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 if (m->m_len < iphlen + sizeof(struct udphdr)) { 425 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 426 UDPSTAT_INC(udps_hdrops); 427 return (IPPROTO_DONE); 428 } 429 } 430 ip = mtod(m, struct ip *); 431 uh = (struct udphdr *)((caddr_t)ip + iphlen); 432 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 433 434 /* 435 * Destination port of 0 is illegal, based on RFC768. 436 */ 437 if (uh->uh_dport == 0) 438 goto badunlocked; 439 440 /* 441 * Construct sockaddr format source address. Stuff source address 442 * and datagram in user buffer. 443 */ 444 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 445 udp_in[0].sin_len = sizeof(struct sockaddr_in); 446 udp_in[0].sin_family = AF_INET; 447 udp_in[0].sin_port = uh->uh_sport; 448 udp_in[0].sin_addr = ip->ip_src; 449 udp_in[1].sin_len = sizeof(struct sockaddr_in); 450 udp_in[1].sin_family = AF_INET; 451 udp_in[1].sin_port = uh->uh_dport; 452 udp_in[1].sin_addr = ip->ip_dst; 453 454 /* 455 * Make mbuf data length reflect UDP length. If not enough data to 456 * reflect UDP length, drop. 457 */ 458 len = ntohs((u_short)uh->uh_ulen); 459 ip_len = ntohs(ip->ip_len) - iphlen; 460 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 461 /* Zero means checksum over the complete packet. */ 462 if (len == 0) 463 len = ip_len; 464 cscov_partial = 0; 465 } 466 if (ip_len != len) { 467 if (len > ip_len || len < sizeof(struct udphdr)) { 468 UDPSTAT_INC(udps_badlen); 469 goto badunlocked; 470 } 471 if (proto == IPPROTO_UDP) 472 m_adj(m, len - ip_len); 473 } 474 475 /* 476 * Save a copy of the IP header in case we want restore it for 477 * sending an ICMP error message in response. 478 */ 479 if (!V_udp_blackhole) 480 save_ip = *ip; 481 else 482 memset(&save_ip, 0, sizeof(save_ip)); 483 484 /* 485 * Checksum extended UDP header and data. 486 */ 487 if (uh->uh_sum) { 488 u_short uh_sum; 489 490 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 491 !cscov_partial) { 492 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 493 uh_sum = m->m_pkthdr.csum_data; 494 else 495 uh_sum = in_pseudo(ip->ip_src.s_addr, 496 ip->ip_dst.s_addr, htonl((u_short)len + 497 m->m_pkthdr.csum_data + proto)); 498 uh_sum ^= 0xffff; 499 } else { 500 char b[9]; 501 502 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 503 bzero(((struct ipovly *)ip)->ih_x1, 9); 504 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 505 uh->uh_ulen : htons(ip_len); 506 uh_sum = in_cksum(m, len + sizeof (struct ip)); 507 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 508 } 509 if (uh_sum) { 510 UDPSTAT_INC(udps_badsum); 511 m_freem(m); 512 return (IPPROTO_DONE); 513 } 514 } else { 515 if (proto == IPPROTO_UDP) { 516 UDPSTAT_INC(udps_nosum); 517 } else { 518 /* UDPLite requires a checksum */ 519 /* XXX: What is the right UDPLite MIB counter here? */ 520 m_freem(m); 521 return (IPPROTO_DONE); 522 } 523 } 524 525 pcbinfo = udp_get_inpcbinfo(proto); 526 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 527 in_broadcast(ip->ip_dst, ifp)) { 528 struct inpcb *last; 529 struct inpcbhead *pcblist; 530 531 INP_INFO_RLOCK_ET(pcbinfo, et); 532 pcblist = udp_get_pcblist(proto); 533 last = NULL; 534 CK_LIST_FOREACH(inp, pcblist, inp_list) { 535 if (inp->inp_lport != uh->uh_dport) 536 continue; 537 #ifdef INET6 538 if ((inp->inp_vflag & INP_IPV4) == 0) 539 continue; 540 #endif 541 if (inp->inp_laddr.s_addr != INADDR_ANY && 542 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 543 continue; 544 if (inp->inp_faddr.s_addr != INADDR_ANY && 545 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 546 continue; 547 if (inp->inp_fport != 0 && 548 inp->inp_fport != uh->uh_sport) 549 continue; 550 551 INP_RLOCK(inp); 552 553 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 554 INP_RUNLOCK(inp); 555 continue; 556 } 557 558 /* 559 * XXXRW: Because we weren't holding either the inpcb 560 * or the hash lock when we checked for a match 561 * before, we should probably recheck now that the 562 * inpcb lock is held. 563 */ 564 565 /* 566 * Handle socket delivery policy for any-source 567 * and source-specific multicast. [RFC3678] 568 */ 569 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 570 struct ip_moptions *imo; 571 struct sockaddr_in group; 572 int blocked; 573 574 imo = inp->inp_moptions; 575 if (imo == NULL) { 576 INP_RUNLOCK(inp); 577 continue; 578 } 579 bzero(&group, sizeof(struct sockaddr_in)); 580 group.sin_len = sizeof(struct sockaddr_in); 581 group.sin_family = AF_INET; 582 group.sin_addr = ip->ip_dst; 583 584 blocked = imo_multi_filter(imo, ifp, 585 (struct sockaddr *)&group, 586 (struct sockaddr *)&udp_in[0]); 587 if (blocked != MCAST_PASS) { 588 if (blocked == MCAST_NOTGMEMBER) 589 IPSTAT_INC(ips_notmember); 590 if (blocked == MCAST_NOTSMEMBER || 591 blocked == MCAST_MUTED) 592 UDPSTAT_INC(udps_filtermcast); 593 INP_RUNLOCK(inp); 594 continue; 595 } 596 } 597 if (last != NULL) { 598 struct mbuf *n; 599 600 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 