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 static VNET_DEFINE(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 int cscov_partial, iphlen; 403 404 m = *mp; 405 iphlen = *offp; 406 ifp = m->m_pkthdr.rcvif; 407 *mp = NULL; 408 UDPSTAT_INC(udps_ipackets); 409 410 /* 411 * Strip IP options, if any; should skip this, make available to 412 * user, and use on returned packets, but we don't yet have a way to 413 * check the checksum with options still present. 414 */ 415 if (iphlen > sizeof (struct ip)) { 416 ip_stripoptions(m); 417 iphlen = sizeof(struct ip); 418 } 419 420 /* 421 * Get IP and UDP header together in first mbuf. 422 */ 423 ip = mtod(m, struct ip *); 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 ip = mtod(m, struct ip *); 430 } 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 struct ip_moptions *imo; 531 532 INP_INFO_RLOCK(pcbinfo); 533 pcblist = udp_get_pcblist(proto); 534 last = NULL; 535 LIST_FOREACH(inp, pcblist, inp_list) { 536 if (inp->inp_lport != uh->uh_dport) 537 continue; 538 #ifdef INET6 539 if ((inp->inp_vflag & INP_IPV4) == 0) 540 continue; 541 #endif 542 if (inp->inp_laddr.s_addr != INADDR_ANY && 543 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 544 continue; 545 if (inp->inp_faddr.s_addr != INADDR_ANY && 546 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 547 continue; 548 if (inp->inp_fport != 0 && 549 inp->inp_fport != uh->uh_sport) 550 continue; 551 552 INP_RLOCK(inp); 553 554 /* 555 * XXXRW: Because we weren't holding either the inpcb 556 * or the hash lock when we checked for a match 557 * before, we should probably recheck now that the 558 * inpcb lock is held. 559 */ 560 561 /* 562 * Handle socket delivery policy for any-source 563 * and source-specific multicast. [RFC3678] 564 */ 565 imo = inp->inp_moptions; 566 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 567 struct sockaddr_in group; 568 int blocked; 569 if (imo == NULL) { 570 INP_RUNLOCK(inp); 571 continue; 572 } 573 bzero(&group, sizeof(struct sockaddr_in)); 574 group.sin_len = sizeof(struct sockaddr_in); 575 group.sin_family = AF_INET; 576 group.sin_addr = ip->ip_dst; 577 578 blocked = imo_multi_filter(imo, ifp, 579 (struct sockaddr *)&group, 580 (struct sockaddr *)&udp_in[0]); 581 if (blocked != MCAST_PASS) { 582 if (blocked == MCAST_NOTGMEMBER) 583 IPSTAT_INC(ips_notmember); 584 if (blocked == MCAST_NOTSMEMBER || 585 blocked == MCAST_MUTED) 586 UDPSTAT_INC(udps_filtermcast); 587 INP_RUNLOCK(inp); 588 continue; 589 } 590 } 591 if (last != NULL) { 592 struct mbuf *n; 593 594 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 595 NULL) { 596 UDP_PROBE(receive, NULL, last, ip, 597 last, uh); 598 if (udp_append(last, ip, n, iphlen, 599 udp_in)) { 600 goto inp_lost; 601 } 602 } 603 INP_RUNLOCK(last); 604 } 605 last = inp; 606 /* 607 * Don't look for additional matches if this one does 608 * not have either the SO_REUSEPORT or SO_REUSEADDR 609 * socket options set. This heuristic avoids 610 * searching through all pcbs in the common case of a 611 * non-shared port. It assumes that an application 612 * will never clear these options after setting them. 613 */ 614 if ((last->inp_socket->so_options & 615 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 616 break; 617 } 618 619 if (last == NULL) { 620 /* 621 * No matching pcb found; discard datagram. (No need 622 * to send an ICMP Port Unreachable for a broadcast 623 * or multicast datgram.) 624 */ 625 UDPSTAT_INC(udps_noportbcast); 626 if (inp) 627 INP_RUNLOCK(inp); 628 INP_INFO_RUNLOCK(pcbinfo); 629 goto badunlocked; 630 } 631 UDP_PROBE(receive, NULL, last, ip, last, uh); 632 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 633 INP_RUNLOCK(last); 634 inp_lost: 635 INP_INFO_RUNLOCK(pcbinfo); 636 return (IPPROTO_DONE); 637 } 638 639 /* 640 * Locate pcb for datagram. 641 */ 642 643 /* 644 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 645 */ 646 if ((m->m_flags & M_IP_NEXTHOP) && 647 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 648 struct sockaddr_in *next_hop; 649 650 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 651 652 /* 653 * Transparently forwarded. Pretend to be the destination. 654 * Already got one like this? 655 */ 656 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 657 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 658 if (!inp) { 659 /* 660 * It's new. Try to find the ambushing socket. 661 * Because we've rewritten the destination address, 662 * any hardware-generated hash is ignored. 