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 if (proto == IPPROTO_UDPLITE) 598 UDPLITE_PROBE(receive, NULL, last, ip, 599 last, uh); 600 else 601 UDP_PROBE(receive, NULL, last, ip, last, 602 uh); 603 if (udp_append(last, ip, n, iphlen, 604 udp_in)) { 605 goto inp_lost; 606 } 607 } 608 INP_RUNLOCK(last); 609 } 610 last = inp; 611 /* 612 * Don't look for additional matches if this one does 613 * not have either the SO_REUSEPORT or SO_REUSEADDR 614 * socket options set. This heuristic avoids 615 * searching through all pcbs in the common case of a 616 * non-shared port. It assumes that an application 617 * will never clear these options after setting them. 618 */ 619 if ((last->inp_socket->so_options & 620 (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) 621 break; 622 } 623 624 if (last == NULL) { 625 /* 626 * No matching pcb found; discard datagram. (No need 627 * to send an ICMP Port Unreachable for a broadcast 628 * or multicast datgram.) 629 */ 630 UDPSTAT_INC(udps_noportbcast); 631 if (inp) 632 INP_RUNLOCK(inp); 633 INP_INFO_RUNLOCK_ET(pcbinfo, et); 634 goto badunlocked; 635 } 636 if (proto == IPPROTO_UDPLITE) 637 UDPLITE_PROBE(receive, NULL, last, ip, last, uh); 638 else 639 UDP_PROBE(receive, NULL, last, ip, last, uh); 640 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 641 INP_RUNLOCK(last); 642 inp_lost: 643 INP_INFO_RUNLOCK_ET(pcbinfo, et); 644 return (IPPROTO_DONE); 645 } 646 647 /* 648 * Locate pcb for datagram. 649 */ 650 651 /* 652 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 653 */ 654 if ((m->m_flags & M_IP_NEXTHOP) && 655 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 656 struct sockaddr_in *next_hop; 657 658 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 659 660 /* 661 * Transparently forwarded. Pretend to be the destination. 662 * Already got one like this? 663 */ 664 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 665 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 666 if (!inp) { 667 /* 668 * It's new. Try to find the ambushing socket. 669 * Because we've rewritten the destination address, 670 * any hardware-generated hash is ignored. 671 */ 672 inp = in_pcblookup(pcbinfo, ip->ip_src, 673 uh->uh_sport, next_hop->sin_addr, 674 next_hop->sin_port ? htons(next_hop->sin_port) : 675 uh->uh_dport, INPLOOKUP_WILDCARD | 676 INPLOOKUP_RLOCKPCB, ifp); 677 } 678 /* Remove the tag from the packet. We don't need it anymore. */ 679 m_tag_delete(m, fwd_tag); 680 m->m_flags &= ~M_IP_NEXTHOP; 681 } else 682 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 683 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 684 INPLOOKUP_RLOCKPCB, ifp, m); 685 if (inp == NULL) { 686 if (udp_log_in_vain) { 687 char src[INET_ADDRSTRLEN]; 688 char dst[INET_ADDRSTRLEN]; 689 690 log(LOG_INFO, 691 "Connection attempt to UDP %s:%d from %s:%d\n", 692 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 693 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 694 } 695 if (proto == IPPROTO_UDPLITE) 696 UDPLITE_PROBE(receive, NULL, NULL, ip, NULL, uh); 697 else 698 UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); 699 UDPSTAT_INC(udps_noport); 700 if (m->m_flags & (M_BCAST | M_MCAST)) { 701 UDPSTAT_INC(udps_noportbcast); 702 goto badunlocked; 703 } 704 if (V_udp_blackhole) 705 goto badunlocked; 706 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 707 goto badunlocked; 708 *ip = save_ip; 709 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 710 return (IPPROTO_DONE); 711 } 712 713 /* 714 * Check the minimum TTL for socket. 