601 NULL) { 602 if (proto == IPPROTO_UDPLITE) 603 UDPLITE_PROBE(receive, NULL, last, ip, 604 last, uh); 605 else 606 UDP_PROBE(receive, NULL, last, ip, last, 607 uh); 608 if (udp_append(last, ip, n, iphlen, 609 udp_in)) { 610 goto inp_lost; 611 } 612 } 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_noportbcast); 636 if (inp) 637 INP_RUNLOCK(inp); 638 INP_INFO_RUNLOCK_ET(pcbinfo, et); 639 goto badunlocked; 640 } 641 if (proto == IPPROTO_UDPLITE) 642 UDPLITE_PROBE(receive, NULL, last, ip, last, uh); 643 else 644 UDP_PROBE(receive, NULL, last, ip, last, uh); 645 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 646 INP_RUNLOCK(last); 647 inp_lost: 648 INP_INFO_RUNLOCK_ET(pcbinfo, et); 649 return (IPPROTO_DONE); 650 } 651 652 /* 653 * Locate pcb for datagram. 654 */ 655 656 /* 657 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 658 */ 659 if ((m->m_flags & M_IP_NEXTHOP) && 660 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 661 struct sockaddr_in *next_hop; 662 663 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 664 665 /* 666 * Transparently forwarded. Pretend to be the destination. 667 * Already got one like this? 668 */ 669 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 670 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 671 if (!inp) { 672 /* 673 * It's new. Try to find the ambushing socket. 674 * Because we've rewritten the destination address, 675 * any hardware-generated hash is ignored. 676 */ 677 inp = in_pcblookup(pcbinfo, ip->ip_src, 678 uh->uh_sport, next_hop->sin_addr, 679 next_hop->sin_port ? htons(next_hop->sin_port) : 680 uh->uh_dport, INPLOOKUP_WILDCARD | 681 INPLOOKUP_RLOCKPCB, ifp); 682 } 683 /* Remove the tag from the packet. We don't need it anymore. */ 684 m_tag_delete(m, fwd_tag); 685 m->m_flags &= ~M_IP_NEXTHOP; 686 } else 687 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 688 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 689 INPLOOKUP_RLOCKPCB, ifp, m); 690 if (inp == NULL) { 691 if (udp_log_in_vain) { 692 char src[INET_ADDRSTRLEN]; 693 char dst[INET_ADDRSTRLEN]; 694 695 log(LOG_INFO, 696 "Connection attempt to UDP %s:%d from %s:%d\n", 697 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 698 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 699 } 700 if (proto == IPPROTO_UDPLITE) 701 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); 702 else 703 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 704 UDPSTAT_INC(udps_noport); 705 if (m->m_flags & (M_BCAST | M_MCAST)) { 706 UDPSTAT_INC(udps_noportbcast); 707 goto badunlocked; 708 } 709 if (V_udp_blackhole) 710 goto badunlocked; 711 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 712 goto badunlocked; 713 *ip = save_ip; 714 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 715 return (IPPROTO_DONE); 716 } 717 718 /* 719 * Check the minimum TTL for socket. 720 */ 721 INP_RLOCK_ASSERT(inp); 722 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 723 if (proto == IPPROTO_UDPLITE) 724 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 725 else 726 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 727 INP_RUNLOCK(inp); 728 m_freem(m); 729 return (IPPROTO_DONE); 730 } 731 if (cscov_partial) { 732 struct udpcb *up; 733 734 up = intoudpcb(inp); 735 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 736 INP_RUNLOCK(inp); 737 m_freem(m); 738 return (IPPROTO_DONE); 739 } 740 } 741 742 if (proto == IPPROTO_UDPLITE) 743 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 744 else 745 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 746 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 747 INP_RUNLOCK(inp); 748 return (IPPROTO_DONE); 749 750 badunlocked: 751 m_freem(m); 752 return (IPPROTO_DONE); 753 } 754 #endif /* INET */ 755 756 /* 757 * Notify a udp user of an asynchronous error; just wake up so that they can 758 * collect error status. 759 */ 760 struct inpcb * 761 udp_notify(struct inpcb *inp, int errno) 762 { 763 764 INP_WLOCK_ASSERT(inp); 765 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 766 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 767 RTFREE(inp->inp_route.ro_rt); 768 inp->inp_route.ro_rt = (struct rtentry *)NULL; 769 } 770 771 inp->inp_socket->so_error = errno; 772 sorwakeup(inp->inp_socket); 773 sowwakeup(inp->inp_socket); 774 return (inp); 775 } 776 777 #ifdef INET 778 static void 779 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 780 struct inpcbinfo *pcbinfo) 781 { 782 struct ip *ip = vip; 783 struct udphdr *uh; 784 struct in_addr faddr; 785 struct inpcb *inp; 786 787 faddr = ((struct sockaddr_in *)sa)->sin_addr; 788 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 789 return; 790 791 if (PRC_IS_REDIRECT(cmd)) { 792 /* signal EHOSTDOWN, as it flushes the cached route */ 793 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 794 return; 795 } 796 797 /* 798 * Hostdead is ugly because it goes linearly through all PCBs. 799 * 800 * XXX: We never get this from ICMP, otherwise it makes an excellent 801 * DoS attack on machines with many connections. 