSG_DATA(cm); 1244 break; 1245 1246 case IP_FLOWTYPE: 1247 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1248 error = EINVAL; 1249 break; 1250 } 1251 flowtype = *(uint32_t *) CMSG_DATA(cm); 1252 break; 1253 1254 #ifdef RSS 1255 case IP_RSSBUCKETID: 1256 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1257 error = EINVAL; 1258 break; 1259 } 1260 /* This is just a placeholder for now */ 1261 break; 1262 #endif /* RSS */ 1263 default: 1264 error = ENOPROTOOPT; 1265 break; 1266 } 1267 if (error) 1268 break; 1269 } 1270 m_freem(control); 1271 control = NULL; 1272 } 1273 if (error) 1274 goto release; 1275 1276 pr = inp->inp_socket->so_proto->pr_protocol; 1277 pcbinfo = udp_get_inpcbinfo(pr); 1278 1279 /* 1280 * If the IP_SENDSRCADDR control message was specified, override the 1281 * source address for this datagram. Its use is invalidated if the 1282 * address thus specified is incomplete or clobbers other inpcbs. 1283 */ 1284 laddr = inp->inp_laddr; 1285 lport = inp->inp_lport; 1286 if (src.sin_family == AF_INET) { 1287 INP_HASH_LOCK_ASSERT(pcbinfo); 1288 if ((lport == 0) || 1289 (laddr.s_addr == INADDR_ANY && 1290 src.sin_addr.s_addr == INADDR_ANY)) { 1291 error = EINVAL; 1292 goto release; 1293 } 1294 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1295 &laddr.s_addr, &lport, td->td_ucred); 1296 if (error) 1297 goto release; 1298 } 1299 1300 /* 1301 * If a UDP socket has been connected, then a local address/port will 1302 * have been selected and bound. 1303 * 1304 * If a UDP socket has not been connected to, then an explicit 1305 * destination address must be used, in which case a local 1306 * address/port may not have been selected and bound. 1307 */ 1308 if (sin != NULL) { 1309 INP_LOCK_ASSERT(inp); 1310 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1311 error = EISCONN; 1312 goto release; 1313 } 1314 1315 /* 1316 * Jail may rewrite the destination address, so let it do 1317 * that before we use it. 1318 */ 1319 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1320 if (error) 1321 goto release; 1322 1323 /* 1324 * If a local address or port hasn't yet been selected, or if 1325 * the destination address needs to be rewritten due to using 1326 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1327 * to do the heavy lifting. Once a port is selected, we 1328 * commit the binding back to the socket; we also commit the 1329 * binding of the address if in jail. 1330 * 1331 * If we already have a valid binding and we're not 1332 * requesting a destination address rewrite, use a fast path. 1333 */ 1334 if (inp->inp_laddr.s_addr == INADDR_ANY || 1335 inp->inp_lport == 0 || 1336 sin->sin_addr.s_addr == INADDR_ANY || 1337 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1338 INP_HASH_LOCK_ASSERT(pcbinfo); 1339 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1340 &lport, &faddr.s_addr, &fport, NULL, 1341 td->td_ucred); 1342 if (error) 1343 goto release; 1344 1345 /* 1346 * XXXRW: Why not commit the port if the address is 1347 * !INADDR_ANY? 1348 */ 1349 /* Commit the local port if newly assigned. */ 1350 if (inp->inp_laddr.s_addr == INADDR_ANY && 1351 inp->inp_lport == 0) { 1352 INP_WLOCK_ASSERT(inp); 1353 /* 1354 * Remember addr if jailed, to prevent 1355 * rebinding. 1356 */ 1357 if (prison_flag(td->td_ucred, PR_IP4)) 1358 inp->inp_laddr = laddr; 1359 inp->inp_lport = lport; 1360 INP_HASH_WLOCK(pcbinfo); 1361 error = in_pcbinshash(inp); 1362 INP_HASH_WUNLOCK(pcbinfo); 1363 if (error != 0) { 1364 inp->inp_lport = 0; 1365 error = EAGAIN; 1366 goto release; 1367 } 1368 inp->inp_flags |= INP_ANONPORT; 1369 } 1370 } else { 1371 faddr = sin->sin_addr; 1372 fport = sin->sin_port; 1373 } 1374 } else { 1375 INP_LOCK_ASSERT(inp); 1376 faddr = inp->inp_faddr; 1377 fport = inp->inp_fport; 1378 if (faddr.s_addr == INADDR_ANY) { 1379 error = ENOTCONN; 1380 goto release; 1381 } 1382 } 1383 1384 /* 1385 * Calculate data length and get a mbuf for UDP, IP, and possible 1386 * link-layer headers. Immediate slide the data pointer back forward 1387 * since we won't use that space at this layer. 