1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. 4 * Copyright (c) 2008 Robert N. M. Watson 5 * Copyright (c) 2010-2011 Juniper Networks, Inc. 6 * Copyright (c) 2014 Kevin Lo 7 * All rights reserved. 8 * 9 * Portions of this software were developed by Robert N. M. Watson under 10 * contract to Juniper Networks, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_ipsec.h" 45 #include "opt_rss.h" 46 47 #include <sys/param.h> 48 #include <sys/domain.h> 49 #include <sys/eventhandler.h> 50 #include <sys/jail.h> 51 #include <sys/kernel.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mbuf.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/protosw.h> 58 #include <sys/sdt.h> 59 #include <sys/signalvar.h> 60 #include <sys/socket.h> 61 #include <sys/socketvar.h> 62 #include <sys/sx.h> 63 #include <sys/sysctl.h> 64 #include <sys/syslog.h> 65 #include <sys/systm.h> 66 67 #include <vm/uma.h> 68 69 #include <net/if.h> 70 #include <net/if_var.h> 71 #include <net/route.h> 72 #include <net/rss_config.h> 73 74 #include <netinet/in.h> 75 #include <netinet/in_kdtrace.h> 76 #include <netinet/in_pcb.h> 77 #include <netinet/in_systm.h> 78 #include <netinet/in_var.h> 79 #include <netinet/ip.h> 80 #ifdef INET6 81 #include <netinet/ip6.h> 82 #endif 83 #include <netinet/ip_icmp.h> 84 #include <netinet/icmp_var.h> 85 #include <netinet/ip_var.h> 86 #include <netinet/ip_options.h> 87 #ifdef INET6 88 #include <netinet6/ip6_var.h> 89 #endif 90 #include <netinet/udp.h> 91 #include <netinet/udp_var.h> 92 #include <netinet/udplite.h> 93 #include <netinet/in_rss.h> 94 95 #include <netipsec/ipsec_support.h> 96 97 #include <machine/in_cksum.h> 98 99 #include <security/mac/mac_framework.h> 100 101 /* 102 * UDP and UDP-Lite protocols implementation. 103 * Per RFC 768, August, 1980. 104 * Per RFC 3828, July, 2004. 105 */ 106 107 /* 108 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 109 * removes the only data integrity mechanism for packets and malformed 110 * packets that would otherwise be discarded due to bad checksums, and may 111 * cause problems (especially for NFS data blocks). 112 */ 113 VNET_DEFINE(int, udp_cksum) = 1; 114 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, 115 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 116 117 int udp_log_in_vain = 0; 118 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 119 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 120 121 VNET_DEFINE(int, udp_blackhole) = 0; 122 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 123 &VNET_NAME(udp_blackhole), 0, 124 "Do not send port unreachables for refused connects"); 125 126 u_long udp_sendspace = 9216; /* really max datagram size */ 127 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 128 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 129 130 u_long udp_recvspace = 40 * (1024 + 131 #ifdef INET6 132 sizeof(struct sockaddr_in6) 133 #else 134 sizeof(struct sockaddr_in) 135 #endif 136 ); /* 40 1K datagrams */ 137 138 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 139 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 140 141 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 142 VNET_DEFINE(struct inpcbinfo, udbinfo); 143 VNET_DEFINE(struct inpcbhead, ulitecb); 144 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 145 static VNET_DEFINE(uma_zone_t, udpcb_zone); 146 #define V_udpcb_zone VNET(udpcb_zone) 147 148 #ifndef UDBHASHSIZE 149 #define UDBHASHSIZE 128 150 #endif 151 152 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 153 VNET_PCPUSTAT_SYSINIT(udpstat); 154 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 155 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 156 157 #ifdef VIMAGE 158 VNET_PCPUSTAT_SYSUNINIT(udpstat); 159 #endif /* VIMAGE */ 160 #ifdef INET 161 static void udp_detach(struct socket *so); 162 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 163 struct mbuf *, struct thread *); 164 #endif 165 166 static void 167 udp_zone_change(void *tag) 168 { 169 170 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 171 uma_zone_set_max(V_udpcb_zone, maxsockets); 172 } 173 174 static int 175 udp_inpcb_init(void *mem, int size, int flags) 176 { 177 struct inpcb *inp; 178 179 inp = mem; 180 INP_LOCK_INIT(inp, "inp", "udpinp"); 181 return (0); 182 } 183 184 static int 185 udplite_inpcb_init(void *mem, int size, int flags) 186 { 187 struct inpcb *inp; 188 189 inp = mem; 190 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 191 return (0); 192 } 193 194 void 195 udp_init(void) 196 { 197 198 /* 199 * For now default to 2-tuple UDP hashing - until the fragment 200 * reassembly code can also update the flowid. 201 * 202 * Once we can calculate the flowid that way and re-establish 203 * a 4-tuple, flip this to 4-tuple. 204 */ 205 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 206 "udp_inpcb", udp_inpcb_init, NULL, 0, 207 IPI_HASHFIELDS_2TUPLE); 208 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 210 uma_zone_set_max(V_udpcb_zone, maxsockets); 211 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 212 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 213 EVENTHANDLER_PRI_ANY); 214 } 215 216 void 217 udplite_init(void) 218 { 219 220 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 221 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL, 222 0, IPI_HASHFIELDS_2TUPLE); 223 } 224 225 /* 226 * Kernel module interface for updating udpstat. The argument is an index 227 * into udpstat treated as an array of u_long. While this encodes the 228 * general layout of udpstat into the caller, it doesn't encode its location, 229 * so that future changes to add, for example, per-CPU stats support won't 230 * cause binary compatibility problems for kernel modules. 