802 */ 803 if (cmd == PRC_HOSTDEAD) 804 ip = NULL; 805 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 806 return; 807 if (ip != NULL) { 808 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 809 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 810 ip->ip_src, uh->uh_sport, INPLOOKUP_WLOCKPCB, NULL); 811 if (inp != NULL) { 812 INP_WLOCK_ASSERT(inp); 813 if (inp->inp_socket != NULL) { 814 udp_notify(inp, inetctlerrmap[cmd]); 815 } 816 INP_WUNLOCK(inp); 817 } else { 818 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 819 ip->ip_src, uh->uh_sport, 820 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 821 if (inp != NULL) { 822 struct udpcb *up; 823 void *ctx; 824 udp_tun_icmp_t func; 825 826 up = intoudpcb(inp); 827 ctx = up->u_tun_ctx; 828 func = up->u_icmp_func; 829 INP_RUNLOCK(inp); 830 if (func != NULL) 831 (*func)(cmd, sa, vip, ctx); 832 } 833 } 834 } else 835 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 836 udp_notify); 837 } 838 void 839 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 840 { 841 842 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 843 } 844 845 void 846 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 847 { 848 849 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 850 } 851 #endif /* INET */ 852 853 static int 854 udp_pcblist(SYSCTL_HANDLER_ARGS) 855 { 856 int error, i, n; 857 struct inpcb *inp, **inp_list; 858 inp_gen_t gencnt; 859 struct xinpgen xig; 860 struct epoch_tracker et; 861 862 /* 863 * The process of preparing the PCB list is too time-consuming and 864 * resource-intensive to repeat twice on every request. 865 */ 866 if (req->oldptr == 0) { 867 n = V_udbinfo.ipi_count; 868 n += imax(n / 8, 10); 869 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 870 return (0); 871 } 872 873 if (req->newptr != 0) 874 return (EPERM); 875 876 /* 877 * OK, now we're committed to doing something. 878 */ 879 INP_INFO_RLOCK_ET(&V_udbinfo, et); 880 gencnt = V_udbinfo.ipi_gencnt; 881 n = V_udbinfo.ipi_count; 882 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 883 884 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 885 + n * sizeof(struct xinpcb)); 886 if (error != 0) 887 return (error); 888 889 bzero(&xig, sizeof(xig)); 890 xig.xig_len = sizeof xig; 891 xig.xig_count = n; 892 xig.xig_gen = gencnt; 893 xig.xig_sogen = so_gencnt; 894 error = SYSCTL_OUT(req, &xig, sizeof xig); 895 if (error) 896 return (error); 897 898 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 899 if (inp_list == NULL) 900 return (ENOMEM); 901 902 INP_INFO_RLOCK_ET(&V_udbinfo, et); 903 for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 904 inp = CK_LIST_NEXT(inp, inp_list)) { 905 INP_WLOCK(inp); 906 if (inp->inp_gencnt <= gencnt && 907 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 908 in_pcbref(inp); 909 inp_list[i++] = inp; 910 } 911 INP_WUNLOCK(inp); 912 } 913 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 914 n = i; 915 916 error = 0; 917 for (i = 0; i < n; i++) { 918 inp = inp_list[i]; 919 INP_RLOCK(inp); 920 if (inp->inp_gencnt <= gencnt) { 921 struct xinpcb xi; 922 923 in_pcbtoxinpcb(inp, &xi); 924 INP_RUNLOCK(inp); 925 error = SYSCTL_OUT(req, &xi, sizeof xi); 926 } else 927 INP_RUNLOCK(inp); 928 } 929 INP_INFO_WLOCK(&V_udbinfo); 930 for (i = 0; i < n; i++) { 931 inp = inp_list[i]; 932 INP_RLOCK(inp); 933 if (!in_pcbrele_rlocked(inp)) 934 INP_RUNLOCK(inp); 935 } 936 INP_INFO_WUNLOCK(&V_udbinfo); 937 938 if (!error) { 939 /* 940 * Give the user an updated idea of our state. If the 941 * generation differs from what we told her before, she knows 942 * that something happened while we were processing this 943 * request, and it might be necessary to retry. 944 */ 945 INP_INFO_RLOCK_ET(&V_udbinfo, et); 946 xig.xig_gen = V_udbinfo.ipi_gencnt; 947 xig.xig_sogen = so_gencnt; 948 xig.xig_count = V_udbinfo.ipi_count; 949 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 950 error = SYSCTL_OUT(req, &xig, sizeof xig); 951 } 952 free(inp_list, M_TEMP); 953 return (error); 954 } 955 956 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 957 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 958 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 959 960 #ifdef INET 961 static int 962 udp_getcred(SYSCTL_HANDLER_ARGS) 963 { 964 struct xucred xuc; 965 struct sockaddr_in addrs[2]; 966 struct inpcb *inp; 967 int error; 968 969 error = priv_check(req->td, PRIV_NETINET_GETCRED); 970 if (error) 971 return (error); 972 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 973 if (error) 974 return (error); 975 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 976 addrs[0].sin_addr, addrs[0].sin_port, 977 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 978 if (inp != NULL) { 979 INP_RLOCK_ASSERT(inp); 980 if (inp->inp_socket == NULL) 981 error = ENOENT; 982 if (error == 0) 983 error = cr_canseeinpcb(req->td->td_ucred, inp); 984 if (error == 0) 985 cru2x(inp->inp_cred, &xuc); 