1388 */ 1389 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1390 if (m == NULL) { 1391 error = ENOBUFS; 1392 goto release; 1393 } 1394 m->m_data += max_linkhdr; 1395 m->m_len -= max_linkhdr; 1396 m->m_pkthdr.len -= max_linkhdr; 1397 1398 /* 1399 * Fill in mbuf with extended UDP header and addresses and length put 1400 * into network format. 1401 */ 1402 ui = mtod(m, struct udpiphdr *); 1403 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1404 ui->ui_v = IPVERSION << 4; 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 /* 1430 * Set the Don't Fragment bit in the IP header. 1431 */ 1432 if (inp->inp_flags & INP_DONTFRAG) { 1433 struct ip *ip; 1434 1435 ip = (struct ip *)&ui->ui_i; 1436 ip->ip_off |= htons(IP_DF); 1437 } 1438 1439 if (inp->inp_socket->so_options & SO_DONTROUTE) 1440 ipflags |= IP_ROUTETOIF; 1441 if (inp->inp_socket->so_options & SO_BROADCAST) 1442 ipflags |= IP_ALLOWBROADCAST; 1443 if (inp->inp_flags & INP_ONESBCAST) 1444 ipflags |= IP_SENDONES; 1445 1446 #ifdef MAC 1447 mac_inpcb_create_mbuf(inp, m); 1448 #endif 1449 1450 /* 1451 * Set up checksum and output datagram. 1452 */ 1453 ui->ui_sum = 0; 1454 if (pr == IPPROTO_UDPLITE) { 1455 if (inp->inp_flags & INP_ONESBCAST) 1456 faddr.s_addr = INADDR_BROADCAST; 1457 if (cscov_partial) { 1458 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1459 ui->ui_sum = 0xffff; 1460 } else { 1461 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1462 ui->ui_sum = 0xffff; 1463 } 1464 } else if (V_udp_cksum) { 1465 if (inp->inp_flags & INP_ONESBCAST) 1466 faddr.s_addr = INADDR_BROADCAST; 1467 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1468 htons((u_short)len + sizeof(struct udphdr) + pr)); 1469 m->m_pkthdr.csum_flags = CSUM_UDP; 1470 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1471 } 1472 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1473 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1474 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1475 UDPSTAT_INC(udps_opackets); 1476 1477 /* 1478 * Setup flowid / RSS information for outbound socket. 1479 * 1480 * Once the UDP code decides to set a flowid some other way, 1481 * this allows the flowid to be overridden by userland. 1482 */ 1483 if (flowtype != M_HASHTYPE_NONE) { 1484 m->m_pkthdr.flowid = flowid; 1485 M_HASHTYPE_SET(m, flowtype); 1486 } 1487 #if defined(ROUTE_MPATH) || defined(RSS) 1488 else if (CALC_FLOWID_OUTBOUND_SENDTO) { 1489 uint32_t hash_val, hash_type; 1490 1491 hash_val = fib4_calc_packet_hash(laddr, faddr, 1492 lport, fport, pr, &hash_type); 1493 m->m_pkthdr.flowid = hash_val; 1494 M_HASHTYPE_SET(m, hash_type); 1495 } 1496 1497 /* 1498 * Don't override with the inp cached flowid value. 1499 * 1500 * Depending upon the kind of send being done, the inp 1501 * flowid/flowtype values may actually not be appropriate 1502 * for this particular socket send. 1503 * 1504 * We should either leave the flowid at zero (which is what is 1505 * currently done) or set it to some software generated 1506 * hash value based on the packet contents. 