663 */ 664 inp = in_pcblookup(pcbinfo, ip->ip_src, 665 uh->uh_sport, next_hop->sin_addr, 666 next_hop->sin_port ? htons(next_hop->sin_port) : 667 uh->uh_dport, INPLOOKUP_WILDCARD | 668 INPLOOKUP_RLOCKPCB, ifp); 669 } 670 /* Remove the tag from the packet. We don't need it anymore. */ 671 m_tag_delete(m, fwd_tag); 672 m->m_flags &= ~M_IP_NEXTHOP; 673 } else 674 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 675 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 676 INPLOOKUP_RLOCKPCB, ifp, m); 677 if (inp == NULL) { 678 if (udp_log_in_vain) { 679 char src[INET_ADDRSTRLEN]; 680 char dst[INET_ADDRSTRLEN]; 681 682 log(LOG_INFO, 683 "Connection attempt to UDP %s:%d from %s:%d\n", 684 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 685 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 686 } 687 UDPSTAT_INC(udps_noport); 688 if (m->m_flags & (M_BCAST | M_MCAST)) { 689 UDPSTAT_INC(udps_noportbcast); 690 goto badunlocked; 691 } 692 if (V_udp_blackhole) 693 goto badunlocked; 694 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 695 goto badunlocked; 696 *ip = save_ip; 697 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 698 return (IPPROTO_DONE); 699 } 700 701 /* 702 * Check the minimum TTL for socket. 703 */ 704 INP_RLOCK_ASSERT(inp); 705 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 706 INP_RUNLOCK(inp); 707 m_freem(m); 708 return (IPPROTO_DONE); 709 } 710 if (cscov_partial) { 711 struct udpcb *up; 712 713 up = intoudpcb(inp); 714 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 715 INP_RUNLOCK(inp); 716 m_freem(m); 717 return (IPPROTO_DONE); 718 } 719 } 720 721 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 722 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 723 INP_RUNLOCK(inp); 724 return (IPPROTO_DONE); 725 726 badunlocked: 727 m_freem(m); 728 return (IPPROTO_DONE); 729 } 730 #endif /* INET */ 731 732 /* 733 * Notify a udp user of an asynchronous error; just wake up so that they can 734 * collect error status. 735 */ 736 struct inpcb * 737 udp_notify(struct inpcb *inp, int errno) 738 { 739 740 /* 741 * While udp_ctlinput() always calls udp_notify() with a read lock 742 * when invoking it directly, in_pcbnotifyall() currently uses write 743 * locks due to sharing code with TCP. For now, accept either a read 744 * or a write lock, but a read lock is sufficient. 745 */ 746 INP_LOCK_ASSERT(inp); 747 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 748 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 749 RTFREE(inp->inp_route.ro_rt); 750 inp->inp_route.ro_rt = (struct rtentry *)NULL; 751 } 752 753 inp->inp_socket->so_error = errno; 754 sorwakeup(inp->inp_socket); 755 sowwakeup(inp->inp_socket); 756 return (inp); 757 } 758 759 #ifdef INET 760 static void 761 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 762 struct inpcbinfo *pcbinfo) 763 { 764 struct ip *ip = vip; 765 struct udphdr *uh; 766 struct in_addr faddr; 767 struct inpcb *inp; 768 769 faddr = ((struct sockaddr_in *)sa)->sin_addr; 770 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 771 return; 772 773 if (PRC_IS_REDIRECT(cmd)) { 774 /* signal EHOSTDOWN, as it flushes the cached route */ 775 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 776 return; 777 } 778 779 /* 780 * Hostdead is ugly because it goes linearly through all PCBs. 781 * 782 * XXX: We never get this from ICMP, otherwise it makes an excellent 783 * DoS attack on machines with many connections. 784 */ 785 if (cmd == PRC_HOSTDEAD) 786 ip = NULL; 787 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 788 return; 789 if (ip != NULL) { 790 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 791 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 792 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 793 if (inp != NULL) { 794 INP_RLOCK_ASSERT(inp); 795 if (inp->inp_socket != NULL) { 796 udp_notify(inp, inetctlerrmap[cmd]); 797 } 798 INP_RUNLOCK(inp); 799 } else { 800 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 801 ip->ip_src, uh->uh_sport, 802 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 803 if (inp != NULL) { 804 struct udpcb *up; 805 806 up = intoudpcb(inp); 807 if (up->u_icmp_func != NULL) { 808 INP_RUNLOCK(inp); 809 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx); 810 } else { 811 INP_RUNLOCK(inp); 812 } 813 } 814 } 815 } else 816 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 817 udp_notify); 818 } 819 void 820 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 821 { 822 823 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 824 } 825 826 void 827 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 828 { 829 830 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 831 } 832 #endif /* INET */ 833 834 static int 835 udp_pcblist(SYSCTL_HANDLER_ARGS) 836 { 837 int error, i, n; 838 struct inpcb *inp, **inp_list; 839 struct in_pcblist *il; 840 inp_gen_t gencnt; 841 struct xinpgen xig; 842 843 /* 844 * The process of preparing the PCB list is too time-consuming and 845 * resource-intensive to repeat twice on every request. 