986 INP_RUNLOCK(inp); 987 } else 988 error = ENOENT; 989 if (error == 0) 990 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 991 return (error); 992 } 993 994 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 995 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 996 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 997 #endif /* INET */ 998 999 int 1000 udp_ctloutput(struct socket *so, struct sockopt *sopt) 1001 { 1002 struct inpcb *inp; 1003 struct udpcb *up; 1004 int isudplite, error, optval; 1005 1006 error = 0; 1007 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 1008 inp = sotoinpcb(so); 1009 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1010 INP_WLOCK(inp); 1011 if (sopt->sopt_level != so->so_proto->pr_protocol) { 1012 #ifdef INET6 1013 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 1014 INP_WUNLOCK(inp); 1015 error = ip6_ctloutput(so, sopt); 1016 } 1017 #endif 1018 #if defined(INET) && defined(INET6) 1019 else 1020 #endif 1021 #ifdef INET 1022 { 1023 INP_WUNLOCK(inp); 1024 error = ip_ctloutput(so, sopt); 1025 } 1026 #endif 1027 return (error); 1028 } 1029 1030 switch (sopt->sopt_dir) { 1031 case SOPT_SET: 1032 switch (sopt->sopt_name) { 1033 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1034 #ifdef INET 1035 case UDP_ENCAP: 1036 if (!IPSEC_ENABLED(ipv4)) { 1037 INP_WUNLOCK(inp); 1038 return (ENOPROTOOPT); 1039 } 1040 error = UDPENCAP_PCBCTL(inp, sopt); 1041 break; 1042 #endif /* INET */ 1043 #endif /* IPSEC */ 1044 case UDPLITE_SEND_CSCOV: 1045 case UDPLITE_RECV_CSCOV: 1046 if (!isudplite) { 1047 INP_WUNLOCK(inp); 1048 error = ENOPROTOOPT; 1049 break; 1050 } 1051 INP_WUNLOCK(inp); 1052 error = sooptcopyin(sopt, &optval, sizeof(optval), 1053 sizeof(optval)); 1054 if (error != 0) 1055 break; 1056 inp = sotoinpcb(so); 1057 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1058 INP_WLOCK(inp); 1059 up = intoudpcb(inp); 1060 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1061 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1062 INP_WUNLOCK(inp); 1063 error = EINVAL; 1064 break; 1065 } 1066 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1067 up->u_txcslen = optval; 1068 else 1069 up->u_rxcslen = optval; 1070 INP_WUNLOCK(inp); 1071 break; 1072 default: 1073 INP_WUNLOCK(inp); 1074 error = ENOPROTOOPT; 1075 break; 1076 } 1077 break; 1078 case SOPT_GET: 1079 switch (sopt->sopt_name) { 1080 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1081 #ifdef INET 1082 case UDP_ENCAP: 1083 if (!IPSEC_ENABLED(ipv4)) { 1084 INP_WUNLOCK(inp); 1085 return (ENOPROTOOPT); 1086 } 1087 error = UDPENCAP_PCBCTL(inp, sopt); 1088 break; 1089 #endif /* INET */ 1090 #endif /* IPSEC */ 1091 case UDPLITE_SEND_CSCOV: 1092 case UDPLITE_RECV_CSCOV: 1093 if (!isudplite) { 1094 INP_WUNLOCK(inp); 1095 error = ENOPROTOOPT; 1096 break; 1097 } 1098 up = intoudpcb(inp); 1099 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1100 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1101 optval = up->u_txcslen; 1102 else 1103 optval = up->u_rxcslen; 1104 INP_WUNLOCK(inp); 1105 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1106 break; 1107 default: 1108 INP_WUNLOCK(inp); 1109 error = ENOPROTOOPT; 1110 break; 1111 } 1112 break; 1113 } 1114 return (error); 1115 } 1116 1117 #ifdef INET 1118 #define UH_WLOCKED 2 1119 #define UH_RLOCKED 1 1120 #define UH_UNLOCKED 0 1121 static int 1122 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1123 struct mbuf *control, struct thread *td) 1124 { 1125 struct udpiphdr *ui; 1126 int len = m->m_pkthdr.len; 1127 struct in_addr faddr, laddr; 1128 struct cmsghdr *cm; 1129 struct inpcbinfo *pcbinfo; 1130 struct sockaddr_in *sin, src; 1131 struct epoch_tracker et; 1132 int cscov_partial = 0; 1133 int error = 0; 1134 int ipflags; 1135 u_short fport, lport; 1136 int unlock_udbinfo, unlock_inp; 1137 u_char tos; 1138 uint8_t pr; 1139 uint16_t cscov = 0; 1140 uint32_t flowid = 0; 1141 uint8_t flowtype = M_HASHTYPE_NONE; 1142 1143 /* 1144 * udp_output() may need to temporarily bind or connect the current 1145 * inpcb. As such, we don't know up front whether we will need the 1146 * pcbinfo lock or not. Do any work to decide what is needed up 1147 * front before acquiring any locks. 1148 */ 1149 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1150 if (control) 1151 m_freem(control); 1152 m_freem(m); 1153 return (EMSGSIZE); 1154 } 1155 1156 src.sin_family = 0; 1157 sin = (struct sockaddr_in *)addr; 1158 retry: 1159 if (sin == NULL || 1160 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1161 INP_WLOCK(inp); 1162 /* 1163 * In case we lost a race and another thread bound addr/port 1164 * on the inp we cannot keep the wlock (which still would be 1165 * fine) as further down, based on these values we make 1166 * decisions for the pcbinfo lock. If the locks are not in 1167 * synch the assertions on unlock will fire, hence we go for 1168 * one retry loop. 1169 */ 1170 if (sin != NULL && (inp->inp_laddr.s_addr != INADDR_ANY || 1171 inp->inp_lport != 0)) { 1172 INP_WUNLOCK(inp); 1173 goto retry; 1174 } 1175 unlock_inp = UH_WLOCKED; 1176 } else { 1177 INP_RLOCK(inp); 1178 unlock_inp = UH_RLOCKED; 1179 } 1180 tos = inp->inp_ip_tos; 1181 if (control != NULL) { 1182 /* 1183 * XXX: Currently, we assume all the optional information is 1184 * stored in a single mbuf. 1185 */ 1186 if (control->m_next) { 1187 if (unlock_inp == UH_WLOCKED) 1188 INP_WUNLOCK(inp); 1189 else 1190 INP_RUNLOCK(inp); 1191 m_freem(control); 1192 m_freem(m); 1193 return (EINVAL); 1194 } 1195 for (; control->m_len > 0; 1196 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1197 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1198 cm = mtod(control, struct cmsghdr *); 1199 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1200 || cm->cmsg_len > control->m_len) { 1201 error = EINVAL; 1202 break; 1203 } 1204 if (cm->cmsg_level != IPPROTO_IP) 1205 continue; 1206 1207 switch (cm->cmsg_type) { 1208 case IP_SENDSRCADDR: 1209 if (cm->cmsg_len != 1210 CMSG_LEN(sizeof(struct in_addr))) { 1211 error = EINVAL; 1212 break; 1213 } 1214 bzero(&src, sizeof(src)); 1215 src.sin_family = AF_INET; 1216 src.sin_len = sizeof(src); 1217 src.sin_port = inp->inp_lport; 1218 src.sin_addr = 1219 *(struct in_addr *)CMSG_DATA(cm); 1220 break; 1221 1222 case IP_TOS: 1223 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1224 error = EINVAL; 1225 break; 1226 } 1227 tos = *(u_char *)CMSG_DATA(cm); 1228 break; 1229 1230 case IP_FLOWID: 1231 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1232 error = EINVAL; 1233 break; 1234 } 1235 flowid = *(uint32_t *) CMSG_DATA(cm); 1236 break; 1237 1238 case IP_FLOWTYPE: 1239 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1240 error = EINVAL; 1241 break; 1242 } 1243 flowtype = *(uint32_t *) CMSG_DATA(cm); 1244 break; 1245 1246 #ifdef RSS 1247 case IP_RSSBUCKETID: 1248 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1249 error = EINVAL; 1250 break; 1251 } 1252 /* This is just a placeholder for now */ 1253 break; 1254 #endif /* RSS */ 1255 default: 1256 error = ENOPROTOOPT; 1257 break; 1258 } 1259 if (error) 1260 break; 1261 } 1262 m_freem(control); 1263 } 1264 if (error) { 1265 if (unlock_inp == UH_WLOCKED) 1266 INP_WUNLOCK(inp); 1267 else 1268 INP_RUNLOCK(inp); 1269 m_freem(m); 1270 return (error); 1271 } 1272 1273 /* 1274 * In the old days, depending on whether or not the application had 1275 * bound or connected the socket, we had to do varying levels of work. 1276 * The optimal case was for a connected UDP socket, as a global lock 1277 * wasn't required at all. 1278 * In order to decide which we need, we required stability of the 1279 * inpcb binding, which we ensured by acquiring a read lock on the 1280 * inpcb. This didn't strictly follow the lock order, so we played 1281 * the trylock and retry game. 1282 * With the re-introduction of the route-cache in some cases, we started 1283 * to acquire an early inp wlock and a possible race during re-lock 1284 * went away. With the introduction of epoch(9) some read locking 1285 * became epoch(9) and the lock-order issues also went away. 1286 * Due to route-cache we may now hold more conservative locks than 1287 * otherwise required and have split up the 2nd case in case 2 and 3 1288 * in order to keep the udpinfo lock level in sync with the inp one 1289 * for the IP_SENDSRCADDR case below. 1290 */ 1291 pr = inp->inp_socket->so_proto->pr_protocol; 1292 pcbinfo = udp_get_inpcbinfo(pr); 1293 if (sin != NULL && 1294 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1295 INP_HASH_WLOCK(pcbinfo); 1296 unlock_udbinfo = UH_WLOCKED; 1297 } else if (sin != NULL && 1298 (sin->sin_addr.s_addr == INADDR_ANY || 1299 sin->sin_addr.s_addr == INADDR_BROADCAST || 1300 inp->inp_laddr.s_addr == INADDR_ANY || 1301 inp->inp_lport == 0)) { 1302 INP_HASH_RLOCK_ET(pcbinfo, et); 1303 unlock_udbinfo = UH_RLOCKED; 1304 } else if (src.sin_family == AF_INET) { 1305 if (unlock_inp == UH_WLOCKED) { 1306 INP_HASH_WLOCK(pcbinfo); 1307 unlock_udbinfo = UH_WLOCKED; 1308 } else { 1309 INP_HASH_RLOCK_ET(pcbinfo, et); 1310 unlock_udbinfo = UH_RLOCKED; 1311 } 1312 } else 1313 unlock_udbinfo = UH_UNLOCKED; 1314 1315 /* 1316 * If the IP_SENDSRCADDR control message was specified, override the 1317 * source address for this datagram. Its use is invalidated if the 1318 * address thus specified is incomplete or clobbers other inpcbs. 1319 */ 1320 laddr = inp->inp_laddr; 1321 lport = inp->inp_lport; 1322 if (src.sin_family == AF_INET) { 1323 INP_HASH_LOCK_ASSERT(pcbinfo); 1324 if ((lport == 0) || 1325 (laddr.s_addr == INADDR_ANY && 1326 src.sin_addr.s_addr == INADDR_ANY)) { 1327 error = EINVAL; 1328 goto release; 1329 } 1330 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1331 &laddr.s_addr, &lport, td->td_ucred); 1332 if (error) 1333 goto release; 1334 } 1335 1336 /* 1337 * If a UDP socket has been connected, then a local address/port will 1338 * have been selected and bound. 