715 */ 716 INP_RLOCK_ASSERT(inp); 717 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 718 if (proto == IPPROTO_UDPLITE) 719 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 720 else 721 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 722 INP_RUNLOCK(inp); 723 m_freem(m); 724 return (IPPROTO_DONE); 725 } 726 if (cscov_partial) { 727 struct udpcb *up; 728 729 up = intoudpcb(inp); 730 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 731 INP_RUNLOCK(inp); 732 m_freem(m); 733 return (IPPROTO_DONE); 734 } 735 } 736 737 if (proto == IPPROTO_UDPLITE) 738 UDPLITE_PROBE(receive, NULL, inp, ip, inp, uh); 739 else 740 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 741 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 742 INP_RUNLOCK(inp); 743 return (IPPROTO_DONE); 744 745 badunlocked: 746 m_freem(m); 747 return (IPPROTO_DONE); 748 } 749 #endif /* INET */ 750 751 /* 752 * Notify a udp user of an asynchronous error; just wake up so that they can 753 * collect error status. 754 */ 755 struct inpcb * 756 udp_notify(struct inpcb *inp, int errno) 757 { 758 759 /* 760 * While udp_ctlinput() always calls udp_notify() with a read lock 761 * when invoking it directly, in_pcbnotifyall() currently uses write 762 * locks due to sharing code with TCP. For now, accept either a read 763 * or a write lock, but a read lock is sufficient. 764 */ 765 INP_LOCK_ASSERT(inp); 766 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 767 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 768 RTFREE(inp->inp_route.ro_rt); 769 inp->inp_route.ro_rt = (struct rtentry *)NULL; 770 } 771 772 inp->inp_socket->so_error = errno; 773 sorwakeup(inp->inp_socket); 774 sowwakeup(inp->inp_socket); 775 return (inp); 776 } 777 778 #ifdef INET 779 static void 780 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 781 struct inpcbinfo *pcbinfo) 782 { 783 struct ip *ip = vip; 784 struct udphdr *uh; 785 struct in_addr faddr; 786 struct inpcb *inp; 787 788 faddr = ((struct sockaddr_in *)sa)->sin_addr; 789 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 790 return; 791 792 if (PRC_IS_REDIRECT(cmd)) { 793 /* signal EHOSTDOWN, as it flushes the cached route */ 794 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 795 return; 796 } 797 798 /* 799 * Hostdead is ugly because it goes linearly through all PCBs. 800 * 801 * XXX: We never get this from ICMP, otherwise it makes an excellent 802 * DoS attack on machines with many connections. 803 */ 804 if (cmd == PRC_HOSTDEAD) 805 ip = NULL; 806 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 807 return; 808 if (ip != NULL) { 809 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 810 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 811 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 812 if (inp != NULL) { 813 INP_RLOCK_ASSERT(inp); 814 if (inp->inp_socket != NULL) { 815 udp_notify(inp, inetctlerrmap[cmd]); 816 } 817 INP_RUNLOCK(inp); 818 } else { 819 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 820 ip->ip_src, uh->uh_sport, 821 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 822 if (inp != NULL) { 823 struct udpcb *up; 824 void *ctx; 825 udp_tun_icmp_t func; 826 827 up = intoudpcb(inp); 828 ctx = up->u_tun_ctx; 829 func = up->u_icmp_func; 830 INP_RUNLOCK(inp); 831 if (func != NULL) 