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_rss.h" 48 49 #include <sys/param.h> 50 #include <sys/domain.h> 51 #include <sys/eventhandler.h> 52 #include <sys/jail.h> 53 #include <sys/kernel.h> 54 #include <sys/lock.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/priv.h> 58 #include <sys/proc.h> 59 #include <sys/protosw.h> 60 #include <sys/sdt.h> 61 #include <sys/signalvar.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <sys/sx.h> 65 #include <sys/sysctl.h> 66 #include <sys/syslog.h> 67 #include <sys/systm.h> 68 69 #include <vm/uma.h> 70 71 #include <net/if.h> 72 #include <net/if_var.h> 73 #include <net/route.h> 74 #include <net/rss_config.h> 75 76 #include <netinet/in.h> 77 #include <netinet/in_kdtrace.h> 78 #include <netinet/in_pcb.h> 79 #include <netinet/in_systm.h> 80 #include <netinet/in_var.h> 81 #include <netinet/ip.h> 82 #ifdef INET6 83 #include <netinet/ip6.h> 84 #endif 85 #include <netinet/ip_icmp.h> 86 #include <netinet/icmp_var.h> 87 #include <netinet/ip_var.h> 88 #include <netinet/ip_options.h> 89 #ifdef INET6 90 #include <netinet6/ip6_var.h> 91 #endif 92 #include <netinet/udp.h> 93 #include <netinet/udp_var.h> 94 #include <netinet/udplite.h> 95 #include <netinet/in_rss.h> 96 97 #include <netipsec/ipsec_support.h> 98 99 #include <machine/in_cksum.h> 100 101 #include <security/mac/mac_framework.h> 102 103 /* 104 * UDP and UDP-Lite protocols implementation. 105 * Per RFC 768, August, 1980. 106 * Per RFC 3828, July, 2004. 107 */ 108 109 /* 110 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 111 * removes the only data integrity mechanism for packets and malformed 112 * packets that would otherwise be discarded due to bad checksums, and may 113 * cause problems (especially for NFS data blocks). 114 */ 115 VNET_DEFINE(int, udp_cksum) = 1; 116 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, 117 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 118 119 int udp_log_in_vain = 0; 120 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 121 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 122 123 VNET_DEFINE(int, udp_blackhole) = 0; 124 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 125 &VNET_NAME(udp_blackhole), 0, 126 "Do not send port unreachables for refused connects"); 127 128 u_long udp_sendspace = 9216; /* really max datagram size */ 129 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 130 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 131 132 u_long udp_recvspace = 40 * (1024 + 133 #ifdef INET6 134 sizeof(struct sockaddr_in6) 135 #else 136 sizeof(struct sockaddr_in) 137 #endif 138 ); /* 40 1K datagrams */ 139 140 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 141 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 142 143 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 144 VNET_DEFINE(struct inpcbinfo, udbinfo); 145 VNET_DEFINE(struct inpcbhead, ulitecb); 146 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 147 VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone); 148 #define V_udpcb_zone VNET(udpcb_zone) 149 150 #ifndef UDBHASHSIZE 151 #define UDBHASHSIZE 128 152 #endif 153 154 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 155 VNET_PCPUSTAT_SYSINIT(udpstat); 156 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 157 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 158 159 #ifdef VIMAGE 160 VNET_PCPUSTAT_SYSUNINIT(udpstat); 161 #endif /* VIMAGE */ 162 #ifdef INET 163 static void udp_detach(struct socket *so); 164 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 165 struct mbuf *, struct thread *); 166 #endif 167 168 static void 169 udp_zone_change(void *tag) 170 { 171 172 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 173 uma_zone_set_max(V_udpcb_zone, maxsockets); 174 } 175 176 static int 177 udp_inpcb_init(void *mem, int size, int flags) 178 { 179 struct inpcb *inp; 180 181 inp = mem; 182 INP_LOCK_INIT(inp, "inp", "udpinp"); 183 return (0); 184 } 185 186 static int 187 udplite_inpcb_init(void *mem, int size, int flags) 188 { 189 struct inpcb *inp; 190 191 inp = mem; 192 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 193 return (0); 194 } 195 196 void 197 udp_init(void) 198 { 199 200 /* 201 * For now default to 2-tuple UDP hashing - until the fragment 202 * reassembly code can also update the flowid. 203 * 204 * Once we can calculate the flowid that way and re-establish 205 * a 4-tuple, flip this to 4-tuple. 206 */ 207 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 208 "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE); 209 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 210 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 211 uma_zone_set_max(V_udpcb_zone, maxsockets); 212 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 213 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 214 EVENTHANDLER_PRI_ANY); 215 } 216 217 void 218 udplite_init(void) 219 { 220 221 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 222 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, 223 IPI_HASHFIELDS_2TUPLE); 224 } 225 226 /* 227 * Kernel module interface for updating udpstat. The argument is an index 228 * into udpstat treated as an array of u_long. While this encodes the 229 * general layout of udpstat into the caller, it doesn't encode its location, 230 * so that future changes to add, for example, per-CPU stats support won't 231 * cause binary compatibility problems for kernel modules. 