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, IPI_HASHFIELDS_2TUPLE); 207 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 208 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 209 uma_zone_set_max(V_udpcb_zone, maxsockets); 210 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 211 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 212 EVENTHANDLER_PRI_ANY); 213 } 214 215 void 216 udplite_init(void) 217 { 218 219 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 220 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, 221 IPI_HASHFIELDS_2TUPLE); 222 } 223 224 /* 225 * Kernel module interface for updating udpstat. The argument is an index 226 * into udpstat treated as an array of u_long. While this encodes the 227 * general layout of udpstat into the caller, it doesn't encode its location, 228 * so that future changes to add, for example, per-CPU stats support won't 229 * cause binary compatibility problems for kernel modules. 230 */ 231 void 232 kmod_udpstat_inc(int statnum) 233 { 234 235 counter_u64_add(VNET(udpstat)[statnum], 1); 236 } 237 238 int 239 udp_newudpcb(struct inpcb *inp) 240 { 241 struct udpcb *up; 242 243 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 244 if (up == NULL) 245 return (ENOBUFS); 246 inp->inp_ppcb = up; 247 return (0); 248 } 249 250 void 251 udp_discardcb(struct udpcb *up) 252 { 253 254 uma_zfree(V_udpcb_zone, up); 255 } 256 257 #ifdef VIMAGE 258 static void 259 udp_destroy(void *unused __unused) 260 { 261 262 in_pcbinfo_destroy(&V_udbinfo); 263 uma_zdestroy(V_udpcb_zone); 264 } 265 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 266 267 static void 268 udplite_destroy(void *unused __unused) 269 { 270 271 in_pcbinfo_destroy(&V_ulitecbinfo); 272 } 273 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 274 NULL); 275 #endif 276 277 #ifdef INET 278 /* 279 * Subroutine of udp_input(), which appends the provided mbuf chain to the 280 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 281 * contains the source address. If the socket ends up being an IPv6 socket, 282 * udp_append() will convert to a sockaddr_in6 before passing the address 283 * into the socket code. 284 * 285 * In the normal case udp_append() will return 0, indicating that you 286 * must unlock the inp. However if a tunneling protocol is in place we increment 287 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 288 * then decrement the reference count. If the inp_rele returns 1, indicating the 289 * inp is gone, we return that to the caller to tell them *not* to unlock 290 * the inp. In the case of multi-cast this will cause the distribution 291 * to stop (though most tunneling protocols known currently do *not* use 292 * multicast). 293 */ 294 static int 295 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 296 struct sockaddr_in *udp_in) 297 { 298 struct sockaddr *append_sa; 299 struct socket *so; 300 struct mbuf *tmpopts, *opts = NULL; 301 #ifdef INET6 302 struct sockaddr_in6 udp_in6; 303 #endif 304 struct udpcb *up; 305 306 INP_LOCK_ASSERT(inp); 307 308 /* 309 * Engage the tunneling protocol. 310 */ 311 up = intoudpcb(inp); 312 if (up->u_tun_func != NULL) { 313 in_pcbref(inp); 314 INP_RUNLOCK(inp); 315 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], 316 up->u_tun_ctx); 317 INP_RLOCK(inp); 318 return (in_pcbrele_rlocked(inp)); 319 } 320 321 off += sizeof(struct udphdr); 322 323 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 324 /* Check AH/ESP integrity. */ 325 if (IPSEC_ENABLED(ipv4) && 326 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 327 m_freem(n); 328 return (0); 329 } 330 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 331 if (IPSEC_ENABLED(ipv4) && 332 UDPENCAP_INPUT(n, off, AF_INET) != 0) 333 return (0); /* Consumed. */ 334 } 335 #endif /* IPSEC */ 336 #ifdef MAC 337 if (mac_inpcb_check_deliver(inp, n) != 0) { 338 m_freem(n); 339 return (0); 340 } 341 #endif /* MAC */ 342 if (inp->inp_flags & INP_CONTROLOPTS || 343 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 344 #ifdef INET6 345 if (inp->inp_vflag & INP_IPV6) 346 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 347 else 348 #endif /* INET6 */ 349 