832 (*func)(cmd, sa, vip, ctx); 833 } 834 } 835 } else 836 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 837 udp_notify); 838 } 839 void 840 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 841 { 842 843 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 844 } 845 846 void 847 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 848 { 849 850 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 851 } 852 #endif /* INET */ 853 854 static int 855 udp_pcblist(SYSCTL_HANDLER_ARGS) 856 { 857 int error, i, n; 858 struct inpcb *inp, **inp_list; 859 inp_gen_t gencnt; 860 struct xinpgen xig; 861 struct epoch_tracker et; 862 863 /* 864 * The process of preparing the PCB list is too time-consuming and 865 * resource-intensive to repeat twice on every request. 866 */ 867 if (req->oldptr == 0) { 868 n = V_udbinfo.ipi_count; 869 n += imax(n / 8, 10); 870 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 871 return (0); 872 } 873 874 if (req->newptr != 0) 875 return (EPERM); 876 877 /* 878 * OK, now we're committed to doing something. 879 */ 880 INP_INFO_RLOCK_ET(&V_udbinfo, et); 881 gencnt = V_udbinfo.ipi_gencnt; 882 n = V_udbinfo.ipi_count; 883 INP_INFO_RUNLOCK_ET(&V_udbinfo, et); 884 885 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 886 + n * sizeof(struct xinpcb)); 887 if (error != 0) 888 return (error); 889 890 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 if (sin == NULL || 1159 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1160 INP_WLOCK(inp); 1161 unlock_inp = UH_WLOCKED; 1162 } else { 1163 INP_RLOCK(inp); 1164 unlock_inp = UH_RLOCKED; 1165 } 1166 tos = inp->inp_ip_tos; 1167 if (control != NULL) { 1168 /* 1169 * XXX: Currently, we assume all the optional information is 1170 * stored in a single mbuf. 1171 */ 1172 if (control->m_next) { 1173 if (unlock_inp == UH_WLOCKED) 1174 INP_WUNLOCK(inp); 1175 else 1176 INP_RUNLOCK(inp); 1177 m_freem(control); 1178 m_freem(m); 1179 return (EINVAL); 1180 } 1181 for (; control->m_len > 0; 1182 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1183 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1184 cm = mtod(control, struct cmsghdr *); 1185 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1186 || cm->cmsg_len > control->m_len) { 1187 error = EINVAL; 1188 break; 1189 } 1190 if (cm->cmsg_level != IPPROTO_IP) 1191 continue; 1192 1193 switch (cm->cmsg_type) { 1194 case IP_SENDSRCADDR: 1195 if (cm->cmsg_len != 1196 CMSG_LEN(sizeof(struct in_addr))) { 1197 error = EINVAL; 1198 break; 1199 } 1200 bzero(&src, sizeof(src)); 1201 src.sin_family = AF_INET; 1202 src.sin_len = sizeof(src); 1203 src.sin_port = inp->inp_lport; 1204 src.sin_addr = 1205 *(struct in_addr *)CMSG_DATA(cm); 1206 break; 1207 1208 case IP_TOS: 1209 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1210 error = EINVAL; 1211 break; 1212 } 1213 tos = *(u_char *)CMSG_DATA(cm); 1214 break; 1215 1216 case IP_FLOWID: 1217 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1218 error = EINVAL; 1219 break; 1220 } 1221 flowid = *(uint32_t *) CMSG_DATA(cm); 1222 break; 1223 1224 case IP_FLOWTYPE: 1225 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1226 error = EINVAL; 1227 break; 1228 } 1229 flowtype = *(uint32_t *) CMSG_DATA(cm); 1230 break; 1231 1232 #ifdef RSS 1233 case IP_RSSBUCKETID: 1234 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1235 error = EINVAL; 1236 break; 1237 } 1238 /* This is just a placeholder