231 */ 232 void 233 kmod_udpstat_inc(int statnum) 234 { 235 236 counter_u64_add(VNET(udpstat)[statnum], 1); 237 } 238 239 int 240 udp_newudpcb(struct inpcb *inp) 241 { 242 struct udpcb *up; 243 244 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 245 if (up == NULL) 246 return (ENOBUFS); 247 inp->inp_ppcb = up; 248 return (0); 249 } 250 251 void 252 udp_discardcb(struct udpcb *up) 253 { 254 255 uma_zfree(V_udpcb_zone, up); 256 } 257 258 #ifdef VIMAGE 259 static void 260 udp_destroy(void *unused __unused) 261 { 262 263 in_pcbinfo_destroy(&V_udbinfo); 264 uma_zdestroy(V_udpcb_zone); 265 } 266 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 267 268 static void 269 udplite_destroy(void *unused __unused) 270 { 271 272 in_pcbinfo_destroy(&V_ulitecbinfo); 273 } 274 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 275 NULL); 276 #endif 277 278 #ifdef INET 279 /* 280 * Subroutine of udp_input(), which appends the provided mbuf chain to the 281 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 282 * contains the source address. If the socket ends up being an IPv6 socket, 283 * udp_append() will convert to a sockaddr_in6 before passing the address 284 * into the socket code. 285 * 286 * In the normal case udp_append() will return 0, indicating that you 287 * must unlock the inp. However if a tunneling protocol is in place we increment 288 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 289 * then decrement the reference count. If the inp_rele returns 1, indicating the 290 * inp is gone, we return that to the caller to tell them *not* to unlock 291 * the inp. In the case of multi-cast this will cause the distribution 292 * to stop (though most tunneling protocols known currently do *not* use 293 * multicast). 294 */ 295 static int 296 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 297 struct sockaddr_in *udp_in) 298 { 299 struct sockaddr *append_sa; 300 struct socket *so; 301 struct mbuf *tmpopts, *opts = NULL; 302 #ifdef INET6 303 struct sockaddr_in6 udp_in6; 304 #endif 305 struct udpcb *up; 306 307 INP_LOCK_ASSERT(inp); 308 309 /* 310 * Engage the tunneling protocol. 311 */ 312 up = intoudpcb(inp); 313 if (up->u_tun_func != NULL) { 314 in_pcbref(inp); 315 INP_RUNLOCK(inp); 316 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], 317 up->u_tun_ctx); 318 INP_RLOCK(inp); 319 return (in_pcbrele_rlocked(inp)); 320 } 321 322 off += sizeof(struct udphdr); 323 324 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 325 /* Check AH/ESP integrity. */ 326 if (IPSEC_ENABLED(ipv4) && 327 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 328 m_freem(n); 329 return (0); 330 } 331 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 332 if (IPSEC_ENABLED(ipv4) && 333 UDPENCAP_INPUT(n, off, AF_INET) != 0) 334 return (0); /* Consumed. */ 335 } 336 #endif /* IPSEC */ 337 #ifdef MAC 338 if (mac_inpcb_check_deliver(inp, n) != 0) { 339 m_freem(n); 340 return (0); 341 } 342 #endif /* MAC */ 343 if (inp->inp_flags & INP_CONTROLOPTS || 344 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 345 #ifdef INET6 346 if (inp->inp_vflag & INP_IPV6) 347 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 348 else 349 #endif /* INET6 */ 350 ip_savecontrol(inp, &opts, ip, n); 351 } 352 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 353 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], 354 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); 355 if (tmpopts) { 356 if (opts) { 357 tmpopts->m_next = opts; 358 opts = tmpopts; 359 } else 360 opts = tmpopts; 361 } 362 } 363 #ifdef INET6 364 if (inp->inp_vflag & INP_IPV6) { 365 bzero(&udp_in6, sizeof(udp_in6)); 366 udp_in6.sin6_len = sizeof(udp_in6); 367 udp_in6.sin6_family = AF_INET6; 368 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 369 append_sa = (struct sockaddr *)&udp_in6; 370 } else 371 #endif /* INET6 */ 372 append_sa = (struct sockaddr *)&udp_in[0]; 373 m_adj(n, off); 374 375 so = inp->inp_socket; 376 SOCKBUF_LOCK(&so->so_rcv); 377 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 378 SOCKBUF_UNLOCK(&so->so_rcv); 379 m_freem(n); 380 if (opts) 381 m_freem(opts); 382 UDPSTAT_INC(udps_fullsock); 383 } else 384 sorwakeup_locked(so); 385 return (0); 386 } 387 388 int 389 udp_input(struct mbuf **mp, int *offp, int proto) 390 { 391 struct ip *ip; 392 struct udphdr *uh; 393 struct ifnet *ifp; 394 struct inpcb *inp; 395 uint16_t len, ip_len; 396 struct inpcbinfo *pcbinfo; 397 struct ip save_ip; 398 struct sockaddr_in udp_in[2]; 399 struct mbuf *m; 400 struct m_tag *fwd_tag; 401 int cscov_partial, iphlen; 402 403 m = *mp; 404 iphlen = *offp; 405 ifp = m->m_pkthdr.rcvif; 406 *mp = NULL; 407 UDPSTAT_INC(udps_ipackets); 408 409 /* 410 * Strip IP options, if any; should skip this, make available to 411 * user, and use on returned packets, but we don't yet have a way to 412 * check the checksum with options still present. 413 */ 414 if (iphlen > sizeof (struct ip)) { 415 ip_stripoptions(m); 416 iphlen = sizeof(struct ip); 417 } 418 419 /* 420 * Get IP and UDP header together in first mbuf. 421 */ 422 ip = mtod(m, struct ip *); 423 if (m->m_len < iphlen + sizeof(struct udphdr)) { 424 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 425 UDPSTAT_INC(udps_hdrops); 426 return (IPPROTO_DONE); 427 } 428 ip = mtod(m, struct ip *); 429 } 430 uh = (struct udphdr *)((caddr_t)ip + iphlen); 431 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 432 433 /* 434 * Destination port of 0 is illegal, based on RFC768. 435 */ 436 if (uh->uh_dport == 0) 437 goto badunlocked; 438 439 /* 440 * Construct sockaddr format source address. Stuff source address 441 * and datagram in user buffer. 442 */ 443 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 444 udp_in[0].sin_len = sizeof(struct sockaddr_in); 445 udp_in[0].sin_family = AF_INET; 446 udp_in[0].sin_port = uh->uh_sport; 447 udp_in[0].sin_addr = ip->ip_src; 448 udp_in[1].sin_len = sizeof(struct sockaddr_in); 449 udp_in[1].sin_family = AF_INET; 450 udp_in[1].sin_port = uh->uh_dport; 451 udp_in[1].sin_addr = ip->ip_dst; 452 453 /* 454 * Make mbuf data length reflect UDP length. If not enough data to 455 * reflect UDP length, drop. 456 */ 457 len = ntohs((u_short)uh->uh_ulen); 458 ip_len = ntohs(ip->ip_len) - iphlen; 459 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 460 /* Zero means checksum over the complete packet. */ 461 if (len == 0) 462 len = ip_len; 463 cscov_partial = 0; 464 } 465 if (ip_len != len) { 466 if (len > ip_len || len < sizeof(struct udphdr)) { 467 UDPSTAT_INC(udps_badlen); 468 goto badunlocked; 469 } 470 if (proto == IPPROTO_UDP) 471 m_adj(m, len - ip_len); 472 } 473 474 /* 475 * Save a copy of the IP header in case we want restore it for 476 * sending an ICMP error message in response. 