ip_savecontrol(inp, &opts, ip, n); 350 } 351 if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { 352 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], 353 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); 354 if (tmpopts) { 355 if (opts) { 356 tmpopts->m_next = opts; 357 opts = tmpopts; 358 } else 359 opts = tmpopts; 360 } 361 } 362 #ifdef INET6 363 if (inp->inp_vflag & INP_IPV6) { 364 bzero(&udp_in6, sizeof(udp_in6)); 365 udp_in6.sin6_len = sizeof(udp_in6); 366 udp_in6.sin6_family = AF_INET6; 367 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 368 append_sa = (struct sockaddr *)&udp_in6; 369 } else 370 #endif /* INET6 */ 371 append_sa = (struct sockaddr *)&udp_in[0]; 372 m_adj(n, off); 373 374 so = inp->inp_socket; 375 SOCKBUF_LOCK(&so->so_rcv); 376 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 377 SOCKBUF_UNLOCK(&so->so_rcv); 378 m_freem(n); 379 if (opts) 380 m_freem(opts); 381 UDPSTAT_INC(udps_fullsock); 382 } else 383 sorwakeup_locked(so); 384 return (0); 385 } 386 387 int 388 udp_input(struct mbuf **mp, int *offp, int proto) 389 { 390 struct ip *ip; 391 struct udphdr *uh; 392 struct ifnet *ifp; 393 struct inpcb *inp; 394 uint16_t len, ip_len; 395 struct inpcbinfo *pcbinfo; 396 struct ip save_ip; 397 struct sockaddr_in udp_in[2]; 398 struct mbuf *m; 399 struct m_tag *fwd_tag; 400 int cscov_partial, iphlen; 401 402 m = *mp; 403 iphlen = *offp; 404 ifp = m->m_pkthdr.rcvif; 405 *mp = NULL; 406 UDPSTAT_INC(udps_ipackets); 407 408 /* 409 * Strip IP options, if any; should skip this, make available to 410 * user, and use on returned packets, but we don't yet have a way to 411 * check the checksum with options still present. 412 */ 413 if (iphlen > sizeof (struct ip)) { 414 ip_stripoptions(m); 415 iphlen = sizeof(struct ip); 416 } 417 418 /* 419 * Get IP and UDP header together in first mbuf. 420 */ 421 ip = mtod(m, struct ip *); 422 if (m->m_len < iphlen + sizeof(struct udphdr)) { 423 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 424 UDPSTAT_INC(udps_hdrops); 425 return (IPPROTO_DONE); 426 } 427 ip = mtod(m, struct ip *); 428 } 429 uh = (struct udphdr *)((caddr_t)ip + iphlen); 430 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 431 432 /* 433 * Destination port of 0 is illegal, based on RFC768. 434 */ 435 if (uh->uh_dport == 0) 436 goto badunlocked; 437 438 /* 439 * Construct sockaddr format source address. Stuff source address 440 * and datagram in user buffer. 441 */ 442 bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); 443 udp_in[0].sin_len = sizeof(struct sockaddr_in); 444 udp_in[0].sin_family = AF_INET; 445 udp_in[0].sin_port = uh->uh_sport; 446 udp_in[0].sin_addr = ip->ip_src; 447 udp_in[1].sin_len = sizeof(struct sockaddr_in); 448 udp_in[1].sin_family = AF_INET; 449 udp_in[1].sin_port = uh->uh_dport; 450 udp_in[1].sin_addr = ip->ip_dst; 451 452 /* 453 * Make mbuf data length reflect UDP length. If not enough data to 454 * reflect UDP length, drop. 455 */ 456 len = ntohs((u_short)uh->uh_ulen); 457 ip_len = ntohs(ip->ip_len) - iphlen; 458 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 459 /* Zero means checksum over the complete packet. */ 460 if (len == 0) 461 len = ip_len; 462 cscov_partial = 0; 463 } 464 if (ip_len != len) { 465 if (len > ip_len || len < sizeof(struct udphdr)) { 466 UDPSTAT_INC(udps_badlen); 467 goto badunlocked; 468 } 469 if (proto == IPPROTO_UDP) 470 m_adj(m, len - ip_len); 471 } 472 473 /* 474 * Save a copy of the IP header in case we want restore it for 475 * sending an ICMP error message in response. 476 */ 477 if (!V_udp_blackhole) 478 save_ip = *ip; 479 else 480 memset(&save_ip, 0, sizeof(save_ip)); 481 482 /* 483 * Checksum extended UDP header and data. 484 */ 485 if (uh->uh_sum) { 486 u_short uh_sum; 487 488 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 489 !cscov_partial) { 490 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 491 uh_sum = m->m_pkthdr.csum_data; 492 else 493 uh_sum = in_pseudo(ip->ip_src.s_addr, 494 ip->ip_dst.s_addr, htonl((u_short)len + 495 m->m_pkthdr.csum_data + proto)); 496 uh_sum ^= 0xffff; 497 } else { 498 char b[9]; 499 500 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 