1507 */ 1508 ipflags |= IP_NODEFAULTFLOWID; 1509 #endif /* RSS */ 1510 1511 if (pr == IPPROTO_UDPLITE) 1512 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1513 else 1514 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1515 error = ip_output(m, inp->inp_options, 1516 INP_WLOCKED(inp) ? &inp->inp_route : NULL, ipflags, 1517 inp->inp_moptions, inp); 1518 INP_UNLOCK(inp); 1519 NET_EPOCH_EXIT(et); 1520 return (error); 1521 1522 release: 1523 INP_UNLOCK(inp); 1524 NET_EPOCH_EXIT(et); 1525 m_freem(m); 1526 return (error); 1527 } 1528 1529 static void 1530 udp_abort(struct socket *so) 1531 { 1532 struct inpcb *inp; 1533 struct inpcbinfo *pcbinfo; 1534 1535 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1536 inp = sotoinpcb(so); 1537 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1538 INP_WLOCK(inp); 1539 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1540 INP_HASH_WLOCK(pcbinfo); 1541 in_pcbdisconnect(inp); 1542 inp->inp_laddr.s_addr = INADDR_ANY; 1543 INP_HASH_WUNLOCK(pcbinfo); 1544 soisdisconnected(so); 1545 } 1546 INP_WUNLOCK(inp); 1547 } 1548 1549 static int 1550 udp_attach(struct socket *so, int proto, struct thread *td) 1551 { 1552 static uint32_t udp_flowid; 1553 struct inpcb *inp; 1554 struct inpcbinfo *pcbinfo; 1555 int error; 1556 1557 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1558 inp = sotoinpcb(so); 1559 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1560 error = soreserve(so, udp_sendspace, udp_recvspace); 1561 if (error) 1562 return (error); 1563 INP_INFO_WLOCK(pcbinfo); 1564 error = in_pcballoc(so, pcbinfo); 1565 if (error) { 1566 INP_INFO_WUNLOCK(pcbinfo); 1567 return (error); 1568 } 1569 1570 inp = sotoinpcb(so); 1571 inp->inp_vflag |= INP_IPV4; 1572 inp->inp_ip_ttl = V_ip_defttl; 1573 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1574 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1575 1576 error = udp_newudpcb(inp); 1577 if (error) { 1578 in_pcbdetach(inp); 1579 in_pcbfree(inp); 1580 INP_INFO_WUNLOCK(pcbinfo); 1581 return (error); 1582 } 1583 1584 INP_WUNLOCK(inp); 1585 INP_INFO_WUNLOCK(pcbinfo); 1586 return (0); 1587 } 1588 #endif /* INET */ 1589 1590 int 1591 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1592 { 1593 struct inpcb *inp; 1594 struct udpcb *up; 1595 1596 KASSERT(so->so_type == SOCK_DGRAM, 1597 ("udp_set_kernel_tunneling: !dgram")); 1598 inp = sotoinpcb(so); 1599 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1600 INP_WLOCK(inp); 1601 up = intoudpcb(inp); 1602 if ((up->u_tun_func != NULL) || 1603 (up->u_icmp_func != NULL)) { 1604 INP_WUNLOCK(inp); 1605 return (EBUSY); 1606 } 1607 up->u_tun_func = f; 1608 up->u_icmp_func = i; 1609 up->u_tun_ctx = ctx; 1610 INP_WUNLOCK(inp); 1611 return (0); 1612 } 1613 1614 #ifdef INET 1615 static int 1616 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1617 { 1618 struct inpcb *inp; 1619 struct inpcbinfo *pcbinfo; 1620 struct sockaddr_in *sinp; 1621 int error; 1622 1623 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1624 inp = sotoinpcb(so); 1625 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1626 1627 sinp = (struct sockaddr_in *)nam; 1628 if (nam->sa_family != AF_INET) { 1629 /* 1630 * Preserve compatibility with old programs. 1631 */ 1632 if (nam->sa_family != AF_UNSPEC || 1633 nam->sa_len < offsetof(struct sockaddr_in, sin_zero) || 1634 sinp->sin_addr.s_addr != INADDR_ANY) 1635 return (EAFNOSUPPORT); 1636 nam->sa_family = AF_INET; 1637 } 1638 if (nam->sa_len != sizeof(struct sockaddr_in)) 1639 return (EINVAL); 1640 1641 INP_WLOCK(inp); 1642 INP_HASH_WLOCK(pcbinfo); 1643 error = in_pcbbind(inp, nam, td->td_ucred); 1644 INP_HASH_WUNLOCK(pcbinfo); 1645 INP_WUNLOCK(inp); 1646 return (error); 1647 } 1648 1649 static void 1650 udp_close(struct socket *so) 1651 { 1652 struct inpcb *inp; 1653 struct inpcbinfo *pcbinfo; 1654 1655 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1656 inp = sotoinpcb(so); 1657 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1658 INP_WLOCK(inp); 1659 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1660 INP_HASH_WLOCK(pcbinfo); 1661 in_pcbdisconnect(inp); 1662 inp->inp_laddr.s_addr = INADDR_ANY; 1663 INP_HASH_WUNLOCK(pcbinfo); 1664 soisdisconnected(so); 1665 } 1666 INP_WUNLOCK(inp); 1667 } 1668 1669 static int 1670 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1671 { 1672 