477 */ 478 if (!V_udp_blackhole) 479 save_ip = *ip; 480 else 481 memset(&save_ip, 0, sizeof(save_ip)); 482 483 /* 484 * Checksum extended UDP header and data. 485 */ 486 if (uh->uh_sum) { 487 u_short uh_sum; 488 489 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 490 !cscov_partial) { 491 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 492 uh_sum = m->m_pkthdr.csum_data; 493 else 494 uh_sum = in_pseudo(ip->ip_src.s_addr, 495 ip->ip_dst.s_addr, htonl((u_short)len + 496 m->m_pkthdr.csum_data + proto)); 497 uh_sum ^= 0xffff; 498 } else { 499 char b[9]; 500 501 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 502 bzero(((struct ipovly *)ip)->ih_x1, 9); 503 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 504 uh->uh_ulen : htons(ip_len); 505 uh_sum = in_cksum(m, len + sizeof (struct ip)); 506 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 507 } 508 if (uh_sum) { 509 UDPSTAT_INC(udps_badsum); 510 m_freem(m); 511 return (IPPROTO_DONE); 512 } 513 } else { 514 if (proto == IPPROTO_UDP) { 515 UDPSTAT_INC(udps_nosum); 516 } else { 517 /* UDPLite requires a checksum */ 518 /* XXX: What is the right UDPLite MIB counter here? */ 519 m_freem(m); 520 return (IPPROTO_DONE); 521 } 522 } 523 524 pcbinfo = udp_get_inpcbinfo(proto); 525 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 526 in_broadcast(ip->ip_dst, ifp)) { 527 struct inpcb *last; 528 struct inpcbhead *pcblist; 529 struct ip_moptions *imo; 530 531 INP_INFO_RLOCK(pcbinfo); 532 pcblist = udp_get_pcblist(proto); 533 last = NULL; 534 LIST_FOREACH(inp, pcblist, inp_list) { 535 if (inp->inp_lport != uh->uh_dport) 536 continue; 537 #ifdef INET6 538 if ((inp->inp_vflag & INP_IPV4) == 0) 539 continue; 540 #endif 541 if (inp->inp_laddr.s_addr != INADDR_ANY && 542 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 543 continue; 544 if (inp->inp_faddr.s_addr != INADDR_ANY && 545 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 546 continue; 547 if (inp->inp_fport != 0 && 548 inp->inp_fport != uh->uh_sport) 549 continue; 550 551 INP_RLOCK(inp); 552 553 /* 554 * XXXRW: Because we weren't holding either the inpcb 555 * or the hash lock when we checked for a match 556 * before, we should probably recheck now that the 557 * inpcb lock is held. 558 */ 559 560 /* 561 * Handle socket delivery policy for any-source 562 * and source-specific multicast. [RFC3678] 563 */ 564 imo = inp->inp_moptions; 565 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 566 struct sockaddr_in group; 567 int blocked; 568 if (imo == NULL) { 569 INP_RUNLOCK(inp); 570 continue; 571 } 572 bzero(&group, sizeof(struct sockaddr_in)); 573 group.sin_len = sizeof(struct sockaddr_in); 574 group.sin_family = AF_INET; 575 group.sin_addr = ip->ip_dst; 576 577 blocked = imo_multi_filter(imo, ifp, 578 (struct sockaddr *)&group, 579 (struct sockaddr *)&udp_in[0]); 580 if (blocked != MCAST_PASS) { 581 if (blocked == MCAST_NOTGMEMBER) 582 IPSTAT_INC(ips_notmember); 583 if (blocked == MCAST_NOTSMEMBER || 584 blocked == MCAST_MUTED) 585 UDPSTAT_INC(udps_filtermcast); 586 INP_RUNLOCK(inp); 587 continue; 588 } 589 } 590 if (last != NULL) { 591 struct mbuf *n; 592 593 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 594 NULL) { 595 UDP_PROBE(receive, NULL, last, ip, 596 last, uh); 597 if (udp_append(last, ip, n, iphlen, 598 udp_in)) { 599 goto inp_lost; 600 } 601 } 602 INP_RUNLOCK(last); 603 } 604 last = inp; 605 /* 606 * Don't look for additional matches if this one does 607 * not have either the SO_REUSEPORT or SO_REUSEADDR 608 * socket options set. This heuristic avoids 609 * searching through all pcbs in the common case of a 610 * non-shared port. It assumes that an application 611 * will never clear these options after setting them. 612 */ 613 if ((last->inp_socket->so_options & 614 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 615 break; 616 } 617 618 if (last == NULL) { 619 /* 620 * No matching pcb found; discard datagram. (No need 621 * to send an ICMP Port Unreachable for a broadcast 622 * or multicast datgram.) 623 */ 624 UDPSTAT_INC(udps_noportbcast); 625 if (inp) 626 INP_RUNLOCK(inp); 627 INP_INFO_RUNLOCK(pcbinfo); 628 goto badunlocked; 629 } 630 UDP_PROBE(receive, NULL, last, ip, last, uh); 631 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 632 INP_RUNLOCK(last); 633 inp_lost: 634 INP_INFO_RUNLOCK(pcbinfo); 635 return (IPPROTO_DONE); 636 } 637 638 /* 639 * Locate pcb for datagram. 640 */ 641 642 /* 643 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 644 */ 645 if ((m->m_flags & M_IP_NEXTHOP) && 646 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 647 struct sockaddr_in *next_hop; 648 649 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 650 651 /* 652 * Transparently forwarded. Pretend to be the destination. 653 * Already got one like this? 654 */ 655 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 656 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 657 if (!inp) { 658 /* 659 * It's new. Try to find the ambushing socket. 660 * Because we've rewritten the destination address, 661 * any hardware-generated hash is ignored. 