232 */ 233 void 234 kmod_udpstat_inc(int statnum) 235 { 236 237 counter_u64_add(VNET(udpstat)[statnum], 1); 238 } 239 240 int 241 udp_newudpcb(struct inpcb *inp) 242 { 243 struct udpcb *up; 244 245 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 246 if (up == NULL) 247 return (ENOBUFS); 248 inp->inp_ppcb = up; 249 return (0); 250 } 251 252 void 253 udp_discardcb(struct udpcb *up) 254 { 255 256 uma_zfree(V_udpcb_zone, up); 257 } 258 259 #ifdef VIMAGE 260 static void 261 udp_destroy(void *unused __unused) 262 { 263 264 in_pcbinfo_destroy(&V_udbinfo); 265 uma_zdestroy(V_udpcb_zone); 266 } 267 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 268 269 static void 270 udplite_destroy(void *unused __unused) 271 { 272 273 in_pcbinfo_destroy(&V_ulitecbinfo); 274 } 275 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 276 NULL); 277 #endif 278 279 #ifdef INET 280 /* 281 * Subroutine of udp_input(), which appends the provided mbuf chain to the 282 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 283 * contains the source address. If the socket ends up being an IPv6 socket, 284 * udp_append() will convert to a sockaddr_in6 before passing the address 285 * into the socket code. 286 * 287 * In the normal case udp_append() will return 0, indicating that you 288 * must unlock the inp. However if a tunneling protocol is in place we increment 289 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 290 * then decrement the reference count. If the inp_rele returns 1, indicating the 291 * inp is gone, we return that to the caller to tell them *not* to unlock 292 * the inp. In the case of multi-cast this will cause the distribution 293 * to stop (though most tunneling protocols known currently do *not* use 294 * multicast). 295 */ 296 static int 297 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 298 struct sockaddr_in *udp_in) 299 { 300 struct sockaddr *append_sa; 301 struct socket *so; 302 struct mbuf *tmpopts, *opts = NULL; 303 #ifdef INET6 304 struct sockaddr_in6 udp_in6; 305 #endif 306 struct udpcb *up; 307 308 INP_LOCK_ASSERT(inp); 309 310 /* 311 * Engage the tunneling protocol. 312 */ 313 up = intoudpcb(inp); 314 if (up->u_tun_func != NULL) { 315 in_pcbref(inp); 316 INP_RUNLOCK(inp); 317 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], 318 up->u_tun_ctx); 319 INP_RLOCK(inp); 320 return (in_pcbrele_rlocked(inp)); 321 } 322 323 off += sizeof(struct udphdr); 324 325 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 326 /* Check AH/ESP integrity. */ 327 if (IPSEC_ENABLED(ipv4) && 328 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 329 m_freem(n); 330 return (0); 331 } 332 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 333 if (IPSEC_ENABLED(ipv4) && 334 UDPENCAP_INPUT(n, off, AF_INET) != 0) 335 return (0); /* Consumed. */ 336 } 337 #endif /* IPSEC */ 338 #ifdef MAC 339 if (mac_inpcb_check_deliver(inp, n) != 0) { 340 m_freem(n); 341 return (0); 342 } 343 #endif /* MAC */ 344 if (inp->inp_flags & INP_CONTROLOPTS || 345 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 346 #ifdef INET6 347 if (inp->inp_vflag & INP_IPV6) 348 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 349 else 350 #endif /* INET6 */ 351 ip_savecontrol(inp, &opts, ip, n); 352 } 353 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 354 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], 355 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); 356 if (tmpopts) { 357 if (opts) { 358 tmpopts->m_next = opts; 359 opts = tmpopts; 360 } else 361 opts = tmpopts; 362 } 363 } 364 #ifdef INET6 365 if (inp->inp_vflag & INP_IPV6) { 366 bzero(&udp_in6, sizeof(udp_in6)); 367 udp_in6.sin6_len = sizeof(udp_in6); 368 udp_in6.sin6_family = AF_INET6; 369 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 370 append_sa = (struct sockaddr *)&udp_in6; 371 } else 372 #endif /* INET6 */ 373 append_sa = (struct sockaddr *)&udp_in[0]; 374 m_adj(n, off); 375 376 so = inp->inp_socket; 377 SOCKBUF_LOCK(&so->so_rcv); 378 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 379 SOCKBUF_UNLOCK(&so->so_rcv); 380 m_freem(n); 381 if (opts) 382 m_freem(opts); 383 UDPSTAT_INC(udps_fullsock); 384 } else 385 sorwakeup_locked(so); 386 return (0); 387 } 388 389 int 390 udp_input(struct mbuf **mp, int *offp, int proto) 391 { 392 struct ip *ip; 393 struct udphdr *uh; 394 struct ifnet *ifp; 395 struct inpcb *inp; 396 uint16_t len, ip_len; 397 struct inpcbinfo *pcbinfo; 398 struct ip save_ip; 399 struct sockaddr_in udp_in[2]; 400 struct mbuf *m; 401 struct m_tag *fwd_tag; 402 struct epoch_tracker et; 403 int cscov_partial, iphlen; 404 405 m = *mp; 406 iphlen = *offp; 407 ifp = m->m_pkthdr.rcvif; 408 *mp = NULL; 409 UDPSTAT_INC(udps_ipackets); 410 411 /* 412 * Strip IP options, if any; should skip this, make available to 413 * user, and use on returned packets, but we don't yet have a way to 414 * check the checksum with options still present. 415 */ 416 if (iphlen > sizeof (struct ip)) { 417 ip_stripoptions(m); 418 iphlen = sizeof(struct ip); 419 } 420 421 /* 422 * Get IP and UDP header together in first mbuf. 423 */ 424 ip = mtod(m, struct ip *); 425 if (m->m_len < iphlen + sizeof(struct udphdr)) { 426 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 427 UDPSTAT_INC(udps_hdrops); 428 return (IPPROTO_DONE); 429 } 430 ip = mtod(m, struct ip *); 431 } 432 uh = (struct udphdr *)((caddr_t)ip + iphlen); 433 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 434 435 /* 436 * Destination port of 0 is illegal, based on RFC768. 437 */ 438 if (uh->uh_dport == 0) 439 goto badunlocked; 440 441 /* 442 * Construct sockaddr format source address. Stuff source address 443 * and datagram in user buffer. 444 */ 445 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 446 udp_in[0].sin_len = sizeof(struct sockaddr_in); 447 udp_in[0].sin_family = AF_INET; 448 udp_in[0].sin_port = uh->uh_sport; 449 udp_in[0].sin_addr = ip->ip_src; 450 udp_in[1].sin_len = sizeof(struct sockaddr_in); 451 udp_in[1].sin_family = AF_INET; 452 udp_in[1].sin_port = uh->uh_dport; 453 udp_in[1].sin_addr = ip->ip_dst; 454 455 /* 456 * Make mbuf data length reflect UDP length. If not enough data to 457 * reflect UDP length, drop. 458 */ 459 len = ntohs((u_short)uh->uh_ulen); 460 ip_len = ntohs(ip->ip_len) - iphlen; 461 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 462 /* Zero means checksum over the complete packet. */ 463 if (len == 0) 464 len = ip_len; 465 cscov_partial = 0; 466 } 467 if (ip_len != len) { 468 if (len > ip_len || len < sizeof(struct udphdr)) { 469 UDPSTAT_INC(udps_badlen); 470 goto badunlocked; 471 } 472 if (proto == IPPROTO_UDP) 473 m_adj(m, len - ip_len); 474 } 475 476 /* 477 * Save a copy of the IP header in case we want restore it for 478 * sending an ICMP error message in response. 