for now */ 1239 break; 1240 #endif /* RSS */ 1241 default: 1242 error = ENOPROTOOPT; 1243 break; 1244 } 1245 if (error) 1246 break; 1247 } 1248 m_freem(control); 1249 } 1250 if (error) { 1251 if (unlock_inp == UH_WLOCKED) 1252 INP_WUNLOCK(inp); 1253 else 1254 INP_RUNLOCK(inp); 1255 m_freem(m); 1256 return (error); 1257 } 1258 1259 /* 1260 * Depending on whether or not the application has bound or connected 1261 * the socket, we may have to do varying levels of work. The optimal 1262 * case is for a connected UDP socket, as a global lock isn't 1263 * required at all. 1264 * 1265 * In order to decide which we need, we require stability of the 1266 * inpcb binding, which we ensure by acquiring a read lock on the 1267 * inpcb. This doesn't strictly follow the lock order, so we play 1268 * the trylock and retry game; note that we may end up with more 1269 * conservative locks than required the second time around, so later 1270 * assertions have to accept that. Further analysis of the number of 1271 * misses under contention is required. 1272 * 1273 * XXXRW: Check that hash locking update here is correct. 1274 */ 1275 pr = inp->inp_socket->so_proto->pr_protocol; 1276 pcbinfo = udp_get_inpcbinfo(pr); 1277 sin = (struct sockaddr_in *)addr; 1278 if (sin != NULL && 1279 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1280 INP_HASH_WLOCK(pcbinfo); 1281 unlock_udbinfo = UH_WLOCKED; 1282 } else if ((sin != NULL && ( 1283 (sin->sin_addr.s_addr == INADDR_ANY) || 1284 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1285 (inp->inp_laddr.s_addr == INADDR_ANY) || 1286 (inp->inp_lport == 0))) || 1287 (src.sin_family == AF_INET)) { 1288 INP_HASH_RLOCK_ET(pcbinfo, et); 1289 unlock_udbinfo = UH_RLOCKED; 1290 } else 1291 unlock_udbinfo = UH_UNLOCKED; 1292 1293 /* 1294 * If the IP_SENDSRCADDR control message was specified, override the 1295 * source address for this datagram. Its use is invalidated if the 1296 * address thus specified is incomplete or clobbers other inpcbs. 1297 */ 1298 laddr = inp->inp_laddr; 1299 lport = inp->inp_lport; 1300 if (src.sin_family == AF_INET) { 1301 INP_HASH_LOCK_ASSERT(pcbinfo); 1302 if ((lport == 0) || 1303 (laddr.s_addr == INADDR_ANY && 1304 src.sin_addr.s_addr == INADDR_ANY)) { 1305 error = EINVAL; 1306 goto release; 1307 } 1308 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1309 &laddr.s_addr, &lport, td->td_ucred); 1310 if (error) 1311 goto release; 1312 } 1313 1314 /* 1315 * If a UDP socket has been connected, then a local address/port will 1316 * have been selected and bound. 1317 * 1318 * If a UDP socket has not been connected to, then an explicit 1319 * destination address must be used, in which case a local 1320 * address/port may not have been selected and bound. 1321 */ 1322 if (sin != NULL) { 1323 INP_LOCK_ASSERT(inp); 1324 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1325 error = EISCONN; 1326 goto release; 1327 } 1328 1329 /* 1330 * Jail may rewrite the destination address, so let it do 1331 * that before we use it. 1332 */ 1333 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1334 if (error) 1335 goto release; 1336 1337 /* 1338 * If a local address or port hasn't yet been selected, or if 1339 * the destination address needs to be rewritten due to using 1340 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1341 * to do the heavy lifting. Once a port is selected, we 1342 * commit the binding back to the socket; we