501 bzero(((struct ipovly *)ip)->ih_x1, 9); 502 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 503 uh->uh_ulen : htons(ip_len); 504 uh_sum = in_cksum(m, len + sizeof (struct ip)); 505 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 506 } 507 if (uh_sum) { 508 UDPSTAT_INC(udps_badsum); 509 m_freem(m); 510 return (IPPROTO_DONE); 511 } 512 } else { 513 if (proto == IPPROTO_UDP) { 514 UDPSTAT_INC(udps_nosum); 515 } else { 516 /* UDPLite requires a checksum */ 517 /* XXX: What is the right UDPLite MIB counter here? */ 518 m_freem(m); 519 return (IPPROTO_DONE); 520 } 521 } 522 523 pcbinfo = udp_get_inpcbinfo(proto); 524 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 525 in_broadcast(ip->ip_dst, ifp)) { 526 struct inpcb *last; 527 struct inpcbhead *pcblist; 528 struct ip_moptions *imo; 529 530 INP_INFO_RLOCK(pcbinfo); 531 pcblist = udp_get_pcblist(proto); 532 last = NULL; 533 LIST_FOREACH(inp, pcblist, inp_list) { 534 if (inp->inp_lport != uh->uh_dport) 535 continue; 536 #ifdef INET6 537 if ((inp->inp_vflag & INP_IPV4) == 0) 538 continue; 539 #endif 540 if (inp->inp_laddr.s_addr != INADDR_ANY && 541 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 542 continue; 543 if (inp->inp_faddr.s_addr != INADDR_ANY && 544 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 545 continue; 546 if (inp->inp_fport != 0 && 547 inp->inp_fport != uh->uh_sport) 548 continue; 549 550 INP_RLOCK(inp); 551 552 /* 553 * XXXRW: Because we weren't holding either the inpcb 554 * or the hash lock when we checked for a match 555 * before, we should probably recheck now that the 556 * inpcb lock is held. 557 */ 558 559 /* 560 * Handle socket delivery policy for any-source 561 * and source-specific multicast. [RFC3678] 562 */ 563 imo = inp->inp_moptions; 564 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 565 struct sockaddr_in group; 566 int blocked; 567 if (imo == NULL) { 568 INP_RUNLOCK(inp); 569 continue; 570 } 571 bzero(&group, sizeof(struct sockaddr_in)); 572 group.sin_len = sizeof(struct sockaddr_in); 573 group.sin_family = AF_INET; 574 group.sin_addr = ip->ip_dst; 575 576 blocked = imo_multi_filter(imo, ifp, 577 (struct sockaddr *)&group, 578 (struct sockaddr *)&udp_in[0]); 579 if (blocked != MCAST_PASS) { 580 if (blocked == MCAST_NOTGMEMBER) 581 IPSTAT_INC(ips_notmember); 582 if (blocked == MCAST_NOTSMEMBER || 583 blocked == MCAST_MUTED) 584 UDPSTAT_INC(udps_filtermcast); 585 INP_RUNLOCK(inp); 586 continue; 587 } 588 } 589 if (last != NULL) { 590 struct mbuf *n; 591 592 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 593 NULL) { 594 UDP_PROBE(receive, NULL, last, ip, 595 last, uh); 596 if (udp_append(last, ip, n, iphlen, 597 udp_in)) { 598 goto inp_lost; 599 } 600 } 601 INP_RUNLOCK(last); 602 } 603 last = inp; 604 /* 605 * Don't look for additional matches if this one does 606 * not have either the SO_REUSEPORT or SO_REUSEADDR 607 * socket options set. This heuristic avoids 608 * searching through all pcbs in the common case of a 609 * non-shared port. It assumes that an application 610 * will never clear these options after setting them. 611 */ 612 if ((last->inp_socket->so_options & 613 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 614 break; 615 } 616 617 if (last == NULL) { 618 /* 619 * No matching pcb found; discard datagram. (No need 620 * to send an ICMP Port Unreachable for a broadcast 621 * or multicast datgram.) 622 */ 623 UDPSTAT_INC(udps_noportbcast); 624 if (inp) 625 INP_RUNLOCK(inp); 626 INP_INFO_RUNLOCK(pcbinfo); 627 goto badunlocked; 628 } 629 UDP_PROBE(receive, NULL, last, ip, last, uh); 630 if (udp_append(last, ip, m, iphlen, udp_in) == 0) 631 INP_RUNLOCK(last); 632 inp_lost: 633 INP_INFO_RUNLOCK(pcbinfo); 634 return (IPPROTO_DONE); 635 } 636 637 /* 638 * Locate pcb for datagram. 639 */ 640 641 /* 642 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 643 */ 644 if ((m->m_flags & M_IP_NEXTHOP) && 645 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 646 struct sockaddr_in *next_hop; 647 648 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 649 650 /* 651 * Transparently forwarded. Pretend to be the destination. 