479 */ 480 if (!V_udp_blackhole) 481 save_ip = *ip; 482 else 483 memset(&save_ip, 0, sizeof(save_ip)); 484 485 /* 486 * Checksum extended UDP header and data. 487 */ 488 if (uh->uh_sum) { 489 u_short uh_sum; 490 491 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 492 !cscov_partial) { 493 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 494 uh_sum = m->m_pkthdr.csum_data; 495 else 496 uh_sum = in_pseudo(ip->ip_src.s_addr, 497 ip->ip_dst.s_addr, htonl((u_short)len + 498 m->m_pkthdr.csum_data + proto)); 499 uh_sum ^= 0xffff; 500 } else { 501 char b[9]; 502 503 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 504 bzero(((struct ipovly *)ip)->ih_x1, 9); 505 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 506 uh->uh_ulen : htons(ip_len); 507 uh_sum = in_cksum(m, len + sizeof (struct ip)); 508 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 509 } 510 if (uh_sum) { 511 UDPSTAT_INC(udps_badsum); 512 m_freem(m); 513 return (IPPROTO_DONE); 514 } 515 } else { 516 if (proto == IPPROTO_UDP) { 517 UDPSTAT_INC(udps_nosum); 518 } else { 519 /* UDPLite requires a checksum */ 520 /* XXX: What is the right UDPLite MIB counter here? */ 521 m_freem(m); 522 return (IPPROTO_DONE); 523 } 524 } 525 526 pcbinfo = udp_get_inpcbinfo(proto); 527 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 528 in_broadcast(ip->ip_dst, ifp)) { 529 struct inpcb *last; 530 struct inpcbhead *pcblist; 531 struct ip_moptions *imo; 532 533 INP_INFO_RLOCK_ET(pcbinfo, et); 534 pcblist = udp_get_pcblist(proto); 535 last = NULL; 536 CK_LIST_FOREACH(inp, pcblist, inp_list) { 537 if (inp->inp_lport != uh->uh_dport) 538 continue; 539 #ifdef INET6 540 if ((inp->inp_vflag & INP_IPV4) == 0) 541 continue; 542 #endif 543 if (inp->inp_laddr.s_addr != INADDR_ANY && 544 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 545 continue; 546 if (inp->inp_faddr.s_addr != INADDR_ANY && 547 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 548 continue; 549 if (inp->inp_fport != 0 && 550 inp->inp_fport != uh->uh_sport) 551 continue; 552 553 INP_RLOCK(inp); 554 555 /* 556 * XXXRW: Because we weren't holding either the inpcb 557 * or the hash lock when we checked for a match 558 * before, we should probably recheck now that the 559 * inpcb lock is held. 560 */ 561 562 /* 563 * Handle socket delivery policy for any-source 564 * and source-specific multicast. [RFC3678] 565 */ 566 imo = inp->inp_moptions; 567 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 568 struct sockaddr_in group; 569 int blocked; 570 if (imo == NULL) { 571 INP_RUNLOCK(inp); 572 continue; 573 } 574 bzero(&group, sizeof(struct sockaddr_in)); 575 group.sin_len = sizeof(struct sockaddr_in); 576 group.sin_family = AF_INET; 577 group.sin_addr = ip->ip_dst; 578 579 blocked = imo_multi_filter(imo, ifp, 580 (struct sockaddr *)&group, 581 (struct sockaddr *)&udp_in[0]); 582 if (blocked != MCAST_PASS) { 583 if (blocked == MCAST_NOTGMEMBER) 584 IPSTAT_INC(ips_notmember); 585 if (blocked == MCAST_NOTSMEMBER || 586 blocked == MCAST_MUTED) 587 UDPSTAT_INC(udps_filtermcast); 588 INP_RUNLOCK(inp); 589 continue; 590 } 591 } 592 if (last != NULL) { 593 struct mbuf *n; 594 595 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 596 NULL) { 597 UDP_PROBE(receive, NULL, last, ip, 598 last, uh); 599 if (udp_append(last, ip, n, iphlen, 600 udp_in)) { 601 goto inp_lost; 602 } 603 } 604 INP_RUNLOCK(last); 605 } 606 last = inp; 607 /* 608 * Don't look for additional matches if this one does 609 * not have either the SO_REUSEPORT or SO_REUSEADDR 610 * socket options set. This heuristic avoids 611 * searching through all pcbs in the common case of a 612 * non-shared port. It assumes that an application 613 * will never clear these options after setting them. 614 */ 615 if ((last->inp_socket->so_options & 616 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) 617 break; 618 } 619 620 if (last == NULL) { 621 /* 622 * No matching pcb found; discard datagram. (No need 623 * to send an ICMP Port Unreachable for a broadcast 624 * or multicast datgram.) 625 */ 626 UDPSTAT_INC(udps_noportbcast); 627 if (inp) 628 INP_RUNLOCK(inp); 629 INP_INFO_RUNLOCK_ET(pcbinfo, et); 630 goto badunlocked; 631 } 632 UDP_PROBE(receive, NULL, last, ip, last, uh); 633 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 634 INP_RUNLOCK(last); 635 inp_lost: 636 INP_INFO_RUNLOCK_ET(pcbinfo, et); 637 return (IPPROTO_DONE); 638 } 639 640 /* 641 * Locate pcb for datagram. 642 */ 643 644 /* 645 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 646 */ 647 if ((m->m_flags & M_IP_NEXTHOP) && 648 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 649 struct sockaddr_in *next_hop; 650 651 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 652 653 /* 654 * Transparently forwarded. Pretend to be the destination. 655 * Already got one like this? 656 */ 657 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 658 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 659 if (!inp) { 660 /* 661 * It's new. Try to find the ambushing socket. 662 * Because we've rewritten the destination address, 663 * any hardware-generated hash is ignored. 664 */ 665 inp = in_pcblookup(pcbinfo, ip->ip_src, 666 uh->uh_sport, next_hop->sin_addr, 667 next_hop->sin_port ? htons(next_hop->sin_port) : 668 uh->uh_dport, INPLOOKUP_WILDCARD | 669 INPLOOKUP_RLOCKPCB, ifp); 670 } 671 /* Remove the tag from the packet. We don't need it anymore. */ 672 m_tag_delete(m, fwd_tag); 673 m->m_flags &= ~M_IP_NEXTHOP; 674 } else 675 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 676 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 677 INPLOOKUP_RLOCKPCB, ifp, m); 678 if (inp == NULL) { 679 if (udp_log_in_vain) { 680 char src[INET_ADDRSTRLEN]; 681 char dst[INET_ADDRSTRLEN]; 682 683 log(LOG_INFO, 684 "Connection attempt to UDP %s:%d from %s:%d\n", 685 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 686 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 687 } 688 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 689 UDPSTAT_INC(udps_noport); 690 if (m->m_flags & (M_BCAST | M_MCAST)) { 691 UDPSTAT_INC(udps_noportbcast); 692 goto badunlocked; 693 } 694 if (V_udp_blackhole) 695 goto badunlocked; 696 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 697 goto badunlocked; 698 *ip = save_ip; 699 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 700 return (IPPROTO_DONE); 701 } 702 703 /* 704 * Check the minimum TTL for socket. 