662 */ 663 inp = in_pcblookup(pcbinfo, ip->ip_src, 664 uh->uh_sport, next_hop->sin_addr, 665 next_hop->sin_port ? htons(next_hop->sin_port) : 666 uh->uh_dport, INPLOOKUP_WILDCARD | 667 INPLOOKUP_RLOCKPCB, ifp); 668 } 669 /* Remove the tag from the packet. We don't need it anymore. */ 670 m_tag_delete(m, fwd_tag); 671 m->m_flags &= ~M_IP_NEXTHOP; 672 } else 673 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 674 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 675 INPLOOKUP_RLOCKPCB, ifp, m); 676 if (inp == NULL) { 677 if (udp_log_in_vain) { 678 char src[INET_ADDRSTRLEN]; 679 char dst[INET_ADDRSTRLEN]; 680 681 log(LOG_INFO, 682 "Connection attempt to UDP %s:%d from %s:%d\n", 683 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 684 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 685 } 686 UDPSTAT_INC(udps_noport); 687 if (m->m_flags & (M_BCAST | M_MCAST)) { 688 UDPSTAT_INC(udps_noportbcast); 689 goto badunlocked; 690 } 691 if (V_udp_blackhole) 692 goto badunlocked; 693 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 694 goto badunlocked; 695 *ip = save_ip; 696 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 697 return (IPPROTO_DONE); 698 } 699 700 /* 701 * Check the minimum TTL for socket. 702 */ 703 INP_RLOCK_ASSERT(inp); 704 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 705 INP_RUNLOCK(inp); 706 m_freem(m); 707 return (IPPROTO_DONE); 708 } 709 if (cscov_partial) { 710 struct udpcb *up; 711 712 up = intoudpcb(inp); 713 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 714 INP_RUNLOCK(inp); 715 m_freem(m); 716 return (IPPROTO_DONE); 717 } 718 } 719 720 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 721 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 722 INP_RUNLOCK(inp); 723 return (IPPROTO_DONE); 724 725 badunlocked: 726 m_freem(m); 727 return (IPPROTO_DONE); 728 } 729 #endif /* INET */ 730 731 /* 732 * Notify a udp user of an asynchronous error; just wake up so that they can 733 * collect error status. 734 */ 735 struct inpcb * 736 udp_notify(struct inpcb *inp, int errno) 737 { 738 739 /* 740 * While udp_ctlinput() always calls udp_notify() with a read lock 741 * when invoking it directly, in_pcbnotifyall() currently uses write 742 * locks due to sharing code with TCP. For now, accept either a read 743 * or a write lock, but a read lock is sufficient. 744 */ 745 INP_LOCK_ASSERT(inp); 746 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 747 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 748 RTFREE(inp->inp_route.ro_rt); 749 inp->inp_route.ro_rt = (struct rtentry *)NULL; 750 } 751 752 inp->inp_socket->so_error = errno; 753 sorwakeup(inp->inp_socket); 754 sowwakeup(inp->inp_socket); 755 return (inp); 756 } 757 758 #ifdef INET 759 static void 760 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 761 struct inpcbinfo *pcbinfo) 762 { 763 struct ip *ip = vip; 764 struct udphdr *uh; 765 struct in_addr faddr; 766 struct inpcb *inp; 767 768 faddr = ((struct sockaddr_in *)sa)->sin_addr; 769 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 770 return; 771 772 if (PRC_IS_REDIRECT(cmd)) { 773 /* signal EHOSTDOWN, as it flushes the cached route */ 774 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 775 return; 776 } 777 778 /* 779 * Hostdead is ugly because it goes linearly through all PCBs. 780 * 781 * XXX: We never get this from ICMP, otherwise it makes an excellent 782 * DoS attack on machines with many connections. 783 */ 784 if (cmd == PRC_HOSTDEAD) 785 ip = NULL; 786 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 787 return; 788 if (ip != NULL) { 789 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 790 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 791 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 792 if (inp != NULL) { 793 INP_RLOCK_ASSERT(inp); 794 if (inp->inp_socket != NULL) { 795 udp_notify(inp, inetctlerrmap[cmd]); 796 } 797 INP_RUNLOCK(inp); 798 } else { 799 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 800 ip->ip_src, uh->uh_sport, 801 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 802 if (inp != NULL) { 803 struct udpcb *up; 804 805 up = intoudpcb(inp); 806 if (up->u_icmp_func != NULL) { 807 INP_RUNLOCK(inp); 808 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx); 809 } else { 810 INP_RUNLOCK(inp); 811 } 812 } 813 } 814 } else 815 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 816 udp_notify); 817 } 818 void 819 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 820 { 821 822 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 823 } 824 825 void 826 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 827 { 828 829 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 830 } 831 #endif /* INET */ 832 833 static int 834 udp_pcblist(SYSCTL_HANDLER_ARGS) 835 { 836 int error, i, n; 837 struct inpcb *inp, **inp_list; 838 inp_gen_t gencnt; 839 struct xinpgen xig; 840 841 /* 842 * The process of preparing the PCB list is too time-consuming and 843 * resource-intensive to repeat twice on every request. 844 */ 845 if (req->oldptr == 0) { 846 n = V_udbinfo.ipi_count; 847 n += imax(n / 8, 10); 848 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 849 return (0); 850 } 851 852 if (req->newptr != 0) 853 return (EPERM); 854 855 /* 856 * OK, now we're committed to doing something. 857 */ 858 INP_INFO_RLOCK(&V_udbinfo); 859 gencnt = V_udbinfo.ipi_gencnt; 860 n = V_udbinfo.ipi_count; 861 INP_INFO_RUNLOCK(&V_udbinfo); 862 863 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 864 + n * sizeof(struct xinpcb)); 865 if (error != 0) 866 return (error); 867 868 xig.xig_len = sizeof xig; 869 xig.xig_count = n; 870 xig.xig_gen = gencnt; 871 xig.xig_sogen = so_gencnt; 872 error = SYSCTL_OUT(req, &xig, sizeof xig); 873 if (error) 874 return (error); 875 876 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 877 if (inp_list == NULL) 878 return (ENOMEM); 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 INP_INFO_WLOCK(&V_udbinfo); 908 for (i = 0; i < n; i++) { 909 inp = inp_list[i]; 910 INP_RLOCK(inp); 911 if (!in_pcbrele_rlocked(inp)) 912 INP_RUNLOCK(inp); 913 } 914 INP_INFO_WUNLOCK(&V_udbinfo); 915 916 if (!error) { 917 /* 918 * Give the user an updated idea of our state. If the 919 * generation differs from what we told her before, she knows 920 * that something happened while we were processing this 921 * request, and it might be necessary to retry. 