652 * Already got one like this? 653 */ 654 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 655 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 656 if (!inp) { 657 /* 658 * It's new. Try to find the ambushing socket. 659 * Because we've rewritten the destination address, 660 * any hardware-generated hash is ignored. 661 */ 662 inp = in_pcblookup(pcbinfo, ip->ip_src, 663 uh->uh_sport, next_hop->sin_addr, 664 next_hop->sin_port ? htons(next_hop->sin_port) : 665 uh->uh_dport, INPLOOKUP_WILDCARD | 666 INPLOOKUP_RLOCKPCB, ifp); 667 } 668 /* Remove the tag from the packet. We don't need it anymore. */ 669 m_tag_delete(m, fwd_tag); 670 m->m_flags &= ~M_IP_NEXTHOP; 671 } else 672 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 673 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 674 INPLOOKUP_RLOCKPCB, ifp, m); 675 if (inp == NULL) { 676 if (udp_log_in_vain) { 677 char src[INET_ADDRSTRLEN]; 678 char dst[INET_ADDRSTRLEN]; 679 680 log(LOG_INFO, 681 "Connection attempt to UDP %s:%d from %s:%d\n", 682 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 683 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 684 } 685 UDPSTAT_INC(udps_noport); 686 if (m->m_flags & (M_BCAST | M_MCAST)) { 687 UDPSTAT_INC(udps_noportbcast); 688 goto badunlocked; 689 } 690 if (V_udp_blackhole) 691 goto badunlocked; 692 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 693 goto badunlocked; 694 *ip = save_ip; 695 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 696 return (IPPROTO_DONE); 697 } 698 699 /* 700 * Check the minimum TTL for socket. 701 */ 702 INP_RLOCK_ASSERT(inp); 703 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 704 INP_RUNLOCK(inp); 705 m_freem(m); 706 return (IPPROTO_DONE); 707 } 708 if (cscov_partial) { 709 struct udpcb *up; 710 711 up = intoudpcb(inp); 712 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 713 INP_RUNLOCK(inp); 714 m_freem(m); 715 return (IPPROTO_DONE); 716 } 717 } 718 719 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 720 if (udp_append(inp, ip, m, iphlen, udp_in) == 0) 721 INP_RUNLOCK(inp); 722 return (IPPROTO_DONE); 723 724 badunlocked: 725 m_freem(m); 726 return (IPPROTO_DONE); 727 } 728 #endif /* INET */ 729 730 /* 731 * Notify a udp user of an asynchronous error; just wake up so that they can 732 * collect error status. 733 */ 734 struct inpcb * 735 udp_notify(struct inpcb *inp, int errno) 736 { 737 738 /* 739 * While udp_ctlinput() always calls udp_notify() with a read lock 740 * when invoking it directly, in_pcbnotifyall() currently uses write 741 * locks due to sharing code with TCP. For now, accept either a read 742 * or a write lock, but a read lock is sufficient. 743 */ 744 INP_LOCK_ASSERT(inp); 745 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 746 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 747 RTFREE(inp->inp_route.ro_rt); 748 inp->inp_route.ro_rt = (struct rtentry *)NULL; 749 } 750 751 inp->inp_socket->so_error = errno; 752 sorwakeup(inp->inp_socket); 753 sowwakeup(inp->inp_socket); 754 return (inp); 755 } 756 757 #ifdef INET 758 static void 759 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 760 struct inpcbinfo *pcbinfo) 761 { 762 struct ip *ip = vip; 763 struct udphdr *uh; 764 struct in_addr faddr; 765 struct inpcb *inp; 766 767 faddr = ((struct sockaddr_in *)sa)->sin_addr; 768 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 769 return; 770 771 if (PRC_IS_REDIRECT(cmd)) { 772 /* signal EHOSTDOWN, as it flushes the cached route */ 773 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 774 return; 775 } 776 777 /* 778 * Hostdead is ugly because it goes linearly through all PCBs. 779 * 780 * XXX: We never get this from ICMP, otherwise it makes an excellent 781 * DoS attack on machines with many connections. 