struct epoch_tracker et; 1673 struct inpcb *inp; 1674 struct inpcbinfo *pcbinfo; 1675 struct sockaddr_in *sin; 1676 int error; 1677 1678 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1679 inp = sotoinpcb(so); 1680 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1681 1682 sin = (struct sockaddr_in *)nam; 1683 if (sin->sin_family != AF_INET) 1684 return (EAFNOSUPPORT); 1685 if (sin->sin_len != sizeof(*sin)) 1686 return (EINVAL); 1687 1688 INP_WLOCK(inp); 1689 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1690 INP_WUNLOCK(inp); 1691 return (EISCONN); 1692 } 1693 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1694 if (error != 0) { 1695 INP_WUNLOCK(inp); 1696 return (error); 1697 } 1698 NET_EPOCH_ENTER(et); 1699 INP_HASH_WLOCK(pcbinfo); 1700 error = in_pcbconnect(inp, nam, td->td_ucred); 1701 INP_HASH_WUNLOCK(pcbinfo); 1702 NET_EPOCH_EXIT(et); 1703 if (error == 0) 1704 soisconnected(so); 1705 INP_WUNLOCK(inp); 1706 return (error); 1707 } 1708 1709 static void 1710 udp_detach(struct socket *so) 1711 { 1712 struct inpcb *inp; 1713 struct inpcbinfo *pcbinfo; 1714 struct udpcb *up; 1715 1716 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1717 inp = sotoinpcb(so); 1718 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1719 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1720 ("udp_detach: not disconnected")); 1721 INP_INFO_WLOCK(pcbinfo); 1722 INP_WLOCK(inp); 1723 up = intoudpcb(inp); 1724 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1725 inp->inp_ppcb = NULL; 1726 in_pcbdetach(inp); 1727 in_pcbfree(inp); 1728 INP_INFO_WUNLOCK(pcbinfo); 1729 udp_discardcb(up); 1730 } 1731 1732 static int 1733 udp_disconnect(struct socket *so) 1734 { 1735 struct inpcb *inp; 1736 struct inpcbinfo *pcbinfo; 1737 1738 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1739 inp = sotoinpcb(so); 1740 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1741 INP_WLOCK(inp); 1742 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1743 INP_WUNLOCK(inp); 1744 return (ENOTCONN); 1745 } 1746 INP_HASH_WLOCK(pcbinfo); 1747 in_pcbdisconnect(inp); 1748 inp->inp_laddr.s_addr = INADDR_ANY; 1749 INP_HASH_WUNLOCK(pcbinfo); 1750 SOCK_LOCK(so); 1751 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1752 SOCK_UNLOCK(so); 1753 INP_WUNLOCK(inp); 1754 return (0); 1755 } 1756 1757 static int 1758 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1759 struct mbuf *control, struct thread *td) 1760 { 1761 struct inpcb *inp; 1762 int error; 1763 1764 inp = sotoinpcb(so); 1765 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1766 1767 if (addr != NULL) { 1768 error = 0; 1769 if (addr->sa_family != AF_INET) 1770 error = EAFNOSUPPORT; 1771 else if (addr->sa_len != sizeof(struct sockaddr_in)) 1772 error = EINVAL; 1773 if (__predict_false(error != 0)) { 1774 m_freem(control); 1775 m_freem(m); 1776 return (error); 1777 } 1778 } 1779 return (udp_output(inp, m, addr, control, td, flags)); 1780 } 1781 #endif /* INET */ 1782 1783 int 1784 udp_shutdown(struct socket *so) 1785 { 1786 struct inpcb *inp; 1787 1788 inp = sotoinpcb(so); 1789 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1790 INP_WLOCK(inp); 1791 socantsendmore(so); 1792 INP_WUNLOCK(inp); 1793 return (0); 1794 } 1795 1796 #ifdef INET 1797 struct pr_usrreqs udp_usrreqs = { 1798 .pru_abort = udp_abort, 1799 .pru_attach = udp_attach, 1800 .pru_bind = udp_bind, 1801 .pru_connect = udp_connect, 1802 .pru_control = in_control, 1803 .pru_detach = udp_detach, 1804 .pru_disconnect = udp_disconnect, 1805 .pru_peeraddr = in_getpeeraddr, 1806 .pru_send = udp_send, 1807 .pru_soreceive = soreceive_dgram, 1808 .pru_sosend = sosend_dgram, 1809 .pru_shutdown = udp_shutdown, 1810 .pru_sockaddr = in_getsockaddr, 1811 .pru_sosetlabel = in_pcbsosetlabel, 1812 .pru_close = udp_close, 1813 }; 1814 #endif /* INET */ 1815