1339 * 1340 * If a UDP socket has not been connected to, then an explicit 1341 * destination address must be used, in which case a local 1342 * address/port may not have been selected and bound. 1343 */ 1344 if (sin != NULL) { 1345 INP_LOCK_ASSERT(inp); 1346 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1347 error = EISCONN; 1348 goto release; 1349 } 1350 1351 /* 1352 * Jail may rewrite the destination address, so let it do 1353 * that before we use it. 1354 */ 1355 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1356 if (error) 1357 goto release; 1358 1359 /* 1360 * If a local address or port hasn't yet been selected, or if 1361 * the destination address needs to be rewritten due to using 1362 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1363 * to do the heavy lifting. Once a port is selected, we 1364 * commit the binding back to the socket; we also commit the 1365 * binding of the address if in jail. 1366 * 1367 * If we already have a valid binding and we're not 1368 * requesting a destination address rewrite, use a fast path. 1369 */ 1370 if (inp->inp_laddr.s_addr == INADDR_ANY || 1371 inp->inp_lport == 0 || 1372 sin->sin_addr.s_addr == INADDR_ANY || 1373 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1374 INP_HASH_LOCK_ASSERT(pcbinfo); 1375 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1376 &lport, &faddr.s_addr, &fport, NULL, 1377 td->td_ucred); 1378 if (error) 1379 goto release; 1380 1381 /* 1382 * XXXRW: Why not commit the port if the address is 1383 * !INADDR_ANY? 1384 */ 1385 /* Commit the local port if newly assigned. */ 1386 if (inp->inp_laddr.s_addr == INADDR_ANY && 1387 inp->inp_lport == 0) { 1388 INP_WLOCK_ASSERT(inp); 1389 INP_HASH_WLOCK_ASSERT(pcbinfo); 1390 /* 1391 * Remember addr if jailed, to prevent 1392 * rebinding. 1393 */ 1394 if (prison_flag(td->td_ucred, PR_IP4)) 1395 inp->inp_laddr = laddr; 1396 inp->inp_lport = lport; 1397 if (in_pcbinshash(inp) != 0) { 1398 inp->inp_lport = 0; 1399 error = EAGAIN; 1400 goto release; 1401 } 1402 inp->inp_flags |= INP_ANONPORT; 1403 } 1404 } else { 1405 faddr = sin->sin_addr; 1406 fport = sin->sin_port; 1407 } 1408 } else { 1409 INP_LOCK_ASSERT(inp); 1410 faddr = inp->inp_faddr; 1411 fport = inp->inp_fport; 1412 if (faddr.s_addr == INADDR_ANY) { 1413 error = ENOTCONN; 1414 goto release; 1415 } 1416 } 1417 1418 /* 1419 * Calculate data length and get a mbuf for UDP, IP, and possible 1420 * link-layer headers. Immediate slide the data pointer back forward 1421 * since we won't use that space at this layer. 1422 */ 1423 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1424 if (m == NULL) { 1425 error = ENOBUFS; 1426 goto release; 1427 } 1428 m->m_data += max_linkhdr; 1429 m->m_len -= max_linkhdr; 1430 m->m_pkthdr.len -= max_linkhdr; 1431 1432 /* 1433 * Fill in mbuf with extended UDP header and addresses and length put 1434 * into network format. 1435 */ 1436 ui = mtod(m, struct udpiphdr *); 1437 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1438 ui->ui_v = IPVERSION << 4; 1439 ui->ui_pr = pr; 1440 ui->ui_src = laddr; 1441 ui->ui_dst = faddr; 1442 ui->ui_sport = lport; 1443 ui->ui_dport = fport; 1444 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1445 if (pr == IPPROTO_UDPLITE) { 1446 struct udpcb *up; 1447 uint16_t plen; 1448 1449 up = intoudpcb(inp); 1450 cscov = up->u_txcslen; 1451 plen = (u_short)len + sizeof(struct udphdr); 1452 if (cscov >= plen) 1453 cscov = 0; 1454 ui->ui_len = htons(plen); 1455 ui->ui_ulen = htons(cscov); 1456 /* 1457 * For UDP-Lite, checksum coverage length of zero means 1458 * the entire UDPLite packet is covered by the checksum. 1459 */ 1460 cscov_partial = (cscov == 0) ? 0 : 1; 1461 } 1462 1463 /* 1464 * Set the Don't Fragment bit in the IP header. 1465 */ 1466 if (inp->inp_flags & INP_DONTFRAG) { 1467 struct ip *ip; 1468 1469 ip = (struct ip *)&ui->ui_i; 1470 ip->ip_off |= htons(IP_DF); 1471 } 1472 1473 ipflags = 0; 1474 if (inp->inp_socket->so_options & SO_DONTROUTE) 1475 ipflags |= IP_ROUTETOIF; 1476 if (inp->inp_socket->so_options & SO_BROADCAST) 1477 ipflags |= IP_ALLOWBROADCAST; 1478 if (inp->inp_flags & INP_ONESBCAST) 1479 ipflags |= IP_SENDONES; 1480 1481 #ifdef MAC 1482 mac_inpcb_create_mbuf(inp, m); 1483 #endif 1484 1485 /* 1486 * Set up checksum and output datagram. 1487 */ 1488 ui->ui_sum = 0; 1489 if (pr == IPPROTO_UDPLITE) { 1490 if (inp->inp_flags & INP_ONESBCAST) 1491 faddr.s_addr = INADDR_BROADCAST; 1492 if (cscov_partial) { 1493 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1494 ui->ui_sum = 0xffff; 1495 } else { 1496 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1497 ui->ui_sum = 0xffff; 1498 } 1499 } else if (V_udp_cksum) { 1500 if (inp->inp_flags & INP_ONESBCAST) 1501 faddr.s_addr = INADDR_BROADCAST; 1502 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1503 htons((u_short)len + sizeof(struct udphdr) + pr)); 1504 m->m_pkthdr.csum_flags = CSUM_UDP; 1505 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1506 } 1507 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1508 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1509 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1510 UDPSTAT_INC(udps_opackets); 1511 1512 /* 1513 * Setup flowid / RSS information for outbound socket. 