782 */ 783 if (cmd == PRC_HOSTDEAD) 784 ip = NULL; 785 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 786 return; 787 if (ip != NULL) { 788 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 789 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 790 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 791 if (inp != NULL) { 792 INP_RLOCK_ASSERT(inp); 793 if (inp->inp_socket != NULL) { 794 udp_notify(inp, inetctlerrmap[cmd]); 795 } 796 INP_RUNLOCK(inp); 797 } else { 798 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 799 ip->ip_src, uh->uh_sport, 800 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 801 if (inp != NULL) { 802 struct udpcb *up; 803 804 up = intoudpcb(inp); 805 if (up->u_icmp_func != NULL) { 806 INP_RUNLOCK(inp); 807 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx); 808 } else { 809 INP_RUNLOCK(inp); 810 } 811 } 812 } 813 } else 814 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 815 udp_notify); 816 } 817 void 818 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 819 { 820 821 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 822 } 823 824 void 825 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 826 { 827 828 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 829 } 830 #endif /* INET */ 831 832 static int 833 udp_pcblist(SYSCTL_HANDLER_ARGS) 834 { 835 int error, i, n; 836 struct inpcb *inp, **inp_list; 837 inp_gen_t gencnt; 838 struct xinpgen xig; 839 840 /* 841 * The process of preparing the PCB list is too time-consuming and 842 * resource-intensive to repeat twice on every request. 843 */ 844 if (req->oldptr == 0) { 845 n = V_udbinfo.ipi_count; 846 n += imax(n / 8, 10); 847 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 848 return (0); 849 } 850 851 if (req->newptr != 0) 852 return (EPERM); 853 854 /* 855 * OK, now we're committed to doing something. 856 */ 857 INP_INFO_RLOCK(&V_udbinfo); 858 gencnt = V_udbinfo.ipi_gencnt; 859 n = V_udbinfo.ipi_count; 860 INP_INFO_RUNLOCK(&V_udbinfo); 861 862 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 863 + n * sizeof(struct xinpcb)); 864 if (error != 0) 865 return (error); 866 867 xig.xig_len = sizeof xig; 868 xig.xig_count = n; 869 xig.xig_gen = gencnt; 870 xig.xig_sogen = so_gencnt; 871 error = SYSCTL_OUT(req, &xig, sizeof xig); 872 if (error) 873 return (error); 874 875 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 876 if (inp_list == NULL) 877 return (ENOMEM); 878 879 INP_INFO_RLOCK(&V_udbinfo); 880 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 881 inp = LIST_NEXT(inp, inp_list)) { 882 INP_WLOCK(inp); 883 if (inp->inp_gencnt <= gencnt && 884 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 885 in_pcbref(inp); 886 inp_list[i++] = inp; 887 } 888 INP_WUNLOCK(inp); 889 } 890 INP_INFO_RUNLOCK(&V_udbinfo); 891 n = i; 892 893 error = 0; 894 for (i = 0; i < n; i++) { 895 inp = inp_list[i]; 896 INP_RLOCK(inp); 897 if (inp->inp_gencnt <= gencnt) { 898 struct xinpcb xi; 899 900 in_pcbtoxinpcb(inp, &xi); 901 INP_RUNLOCK(inp); 902 error = SYSCTL_OUT(req, &xi, sizeof xi); 903 } else 904 INP_RUNLOCK(inp); 905 } 906 INP_INFO_WLOCK(&V_udbinfo); 907 for (i = 0; i < n; i++) { 908 inp = inp_list[i]; 909 INP_RLOCK(inp); 910 if (!in_pcbrele_rlocked(inp)) 911 INP_RUNLOCK(inp); 912 } 913 INP_INFO_WUNLOCK(&V_udbinfo); 914 915 if (!error) { 916 /* 917 * Give the user an updated idea of our state. If the 918 * generation differs from what we told her before, she knows 919 * that something happened while we were processing this 920 * request, and it might be necessary to retry. 921 */ 922 INP_INFO_RLOCK(&V_udbinfo); 923 xig.xig_gen = V_udbinfo.ipi_gencnt; 924 xig.xig_sogen = so_gencnt; 925 xig.xig_count = V_udbinfo.ipi_count; 926 INP_INFO_RUNLOCK(&V_udbinfo); 927 error = SYSCTL_OUT(req, &xig, sizeof xig); 928 } 929 free(inp_list, M_TEMP); 930 return (error); 931 } 932 933 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 934 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 935 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 936 937 #ifdef INET 938 static int 939 udp_getcred(SYSCTL_HANDLER_ARGS) 940 { 941 struct xucred xuc; 942 struct sockaddr_in addrs[2]; 943 struct inpcb *inp; 944 int error; 945 946 error = priv_check(req->td, PRIV_NETINET_GETCRED); 947 if (error) 948 return (error); 949 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 950 if (error) 951 return (error); 952 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 953 addrs[0].sin_addr, addrs[0].sin_port, 954 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 955 if (inp != NULL) { 956 INP_RLOCK_ASSERT(inp); 957 if (inp->inp_socket == NULL) 958 error = ENOENT; 959 if (error == 0) 960 error = cr_canseeinpcb(req->td->td_ucred, inp); 961 if (error == 0) 962 cru2x(inp->inp_cred, &xuc); 963 INP_RUNLOCK(inp); 964 } else 965 error = ENOENT; 966 if (error == 0) 967 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 968 return (error); 969 } 970 971 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 972 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 973 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 974 #endif /* INET */ 975 976 int 977 udp_ctloutput(struct socket *so, struct sockopt *sopt) 978 { 979 struct inpcb *inp; 980 struct udpcb *up; 981 int isudplite, error, optval; 982 983 error = 0; 984 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 985 inp = sotoinpcb(so); 986 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 987 INP_WLOCK(inp); 988 if (sopt->sopt_level != so->so_proto->pr_protocol) { 989 #ifdef INET6 990 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 991 INP_WUNLOCK(inp); 992 error = ip6_ctloutput(so, sopt); 993 } 994 #endif 995 #if defined(INET) && defined(INET6) 996 else 997 #endif 998 #ifdef INET 999 { 1000 INP_WUNLOCK(inp); 1001 error = ip_ctloutput(so, sopt); 1002 } 1003 #endif 1004 return (error); 1005 } 1006 1007 switch (sopt->sopt_dir) { 1008 case SOPT_SET: 1009 switch (sopt->sopt_name) { 1010 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1011 #ifdef INET 1012 case UDP_ENCAP: 1013 if (!IPSEC_ENABLED(ipv4)) { 1014 INP_WUNLOCK(inp); 1015 return (ENOPROTOOPT); 1016 } 1017 error = UDPENCAP_PCBCTL(inp, sopt); 1018 break; 1019 #endif /* INET */ 1020 #endif /* IPSEC */ 1021 case UDPLITE_SEND_CSCOV: 1022 case UDPLITE_RECV_CSCOV: 1023 if (!isudplite) { 1024 INP_WUNLOCK(inp); 1025 error = ENOPROTOOPT; 1026 break; 1027 } 1028 INP_WUNLOCK(inp); 1029 error = sooptcopyin(sopt, &optval, sizeof(optval), 1030 sizeof(optval)); 1031 if (error != 0) 1032 break; 1033 inp = sotoinpcb(so); 1034 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1035 INP_WLOCK(inp); 1036 up = intoudpcb(inp); 1037 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1038 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1039 INP_WUNLOCK(inp); 1040 error = EINVAL; 1041 break; 1042 } 1043 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1044 up->u_txcslen = optval; 1045 else 1046 up->u_rxcslen = optval; 1047 INP_WUNLOCK(inp); 1048 break; 1049 default: 1050 INP_WUNLOCK(inp); 1051 error = ENOPROTOOPT; 1052 break; 1053 } 1054 break; 1055 case SOPT_GET: 1056 switch (sopt->sopt_name) { 1057 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1058 #ifdef INET 1059 case UDP_ENCAP: 1060 if (!IPSEC_ENABLED(ipv4)) { 1061 INP_WUNLOCK(inp); 1062 return (ENOPROTOOPT); 1063 } 1064 error = UDPENCAP_PCBCTL(inp, sopt); 1065 break; 1066 #endif /* INET */ 1067 #endif /* IPSEC */ 1068 case UDPLITE_SEND_CSCOV: 1069 case UDPLITE_RECV_CSCOV: 1070 if (!isudplite) { 1071 INP_WUNLOCK(inp); 1072 error = ENOPROTOOPT; 1073 break; 1074 } 1075 up = intoudpcb(inp); 1076 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1077 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1078 optval = up->u_txcslen; 1079 else 1080 optval = up->u_rxcslen; 1081 INP_WUNLOCK(inp); 1082 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1083 break; 1084 default: 1085 INP_WUNLOCK(inp); 1086 error = ENOPROTOOPT; 1087 break; 1088 } 1089 break; 1090 } 1091 return (error); 1092 } 1093 1094 #ifdef INET 1095 #define UH_WLOCKED 2 1096 #define UH_RLOCKED 1 1097 #define UH_UNLOCKED 0 1098 static int 1099 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1100 struct mbuf *control, struct thread *td) 1101 { 1102 struct udpiphdr *ui; 1103 int len = m->m_pkthdr.len; 1104 struct in_addr faddr, laddr; 1105 struct cmsghdr *cm; 1106 struct inpcbinfo *pcbinfo; 1107 struct sockaddr_in *sin, src; 1108 int cscov_partial = 0; 1109 int error = 0; 1110 int ipflags; 1111 u_short fport, lport; 1112 int unlock_udbinfo, unlock_inp; 1113 u_char tos; 1114 uint8_t pr; 1115 uint16_t cscov = 0; 1116 uint32_t flowid = 0; 1117 uint8_t flowtype = M_HASHTYPE_NONE; 1118 1119 /* 1120 * udp_output() may need to temporarily bind or connect the current 1121 * inpcb. As such, we don't know up front whether we will need the 1122 * pcbinfo lock or not. Do any work to decide what is needed up 1123 * front before acquiring any locks. 