also commit the 1343 * binding of the address if in jail. 1344 * 1345 * If we already have a valid binding and we're not 1346 * requesting a destination address rewrite, use a fast path. 1347 */ 1348 if (inp->inp_laddr.s_addr == INADDR_ANY || 1349 inp->inp_lport == 0 || 1350 sin->sin_addr.s_addr == INADDR_ANY || 1351 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1352 INP_HASH_LOCK_ASSERT(pcbinfo); 1353 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1354 &lport, &faddr.s_addr, &fport, NULL, 1355 td->td_ucred); 1356 if (error) 1357 goto release; 1358 1359 /* 1360 * XXXRW: Why not commit the port if the address is 1361 * !INADDR_ANY? 1362 */ 1363 /* Commit the local port if newly assigned. */ 1364 if (inp->inp_laddr.s_addr == INADDR_ANY && 1365 inp->inp_lport == 0) { 1366 INP_WLOCK_ASSERT(inp); 1367 INP_HASH_WLOCK_ASSERT(pcbinfo); 1368 /* 1369 * Remember addr if jailed, to prevent 1370 * rebinding. 1371 */ 1372 if (prison_flag(td->td_ucred, PR_IP4)) 1373 inp->inp_laddr = laddr; 1374 inp->inp_lport = lport; 1375 if (in_pcbinshash(inp) != 0) { 1376 inp->inp_lport = 0; 1377 error = EAGAIN; 1378 goto release; 1379 } 1380 inp->inp_flags |= INP_ANONPORT; 1381 } 1382 } else { 1383 faddr = sin->sin_addr; 1384 fport = sin->sin_port; 1385 } 1386 } else { 1387 INP_LOCK_ASSERT(inp); 1388 faddr = inp->inp_faddr; 1389 fport = inp->inp_fport; 1390 if (faddr.s_addr == INADDR_ANY) { 1391 error = ENOTCONN; 1392 goto release; 1393 } 1394 } 1395 1396 /* 1397 * Calculate data length and get a mbuf for UDP, IP, and possible 1398 * link-layer headers. Immediate slide the data pointer back forward 1399 * since we won't use that space at this layer. 1400 */ 1401 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1402 if (m == NULL) { 1403 error = ENOBUFS; 1404 goto release; 1405 } 1406 m->m_data += max_linkhdr; 1407 m->m_len -= max_linkhdr; 1408 m->m_pkthdr.len -= max_linkhdr; 1409 1410 /* 1411 * Fill in mbuf with extended UDP header and addresses and length put 1412 * into network format. 1413 */ 1414 ui = mtod(m, struct udpiphdr *); 1415 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1416 ui->ui_v = IPVERSION << 4; 1417 ui->ui_pr = pr; 1418 ui->ui_src = laddr; 1419 ui->ui_dst = faddr; 1420 ui->ui_sport = lport; 1421 ui->ui_dport = fport; 1422 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1423 if (pr == IPPROTO_UDPLITE) { 1424 struct udpcb *up; 1425 uint16_t plen; 1426 1427 up = intoudpcb(inp); 1428 cscov = up->u_txcslen; 1429 plen = (u_short)len + sizeof(struct udphdr); 1430 if (cscov >= plen) 1431 cscov = 0; 1432 ui->ui_len = htons(plen); 1433 ui->ui_ulen = htons(cscov); 1434 /* 1435 * For UDP-Lite, checksum coverage length of zero means 1436 * the entire UDPLite packet is covered by the checksum. 1437 */ 1438 cscov_partial = (cscov == 0) ? 0 : 1; 1439 } 1440 1441 /* 1442 * Set the Don't Fragment bit in the IP header. 1443 */ 1444 if (inp->inp_flags & INP_DONTFRAG) { 1445 struct ip *ip; 1446 1447 ip = (struct ip *)&ui->ui_i; 1448 ip->ip_off |= htons(IP_DF); 1449 } 1450 1451 ipflags = 0; 1452 if (inp->inp_socket->so_options & SO_DONTROUTE) 1453 ipflags |= IP_ROUTETOIF; 1454 if (inp->inp_socket->so_options & SO_BROADCAST) 1455 ipflags |= IP_ALLOWBROADCAST; 1456 if (inp->inp_flags & INP_ONESBCAST) 1457 ipflags |= IP_SENDONES; 1458 1459 #ifdef MAC 1460 mac_inpcb_create_mbuf(inp, m); 1461 #endif 1462 1463 /* 1464 * Set up checksum and output datagram. 