846 */ 847 if (req->oldptr == 0) { 848 n = V_udbinfo.ipi_count; 849 n += imax(n / 8, 10); 850 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 851 return (0); 852 } 853 854 if (req->newptr != 0) 855 return (EPERM); 856 857 /* 858 * OK, now we're committed to doing something. 859 */ 860 INP_INFO_RLOCK(&V_udbinfo); 861 gencnt = V_udbinfo.ipi_gencnt; 862 n = V_udbinfo.ipi_count; 863 INP_INFO_RUNLOCK(&V_udbinfo); 864 865 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 866 + n * sizeof(struct xinpcb)); 867 if (error != 0) 868 return (error); 869 870 xig.xig_len = sizeof xig; 871 xig.xig_count = n; 872 xig.xig_gen = gencnt; 873 xig.xig_sogen = so_gencnt; 874 error = SYSCTL_OUT(req, &xig, sizeof xig); 875 if (error) 876 return (error); 877 il = malloc(sizeof(struct in_pcblist) + n * sizeof(struct inpcb *), M_TEMP, M_WAITOK|M_ZERO); 878 inp_list = il->il_inp_list; 879 880 INP_INFO_RLOCK(&V_udbinfo); 881 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 882 inp = LIST_NEXT(inp, inp_list)) { 883 INP_WLOCK(inp); 884 if (inp->inp_gencnt <= gencnt && 885 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 886 in_pcbref(inp); 887 inp_list[i++] = inp; 888 } 889 INP_WUNLOCK(inp); 890 } 891 INP_INFO_RUNLOCK(&V_udbinfo); 892 n = i; 893 894 error = 0; 895 for (i = 0; i < n; i++) { 896 inp = inp_list[i]; 897 INP_RLOCK(inp); 898 if (inp->inp_gencnt <= gencnt) { 899 struct xinpcb xi; 900 901 in_pcbtoxinpcb(inp, &xi); 902 INP_RUNLOCK(inp); 903 error = SYSCTL_OUT(req, &xi, sizeof xi); 904 } else 905 INP_RUNLOCK(inp); 906 } 907 il->il_count = n; 908 il->il_pcbinfo = &V_udbinfo; 909 epoch_call(net_epoch_preempt, &il->il_epoch_ctx, in_pcblist_rele_rlocked); 910 911 if (!error) { 912 /* 913 * Give the user an updated idea of our state. If the 914 * generation differs from what we told her before, she knows 915 * that something happened while we were processing this 916 * request, and it might be necessary to retry. 917 */ 918 INP_INFO_RLOCK(&V_udbinfo); 919 xig.xig_gen = V_udbinfo.ipi_gencnt; 920 xig.xig_sogen = so_gencnt; 921 xig.xig_count = V_udbinfo.ipi_count; 922 INP_INFO_RUNLOCK(&V_udbinfo); 923 error = SYSCTL_OUT(req, &xig, sizeof xig); 924 } 925 return (error); 926 } 927 928 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 929 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 930 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 931 932 #ifdef INET 933 static int 934 udp_getcred(SYSCTL_HANDLER_ARGS) 935 { 936 struct xucred xuc; 937 struct sockaddr_in addrs[2]; 938 struct inpcb *inp; 939 int error; 940 941 error = priv_check(req->td, PRIV_NETINET_GETCRED); 942 if (error) 943 return (error); 944 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 945 if (error) 946 return (error); 947 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 948 addrs[0].sin_addr, addrs[0].sin_port, 949 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 950 if (inp != NULL) { 951 INP_RLOCK_ASSERT(inp); 952 if (inp->inp_socket == NULL) 953 error = ENOENT; 954 if (error == 0) 955 error = cr_canseeinpcb(req->td->td_ucred, inp); 956 if (error == 0) 957 cru2x(inp->inp_cred, &xuc); 958 INP_RUNLOCK(inp); 959 } else 960 error = ENOENT; 961 if (error == 0) 962 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 963 return (error); 964 } 965 966 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 967 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 968 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 969 #endif /* INET */ 970 971 int 972 udp_ctloutput(struct socket *so, struct sockopt *sopt) 973 { 974 struct inpcb *inp; 975 struct udpcb *up; 976 int isudplite, error, optval; 977 978 error = 0; 979 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 980 inp = sotoinpcb(so); 981 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 982 INP_WLOCK(inp); 983 if (sopt->sopt_level != so->so_proto->pr_protocol) { 984 #ifdef INET6 985 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 