1124 */ 1125 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1126 if (control) 1127 m_freem(control); 1128 m_freem(m); 1129 return (EMSGSIZE); 1130 } 1131 1132 src.sin_family = 0; 1133 sin = (struct sockaddr_in *)addr; 1134 if (sin == NULL || 1135 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1136 INP_WLOCK(inp); 1137 unlock_inp = UH_WLOCKED; 1138 } else { 1139 INP_RLOCK(inp); 1140 unlock_inp = UH_RLOCKED; 1141 } 1142 tos = inp->inp_ip_tos; 1143 if (control != NULL) { 1144 /* 1145 * XXX: Currently, we assume all the optional information is 1146 * stored in a single mbuf. 1147 */ 1148 if (control->m_next) { 1149 if (unlock_inp == UH_WLOCKED) 1150 INP_WUNLOCK(inp); 1151 else 1152 INP_RUNLOCK(inp); 1153 m_freem(control); 1154 m_freem(m); 1155 return (EINVAL); 1156 } 1157 for (; control->m_len > 0; 1158 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1159 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1160 cm = mtod(control, struct cmsghdr *); 1161 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1162 || cm->cmsg_len > control->m_len) { 1163 error = EINVAL; 1164 break; 1165 } 1166 if (cm->cmsg_level != IPPROTO_IP) 1167 continue; 1168 1169 switch (cm->cmsg_type) { 1170 case IP_SENDSRCADDR: 1171 if (cm->cmsg_len != 1172 CMSG_LEN(sizeof(struct in_addr))) { 1173 error = EINVAL; 1174 break; 1175 } 1176 bzero(&src, sizeof(src)); 1177 src.sin_family = AF_INET; 1178 src.sin_len = sizeof(src); 1179 src.sin_port = inp->inp_lport; 1180 src.sin_addr = 1181 *(struct in_addr *)CMSG_DATA(cm); 1182 break; 1183 1184 case IP_TOS: 1185 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1186 error = EINVAL; 1187 break; 1188 } 1189 tos = *(u_char *)CMSG_DATA(cm); 1190 break; 1191 1192 case IP_FLOWID: 1193 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1194 error = EINVAL; 1195 break; 1196 } 1197 flowid = *(uint32_t *) CMSG_DATA(cm); 1198 break; 1199 1200 case IP_FLOWTYPE: 1201 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1202 error = EINVAL; 1203 break; 1204 } 1205 flowtype = *(uint32_t *) CMSG_DATA(cm); 1206 break; 1207 1208 #ifdef RSS 1209 case IP_RSSBUCKETID: 1210 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1211 error = EINVAL; 1212 break; 1213 } 1214 /* This is just a placeholder for now */ 1215 break; 1216 #endif /* RSS */ 1217 default: 1218 error = ENOPROTOOPT; 1219 break; 1220 } 1221 if (error) 1222 break; 1223 } 1224 m_freem(control); 1225 } 1226 if (error) { 1227 if (unlock_inp == UH_WLOCKED) 1228 INP_WUNLOCK(inp); 1229 else 1230 INP_RUNLOCK(inp); 1231 m_freem(m); 1232 return (error); 1233 } 1234 1235 /* 1236 * Depending on whether or not the application has bound or connected 1237 * the socket, we may have to do varying levels of work. The optimal 1238 * case is for a connected UDP socket, as a global lock isn't 1239 * required at all. 1240 * 1241 * In order to decide which we need, we require stability of the 1242 * inpcb binding, which we ensure by acquiring a read lock on the 1243 * inpcb. This doesn't strictly follow the lock order, so we play 1244 * the trylock and retry game; note that we may end up with more 1245 * conservative locks than required the second time around, so later 1246 * assertions have to accept that. Further analysis of the number of 1247 * misses under contention is required. 1248 * 1249 * XXXRW: Check that hash locking update here is correct. 1250 */ 1251 pr = inp->inp_socket->so_proto->pr_protocol; 1252 pcbinfo = udp_get_inpcbinfo(pr); 1253 sin = (struct sockaddr_in *)addr; 1254 if (sin != NULL && 1255 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1256 INP_HASH_WLOCK(pcbinfo); 1257 unlock_udbinfo = UH_WLOCKED; 1258 } else if ((sin != NULL && ( 1259 (sin->sin_addr.s_addr == INADDR_ANY) || 1260 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1261 (inp->inp_laddr.s_addr == INADDR_ANY) || 1262 (inp->inp_lport == 0))) || 1263 (src.sin_family == AF_INET)) { 1264 INP_HASH_RLOCK(pcbinfo); 1265 unlock_udbinfo = UH_RLOCKED; 1266 } else 1267 unlock_udbinfo = UH_UNLOCKED; 1268 1269 /* 1270 * If the IP_SENDSRCADDR control message was specified, override the 1271 * source address for this datagram. Its use is invalidated if the 1272 * address thus specified is incomplete or clobbers other inpcbs. 