1514 * 1515 * Once the UDP code decides to set a flowid some other way, 1516 * this allows the flowid to be overridden by userland. 1517 */ 1518 if (flowtype != M_HASHTYPE_NONE) { 1519 m->m_pkthdr.flowid = flowid; 1520 M_HASHTYPE_SET(m, flowtype); 1521 } 1522 #ifdef RSS 1523 else { 1524 uint32_t hash_val, hash_type; 1525 /* 1526 * Calculate an appropriate RSS hash for UDP and 1527 * UDP Lite. 1528 * 1529 * The called function will take care of figuring out 1530 * whether a 2-tuple or 4-tuple hash is required based 1531 * on the currently configured scheme. 1532 * 1533 * Later later on connected socket values should be 1534 * cached in the inpcb and reused, rather than constantly 1535 * re-calculating it. 1536 * 1537 * UDP Lite is a different protocol number and will 1538 * likely end up being hashed as a 2-tuple until 1539 * RSS / NICs grow UDP Lite protocol awareness. 1540 */ 1541 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1542 pr, &hash_val, &hash_type) == 0) { 1543 m->m_pkthdr.flowid = hash_val; 1544 M_HASHTYPE_SET(m, hash_type); 1545 } 1546 } 1547 1548 /* 1549 * Don't override with the inp cached flowid value. 1550 * 1551 * Depending upon the kind of send being done, the inp 1552 * flowid/flowtype values may actually not be appropriate 1553 * for this particular socket send. 1554 * 1555 * We should either leave the flowid at zero (which is what is 1556 * currently done) or set it to some software generated 1557 * hash value based on the packet contents. 1558 */ 1559 ipflags |= IP_NODEFAULTFLOWID; 1560 #endif /* RSS */ 1561 1562 if (unlock_udbinfo == UH_WLOCKED) 1563 INP_HASH_WUNLOCK(pcbinfo); 1564 else if (unlock_udbinfo == UH_RLOCKED) 1565 INP_HASH_RUNLOCK_ET(pcbinfo, et); 1566 if (pr == IPPROTO_UDPLITE) 1567 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1568 else 1569 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1570 error = ip_output(m, inp->inp_options, 1571 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1572 inp->inp_moptions, inp); 1573 if (unlock_inp == UH_WLOCKED) 1574 INP_WUNLOCK(inp); 1575 else 1576 INP_RUNLOCK(inp); 1577 return (error); 1578 1579 release: 1580 if (unlock_udbinfo == UH_WLOCKED) { 1581 KASSERT(unlock_inp == UH_WLOCKED, 1582 ("%s: excl udbinfo lock %#03x, shared inp lock %#03x, " 1583 "sin %p daddr %#010x inp %p laddr %#010x lport %#06x " 1584 "src fam %#04x", 1585 __func__, unlock_udbinfo, unlock_inp, sin, 1586 (sin != NULL) ? sin->sin_addr.s_addr : 0xfefefefe, inp, 1587 inp->inp_laddr.s_addr, inp->inp_lport, src.sin_family)); 1588 INP_HASH_WUNLOCK(pcbinfo); 1589 INP_WUNLOCK(inp); 1590 } else if (unlock_udbinfo == UH_RLOCKED) { 1591 KASSERT(unlock_inp == UH_RLOCKED, 1592 ("%s: shared udbinfo lock %#03x, excl inp lock %#03x, " 1593 "sin %p daddr %#010x inp %p laddr %#010x lport %#06x " 1594 "src fam %#04x", 1595 __func__, unlock_udbinfo, unlock_inp, sin, 1596 (sin != NULL) ? sin->sin_addr.s_addr : 0xfefefefe, inp, 1597 inp->inp_laddr.s_addr, inp->inp_lport, src.sin_family)); 1598 INP_HASH_RUNLOCK_ET(pcbinfo, et); 1599 INP_RUNLOCK(inp); 1600 } else if (unlock_inp == UH_WLOCKED) 1601 INP_WUNLOCK(inp); 1602 else 1603 INP_RUNLOCK(inp); 1604 m_freem(m); 1605 return (error); 1606 } 1607 1608 static void 1609 udp_abort(struct socket *so) 1610 { 1611 struct inpcb *inp; 1612 struct inpcbinfo *pcbinfo; 1613 1614 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1615 inp = sotoinpcb(so); 1616 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1617 INP_WLOCK(inp); 1618 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1619 INP_HASH_WLOCK(pcbinfo); 1620 in_pcbdisconnect(inp); 1621 inp->inp_laddr.s_addr = INADDR_ANY; 1622 INP_HASH_WUNLOCK(pcbinfo); 1623 soisdisconnected(so); 1624 } 1625 INP_WUNLOCK(inp); 1626 } 1627 1628 static int 1629 udp_attach(struct socket *so, int proto, struct thread *td) 1630 { 1631 static uint32_t udp_flowid; 1632 struct inpcb *inp; 1633 struct inpcbinfo *pcbinfo; 1634 int error; 1635 1636 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1637 inp = sotoinpcb(so); 1638 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1639 error = soreserve(so, udp_sendspace, udp_recvspace); 1640 if (error) 1641 return (error); 1642 INP_INFO_WLOCK(pcbinfo); 1643 error = in_pcballoc(so, pcbinfo); 1644 if (error) { 1645 INP_INFO_WUNLOCK(pcbinfo); 1646 return (error); 1647 } 1648 1649 inp = sotoinpcb(so); 1650 inp->inp_vflag |= INP_IPV4; 1651 inp->inp_ip_ttl = V_ip_defttl; 1652 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1653 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1654 1655 error = udp_newudpcb(inp); 1656 if (error) { 1657 in_pcbdetach(inp); 1658 in_pcbfree(inp); 1659 INP_INFO_WUNLOCK(pcbinfo); 1660 return (error); 1661 } 1662 1663 INP_WUNLOCK(inp); 1664 INP_INFO_WUNLOCK(pcbinfo); 1665 return (0); 1666 } 1667 #endif /* INET */ 1668 1669 