705 */ 706 INP_RLOCK_ASSERT(inp); 707 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 708 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 709 INP_RUNLOCK(inp); 710 m_freem(m); 711 return (IPPROTO_DONE); 712 } 713 if (cscov_partial) { 714 struct udpcb *up; 715 716 up = intoudpcb(inp); 717 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 718 INP_RUNLOCK(inp); 719 m_freem(m); 720 return (IPPROTO_DONE); 721 } 722 } 723 724 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 725 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 726 INP_RUNLOCK(inp); 727 return (IPPROTO_DONE); 728 729 badunlocked: 730 m_freem(m); 731 return (IPPROTO_DONE); 732 } 733 #endif /* INET */ 734 735 /* 736 * Notify a udp user of an asynchronous error; just wake up so that they can 737 * collect error status. 738 */ 739 struct inpcb * 740 udp_notify(struct inpcb *inp, int errno) 741 { 742 743 /* 744 * While udp_ctlinput() always calls udp_notify() with a read lock 745 * when invoking it directly, in_pcbnotifyall() currently uses write 746 * locks due to sharing code with TCP. For now, accept either a read 747 * or a write lock, but a read lock is sufficient. 748 */ 749 INP_LOCK_ASSERT(inp); 750 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 751 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 752 RTFREE(inp->inp_route.ro_rt); 753 inp->inp_route.ro_rt = (struct rtentry *)NULL; 754 } 755 756 inp->inp_socket->so_error = errno; 757 sorwakeup(inp->inp_socket); 758 sowwakeup(inp->inp_socket); 759 return (inp); 760 } 761 762 #ifdef INET 763 static void 764 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 765 struct inpcbinfo *pcbinfo) 766 { 767 struct ip *ip = vip; 768 struct udphdr *uh; 769 struct in_addr faddr; 770 struct inpcb *inp; 771 772 faddr = ((struct sockaddr_in *)sa)->sin_addr; 773 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 774 return; 775 776 if (PRC_IS_REDIRECT(cmd)) { 777 /* signal EHOSTDOWN, as it flushes the cached route */ 778 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 779 return; 780 } 781 782 /* 783 * Hostdead is ugly because it goes linearly through all PCBs. 784 * 785 * XXX: We never get this from ICMP, otherwise it makes an excellent 786 * DoS attack on machines with many connections. 787 */ 788 if (cmd == PRC_HOSTDEAD) 789 ip = NULL; 790 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 791 return; 792 if (ip != NULL) { 793 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 794 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 795 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 796 if (inp != NULL) { 797 INP_RLOCK_ASSERT(inp); 798 if (inp->inp_socket != NULL) { 799 udp_notify(inp, inetctlerrmap[cmd]); 800 } 801 INP_RUNLOCK(inp); 802 } else { 803 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 804 ip->ip_src, uh->uh_sport, 805 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 806 if (inp != NULL) { 807 struct udpcb *up; 808 void *ctx; 809 udp_tun_icmp_t func; 810 811 up = intoudpcb(inp); 812 ctx = up->u_tun_ctx; 813 func = up->u_icmp_func; 814 INP_RUNLOCK(inp); 815 if (func != NULL) 816 (*func)(cmd, sa, vip, ctx); 817 } 818 } 819 } else 820 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 821 udp_notify); 822 } 823 void 824 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 825 { 826 827 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 828 } 829 830 void 831 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 832 { 833 834 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 835 } 836 #endif /* INET */ 837 838 static int 839 udp_pcblist(SYSCTL_HANDLER_ARGS) 840 { 841 int error, i, n; 842 struct inpcb *inp, **inp_list; 843 inp_gen_t gencnt; 844 struct xinpgen xig; 845 struct epoch_tracker et; 846 847 /* 848 * The process of preparing the PCB list is too time-consuming and 849 * resource-intensive to repeat twice on every request. 850 */ 851 if (req->oldptr == 0) { 852 n = V_udbinfo.ipi_count; 853 n += imax(n / 8, 10); 854 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 855 return (0); 856 } 857 858 if (req->newptr != 0) 859 return (EPERM); 860 861 /* 862 * OK, now we're committed to doing something. 863 */ 864 INP_INFO_RLOCK_ET(&V_udbinfo, et); 865 gencnt = V_udbinfo.ipi_gencnt; 866 n = V_udbinfo.ipi_count; 867 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 868 869 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 870 + n * sizeof(struct xinpcb)); 871 if (error != 0) 872 return (error); 873 874 xig.xig_len = sizeof xig; 875 xig.xig_count = n; 876 xig.xig_gen = gencnt; 877 xig.xig_sogen = so_gencnt; 878 error = SYSCTL_OUT(req, &xig, sizeof xig); 879 if (error) 880 return (error); 881 882 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 883 if (inp_list == NULL) 884 return (ENOMEM); 885 886 INP_INFO_RLOCK_ET(&V_udbinfo, et); 887 for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 888 inp = CK_LIST_NEXT(inp, inp_list)) { 889 INP_WLOCK(inp); 890 if (inp->inp_gencnt <= gencnt && 891 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 892 in_pcbref(inp); 893 inp_list[i++] = inp; 894 } 895 INP_WUNLOCK(inp); 896 } 897 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 898 n = i; 899 900 error = 0; 901 for (i = 0; i < n; i++) { 902 inp = inp_list[i]; 903 INP_RLOCK(inp); 904 if (inp->inp_gencnt <= gencnt) { 905 struct xinpcb xi; 906 907 in_pcbtoxinpcb(inp, &xi); 908 INP_RUNLOCK(inp); 909 error = SYSCTL_OUT(req, &xi, sizeof xi); 910 } else 911 INP_RUNLOCK(inp); 912 } 913 INP_INFO_WLOCK(&V_udbinfo); 914 for (i = 0; i < n; i++) { 915 inp = inp_list[i]; 916 INP_RLOCK(inp); 917 if (!in_pcbrele_rlocked(inp)) 918 INP_RUNLOCK(inp); 919 } 920 INP_INFO_WUNLOCK(&V_udbinfo); 921 922 if (!error) { 923 /* 924 * Give the user an updated idea of our state. If the 925 * generation differs from what we told her before, she knows 926 * that something happened while we were processing this 927 * request, and it might be necessary to retry. 