986 INP_WUNLOCK(inp); 987 error = ip6_ctloutput(so, sopt); 988 } 989 #endif 990 #if defined(INET) && defined(INET6) 991 else 992 #endif 993 #ifdef INET 994 { 995 INP_WUNLOCK(inp); 996 error = ip_ctloutput(so, sopt); 997 } 998 #endif 999 return (error); 1000 } 1001 1002 switch (sopt->sopt_dir) { 1003 case SOPT_SET: 1004 switch (sopt->sopt_name) { 1005 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1006 #ifdef INET 1007 case UDP_ENCAP: 1008 if (!IPSEC_ENABLED(ipv4)) { 1009 INP_WUNLOCK(inp); 1010 return (ENOPROTOOPT); 1011 } 1012 error = UDPENCAP_PCBCTL(inp, sopt); 1013 break; 1014 #endif /* INET */ 1015 #endif /* IPSEC */ 1016 case UDPLITE_SEND_CSCOV: 1017 case UDPLITE_RECV_CSCOV: 1018 if (!isudplite) { 1019 INP_WUNLOCK(inp); 1020 error = ENOPROTOOPT; 1021 break; 1022 } 1023 INP_WUNLOCK(inp); 1024 error = sooptcopyin(sopt, &optval, sizeof(optval), 1025 sizeof(optval)); 1026 if (error != 0) 1027 break; 1028 inp = sotoinpcb(so); 1029 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1030 INP_WLOCK(inp); 1031 up = intoudpcb(inp); 1032 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1033 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1034 INP_WUNLOCK(inp); 1035 error = EINVAL; 1036 break; 1037 } 1038 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1039 up->u_txcslen = optval; 1040 else 1041 up->u_rxcslen = optval; 1042 INP_WUNLOCK(inp); 1043 break; 1044 default: 1045 INP_WUNLOCK(inp); 1046 error = ENOPROTOOPT; 1047 break; 1048 } 1049 break; 1050 case SOPT_GET: 1051 switch (sopt->sopt_name) { 1052 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1053 #ifdef INET 1054 case UDP_ENCAP: 1055 if (!IPSEC_ENABLED(ipv4)) { 1056 INP_WUNLOCK(inp); 1057 return (ENOPROTOOPT); 1058 } 1059 error = UDPENCAP_PCBCTL(inp, sopt); 1060 break; 1061 #endif /* INET */ 1062 #endif /* IPSEC */ 1063 case UDPLITE_SEND_CSCOV: 1064 case UDPLITE_RECV_CSCOV: 1065 if (!isudplite) { 1066 INP_WUNLOCK(inp); 1067 error = ENOPROTOOPT; 1068 break; 1069 } 1070 up = intoudpcb(inp); 1071 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1072 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1073 optval = up->u_txcslen; 1074 else 1075 optval = up->u_rxcslen; 1076 INP_WUNLOCK(inp); 1077 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1078 break; 1079 default: 1080 INP_WUNLOCK(inp); 1081 error = ENOPROTOOPT; 1082 break; 1083 } 1084 break; 1085 } 1086 return (error); 1087 } 1088 1089 #ifdef INET 1090 #define UH_WLOCKED 2 1091 #define UH_RLOCKED 1 1092 #define UH_UNLOCKED 0 1093 static int 1094 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1095 struct mbuf *control, struct thread *td) 1096 { 1097 struct udpiphdr *ui; 1098 int len = m->m_pkthdr.len; 1099 struct in_addr faddr, laddr; 1100 struct cmsghdr *cm; 1101 struct inpcbinfo *pcbinfo; 1102 struct sockaddr_in *sin, src; 1103 int cscov_partial = 0; 1104 int error = 0; 1105 int ipflags; 1106 u_short fport, lport; 1107 int unlock_udbinfo, unlock_inp; 1108 u_char tos; 1109 uint8_t pr; 1110 uint16_t cscov = 0; 1111 uint32_t flowid = 0; 1112 uint8_t flowtype = M_HASHTYPE_NONE; 1113 1114 /* 1115 * udp_output() may need to temporarily bind or connect the current 1116 * inpcb. As such, we don't know up front whether we will need the 1117 * pcbinfo lock or not. Do any work to decide what is needed up 1118 * front before acquiring any locks. 1119 */ 1120 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1121 if (control) 1122 m_freem(control); 1123 m_freem(m); 1124 return (EMSGSIZE); 1125 } 1126 1127 src.sin_family = 0; 1128 sin = (struct sockaddr_in *)addr; 1129 if (sin == NULL || 1130 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1131 INP_WLOCK(inp); 1132 unlock_inp = UH_WLOCKED; 1133 } else { 1134 INP_RLOCK(inp); 1135 unlock_inp = UH_RLOCKED; 1136 } 1137 tos = inp->inp_ip_tos; 1138 if (control != NULL) { 1139 /* 1140 * XXX: Currently, we assume all the optional information is 1141 * stored in a single mbuf. 1142 */ 1143 if (control->m_next) { 1144 if (unlock_inp == UH_WLOCKED) 1145 INP_WUNLOCK(inp); 1146 else 1147 INP_RUNLOCK(inp); 1148 m_freem(control); 1149 m_freem(m); 1150 return (EINVAL); 1151 } 1152 for (; control->m_len > 0; 1153 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1154 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1155 cm = mtod(control, struct cmsghdr *); 1156 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1157 || cm->cmsg_len > control->m_len) { 1158 error = EINVAL; 1159 break; 1160 } 1161 if (cm->cmsg_level != IPPROTO_IP) 