int 1670 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1671 { 1672 struct inpcb *inp; 1673 struct udpcb *up; 1674 1675 KASSERT(so->so_type == SOCK_DGRAM, 1676 ("udp_set_kernel_tunneling: !dgram")); 1677 inp = sotoinpcb(so); 1678 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1679 INP_WLOCK(inp); 1680 up = intoudpcb(inp); 1681 if ((up->u_tun_func != NULL) || 1682 (up->u_icmp_func != NULL)) { 1683 INP_WUNLOCK(inp); 1684 return (EBUSY); 1685 } 1686 up->u_tun_func = f; 1687 up->u_icmp_func = i; 1688 up->u_tun_ctx = ctx; 1689 INP_WUNLOCK(inp); 1690 return (0); 1691 } 1692 1693 #ifdef INET 1694 static int 1695 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1696 { 1697 struct inpcb *inp; 1698 struct inpcbinfo *pcbinfo; 1699 int error; 1700 1701 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1702 inp = sotoinpcb(so); 1703 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1704 INP_WLOCK(inp); 1705 INP_HASH_WLOCK(pcbinfo); 1706 error = in_pcbbind(inp, nam, td->td_ucred); 1707 INP_HASH_WUNLOCK(pcbinfo); 1708 INP_WUNLOCK(inp); 1709 return (error); 1710 } 1711 1712 static void 1713 udp_close(struct socket *so) 1714 { 1715 struct inpcb *inp; 1716 struct inpcbinfo *pcbinfo; 1717 1718 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1719 inp = sotoinpcb(so); 1720 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1721 INP_WLOCK(inp); 1722 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1723 INP_HASH_WLOCK(pcbinfo); 1724 in_pcbdisconnect(inp); 1725 inp->inp_laddr.s_addr = INADDR_ANY; 1726 INP_HASH_WUNLOCK(pcbinfo); 1727 soisdisconnected(so); 1728 } 1729 INP_WUNLOCK(inp); 1730 } 1731 1732 static int 1733 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1734 { 1735 struct inpcb *inp; 1736 struct inpcbinfo *pcbinfo; 1737 struct sockaddr_in *sin; 1738 int error; 1739 1740 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1741 inp = sotoinpcb(so); 1742 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1743 INP_WLOCK(inp); 1744 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1745 INP_WUNLOCK(inp); 1746 return (EISCONN); 1747 } 1748 sin = (struct sockaddr_in *)nam; 1749 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1750 if (error != 0) { 1751 INP_WUNLOCK(inp); 1752 return (error); 1753 } 1754 INP_HASH_WLOCK(pcbinfo); 1755 error = in_pcbconnect(inp, nam, td->td_ucred); 1756 INP_HASH_WUNLOCK(pcbinfo); 1757 if (error == 0) 1758 soisconnected(so); 1759 INP_WUNLOCK(inp); 1760 return (error); 1761 } 1762 1763 static void 1764 udp_detach(struct socket *so) 1765 { 1766 struct inpcb *inp; 1767 struct inpcbinfo *pcbinfo; 1768 struct udpcb *up; 1769 1770 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1771 inp = sotoinpcb(so); 1772 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1773 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1774 ("udp_detach: not disconnected")); 1775 INP_INFO_WLOCK(pcbinfo); 1776 INP_WLOCK(inp); 1777 up = intoudpcb(inp); 1778 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1779 inp->inp_ppcb = NULL; 1780 in_pcbdetach(inp); 1781 in_pcbfree(inp); 1782 INP_INFO_WUNLOCK(pcbinfo); 1783 udp_discardcb(up); 1784 } 1785 1786 static int 1787 udp_disconnect(struct socket *so) 1788 { 1789 struct inpcb *inp; 1790 struct inpcbinfo *pcbinfo; 1791 1792 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1793 inp = sotoinpcb(so); 1794 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1795 INP_WLOCK(inp); 1796 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1797 INP_WUNLOCK(inp); 1798 return (ENOTCONN); 1799 } 1800 INP_HASH_WLOCK(pcbinfo); 1801 in_pcbdisconnect(inp); 1802 inp->inp_laddr.s_addr = INADDR_ANY; 1803 INP_HASH_WUNLOCK(pcbinfo); 1804 SOCK_LOCK(so); 1805 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1806 SOCK_UNLOCK(so); 1807 INP_WUNLOCK(inp); 1808 return (0); 1809 } 1810 1811 static int 1812 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1813 struct mbuf *control, struct thread *td) 1814 { 1815 struct inpcb *inp; 1816 1817 inp = sotoinpcb(so); 1818 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1819 return (udp_output(inp, m, addr, control, td)); 1820 } 1821 #endif /* INET */ 1822 1823 int 1824 udp_shutdown(struct socket *so) 1825 { 1826 struct inpcb *inp; 1827 1828 inp = sotoinpcb(so); 1829 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1830 INP_WLOCK(inp); 1831 socantsendmore(so); 1832 INP_WUNLOCK(inp); 1833 return (0); 1834 } 1835 1836 #ifdef INET 1837 struct pr_usrreqs udp_usrreqs = { 1838 .pru_abort = udp_abort, 1839 .pru_attach = udp_attach, 1840 .pru_bind = udp_bind, 1841 .pru_connect = udp_connect, 1842 .pru_control = in_control, 1843 .pru_detach = udp_detach, 1844 .pru_disconnect = udp_disconnect, 1845 .pru_peeraddr = in_getpeeraddr, 1846 .pru_send = udp_send, 1847 .pru_soreceive = soreceive_dgram, 1848 .pru_sosend = sosend_dgram, 1849 .pru_shutdown = udp_shutdown, 1850 .pru_sockaddr = in_getsockaddr, 1851 .pru_sosetlabel = in_pcbsosetlabel, 1852 .pru_close = udp_close, 1853 }; 1854 #endif /* INET */ 1855