928 */ 929 INP_INFO_RLOCK_ET(&V_udbinfo, et); 930 xig.xig_gen = V_udbinfo.ipi_gencnt; 931 xig.xig_sogen = so_gencnt; 932 xig.xig_count = V_udbinfo.ipi_count; 933 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 934 error = SYSCTL_OUT(req, &xig, sizeof xig); 935 } 936 free(inp_list, M_TEMP); 937 return (error); 938 } 939 940 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 941 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 942 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 943 944 #ifdef INET 945 static int 946 udp_getcred(SYSCTL_HANDLER_ARGS) 947 { 948 struct xucred xuc; 949 struct sockaddr_in addrs[2]; 950 struct inpcb *inp; 951 int error; 952 953 error = priv_check(req->td, PRIV_NETINET_GETCRED); 954 if (error) 955 return (error); 956 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 957 if (error) 958 return (error); 959 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 960 addrs[0].sin_addr, addrs[0].sin_port, 961 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 962 if (inp != NULL) { 963 INP_RLOCK_ASSERT(inp); 964 if (inp->inp_socket == NULL) 965 error = ENOENT; 966 if (error == 0) 967 error = cr_canseeinpcb(req->td->td_ucred, inp); 968 if (error == 0) 969 cru2x(inp->inp_cred, &xuc); 970 INP_RUNLOCK(inp); 971 } else 972 error = ENOENT; 973 if (error == 0) 974 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 975 return (error); 976 } 977 978 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 979 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 980 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 981 #endif /* INET */ 982 983 int 984 udp_ctloutput(struct socket *so, struct sockopt *sopt) 985 { 986 struct inpcb *inp; 987 struct udpcb *up; 988 int isudplite, error, optval; 989 990 error = 0; 991 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 992 inp = sotoinpcb(so); 993 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 994 INP_WLOCK(inp); 995 if (sopt->sopt_level != so->so_proto->pr_protocol) { 996 #ifdef INET6 997 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 998 INP_WUNLOCK(inp); 999 error = ip6_ctloutput(so, sopt); 1000 } 1001 #endif 1002 #if defined(INET) && defined(INET6) 1003 else 1004 #endif 1005 #ifdef INET 1006 { 1007 INP_WUNLOCK(inp); 1008 error = ip_ctloutput(so, sopt); 1009 } 1010 #endif 1011 return (error); 1012 } 1013 1014 switch (sopt->sopt_dir) { 1015 case SOPT_SET: 1016 switch (sopt->sopt_name) { 1017 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1018 #ifdef INET 1019 case UDP_ENCAP: 1020 if (!IPSEC_ENABLED(ipv4)) { 1021 INP_WUNLOCK(inp); 1022 return (ENOPROTOOPT); 1023 } 1024 error = UDPENCAP_PCBCTL(inp, sopt); 1025 break; 1026 #endif /* INET */ 1027 #endif /* IPSEC */ 1028 case UDPLITE_SEND_CSCOV: 1029 case UDPLITE_RECV_CSCOV: 1030 if (!isudplite) { 1031 INP_WUNLOCK(inp); 1032 error = ENOPROTOOPT; 1033 break; 1034 } 1035 INP_WUNLOCK(inp); 1036 error = sooptcopyin(sopt, &optval, sizeof(optval), 1037 sizeof(optval)); 1038 if (error != 0) 1039 break; 1040 inp = sotoinpcb(so); 1041 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1042 INP_WLOCK(inp); 1043 up = intoudpcb(inp); 1044 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1045 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1046 INP_WUNLOCK(inp); 1047 error = EINVAL; 1048 break; 1049 } 1050 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1051 up->u_txcslen = optval; 1052 else 1053 up->u_rxcslen = optval; 1054 INP_WUNLOCK(inp); 1055 break; 1056 default: 1057 INP_WUNLOCK(inp); 1058 error = ENOPROTOOPT; 1059 break; 1060 } 1061 break; 1062 case SOPT_GET: 1063 switch (sopt->sopt_name) { 1064 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1065 #ifdef INET 1066 case UDP_ENCAP: 1067 if (!IPSEC_ENABLED(ipv4)) { 1068 INP_WUNLOCK(inp); 1069 return (ENOPROTOOPT); 1070 } 1071 error = UDPENCAP_PCBCTL(inp, sopt); 1072 break; 1073 #endif /* INET */ 1074 #endif /* IPSEC */ 1075 case UDPLITE_SEND_CSCOV: 1076 case UDPLITE_RECV_CSCOV: 1077 if (!isudplite) { 1078 INP_WUNLOCK(inp); 1079 error = ENOPROTOOPT; 1080 break; 1081 } 1082 up = intoudpcb(inp); 1083 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1084 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1085 optval = up->u_txcslen; 1086 else 1087 optval = up->u_rxcslen; 1088 INP_WUNLOCK(inp); 1089 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1090 break; 1091 default: 1092 INP_WUNLOCK(inp); 1093 error = ENOPROTOOPT; 1094 break; 1095 } 1096 break; 1097 } 1098 return (error); 1099 } 1100 1101 #ifdef INET 1102 #define UH_WLOCKED 2 1103 #define UH_RLOCKED 1 1104 #define UH_UNLOCKED 0 1105 static int 1106 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1107 struct mbuf *control, struct thread *td) 1108 { 1109 struct udpiphdr *ui; 1110 int len = m->m_pkthdr.len; 1111 struct in_addr faddr, laddr; 1112 struct cmsghdr *cm; 1113 struct inpcbinfo *pcbinfo; 1114 struct sockaddr_in *sin, src; 1115 struct epoch_tracker et; 1116 int cscov_partial = 0; 1117 int error = 0; 1118 int ipflags; 1119 u_short fport, lport; 1120 int unlock_udbinfo, unlock_inp; 1121 u_char tos; 1122 uint8_t pr; 1123 uint16_t cscov = 0; 1124 uint32_t flowid = 0; 1125 uint8_t flowtype = M_HASHTYPE_NONE; 1126 1127 /* 1128 * udp_output() may need to temporarily bind or connect the current 1129 * inpcb. As such, we don't know up front whether we will need the 1130 * pcbinfo lock or not. Do any work to decide what is needed up 1131 * front before acquiring any locks. 