1162 continue; 1163 1164 switch (cm->cmsg_type) { 1165 case IP_SENDSRCADDR: 1166 if (cm->cmsg_len != 1167 CMSG_LEN(sizeof(struct in_addr))) { 1168 error = EINVAL; 1169 break; 1170 } 1171 bzero(&src, sizeof(src)); 1172 src.sin_family = AF_INET; 1173 src.sin_len = sizeof(src); 1174 src.sin_port = inp->inp_lport; 1175 src.sin_addr = 1176 *(struct in_addr *)CMSG_DATA(cm); 1177 break; 1178 1179 case IP_TOS: 1180 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1181 error = EINVAL; 1182 break; 1183 } 1184 tos = *(u_char *)CMSG_DATA(cm); 1185 break; 1186 1187 case IP_FLOWID: 1188 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1189 error = EINVAL; 1190 break; 1191 } 1192 flowid = *(uint32_t *) CMSG_DATA(cm); 1193 break; 1194 1195 case IP_FLOWTYPE: 1196 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1197 error = EINVAL; 1198 break; 1199 } 1200 flowtype = *(uint32_t *) CMSG_DATA(cm); 1201 break; 1202 1203 #ifdef RSS 1204 case IP_RSSBUCKETID: 1205 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1206 error = EINVAL; 1207 break; 1208 } 1209 /* This is just a placeholder for now */ 1210 break; 1211 #endif /* RSS */ 1212 default: 1213 error = ENOPROTOOPT; 1214 break; 1215 } 1216 if (error) 1217 break; 1218 } 1219 m_freem(control); 1220 } 1221 if (error) { 1222 if (unlock_inp == UH_WLOCKED) 1223 INP_WUNLOCK(inp); 1224 else 1225 INP_RUNLOCK(inp); 1226 m_freem(m); 1227 return (error); 1228 } 1229 1230 /* 1231 * Depending on whether or not the application has bound or connected 1232 * the socket, we may have to do varying levels of work. The optimal 1233 * case is for a connected UDP socket, as a global lock isn't 1234 * required at all. 1235 * 1236 * In order to decide which we need, we require stability of the 1237 * inpcb binding, which we ensure by acquiring a read lock on the 1238 * inpcb. This doesn't strictly follow the lock order, so we play 1239 * the trylock and retry game; note that we may end up with more 1240 * conservative locks than required the second time around, so later 1241 * assertions have to accept that. Further analysis of the number of 1242 * misses under contention is required. 1243 * 1244 * XXXRW: Check that hash locking update here is correct. 1245 */ 1246 pr = inp->inp_socket->so_proto->pr_protocol; 1247 pcbinfo = udp_get_inpcbinfo(pr); 1248 sin = (struct sockaddr_in *)addr; 1249 if (sin != NULL && 1250 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1251 INP_HASH_WLOCK(pcbinfo); 1252 unlock_udbinfo = UH_WLOCKED; 1253 } else if ((sin != NULL && ( 1254 (sin->sin_addr.s_addr == INADDR_ANY) || 1255 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1256 (inp->inp_laddr.s_addr == INADDR_ANY) || 1257 (inp->inp_lport == 0))) || 1258 (src.sin_family == AF_INET)) { 1259 INP_HASH_RLOCK(pcbinfo); 1260 unlock_udbinfo = UH_RLOCKED; 1261 } else 1262 unlock_udbinfo = UH_UNLOCKED; 1263 1264 /* 1265 * If the IP_SENDSRCADDR control message was specified, override the 1266 * source address for this datagram. Its use is invalidated if the 1267 * address thus specified is incomplete or clobbers other inpcbs. 1268 */ 1269 laddr = inp->inp_laddr; 1270 lport = inp->inp_lport; 1271 if (src.sin_family == AF_INET) { 1272 INP_HASH_LOCK_ASSERT(pcbinfo); 1273 if ((lport == 0) || 1274 (laddr.s_addr == INADDR_ANY && 1275 src.sin_addr.s_addr == INADDR_ANY)) { 1276 error = EINVAL; 1277 goto release; 1278 } 1279 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1280 &laddr.s_addr, &lport, td->td_ucred); 1281 if (error) 1282 goto release; 1283 } 1284 1285 /* 1286 * If a UDP socket has been connected, then a local address/port will 1287 * have been selected and bound. 1288 * 1289 * If a UDP socket has not been connected to, then an explicit 1290 * destination address must be used, in which case a local 1291 * address/port may not have been selected and bound. 