922 */ 923 INP_INFO_RLOCK(&V_udbinfo); 924 xig.xig_gen = V_udbinfo.ipi_gencnt; 925 xig.xig_sogen = so_gencnt; 926 xig.xig_count = V_udbinfo.ipi_count; 927 INP_INFO_RUNLOCK(&V_udbinfo); 928 error = SYSCTL_OUT(req, &xig, sizeof xig); 929 } 930 free(inp_list, M_TEMP); 931 return (error); 932 } 933 934 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 935 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 936 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 937 938 #ifdef INET 939 static int 940 udp_getcred(SYSCTL_HANDLER_ARGS) 941 { 942 struct xucred xuc; 943 struct sockaddr_in addrs[2]; 944 struct inpcb *inp; 945 int error; 946 947 error = priv_check(req->td, PRIV_NETINET_GETCRED); 948 if (error) 949 return (error); 950 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 951 if (error) 952 return (error); 953 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 954 addrs[0].sin_addr, addrs[0].sin_port, 955 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 956 if (inp != NULL) { 957 INP_RLOCK_ASSERT(inp); 958 if (inp->inp_socket == NULL) 959 error = ENOENT; 960 if (error == 0) 961 error = cr_canseeinpcb(req->td->td_ucred, inp); 962 if (error == 0) 963 cru2x(inp->inp_cred, &xuc); 964 INP_RUNLOCK(inp); 965 } else 966 error = ENOENT; 967 if (error == 0) 968 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 969 return (error); 970 } 971 972 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 973 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 974 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 975 #endif /* INET */ 976 977 int 978 udp_ctloutput(struct socket *so, struct sockopt *sopt) 979 { 980 struct inpcb *inp; 981 struct udpcb *up; 982 int isudplite, error, optval; 983 984 error = 0; 985 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 986 inp = sotoinpcb(so); 987 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 988 INP_WLOCK(inp); 989 if (sopt->sopt_level != so->so_proto->pr_protocol) { 990 #ifdef INET6 991 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 992 INP_WUNLOCK(inp); 993 error = ip6_ctloutput(so, sopt); 994 } 995 #endif 996 #if defined(INET) && defined(INET6) 997 else 998 #endif 999 #ifdef INET 1000 { 1001 INP_WUNLOCK(inp); 1002 error = ip_ctloutput(so, sopt); 1003 } 1004 #endif 1005 return (error); 1006 } 1007 1008 switch (sopt->sopt_dir) { 1009 case SOPT_SET: 1010 switch (sopt->sopt_name) { 1011 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1012 #ifdef INET 1013 case UDP_ENCAP: 1014 if (!IPSEC_ENABLED(ipv4)) { 1015 INP_WUNLOCK(inp); 1016 return (ENOPROTOOPT); 1017 } 1018 error = UDPENCAP_PCBCTL(inp, sopt); 1019 break; 1020 #endif /* INET */ 1021 #endif /* IPSEC */ 1022 case UDPLITE_SEND_CSCOV: 1023 case UDPLITE_RECV_CSCOV: 1024 if (!isudplite) { 1025 INP_WUNLOCK(inp); 1026 error = ENOPROTOOPT; 1027 break; 1028 } 1029 INP_WUNLOCK(inp); 1030 error = sooptcopyin(sopt, &optval, sizeof(optval), 1031 sizeof(optval)); 1032 if (error != 0) 1033 break; 1034 inp = sotoinpcb(so); 1035 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1036 INP_WLOCK(inp); 1037 up = intoudpcb(inp); 1038 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1039 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1040 INP_WUNLOCK(inp); 1041 error = EINVAL; 1042 break; 1043 } 1044 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1045 up->u_txcslen = optval; 1046 else 1047 up->u_rxcslen = optval; 1048 INP_WUNLOCK(inp); 1049 break; 1050 default: 1051 INP_WUNLOCK(inp); 1052 error = ENOPROTOOPT; 1053 break; 1054 } 1055 break; 1056 case SOPT_GET: 1057 switch (sopt->sopt_name) { 1058 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1059 #ifdef INET 1060 case UDP_ENCAP: 1061 if (!IPSEC_ENABLED(ipv4)) { 1062 INP_WUNLOCK(inp); 1063 return (ENOPROTOOPT); 1064 } 1065 error = UDPENCAP_PCBCTL(inp, sopt); 1066 break; 1067 #endif /* INET */ 1068 #endif /* IPSEC */ 1069 case UDPLITE_SEND_CSCOV: 1070 case UDPLITE_RECV_CSCOV: 1071 if (!isudplite) { 1072 INP_WUNLOCK(inp); 1073 error = ENOPROTOOPT; 1074 break; 1075 } 1076 up = intoudpcb(inp); 1077 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1078 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1079 optval = up->u_txcslen; 1080 else 1081 optval = up->u_rxcslen; 1082 INP_WUNLOCK(inp); 1083 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1084 break; 1085 default: 1086 INP_WUNLOCK(inp); 1087 error = ENOPROTOOPT; 1088 break; 1089 } 1090 break; 1091 } 1092 return (error); 1093 } 1094 1095 #ifdef INET 1096 #define UH_WLOCKED 2 1097 #define UH_RLOCKED 1 1098 #define UH_UNLOCKED 0 1099 static int 1100 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1101 struct mbuf *control, struct thread *td) 1102 { 1103 struct udpiphdr *ui; 1104 int len = m->m_pkthdr.len; 1105 struct in_addr faddr, laddr; 1106 struct cmsghdr *cm; 1107 struct inpcbinfo *pcbinfo; 1108 struct sockaddr_in *sin, src; 1109 int cscov_partial = 0; 1110 int error = 0; 1111 int ipflags; 1112 u_short fport, lport; 1113 int unlock_udbinfo, unlock_inp; 1114 u_char tos; 1115 uint8_t pr; 1116 uint16_t cscov = 0; 1117 uint32_t flowid = 0; 1118 uint8_t flowtype = M_HASHTYPE_NONE; 1119 1120 /* 1121 * udp_output() may need to temporarily bind or connect the current 1122 * inpcb. As such, we don't know up front whether we will need the 1123 * pcbinfo lock or not. Do any work to decide what is needed up 1124 * front before acquiring any locks. 1125 */ 1126 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1127 if (control) 1128 m_freem(control); 1129 m_freem(m); 1130 return (EMSGSIZE); 1131 } 1132 1133 src.sin_family = 0; 1134 sin = (struct sockaddr_in *)addr; 1135 if (sin == NULL || 1136 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1137 INP_WLOCK(inp); 1138 unlock_inp = UH_WLOCKED; 1139 } else { 1140 INP_RLOCK(inp); 1141 unlock_inp = UH_RLOCKED; 1142 } 1143 tos = inp->inp_ip_tos; 1144 if (control != NULL) { 1145 /* 1146 * XXX: Currently, we assume all the optional information is 1147 * stored in a single mbuf. 1148 */ 1149 if (control->m_next) { 1150 if (unlock_inp == UH_WLOCKED) 1151 INP_WUNLOCK(inp); 1152 else 1153 INP_RUNLOCK(inp); 1154 m_freem(control); 1155 m_freem(m); 1156 return (EINVAL); 1157 } 1158 for (; control->m_len > 0; 1159 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1160 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1161 cm = mtod(control, struct cmsghdr *); 1162 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1163 || cm->cmsg_len > control->m_len) { 1164 error = EINVAL; 1165 break; 1166 } 1167 if (cm->cmsg_level != IPPROTO_IP) 1168 continue; 1169 1170 switch (cm->cmsg_type) { 1171 case IP_SENDSRCADDR: 1172 if (cm->cmsg_len != 1173 CMSG_LEN(sizeof(struct in_addr))) { 1174 error = EINVAL; 1175 break; 1176 } 1177 bzero(&src, sizeof(src)); 1178 src.sin_family = AF_INET; 1179 src.sin_len = sizeof(src); 1180 src.sin_port = inp->inp_lport; 1181 src.sin_addr = 1182 *(struct in_addr *)CMSG_DATA(cm); 1183 break; 1184 1185 case IP_TOS: 1186 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1187 error = EINVAL; 1188 break; 1189 } 1190 tos = *(u_char *)CMSG_DATA(cm); 1191 break; 1192 1193 case IP_FLOWID: 1194 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1195 error = EINVAL; 1196 break; 1197 } 1198 flowid = *(uint32_t *) CMSG_DATA(cm); 1199 break; 1200 1201 case IP_FLOWTYPE: 1202 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1203 error = EINVAL; 1204 break; 1205 } 1206 flowtype = *(uint32_t *) CMSG_DATA(cm); 1207 break; 1208 1209 #ifdef RSS 1210 case IP_RSSBUCKETID: 1211 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1212 error = EINVAL; 1213 break; 1214 } 1215 /* This is just a placeholder for now */ 1216 break; 1217 #endif /* RSS */ 1218 default: 1219 error = ENOPROTOOPT; 1220 break; 1221 } 1222 if (error) 1223 break; 1224 } 1225 m_freem(control); 1226 } 1227 if (error) { 1228 if (unlock_inp == UH_WLOCKED) 1229 INP_WUNLOCK(inp); 1230 else 1231 INP_RUNLOCK(inp); 1232 m_freem(m); 1233 return (error); 1234 } 1235 1236 /* 1237 * Depending on whether or not the application has bound or connected 1238 * the socket, we may have to do varying levels of work. The optimal 1239 * case is for a connected UDP socket, as a global lock isn't 1240 * required at all. 1241 * 1242 * In order to decide which we need, we require stability of the 1243 * inpcb binding, which we ensure by acquiring a read lock on the 1244 * inpcb. This doesn't strictly follow the lock order, so we play 1245 * the trylock and retry game; note that we may end up with more 1246 * conservative locks than required the second time around, so later 1247 * assertions have to accept that. Further analysis of the number of 1248 * misses under contention is required. 1249 * 1250 * XXXRW: Check that hash locking update here is correct. 1251 */ 1252 pr = inp->inp_socket->so_proto->pr_protocol; 1253 pcbinfo = udp_get_inpcbinfo(pr); 1254 sin = (struct sockaddr_in *)addr; 1255 if (sin != NULL && 1256 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1257 INP_HASH_WLOCK(pcbinfo); 1258 unlock_udbinfo = UH_WLOCKED; 1259 } else if ((sin != NULL && ( 1260 (sin->sin_addr.s_addr == INADDR_ANY) || 1261 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1262 (inp->inp_laddr.s_addr == INADDR_ANY) || 1263 (inp->inp_lport == 0))) || 1264 (src.sin_family == AF_INET)) { 1265 INP_HASH_RLOCK(pcbinfo); 1266 unlock_udbinfo = UH_RLOCKED; 1267 } else 1268 unlock_udbinfo = UH_UNLOCKED; 1269 1270 /* 1271 * If the IP_SENDSRCADDR control message was specified, override the 1272 * source address for this datagram. Its use is invalidated if the 1273 * address thus specified is incomplete or clobbers other inpcbs. 1274 */ 1275 laddr = inp->inp_laddr; 1276 lport = inp->inp_lport; 1277 if (src.sin_family == AF_INET) { 1278 INP_HASH_LOCK_ASSERT(pcbinfo); 1279 if ((lport == 0) || 1280 (laddr.s_addr == INADDR_ANY && 1281 src.sin_addr.s_addr == INADDR_ANY)) { 1282 error = EINVAL; 1283 goto release; 1284 } 1285 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1286 &laddr.s_addr, &lport, td->td_ucred); 1287 if (error) 1288 goto release; 1289 } 1290 1291 /* 1292 * If a UDP socket has been connected, then a local address/port will 1293 * have been selected and bound. 1294 * 1295 * If a UDP socket has not been connected to, then an explicit 1296 * destination address must be used, in which case a local 1297 * address/port may not have been selected and bound. 1298 */ 1299 if (sin != NULL) { 1300 INP_LOCK_ASSERT(inp); 1301 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1302 error = EISCONN; 1303 goto release; 1304 } 1305 1306 /* 1307 * Jail may rewrite the destination address, so let it do 1308 * that before we use it. 1309 */ 1310 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1311 if (error) 1312 goto release; 1313 1314 /* 1315 * If a local address or port hasn't yet been selected, or if 1316 * the destination address needs to be rewritten due to using 1317 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1318 * to do the heavy lifting. Once a port is selected, we 1319 * commit the binding back to the socket; we also commit the 1320 * binding of the address if in jail. 1321 * 1322 * If we already have a valid binding and we're not 1323 * requesting a destination address rewrite, use a fast path. 1324 */ 1325 if (inp->inp_laddr.s_addr == INADDR_ANY || 1326 inp->inp_lport == 0 || 1327 sin->sin_addr.s_addr == INADDR_ANY || 1328 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1329 INP_HASH_LOCK_ASSERT(pcbinfo); 1330 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1331 &lport, &faddr.s_addr, &fport, NULL, 1332 td->td_ucred); 1333 if (error) 1334 goto release; 1335 1336 /* 1337 * XXXRW: Why not commit the port if the address is 1338 * !INADDR_ANY? 1339 */ 1340 /* Commit the local port if newly assigned. */ 1341 if (inp->inp_laddr.s_addr == INADDR_ANY && 1342 inp->inp_lport == 0) { 1343 INP_WLOCK_ASSERT(inp); 1344 INP_HASH_WLOCK_ASSERT(pcbinfo); 1345 /* 1346 * Remember addr if jailed, to prevent 1347 * rebinding. 1348 */ 1349 if (prison_flag(td->td_ucred, PR_IP4)) 1350 inp->inp_laddr = laddr; 1351 inp->inp_lport = lport; 1352 if (in_pcbinshash(inp) != 0) { 1353 inp->inp_lport = 0; 1354 error = EAGAIN; 1355 goto release; 1356 } 1357 inp->inp_flags |= INP_ANONPORT; 1358 } 1359 } else { 1360 faddr = sin->sin_addr; 1361 fport = sin->sin_port; 1362 } 1363 } else { 1364 INP_LOCK_ASSERT(inp); 1365 faddr = inp->inp_faddr; 1366 fport = inp->inp_fport; 1367 if (faddr.s_addr == INADDR_ANY) { 1368 error = ENOTCONN; 1369 goto release; 1370 } 1371 } 1372 1373 /* 1374 * Calculate data length and get a mbuf for UDP, IP, and possible 1375 * link-layer headers. Immediate slide the data pointer back forward 1376 * since we won't use that space at this layer. 