1132 */ 1133 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1134 if (control) 1135 m_freem(control); 1136 m_freem(m); 1137 return (EMSGSIZE); 1138 } 1139 1140 src.sin_family = 0; 1141 sin = (struct sockaddr_in *)addr; 1142 if (sin == NULL || 1143 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1144 INP_WLOCK(inp); 1145 unlock_inp = UH_WLOCKED; 1146 } else { 1147 INP_RLOCK(inp); 1148 unlock_inp = UH_RLOCKED; 1149 } 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 if (unlock_inp == UH_WLOCKED) 1158 INP_WUNLOCK(inp); 1159 else 1160 INP_RUNLOCK(inp); 1161 m_freem(control); 1162 m_freem(m); 1163 return (EINVAL); 1164 } 1165 for (; control->m_len > 0; 1166 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1167 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1168 cm = mtod(control, struct cmsghdr *); 1169 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1170 || cm->cmsg_len > control->m_len) { 1171 error = EINVAL; 1172 break; 1173 } 1174 if (cm->cmsg_level != IPPROTO_IP) 1175 continue; 1176 1177 switch (cm->cmsg_type) { 1178 case IP_SENDSRCADDR: 1179 if (cm->cmsg_len != 1180 CMSG_LEN(sizeof(struct in_addr))) { 1181 error = EINVAL; 1182 break; 1183 } 1184 bzero(&src, sizeof(src)); 1185 src.sin_family = AF_INET; 1186 src.sin_len = sizeof(src); 1187 src.sin_port = inp->inp_lport; 1188 src.sin_addr = 1189 *(struct in_addr *)CMSG_DATA(cm); 1190 break; 1191 1192 case IP_TOS: 1193 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1194 error = EINVAL; 1195 break; 1196 } 1197 tos = *(u_char *)CMSG_DATA(cm); 1198 break; 1199 1200 case IP_FLOWID: 1201 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1202 error = EINVAL; 1203 break; 1204 } 1205 flowid = *(uint32_t *) CMSG_DATA(cm); 1206 break; 1207 1208 case IP_FLOWTYPE: 1209 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1210 error = EINVAL; 1211 break; 1212 } 1213 flowtype = *(uint32_t *) CMSG_DATA(cm); 1214 break; 1215 1216 #ifdef RSS 1217 case IP_RSSBUCKETID: 1218 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1219 error = EINVAL; 1220 break; 1221 } 1222 /* This is just a placeholder for now */ 1223 break; 1224 #endif /* RSS */ 1225 default: 1226 error = ENOPROTOOPT; 1227 break; 1228 } 1229 if (error) 1230 break; 1231 } 1232 m_freem(control); 1233 } 1234 if (error) { 1235 if (unlock_inp == UH_WLOCKED) 1236 INP_WUNLOCK(inp); 1237 else 1238 INP_RUNLOCK(inp); 1239 m_freem(m); 1240 return (error); 1241 } 1242 1243 /* 1244 * Depending on whether or not the application has bound or connected 1245 * the socket, we may have to do varying levels of work. The optimal 1246 * case is for a connected UDP socket, as a global lock isn't 1247 * required at all. 1248 * 1249 * In order to decide which we need, we require stability of the 1250 * inpcb binding, which we ensure by acquiring a read lock on the 1251 * inpcb. This doesn't strictly follow the lock order, so we play 1252 * the trylock and retry game; note that we may end up with more 1253 * conservative locks than required the second time around, so later 1254 * assertions have to accept that. Further analysis of the number of 1255 * misses under contention is required. 1256 * 1257 * XXXRW: Check that hash locking update here is correct. 1258 */ 1259 pr = inp->inp_socket->so_proto->pr_protocol; 1260 pcbinfo = udp_get_inpcbinfo(pr); 1261 sin = (struct sockaddr_in *)addr; 1262 if (sin != NULL && 1263 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1264 INP_HASH_WLOCK(pcbinfo); 1265 unlock_udbinfo = UH_WLOCKED; 1266 } else if ((sin != NULL && ( 1267 (sin->sin_addr.s_addr == INADDR_ANY) || 1268 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1269 (inp->inp_laddr.s_addr == INADDR_ANY) || 1270 (inp->inp_lport == 0))) || 1271 (src.sin_family == AF_INET)) { 1272 INP_HASH_RLOCK_ET(pcbinfo, et); 1273 unlock_udbinfo = UH_RLOCKED; 1274 } else 1275 unlock_udbinfo = UH_UNLOCKED; 1276 1277 /* 1278 * If the IP_SENDSRCADDR control message was specified, override the 1279 * source address for this datagram. Its use is invalidated if the 1280 * address thus specified is incomplete or clobbers other inpcbs. 1281 */ 1282 laddr = inp->inp_laddr; 1283 lport = inp->inp_lport; 1284 if (src.sin_family == AF_INET) { 1285 INP_HASH_LOCK_ASSERT(pcbinfo); 1286 if ((lport == 0) || 1287 (laddr.s_addr == INADDR_ANY && 1288 src.sin_addr.s_addr == INADDR_ANY)) { 1289 error = EINVAL; 1290 goto release; 1291 } 1292 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1293 &laddr.s_addr, &lport, td->td_ucred); 1294 if (error) 1295 goto release; 1296 } 1297 1298 /* 1299 * If a UDP socket has been connected, then a local address/port will 1300 * have been selected and bound. 1301 * 1302 * If a UDP socket has not been connected to, then an explicit 1303 * destination address must be used, in which case a local 1304 * address/port may not have been selected and bound. 1305 */ 1306 if (sin != NULL) { 1307 INP_LOCK_ASSERT(inp); 1308 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1309 error = EISCONN; 1310 goto release; 1311 } 1312 1313 /* 1314 * Jail may rewrite the destination address, so let it do 1315 * that before we use it. 1316 */ 1317 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1318 if (error) 1319 goto release; 1320 1321 /* 1322 * If a local address or port hasn't yet been selected, or if 1323 * the destination address needs to be rewritten due to using 1324 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1325 * to do the heavy lifting. Once a port is selected, we 1326 * commit the binding back to the socket; we also commit the 1327 * binding of the address if in jail. 1328 * 1329 * If we already have a valid binding and we're not 1330 * requesting a destination address rewrite, use a fast path. 1331 */ 1332 if (inp->inp_laddr.s_addr == INADDR_ANY || 1333 inp->inp_lport == 0 || 1334 sin->sin_addr.s_addr == INADDR_ANY || 1335 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1336 INP_HASH_LOCK_ASSERT(pcbinfo); 1337 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1338 &lport, &faddr.s_addr, &fport, NULL, 1339 td->td_ucred); 1340 if (error) 1341 goto release; 1342 1343 /* 1344 * XXXRW: Why not commit the port if the address is 1345 * !INADDR_ANY? 1346 */ 1347 /* Commit the local port if newly assigned. */ 1348 if (inp->inp_laddr.s_addr == INADDR_ANY && 1349 inp->inp_lport == 0) { 1350 INP_WLOCK_ASSERT(inp); 1351 INP_HASH_WLOCK_ASSERT(pcbinfo); 1352 /* 1353 * Remember addr if jailed, to prevent 1354 * rebinding. 1355 */ 1356 if (prison_flag(td->td_ucred, PR_IP4)) 1357 inp->inp_laddr = laddr; 1358 inp->inp_lport = lport; 1359 if (in_pcbinshash(inp) != 0) { 1360 inp->inp_lport = 0; 1361 error = EAGAIN; 1362 goto release; 1363 } 1364 inp->inp_flags |= INP_ANONPORT; 1365 } 1366 } else { 1367 faddr = sin->sin_addr; 1368 fport = sin->sin_port; 1369 } 1370 } else { 1371 INP_LOCK_ASSERT(inp); 1372 faddr = inp->inp_faddr; 1373 fport = inp->inp_fport; 1374 if (faddr.s_addr == INADDR_ANY) { 1375 error = ENOTCONN; 1376 goto release; 1377 } 1378 } 1379 1380 /* 1381 * Calculate data length and get a mbuf for UDP, IP, and possible 1382 * link-layer headers. Immediate slide the data pointer back forward 1383 * since we won't use that space at this layer. 