1292 */ 1293 if (sin != NULL) { 1294 INP_LOCK_ASSERT(inp); 1295 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1296 error = EISCONN; 1297 goto release; 1298 } 1299 1300 /* 1301 * Jail may rewrite the destination address, so let it do 1302 * that before we use it. 1303 */ 1304 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1305 if (error) 1306 goto release; 1307 1308 /* 1309 * If a local address or port hasn't yet been selected, or if 1310 * the destination address needs to be rewritten due to using 1311 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1312 * to do the heavy lifting. Once a port is selected, we 1313 * commit the binding back to the socket; we also commit the 1314 * binding of the address if in jail. 1315 * 1316 * If we already have a valid binding and we're not 1317 * requesting a destination address rewrite, use a fast path. 1318 */ 1319 if (inp->inp_laddr.s_addr == INADDR_ANY || 1320 inp->inp_lport == 0 || 1321 sin->sin_addr.s_addr == INADDR_ANY || 1322 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1323 INP_HASH_LOCK_ASSERT(pcbinfo); 1324 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1325 &lport, &faddr.s_addr, &fport, NULL, 1326 td->td_ucred); 1327 if (error) 1328 goto release; 1329 1330 /* 1331 * XXXRW: Why not commit the port if the address is 1332 * !INADDR_ANY? 1333 */ 1334 /* Commit the local port if newly assigned. */ 1335 if (inp->inp_laddr.s_addr == INADDR_ANY && 1336 inp->inp_lport == 0) { 1337 INP_WLOCK_ASSERT(inp); 1338 INP_HASH_WLOCK_ASSERT(pcbinfo); 1339 /* 1340 * Remember addr if jailed, to prevent 1341 * rebinding. 1342 */ 1343 if (prison_flag(td->td_ucred, PR_IP4)) 1344 inp->inp_laddr = laddr; 1345 inp->inp_lport = lport; 1346 if (in_pcbinshash(inp) != 0) { 1347 inp->inp_lport = 0; 1348 error = EAGAIN; 1349 goto release; 1350 } 1351 inp->inp_flags |= INP_ANONPORT; 1352 } 1353 } else { 1354 faddr = sin->sin_addr; 1355 fport = sin->sin_port; 1356 } 1357 } else { 1358 INP_LOCK_ASSERT(inp); 1359 faddr = inp->inp_faddr; 1360 fport = inp->inp_fport; 1361 if (faddr.s_addr == INADDR_ANY) { 1362 error = ENOTCONN; 1363 goto release; 1364 } 1365 } 1366 1367 /* 1368 * Calculate data length and get a mbuf for UDP, IP, and possible 1369 * link-layer headers. Immediate slide the data pointer back forward 1370 * since we won't use that space at this layer. 1371 */ 1372 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1373 if (m == NULL) { 1374 error = ENOBUFS; 1375 goto release; 1376 } 1377 m->m_data += max_linkhdr; 1378 m->m_len -= max_linkhdr; 1379 m->m_pkthdr.len -= max_linkhdr; 1380 1381 /* 1382 * Fill in mbuf with extended UDP header and addresses and length put 1383 * into network format. 1384 */ 1385 ui = mtod(m, struct udpiphdr *); 1386 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1387 ui->ui_pr = pr; 1388 ui->ui_src = laddr; 1389 ui->ui_dst = faddr; 1390 ui->ui_sport = lport; 1391 ui->ui_dport = fport; 1392 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1393 if (pr == IPPROTO_UDPLITE) { 1394 struct udpcb *up; 1395 uint16_t plen; 1396 1397 up = intoudpcb(inp); 1398 cscov = up->u_txcslen; 1399 plen = (u_short)len + sizeof(struct udphdr); 1400 if (cscov >= plen) 1401 cscov = 0; 1402 ui->ui_len = htons(plen); 1403 ui->ui_ulen = htons(cscov); 1404 /* 1405 * For UDP-Lite, checksum coverage length of zero means 1406 * the entire UDPLite packet is covered by the checksum. 1407 */ 1408 cscov_partial = (cscov == 0) ? 0 : 1; 1409 } else 1410 ui->ui_v = IPVERSION << 4; 1411 1412 /* 1413 * Set the Don't Fragment bit in the IP header. 1414 */ 1415 if (inp->inp_flags & INP_DONTFRAG) { 1416 struct ip *ip; 1417 1418 ip = (struct ip *)&ui->ui_i; 1419 ip->ip_off |= htons(IP_DF); 1420 } 1421 1422 ipflags = 0; 1423 if (inp->inp_socket->so_options & SO_DONTROUTE) 1424 ipflags |= IP_ROUTETOIF; 1425 if (inp->inp_socket->so_options & SO_BROADCAST) 1426 ipflags |= IP_ALLOWBROADCAST; 1427 if (inp->inp_flags & INP_ONESBCAST) 1428 ipflags |= IP_SENDONES; 1429 1430 #ifdef MAC 1431 mac_inpcb_create_mbuf(inp, m); 1432 #endif 1433 1434 /* 1435 * Set up checksum and output datagram. 1436 */ 1437 ui->ui_sum = 0; 1438 if (pr == IPPROTO_UDPLITE) { 1439 if (inp->inp_flags & INP_ONESBCAST) 1440 faddr.s_addr = INADDR_BROADCAST; 1441 if (cscov_partial) { 1442 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1443 ui->ui_sum = 0xffff; 1444 } else { 1445 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1446 ui->ui_sum = 0xffff; 1447 } 1448 } else if (V_udp_cksum) { 1449 if (inp->inp_flags & INP_ONESBCAST) 1450 faddr.s_addr = INADDR_BROADCAST; 1451 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1452 htons((u_short)len + sizeof(struct udphdr) + pr)); 1453 m->m_pkthdr.csum_flags = CSUM_UDP; 1454 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1455 } 1456 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1457 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1458 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1459 UDPSTAT_INC(udps_opackets); 1460 1461 /* 1462 * Setup flowid / RSS information for outbound socket. 1463 * 1464 * Once the UDP code decides to set a flowid some other way, 1465 * this allows the flowid to be overridden by userland. 