1377 */ 1378 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1379 if (m == NULL) { 1380 error = ENOBUFS; 1381 goto release; 1382 } 1383 m->m_data += max_linkhdr; 1384 m->m_len -= max_linkhdr; 1385 m->m_pkthdr.len -= max_linkhdr; 1386 1387 /* 1388 * Fill in mbuf with extended UDP header and addresses and length put 1389 * into network format. 1390 */ 1391 ui = mtod(m, struct udpiphdr *); 1392 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1393 ui->ui_pr = pr; 1394 ui->ui_src = laddr; 1395 ui->ui_dst = faddr; 1396 ui->ui_sport = lport; 1397 ui->ui_dport = fport; 1398 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1399 if (pr == IPPROTO_UDPLITE) { 1400 struct udpcb *up; 1401 uint16_t plen; 1402 1403 up = intoudpcb(inp); 1404 cscov = up->u_txcslen; 1405 plen = (u_short)len + sizeof(struct udphdr); 1406 if (cscov >= plen) 1407 cscov = 0; 1408 ui->ui_len = htons(plen); 1409 ui->ui_ulen = htons(cscov); 1410 /* 1411 * For UDP-Lite, checksum coverage length of zero means 1412 * the entire UDPLite packet is covered by the checksum. 1413 */ 1414 cscov_partial = (cscov == 0) ? 0 : 1; 1415 } else 1416 ui->ui_v = IPVERSION << 4; 1417 1418 /* 1419 * Set the Don't Fragment bit in the IP header. 1420 */ 1421 if (inp->inp_flags & INP_DONTFRAG) { 1422 struct ip *ip; 1423 1424 ip = (struct ip *)&ui->ui_i; 1425 ip->ip_off |= htons(IP_DF); 1426 } 1427 1428 ipflags = 0; 1429 if (inp->inp_socket->so_options & SO_DONTROUTE) 1430 ipflags |= IP_ROUTETOIF; 1431 if (inp->inp_socket->so_options & SO_BROADCAST) 1432 ipflags |= IP_ALLOWBROADCAST; 1433 if (inp->inp_flags & INP_ONESBCAST) 1434 ipflags |= IP_SENDONES; 1435 1436 #ifdef MAC 1437 mac_inpcb_create_mbuf(inp, m); 1438 #endif 1439 1440 /* 1441 * Set up checksum and output datagram. 1442 */ 1443 ui->ui_sum = 0; 1444 if (pr == IPPROTO_UDPLITE) { 1445 if (inp->inp_flags & INP_ONESBCAST) 1446 faddr.s_addr = INADDR_BROADCAST; 1447 if (cscov_partial) { 1448 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1449 ui->ui_sum = 0xffff; 1450 } else { 1451 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1452 ui->ui_sum = 0xffff; 1453 } 1454 } else if (V_udp_cksum) { 1455 if (inp->inp_flags & INP_ONESBCAST) 1456 faddr.s_addr = INADDR_BROADCAST; 1457 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1458 htons((u_short)len + sizeof(struct udphdr) + pr)); 1459 m->m_pkthdr.csum_flags = CSUM_UDP; 1460 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1461 } 1462 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1463 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1464 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1465 UDPSTAT_INC(udps_opackets); 1466 1467 /* 1468 * Setup flowid / RSS information for outbound socket. 1469 * 1470 * Once the UDP code decides to set a flowid some other way, 1471 * this allows the flowid to be overridden by userland. 1472 */ 1473 if (flowtype != M_HASHTYPE_NONE) { 1474 m->m_pkthdr.flowid = flowid; 1475 M_HASHTYPE_SET(m, flowtype); 1476 #ifdef RSS 1477 } else { 1478 uint32_t hash_val, hash_type; 1479 /* 1480 * Calculate an appropriate RSS hash for UDP and 1481 * UDP Lite. 1482 * 1483 * The called function will take care of figuring out 1484 * whether a 2-tuple or 4-tuple hash is required based 1485 * on the currently configured scheme. 1486 * 1487 * Later later on connected socket values should be 1488 * cached in the inpcb and reused, rather than constantly 1489 * re-calculating it. 1490 * 1491 * UDP Lite is a different protocol number and will 1492 * likely end up being hashed as a 2-tuple until 1493 * RSS / NICs grow UDP Lite protocol awareness. 1494 */ 1495 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1496 pr, &hash_val, &hash_type) == 0) { 1497 m->m_pkthdr.flowid = hash_val; 1498 M_HASHTYPE_SET(m, hash_type); 1499 } 1500 #endif 1501 } 1502 1503 #ifdef RSS 1504 /* 1505 * Don't override with the inp cached flowid value. 1506 * 1507 * Depending upon the kind of send being done, the inp 1508 * flowid/flowtype values may actually not be appropriate 1509 * for this particular socket send. 1510 * 1511 * We should either leave the flowid at zero (which is what is 1512 * currently done) or set it to some software generated 1513 * hash value based on the packet contents. 1514 */ 1515 ipflags |= IP_NODEFAULTFLOWID; 1516 #endif /* RSS */ 1517 1518 if (unlock_udbinfo == UH_WLOCKED) 1519 INP_HASH_WUNLOCK(pcbinfo); 1520 else if (unlock_udbinfo == UH_RLOCKED) 1521 INP_HASH_RUNLOCK(pcbinfo); 1522 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1523 error = ip_output(m, inp->inp_options, 1524 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1525 inp->inp_moptions, inp); 1526 if (unlock_inp == UH_WLOCKED) 1527 INP_WUNLOCK(inp); 1528 else 1529 INP_RUNLOCK(inp); 1530 return (error); 1531 1532 release: 1533 if (unlock_udbinfo == UH_WLOCKED) { 1534 KASSERT(unlock_inp == UH_WLOCKED, 1535 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1536 INP_HASH_WUNLOCK(pcbinfo); 1537 INP_WUNLOCK(inp); 1538 } else if (unlock_udbinfo == UH_RLOCKED) { 1539 KASSERT(unlock_inp == UH_RLOCKED, 1540 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1541 INP_HASH_RUNLOCK(pcbinfo); 1542 INP_RUNLOCK(inp); 1543 } else if (unlock_inp == UH_WLOCKED) 1544 INP_WUNLOCK(inp); 1545 else 1546 INP_RUNLOCK(inp); 1547 m_freem(m); 1548 return (error); 1549 } 1550 1551 static void 1552 udp_abort(struct socket *so) 1553 { 1554 struct inpcb *inp; 1555 struct inpcbinfo *pcbinfo; 1556 1557 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1558 inp = sotoinpcb(so); 1559 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1560 INP_WLOCK(inp); 1561 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1562 INP_HASH_WLOCK(pcbinfo); 1563 in_pcbdisconnect(inp); 1564 