1465 */ 1466 ui->ui_sum = 0; 1467 if (pr == IPPROTO_UDPLITE) { 1468 if (inp->inp_flags & INP_ONESBCAST) 1469 faddr.s_addr = INADDR_BROADCAST; 1470 if (cscov_partial) { 1471 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1472 ui->ui_sum = 0xffff; 1473 } else { 1474 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1475 ui->ui_sum = 0xffff; 1476 } 1477 } else if (V_udp_cksum) { 1478 if (inp->inp_flags & INP_ONESBCAST) 1479 faddr.s_addr = INADDR_BROADCAST; 1480 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1481 htons((u_short)len + sizeof(struct udphdr) + pr)); 1482 m->m_pkthdr.csum_flags = CSUM_UDP; 1483 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1484 } 1485 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1486 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1487 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1488 UDPSTAT_INC(udps_opackets); 1489 1490 /* 1491 * Setup flowid / RSS information for outbound socket. 1492 * 1493 * Once the UDP code decides to set a flowid some other way, 1494 * this allows the flowid to be overridden by userland. 1495 */ 1496 if (flowtype != M_HASHTYPE_NONE) { 1497 m->m_pkthdr.flowid = flowid; 1498 M_HASHTYPE_SET(m, flowtype); 1499 #ifdef RSS 1500 } else { 1501 uint32_t hash_val, hash_type; 1502 /* 1503 * Calculate an appropriate RSS hash for UDP and 1504 * UDP Lite. 1505 * 1506 * The called function will take care of figuring out 1507 * whether a 2-tuple or 4-tuple hash is required based 1508 * on the currently configured scheme. 1509 * 1510 * Later later on connected socket values should be 1511 * cached in the inpcb and reused, rather than constantly 1512 * re-calculating it. 1513 * 1514 * UDP Lite is a different protocol number and will 1515 * likely end up being hashed as a 2-tuple until 1516 * RSS / NICs grow UDP Lite protocol awareness. 1517 */ 1518 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1519 pr, &hash_val, &hash_type) == 0) { 1520 m->m_pkthdr.flowid = hash_val; 1521 M_HASHTYPE_SET(m, hash_type); 1522 } 1523 #endif 1524 } 1525 1526 #ifdef RSS 1527 /* 1528 * Don't override with the inp cached flowid value. 1529 * 1530 * Depending upon the kind of send being done, the inp 1531 * flowid/flowtype values may actually not be appropriate 1532 * for this particular socket send. 1533 * 1534 * We should either leave the flowid at zero (which is what is 1535 * currently done) or set it to some software generated 1536 * hash value based on the packet contents. 1537 */ 1538 ipflags |= IP_NODEFAULTFLOWID; 1539 #endif /* RSS */ 1540 1541 if (unlock_udbinfo == UH_WLOCKED) 1542 INP_HASH_WUNLOCK(pcbinfo); 1543 else if (unlock_udbinfo == UH_RLOCKED) 1544 INP_HASH_RUNLOCK_ET(pcbinfo, et); 1545 if (pr == IPPROTO_UDPLITE) 1546 UDPLITE_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1547 else 1548 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1549 error = ip_output(m, inp->inp_options, 1550 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1551 inp->inp_moptions, inp); 1552 if (unlock_inp == UH_WLOCKED) 1553 INP_WUNLOCK(inp); 1554 else 1555 INP_RUNLOCK(inp); 1556 return (error); 1557 1558 release: 1559 if (unlock_udbinfo == UH_WLOCKED) { 1560 KASSERT(unlock_inp == UH_WLOCKED, 1561 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1562 INP_HASH_WUNLOCK(pcbinfo); 1563 INP_WUNLOCK(inp); 1564 } else if (unlock_udbinfo == UH_RLOCKED) { 1565 KASSERT(unlock_inp == UH_RLOCKED, 1566 