1384 */ 1385 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1386 if (m == NULL) { 1387 error = ENOBUFS; 1388 goto release; 1389 } 1390 m->m_data += max_linkhdr; 1391 m->m_len -= max_linkhdr; 1392 m->m_pkthdr.len -= max_linkhdr; 1393 1394 /* 1395 * Fill in mbuf with extended UDP header and addresses and length put 1396 * into network format. 1397 */ 1398 ui = mtod(m, struct udpiphdr *); 1399 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1400 ui->ui_v = IPVERSION << 4; 1401 ui->ui_pr = pr; 1402 ui->ui_src = laddr; 1403 ui->ui_dst = faddr; 1404 ui->ui_sport = lport; 1405 ui->ui_dport = fport; 1406 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1407 if (pr == IPPROTO_UDPLITE) { 1408 struct udpcb *up; 1409 uint16_t plen; 1410 1411 up = intoudpcb(inp); 1412 cscov = up->u_txcslen; 1413 plen = (u_short)len + sizeof(struct udphdr); 1414 if (cscov >= plen) 1415 cscov = 0; 1416 ui->ui_len = htons(plen); 1417 ui->ui_ulen = htons(cscov); 1418 /* 1419 * For UDP-Lite, checksum coverage length of zero means 1420 * the entire UDPLite packet is covered by the checksum. 1421 */ 1422 cscov_partial = (cscov == 0) ? 0 : 1; 1423 } 1424 1425 /* 1426 * Set the Don't Fragment bit in the IP header. 1427 */ 1428 if (inp->inp_flags & INP_DONTFRAG) { 1429 struct ip *ip; 1430 1431 ip = (struct ip *)&ui->ui_i; 1432 ip->ip_off |= htons(IP_DF); 1433 } 1434 1435 ipflags = 0; 1436 if (inp->inp_socket->so_options & SO_DONTROUTE) 1437 ipflags |= IP_ROUTETOIF; 1438 if (inp->inp_socket->so_options & SO_BROADCAST) 1439 ipflags |= IP_ALLOWBROADCAST; 1440 if (inp->inp_flags & INP_ONESBCAST) 1441 ipflags |= IP_SENDONES; 1442 1443 #ifdef MAC 1444 mac_inpcb_create_mbuf(inp, m); 1445 #endif 1446 1447 /* 1448 * Set up checksum and output datagram. 1449 */ 1450 ui->ui_sum = 0; 1451 if (pr == IPPROTO_UDPLITE) { 1452 if (inp->inp_flags & INP_ONESBCAST) 1453 faddr.s_addr = INADDR_BROADCAST; 1454 if (cscov_partial) { 1455 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1456 ui->ui_sum = 0xffff; 1457 } else { 1458 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1459 ui->ui_sum = 0xffff; 1460 } 1461 } else if (V_udp_cksum) { 1462 if (inp->inp_flags & INP_ONESBCAST) 1463 faddr.s_addr = INADDR_BROADCAST; 1464 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1465 htons((u_short)len + sizeof(struct udphdr) + pr)); 1466 m->m_pkthdr.csum_flags = CSUM_UDP; 1467 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1468 } 1469 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1470 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1471 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1472 UDPSTAT_INC(udps_opackets); 1473 1474 /* 1475 * Setup flowid / RSS information for outbound socket. 1476 * 1477 * Once the UDP code decides to set a flowid some other way, 1478 * this allows the flowid to be overridden by userland. 1479 */ 1480 if (flowtype != M_HASHTYPE_NONE) { 1481 m->m_pkthdr.flowid = flowid; 1482 M_HASHTYPE_SET(m, flowtype); 1483 #ifdef RSS 1484 } else { 1485 uint32_t hash_val, hash_type; 1486 /* 1487 * Calculate an appropriate RSS hash for UDP and 1488 * UDP Lite. 1489 * 1490 * The called function will take care of figuring out 1491 * whether a 2-tuple or 4-tuple hash is required based 1492 * on the currently configured scheme. 1493 * 1494 * Later later on connected socket values should be 1495 * cached in the inpcb and reused, rather than constantly 1496 * re-calculating it. 1497 * 1498 * UDP Lite is a different protocol number and will 1499 * likely end up being hashed as a 2-tuple until 1500 * RSS / NICs grow UDP Lite protocol awareness. 1501 */ 1502 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1503 pr, &hash_val, &hash_type) == 0) { 1504 m->m_pkthdr.flowid = hash_val; 1505 M_HASHTYPE_SET(m, hash_type); 1506 } 1507 #endif 1508 } 1509 1510 #ifdef RSS 1511 /* 1512 * Don't override with the inp cached flowid value. 1513 * 1514 * Depending upon the kind of send being done, the inp 1515 * flowid/flowtype values may actually not be appropriate 1516 * for this particular socket send. 1517 * 1518 * We should either leave the flowid at zero (which is what is 1519 * currently done) or set it to some software generated 1520 * hash value based on the packet contents. 1521 */ 1522 ipflags |= IP_NODEFAULTFLOWID; 1523 #endif /* RSS */ 1524 1525 if (unlock_udbinfo == UH_WLOCKED) 1526 INP_HASH_WUNLOCK(pcbinfo); 1527 else if (unlock_udbinfo == UH_RLOCKED) 1528 INP_HASH_RUNLOCK_ET(pcbinfo, et); 1529 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1530 error = ip_output(m, inp->inp_options, 1531 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1532 inp->inp_moptions, inp); 1533 if (unlock_inp == UH_WLOCKED) 1534 INP_WUNLOCK(inp); 1535 else 1536 INP_RUNLOCK(inp); 1537 return (error); 1538 1539 release: 1540 if (unlock_udbinfo == UH_WLOCKED) { 1541 KASSERT(unlock_inp == UH_WLOCKED, 1542 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1543 INP_HASH_WUNLOCK(pcbinfo); 1544 INP_WUNLOCK(inp); 1545 } else if (unlock_udbinfo == UH_RLOCKED) { 1546 KASSERT(unlock_inp == UH_RLOCKED, 1547 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1548 INP_HASH_RUNLOCK_ET(pcbinfo, et); 1549 INP_RUNLOCK(inp); 1550 } else if (unlock_inp == UH_WLOCKED) 1551 INP_WUNLOCK(inp); 1552 else 1553 INP_RUNLOCK(inp); 1554 m_freem(m); 1555 return (error); 1556 } 1557 1558 static void 1559 udp_abort(struct socket *so) 1560 { 1561 struct inpcb *inp; 1562 struct inpcbinfo *pcbinfo; 1563 1564 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1565 inp = sotoinpcb(so); 1566 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1567 INP_WLOCK(inp); 1568 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1569 INP_HASH_WLOCK(pcbinfo); 1570 in_pcbdisconnect(inp); 1571 inp->inp_laddr.s_addr = INADDR_ANY; 1572 INP_HASH_WUNLOCK(pcbinfo); 