1466 */ 1467 if (flowtype != M_HASHTYPE_NONE) { 1468 m->m_pkthdr.flowid = flowid; 1469 M_HASHTYPE_SET(m, flowtype); 1470 #ifdef RSS 1471 } else { 1472 uint32_t hash_val, hash_type; 1473 /* 1474 * Calculate an appropriate RSS hash for UDP and 1475 * UDP Lite. 1476 * 1477 * The called function will take care of figuring out 1478 * whether a 2-tuple or 4-tuple hash is required based 1479 * on the currently configured scheme. 1480 * 1481 * Later later on connected socket values should be 1482 * cached in the inpcb and reused, rather than constantly 1483 * re-calculating it. 1484 * 1485 * UDP Lite is a different protocol number and will 1486 * likely end up being hashed as a 2-tuple until 1487 * RSS / NICs grow UDP Lite protocol awareness. 1488 */ 1489 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1490 pr, &hash_val, &hash_type) == 0) { 1491 m->m_pkthdr.flowid = hash_val; 1492 M_HASHTYPE_SET(m, hash_type); 1493 } 1494 #endif 1495 } 1496 1497 #ifdef RSS 1498 /* 1499 * Don't override with the inp cached flowid value. 1500 * 1501 * Depending upon the kind of send being done, the inp 1502 * flowid/flowtype values may actually not be appropriate 1503 * for this particular socket send. 1504 * 1505 * We should either leave the flowid at zero (which is what is 1506 * currently done) or set it to some software generated 1507 * hash value based on the packet contents. 1508 */ 1509 ipflags |= IP_NODEFAULTFLOWID; 1510 #endif /* RSS */ 1511 1512 if (unlock_udbinfo == UH_WLOCKED) 1513 INP_HASH_WUNLOCK(pcbinfo); 1514 else if (unlock_udbinfo == UH_RLOCKED) 1515 INP_HASH_RUNLOCK(pcbinfo); 1516 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1517 error = ip_output(m, inp->inp_options, 1518 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1519 inp->inp_moptions, inp); 1520 if (unlock_inp == UH_WLOCKED) 1521 INP_WUNLOCK(inp); 1522 else 1523 INP_RUNLOCK(inp); 1524 return (error); 1525 1526 release: 1527 if (unlock_udbinfo == UH_WLOCKED) { 1528 KASSERT(unlock_inp == UH_WLOCKED, 1529 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1530 INP_HASH_WUNLOCK(pcbinfo); 1531 INP_WUNLOCK(inp); 1532 } else if (unlock_udbinfo == UH_RLOCKED) { 1533 KASSERT(unlock_inp == UH_RLOCKED, 1534 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1535 INP_HASH_RUNLOCK(pcbinfo); 1536 INP_RUNLOCK(inp); 1537 } else if (unlock_inp == UH_WLOCKED) 1538 INP_WUNLOCK(inp); 1539 else 1540 INP_RUNLOCK(inp); 1541 m_freem(m); 1542 return (error); 1543 } 1544 1545 static void 1546 udp_abort(struct socket *so) 1547 { 1548 struct inpcb *inp; 1549 struct inpcbinfo *pcbinfo; 1550 1551 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1552 inp = sotoinpcb(so); 1553 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1554 INP_WLOCK(inp); 1555 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1556 INP_HASH_WLOCK(pcbinfo); 1557 in_pcbdisconnect(inp); 1558 inp->inp_laddr.s_addr = INADDR_ANY; 1559 INP_HASH_WUNLOCK(pcbinfo); 1560 soisdisconnected(so); 1561 } 1562 INP_WUNLOCK(inp); 1563 } 1564 1565 static int 1566 udp_attach(struct socket *so, int proto, struct thread *td) 1567 { 1568 struct inpcb *inp; 1569 struct inpcbinfo *pcbinfo; 1570 int error; 1571 1572 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1573 inp = sotoinpcb(so); 1574 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1575 error = soreserve(so, udp_sendspace, udp_recvspace); 1576 if (error) 1577 return (error); 1578 INP_INFO_WLOCK(pcbinfo); 1579 error = in_pcballoc(so, pcbinfo); 1580 if (error) { 1581 INP_INFO_WUNLOCK(pcbinfo); 1582 return (error); 1583 } 1584 1585 inp = sotoinpcb(so); 1586 inp->inp_vflag |= INP_IPV4; 1587 inp->inp_ip_ttl = V_ip_defttl; 1588 1589 error = udp_newudpcb(inp); 1590 if (error) { 1591 in_pcbdetach(inp); 1592 in_pcbfree(inp); 1593 INP_INFO_WUNLOCK(pcbinfo); 1594 return (error); 1595 } 1596 1597 INP_WUNLOCK(inp); 1598 INP_INFO_WUNLOCK(pcbinfo); 1599 return (0); 1600 } 1601 #endif /* INET */ 1602 1603 int 1604 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1605 { 1606 struct inpcb *inp; 1607 struct udpcb *up; 1608 1609 KASSERT(so->so_type == SOCK_DGRAM, 1610 ("udp_set_kernel_tunneling: !dgram")); 1611 inp = sotoinpcb(so); 1612 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1613 INP_WLOCK(inp); 1614 up = intoudpcb(inp); 1615 if ((up->u_tun_func != NULL) || 1616 (up->u_icmp_func != NULL)) { 1617 INP_WUNLOCK(inp); 1618 return (EBUSY); 1619 } 1620 up->u_tun_func = f; 1621 up->u_icmp_func = i; 1622 up->u_tun_ctx = ctx; 