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1567 INP_HASH_RUNLOCK_ET(pcbinfo, et); 1568 INP_RUNLOCK(inp); 1569 } else if (unlock_inp == UH_WLOCKED) 1570 INP_WUNLOCK(inp); 1571 else 1572 INP_RUNLOCK(inp); 1573 m_freem(m); 1574 return (error); 1575 } 1576 1577 static void 1578 udp_abort(struct socket *so) 1579 { 1580 struct inpcb *inp; 1581 struct inpcbinfo *pcbinfo; 1582 1583 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1584 inp = sotoinpcb(so); 1585 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1586 INP_WLOCK(inp); 1587 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1588 INP_HASH_WLOCK(pcbinfo); 1589 in_pcbdisconnect(inp); 1590 inp->inp_laddr.s_addr = INADDR_ANY; 1591 INP_HASH_WUNLOCK(pcbinfo); 1592 soisdisconnected(so); 1593 } 1594 INP_WUNLOCK(inp); 1595 } 1596 1597 static int 1598 udp_attach(struct socket *so, int proto, struct thread *td) 1599 { 1600 static uint32_t udp_flowid; 1601 struct inpcb *inp; 1602 struct inpcbinfo *pcbinfo; 1603 int error; 1604 1605 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1606 inp = sotoinpcb(so); 1607 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1608 error = soreserve(so, udp_sendspace, udp_recvspace); 1609 if (error) 1610 return (error); 1611 INP_INFO_WLOCK(pcbinfo); 1612 error = in_pcballoc(so, pcbinfo); 1613 if (error) { 1614 INP_INFO_WUNLOCK(pcbinfo); 1615 return (error); 1616 } 1617 1618 inp = sotoinpcb(so); 1619 inp->inp_vflag |= INP_IPV4; 1620 inp->inp_ip_ttl = V_ip_defttl; 1621 inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); 1622 inp->inp_flowtype = M_HASHTYPE_OPAQUE; 1623 1624 error = udp_newudpcb(inp); 1625 if (error) { 1626 in_pcbdetach(inp); 1627 in_pcbfree(inp); 1628 INP_INFO_WUNLOCK(pcbinfo); 1629 return (error); 1630 } 1631 1632 INP_WUNLOCK(inp); 1633 INP_INFO_WUNLOCK(pcbinfo); 1634 return (0); 1635 } 1636 #endif /* INET */ 1637 1638 int 1639 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1640 { 1641 struct inpcb *inp; 1642 struct udpcb *up; 1643 1644 KASSERT(so->so_type == SOCK_DGRAM, 1645 ("udp_set_kernel_tunneling: !dgram")); 1646 inp = sotoinpcb(so); 1647 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1648 INP_WLOCK(inp); 1649 up = intoudpcb(inp); 1650 if ((up->u_tun_func != NULL) || 1651 (up->u_icmp_func != NULL)) { 1652 INP_WUNLOCK(inp); 1653 return (EBUSY); 1654 } 1655 up->u_tun_func = f; 1656 up->u_icmp_func = i; 1657 up->u_tun_ctx = ctx; 1658 INP_WUNLOCK(inp); 1659 return (0); 1660 } 1661 1662 #ifdef INET 1663 static int 1664 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1665 { 1666 struct inpcb *inp; 1667 struct inpcbinfo *pcbinfo; 1668 int error; 1669 1670 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1671 inp = sotoinpcb(so); 1672 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1673 INP_WLOCK(inp); 1674 INP_HASH_WLOCK(pcbinfo); 1675 error = in_pcbbind(inp, nam, td->td_ucred); 1676 INP_HASH_WUNLOCK(pcbinfo); 1677 INP_WUNLOCK(inp); 1678 return (error); 1679 } 1680 1681 static void 1682 udp_close(struct socket *so) 1683 { 1684 struct inpcb *inp; 1685 struct inpcbinfo *pcbinfo; 1686 1687 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1688 inp = sotoinpcb(so); 1689 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1690 INP_WLOCK(inp); 1691 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1692 INP_HASH_WLOCK(pcbinfo); 1693 in_pcbdisconnect(inp); 1694 inp->inp_laddr.s_addr = INADDR_ANY; 1695 INP_HASH_WUNLOCK(pcbinfo); 1696 