inp->inp_laddr.s_addr = INADDR_ANY; 1565 INP_HASH_WUNLOCK(pcbinfo); 1566 soisdisconnected(so); 1567 } 1568 INP_WUNLOCK(inp); 1569 } 1570 1571 static int 1572 udp_attach(struct socket *so, int proto, struct thread *td) 1573 { 1574 struct inpcb *inp; 1575 struct inpcbinfo *pcbinfo; 1576 int error; 1577 1578 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1579 inp = sotoinpcb(so); 1580 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1581 error = soreserve(so, udp_sendspace, udp_recvspace); 1582 if (error) 1583 return (error); 1584 INP_INFO_WLOCK(pcbinfo); 1585 error = in_pcballoc(so, pcbinfo); 1586 if (error) { 1587 INP_INFO_WUNLOCK(pcbinfo); 1588 return (error); 1589 } 1590 1591 inp = sotoinpcb(so); 1592 inp->inp_vflag |= INP_IPV4; 1593 inp->inp_ip_ttl = V_ip_defttl; 1594 1595 error = udp_newudpcb(inp); 1596 if (error) { 1597 in_pcbdetach(inp); 1598 in_pcbfree(inp); 1599 INP_INFO_WUNLOCK(pcbinfo); 1600 return (error); 1601 } 1602 1603 INP_WUNLOCK(inp); 1604 INP_INFO_WUNLOCK(pcbinfo); 1605 return (0); 1606 } 1607 #endif /* INET */ 1608 1609 int 1610 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1611 { 1612 struct inpcb *inp; 1613 struct udpcb *up; 1614 1615 KASSERT(so->so_type == SOCK_DGRAM, 1616 ("udp_set_kernel_tunneling: !dgram")); 1617 inp = sotoinpcb(so); 1618 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1619 INP_WLOCK(inp); 1620 up = intoudpcb(inp); 1621 if ((up->u_tun_func != NULL) || 1622 (up->u_icmp_func != NULL)) { 1623 INP_WUNLOCK(inp); 1624 return (EBUSY); 1625 } 1626 up->u_tun_func = f; 1627 up->u_icmp_func = i; 1628 up->u_tun_ctx = ctx; 1629 INP_WUNLOCK(inp); 1630 return (0); 1631 } 1632 1633 #ifdef INET 1634 static int 1635 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1636 { 1637 struct inpcb *inp; 1638 struct inpcbinfo *pcbinfo; 1639 int error; 1640 1641 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1642 inp = sotoinpcb(so); 1643 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1644 INP_WLOCK(inp); 1645 INP_HASH_WLOCK(pcbinfo); 1646 error = in_pcbbind(inp, nam, td->td_ucred); 1647 INP_HASH_WUNLOCK(pcbinfo); 1648 INP_WUNLOCK(inp); 1649 return (error); 1650 } 1651 1652 static void 1653 udp_close(struct socket *so) 1654 { 1655 struct inpcb *inp; 1656 struct inpcbinfo *pcbinfo; 1657 1658 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1659 inp = sotoinpcb(so); 1660 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1661 INP_WLOCK(inp); 1662 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1663 INP_HASH_WLOCK(pcbinfo); 1664 in_pcbdisconnect(inp); 1665 inp->inp_laddr.s_addr = INADDR_ANY; 1666 INP_HASH_WUNLOCK(pcbinfo); 1667 soisdisconnected(so); 1668 } 1669 INP_WUNLOCK(inp); 1670 } 1671 1672 static int 1673 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1674 { 1675 struct inpcb *inp; 1676 struct inpcbinfo *pcbinfo; 1677 struct sockaddr_in *sin; 1678 int error; 1679 1680 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1681 inp = sotoinpcb(so); 1682 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1683 INP_WLOCK(inp); 1684 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1685 INP_WUNLOCK(inp); 1686 return (EISCONN); 1687 } 1688 sin = (struct sockaddr_in *)nam; 1689 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1690 if (error != 0) { 1691 INP_WUNLOCK(inp); 1692 return (error); 1693 } 1694 INP_HASH_WLOCK(pcbinfo); 1695 error = in_pcbconnect(inp, nam, td->td_ucred); 1696 INP_HASH_WUNLOCK(pcbinfo); 1697 if (error == 0) 1698 soisconnected(so); 1699 INP_WUNLOCK(inp); 1700 return (error); 1701 } 1702 1703 static void 1704 udp_detach(struct socket *so) 1705 { 1706 struct inpcb *inp; 1707 struct inpcbinfo *pcbinfo; 1708 struct udpcb *up; 1709 1710 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1711 inp = sotoinpcb(so); 1712 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1713 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1714 ("udp_detach: not disconnected")); 1715 INP_INFO_WLOCK(pcbinfo); 1716 INP_WLOCK(inp); 1717 up = intoudpcb(inp); 1718 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1719 inp->inp_ppcb = NULL; 1720 in_pcbdetach(inp); 1721 in_pcbfree(inp); 1722 INP_INFO_WUNLOCK(pcbinfo); 1723 udp_discardcb(up); 1724 } 1725 1726 static int 1727 udp_disconnect(struct socket *so) 1728 { 1729 struct inpcb *inp; 1730 struct inpcbinfo *pcbinfo; 1731 1732 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1733 inp = sotoinpcb(so); 1734 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1735 INP_WLOCK(inp); 1736 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1737 INP_WUNLOCK(inp); 1738 return (ENOTCONN); 1739 } 1740 INP_HASH_WLOCK(pcbinfo); 1741 in_pcbdisconnect(inp); 1742 inp->inp_laddr.s_addr = INADDR_ANY; 1743 INP_HASH_WUNLOCK(pcbinfo); 1744 SOCK_LOCK(so); 1745 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1746 SOCK_UNLOCK(so); 1747 INP_WUNLOCK(inp); 1748 return (0); 1749 } 1750 1751 static int 1752 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1753 struct mbuf *control, struct thread *td) 1754 { 1755 struct inpcb *inp; 1756 1757 inp = sotoinpcb(so); 1758 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1759 return (udp_output(inp, m, addr, control, td)); 1760 } 1761 #endif /* INET */ 1762 1763 int 1764 udp_shutdown(struct socket *so) 1765 { 1766 struct inpcb *inp; 1767 1768 inp = sotoinpcb(so); 1769 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1770 INP_WLOCK(inp); 1771 socantsendmore(so); 1772 INP_WUNLOCK(inp); 1773 return (0); 1774 } 1775 1776 #ifdef INET 1777 struct pr_usrreqs udp_usrreqs = { 1778 .pru_abort = udp_abort, 1779 .pru_attach = udp_attach, 1780 .pru_bind = udp_bind, 1781 .pru_connect = udp_connect, 1782 .pru_control = in_control, 1783 .pru_detach = udp_detach, 1784 .pru_disconnect = udp_disconnect, 1785 .pru_peeraddr = in_getpeeraddr, 1786 .pru_send = udp_send, 1787 .pru_soreceive = soreceive_dgram, 1788 .pru_sosend = sosend_dgram, 1789 .pru_shutdown = udp_shutdown, 1790 .pru_sockaddr = in_getsockaddr, 1791 .pru_sosetlabel = in_pcbsosetlabel, 1792 .pru_close = udp_close, 1793 }; 1794 #endif /* INET */ 1795