1623 INP_WUNLOCK(inp); 1624 return (0); 1625 } 1626 1627 #ifdef INET 1628 static int 1629 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1630 { 1631 struct inpcb *inp; 1632 struct inpcbinfo *pcbinfo; 1633 int error; 1634 1635 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1636 inp = sotoinpcb(so); 1637 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1638 INP_WLOCK(inp); 1639 INP_HASH_WLOCK(pcbinfo); 1640 error = in_pcbbind(inp, nam, td->td_ucred); 1641 INP_HASH_WUNLOCK(pcbinfo); 1642 INP_WUNLOCK(inp); 1643 return (error); 1644 } 1645 1646 static void 1647 udp_close(struct socket *so) 1648 { 1649 struct inpcb *inp; 1650 struct inpcbinfo *pcbinfo; 1651 1652 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1653 inp = sotoinpcb(so); 1654 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1655 INP_WLOCK(inp); 1656 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1657 INP_HASH_WLOCK(pcbinfo); 1658 in_pcbdisconnect(inp); 1659 inp->inp_laddr.s_addr = INADDR_ANY; 1660 INP_HASH_WUNLOCK(pcbinfo); 1661 soisdisconnected(so); 1662 } 1663 INP_WUNLOCK(inp); 1664 } 1665 1666 static int 1667 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1668 { 1669 struct inpcb *inp; 1670 struct inpcbinfo *pcbinfo; 1671 struct sockaddr_in *sin; 1672 int error; 1673 1674 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1675 inp = sotoinpcb(so); 1676 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1677 INP_WLOCK(inp); 1678 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1679 INP_WUNLOCK(inp); 1680 return (EISCONN); 1681 } 1682 sin = (struct sockaddr_in *)nam; 1683 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1684 if (error != 0) { 1685 INP_WUNLOCK(inp); 1686 return (error); 1687 } 1688 INP_HASH_WLOCK(pcbinfo); 1689 error = in_pcbconnect(inp, nam, td->td_ucred); 1690 INP_HASH_WUNLOCK(pcbinfo); 1691 if (error == 0) 1692 soisconnected(so); 1693 INP_WUNLOCK(inp); 1694 return (error); 1695 } 1696 1697 static void 1698 udp_detach(struct socket *so) 1699 { 1700 struct inpcb *inp; 1701 struct inpcbinfo *pcbinfo; 1702 struct udpcb *up; 1703 1704 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1705 inp = sotoinpcb(so); 1706 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1707 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1708 ("udp_detach: not disconnected")); 1709 INP_INFO_WLOCK(pcbinfo); 1710 INP_WLOCK(inp); 1711 up = intoudpcb(inp); 1712 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1713 /* XXX defer to epoch_call */ 1714 inp->inp_ppcb = NULL; 1715 in_pcbdetach(inp); 1716 in_pcbfree(inp); 1717 INP_INFO_WUNLOCK(pcbinfo); 1718 udp_discardcb(up); 1719 } 1720 1721 static int 1722 udp_disconnect(struct socket *so) 1723 { 1724 struct inpcb *inp; 1725 struct inpcbinfo *pcbinfo; 1726 1727 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1728 inp = sotoinpcb(so); 1729 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1730 INP_WLOCK(inp); 1731 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1732 INP_WUNLOCK(inp); 1733 return (ENOTCONN); 1734 } 1735 INP_HASH_WLOCK(pcbinfo); 1736 in_pcbdisconnect(inp); 1737 inp->inp_laddr.s_addr = INADDR_ANY; 1738 INP_HASH_WUNLOCK(pcbinfo); 1739 SOCK_LOCK(so); 1740 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1741 SOCK_UNLOCK(so); 1742 INP_WUNLOCK(inp); 1743 return (0); 1744 } 1745 1746 static int 1747 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1748 struct mbuf *control, struct thread *td) 1749 { 1750 struct inpcb *inp; 1751 1752 inp = sotoinpcb(so); 1753 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1754 return (udp_output(inp, m, addr, control, td)); 1755 } 1756 #endif /* INET */ 1757 1758 int 1759 udp_shutdown(struct socket *so) 1760 { 1761 struct inpcb *inp; 1762 1763 inp = sotoinpcb(so); 1764 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1765 INP_WLOCK(inp); 1766 socantsendmore(so); 1767 INP_WUNLOCK(inp); 1768 return (0); 1769 } 1770 1771 #ifdef INET 1772 struct pr_usrreqs udp_usrreqs = { 1773 .pru_abort = udp_abort, 1774 .pru_attach = udp_attach, 1775 .pru_bind = udp_bind, 1776 .pru_connect = udp_connect, 1777 .pru_control = in_control, 1778 .pru_detach = udp_detach, 1779 .pru_disconnect = udp_disconnect, 1780 .pru_peeraddr = in_getpeeraddr, 1781 .pru_send = udp_send, 1782 .pru_soreceive = soreceive_dgram, 1783 .pru_sosend = sosend_dgram, 1784 .pru_shutdown = udp_shutdown, 1785 .pru_sockaddr = in_getsockaddr, 1786 .pru_sosetlabel = in_pcbsosetlabel, 1787 .pru_close = udp_close, 1788 }; 1789 #endif /* INET */ 1790