soisdisconnected(so); 1697 } 1698 INP_WUNLOCK(inp); 1699 } 1700 1701 static int 1702 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1703 { 1704 struct inpcb *inp; 1705 struct inpcbinfo *pcbinfo; 1706 struct sockaddr_in *sin; 1707 int error; 1708 1709 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1710 inp = sotoinpcb(so); 1711 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1712 INP_WLOCK(inp); 1713 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1714 INP_WUNLOCK(inp); 1715 return (EISCONN); 1716 } 1717 sin = (struct sockaddr_in *)nam; 1718 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1719 if (error != 0) { 1720 INP_WUNLOCK(inp); 1721 return (error); 1722 } 1723 INP_HASH_WLOCK(pcbinfo); 1724 error = in_pcbconnect(inp, nam, td->td_ucred); 1725 INP_HASH_WUNLOCK(pcbinfo); 1726 if (error == 0) 1727 soisconnected(so); 1728 INP_WUNLOCK(inp); 1729 return (error); 1730 } 1731 1732 static void 1733 udp_detach(struct socket *so) 1734 { 1735 struct inpcb *inp; 1736 struct inpcbinfo *pcbinfo; 1737 struct udpcb *up; 1738 1739 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1740 inp = sotoinpcb(so); 1741 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1742 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1743 ("udp_detach: not disconnected")); 1744 INP_INFO_WLOCK(pcbinfo); 1745 INP_WLOCK(inp); 1746 up = intoudpcb(inp); 1747 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1748 inp->inp_ppcb = NULL; 1749 in_pcbdetach(inp); 1750 in_pcbfree(inp); 1751 INP_INFO_WUNLOCK(pcbinfo); 1752 udp_discardcb(up); 1753 } 1754 1755 static int 1756 udp_disconnect(struct socket *so) 1757 { 1758 struct inpcb *inp; 1759 struct inpcbinfo *pcbinfo; 1760 1761 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1762 inp = sotoinpcb(so); 1763 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1764 INP_WLOCK(inp); 1765 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1766 INP_WUNLOCK(inp); 1767 return (ENOTCONN); 1768 } 1769 INP_HASH_WLOCK(pcbinfo); 1770 in_pcbdisconnect(inp); 1771 inp->inp_laddr.s_addr = INADDR_ANY; 1772 INP_HASH_WUNLOCK(pcbinfo); 1773 SOCK_LOCK(so); 1774 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1775 SOCK_UNLOCK(so); 1776 INP_WUNLOCK(inp); 1777 return (0); 1778 } 1779 1780 static int 1781 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1782 struct mbuf *control, struct thread *td) 1783 { 1784 struct inpcb *inp; 1785 1786 inp = sotoinpcb(so); 1787 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1788 return (udp_output(inp, m, addr, control, td)); 1789 } 1790 #endif /* INET */ 1791 1792 int 1793 udp_shutdown(struct socket *so) 1794 { 1795 struct inpcb *inp; 1796 1797 inp = sotoinpcb(so); 1798 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1799 INP_WLOCK(inp); 1800 socantsendmore(so); 1801 INP_WUNLOCK(inp); 1802 return (0); 1803 } 1804 1805 #ifdef INET 1806 struct pr_usrreqs udp_usrreqs = { 1807 .pru_abort = udp_abort, 1808 .pru_attach = udp_attach, 1809 .pru_bind = udp_bind, 1810 .pru_connect = udp_connect, 1811 .pru_control = in_control, 1812 .pru_detach = udp_detach, 1813 .pru_disconnect = udp_disconnect, 1814 .pru_peeraddr = in_getpeeraddr, 1815 .pru_send = udp_send, 1816 .pru_soreceive = soreceive_dgram, 1817 .pru_sosend = sosend_dgram, 1818 .pru_shutdown = udp_shutdown, 1819 .pru_sockaddr = in_getsockaddr, 1820 .pru_sosetlabel = in_pcbsosetlabel, 1821 .pru_close = udp_close, 1822 }; 1823 #endif /* INET */ 1824