1573 soisdisconnected(so); 1574 } 1575 INP_WUNLOCK(inp); 1576 } 1577 1578 static int 1579 udp_attach(struct socket *so, int proto, struct thread *td) 1580 { 1581 static uint32_t udp_flowid; 1582 struct inpcb *inp; 1583 struct inpcbinfo *pcbinfo; 1584 int error; 1585 1586 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1587 inp = sotoinpcb(so); 1588 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1589 error = soreserve(so, udp_sendspace, udp_recvspace); 1590 if (error) 1591 return (error); 1592 INP_INFO_WLOCK(pcbinfo); 1593 error = in_pcballoc(so, pcbinfo); 1594 if (error) { 1595 INP_INFO_WUNLOCK(pcbinfo); 1596 return (error); 1597 } 1598 1599 inp = sotoinpcb(so); 1600 inp->inp_vflag |= INP_IPV4; 1601 inp->inp_ip_ttl = V_ip_defttl; 1602 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1603 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1604 1605 error = udp_newudpcb(inp); 1606 if (error) { 1607 in_pcbdetach(inp); 1608 in_pcbfree(inp); 1609 INP_INFO_WUNLOCK(pcbinfo); 1610 return (error); 1611 } 1612 1613 INP_WUNLOCK(inp); 1614 INP_INFO_WUNLOCK(pcbinfo); 1615 return (0); 1616 } 1617 #endif /* INET */ 1618 1619 int 1620 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1621 { 1622 struct inpcb *inp; 1623 struct udpcb *up; 1624 1625 KASSERT(so->so_type == SOCK_DGRAM, 1626 ("udp_set_kernel_tunneling: !dgram")); 1627 inp = sotoinpcb(so); 1628 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1629 INP_WLOCK(inp); 1630 up = intoudpcb(inp); 1631 if ((up->u_tun_func != NULL) || 1632 (up->u_icmp_func != NULL)) { 1633 INP_WUNLOCK(inp); 1634 return (EBUSY); 1635 } 1636 up->u_tun_func = f; 1637 up->u_icmp_func = i; 1638 up->u_tun_ctx = ctx; 1639 INP_WUNLOCK(inp); 1640 return (0); 1641 } 1642 1643 #ifdef INET 1644 static int 1645 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1646 { 1647 struct inpcb *inp; 1648 struct inpcbinfo *pcbinfo; 1649 int error; 1650 1651 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1652 inp = sotoinpcb(so); 1653 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1654 INP_WLOCK(inp); 1655 INP_HASH_WLOCK(pcbinfo); 1656 error = in_pcbbind(inp, nam, td->td_ucred); 1657 INP_HASH_WUNLOCK(pcbinfo); 1658 INP_WUNLOCK(inp); 1659 return (error); 1660 } 1661 1662 static void 1663 udp_close(struct socket *so) 1664 { 1665 struct inpcb *inp; 1666 struct inpcbinfo *pcbinfo; 1667 1668 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1669 inp = sotoinpcb(so); 1670 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1671 INP_WLOCK(inp); 1672 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1673 INP_HASH_WLOCK(pcbinfo); 1674 in_pcbdisconnect(inp); 1675 inp->inp_laddr.s_addr = INADDR_ANY; 1676 INP_HASH_WUNLOCK(pcbinfo); 1677 soisdisconnected(so); 1678 } 1679 INP_WUNLOCK(inp); 1680 } 1681 1682 static int 1683 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1684 { 1685 struct inpcb *inp; 1686 struct inpcbinfo *pcbinfo; 1687 struct sockaddr_in *sin; 1688 int error; 1689 1690 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1691 inp = sotoinpcb(so); 1692 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1693 INP_WLOCK(inp); 1694 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1695 INP_WUNLOCK(inp); 1696 return (EISCONN); 1697 } 1698 sin = (struct sockaddr_in *)nam; 1699 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1700 if (error != 0) { 1701 INP_WUNLOCK(inp); 1702 return (error); 1703 } 1704 INP_HASH_WLOCK(pcbinfo); 1705 error = in_pcbconnect(inp, nam, td->td_ucred); 1706 INP_HASH_WUNLOCK(pcbinfo); 1707 if (error == 0) 1708 soisconnected(so); 1709 INP_WUNLOCK(inp); 1710 return (error); 1711 } 1712 1713 static void 1714 udp_detach(struct socket *so) 1715 { 1716 struct inpcb *inp; 1717 struct inpcbinfo *pcbinfo; 1718 struct udpcb *up; 1719 1720 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1721 inp = sotoinpcb(so); 1722 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1723 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1724 ("udp_detach: not disconnected")); 1725 INP_INFO_WLOCK(pcbinfo); 1726 INP_WLOCK(inp); 1727 up = intoudpcb(inp); 1728 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1729 inp->inp_ppcb = NULL; 1730 in_pcbdetach(inp); 1731 in_pcbfree(inp); 1732 INP_INFO_WUNLOCK(pcbinfo); 1733 udp_discardcb(up); 1734 } 1735 1736 static int 1737 udp_disconnect(struct socket *so) 1738 { 1739 struct inpcb *inp; 1740 struct inpcbinfo *pcbinfo; 1741 1742 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1743 inp = sotoinpcb(so); 1744 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1745 INP_WLOCK(inp); 1746 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1747 INP_WUNLOCK(inp); 1748 return (ENOTCONN); 1749 } 1750 INP_HASH_WLOCK(pcbinfo); 1751 in_pcbdisconnect(inp); 1752 inp->inp_laddr.s_addr = INADDR_ANY; 1753 INP_HASH_WUNLOCK(pcbinfo); 1754 SOCK_LOCK(so); 1755 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1756 SOCK_UNLOCK(so); 1757 INP_WUNLOCK(inp); 1758 return (0); 1759 } 1760 1761 static int 1762 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1763 struct mbuf *control, struct thread *td) 1764 { 1765 struct inpcb *inp; 1766 1767 inp = sotoinpcb(so); 1768 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1769 return (udp_output(inp, m, addr, control, td)); 1770 } 1771 #endif /* INET */ 1772 1773 int 1774 udp_shutdown(struct socket *so) 1775 { 1776 struct inpcb *inp; 1777 1778 inp = sotoinpcb(so); 1779 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1780 INP_WLOCK(inp); 1781 socantsendmore(so); 1782 INP_WUNLOCK(inp); 1783 return (0); 1784 } 1785 1786 #ifdef INET 1787 struct pr_usrreqs udp_usrreqs = { 1788 .pru_abort = udp_abort, 1789 .pru_attach = udp_attach, 1790 .pru_bind = udp_bind, 1791 .pru_connect = udp_connect, 1792 .pru_control = in_control, 1793 .pru_detach = udp_detach, 1794 .pru_disconnect = udp_disconnect, 1795 .pru_peeraddr = in_getpeeraddr, 1796 .pru_send = udp_send, 1797 .pru_soreceive = soreceive_dgram, 1798 .pru_sosend = sosend_dgram, 1799 .pru_shutdown = udp_shutdown, 1800 .pru_sockaddr = in_getsockaddr, 1801 .pru_sosetlabel = in_pcbsosetlabel, 1802 .pru_close = udp_close, 1803 }; 1804 #endif /* INET */ 1805