1273 */ 1274 laddr = inp->inp_laddr; 1275 lport = inp->inp_lport; 1276 if (src.sin_family == AF_INET) { 1277 INP_HASH_LOCK_ASSERT(pcbinfo); 1278 if ((lport == 0) || 1279 (laddr.s_addr == INADDR_ANY && 1280 src.sin_addr.s_addr == INADDR_ANY)) { 1281 error = EINVAL; 1282 goto release; 1283 } 1284 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1285 &laddr.s_addr, &lport, td->td_ucred); 1286 if (error) 1287 goto release; 1288 } 1289 1290 /* 1291 * If a UDP socket has been connected, then a local address/port will 1292 * have been selected and bound. 1293 * 1294 * If a UDP socket has not been connected to, then an explicit 1295 * destination address must be used, in which case a local 1296 * address/port may not have been selected and bound. 1297 */ 1298 if (sin != NULL) { 1299 INP_LOCK_ASSERT(inp); 1300 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1301 error = EISCONN; 1302 goto release; 1303 } 1304 1305 /* 1306 * Jail may rewrite the destination address, so let it do 1307 * that before we use it. 1308 */ 1309 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1310 if (error) 1311 goto release; 1312 1313 /* 1314 * If a local address or port hasn't yet been selected, or if 1315 * the destination address needs to be rewritten due to using 1316 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1317 * to do the heavy lifting. Once a port is selected, we 1318 * commit the binding back to the socket; we also commit the 1319 * binding of the address if in jail. 1320 * 1321 * If we already have a valid binding and we're not 1322 * requesting a destination address rewrite, use a fast path. 1323 */ 1324 if (inp->inp_laddr.s_addr == INADDR_ANY || 1325 inp->inp_lport == 0 || 1326 sin->sin_addr.s_addr == INADDR_ANY || 1327 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1328 INP_HASH_LOCK_ASSERT(pcbinfo); 1329 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1330 &lport, &faddr.s_addr, &fport, NULL, 1331 td->td_ucred); 1332 if (error) 1333 goto release; 1334 1335 /* 1336 * XXXRW: Why not commit the port if the address is 1337 * !INADDR_ANY? 1338 */ 1339 /* Commit the local port if newly assigned. */ 1340 if (inp->inp_laddr.s_addr == INADDR_ANY && 1341 inp->inp_lport == 0) { 1342 INP_WLOCK_ASSERT(inp); 1343 INP_HASH_WLOCK_ASSERT(pcbinfo); 1344 /* 1345 * Remember addr if jailed, to prevent 1346 * rebinding. 1347 */ 1348 if (prison_flag(td->td_ucred, PR_IP4)) 1349 inp->inp_laddr = laddr; 1350 inp->inp_lport = lport; 1351 if (in_pcbinshash(inp) != 0) { 1352 inp->inp_lport = 0; 1353 error = EAGAIN; 1354 goto release; 1355 } 1356 inp->inp_flags |= INP_ANONPORT; 1357 } 1358 } else { 1359 faddr = sin->sin_addr; 1360 fport = sin->sin_port; 1361 } 1362 } else { 1363 INP_LOCK_ASSERT(inp); 1364 faddr = inp->inp_faddr; 1365 fport = inp->inp_fport; 1366 if (faddr.s_addr == INADDR_ANY) { 1367 error = ENOTCONN; 1368 goto release; 1369 } 1370 } 1371 1372 /* 1373 * Calculate data length and get a mbuf for UDP, IP, and possible 1374 * link-layer headers. Immediate slide the data pointer back forward 1375 * since we won't use that space at this layer. 1376 */ 1377 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1378 if (m == NULL) { 1379 error = ENOBUFS; 1380 goto release; 1381 } 1382 m->m_data += max_linkhdr; 1383 m->m_len -= max_linkhdr; 1384 m->m_pkthdr.len -= max_linkhdr; 1385 1386 /* 1387 * Fill in mbuf with extended UDP header and addresses and length put 1388 * into network format. 1389 */ 1390 ui = mtod(m, struct udpiphdr *); 1391 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1392 ui->ui_pr = pr; 1393 ui->ui_src = laddr; 1394 ui->ui_dst = faddr; 1395 ui->ui_sport = lport; 1396 ui->ui_dport = fport; 1397 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1398 if (pr == IPPROTO_UDPLITE) { 1399 struct udpcb *up; 1400 uint16_t plen; 1401 1402 up = intoudpcb(inp); 1403 cscov = up->u_txcslen; 1404 plen = (u_short)len + sizeof(struct udphdr); 1405 if (cscov >= plen) 1406 cscov = 0; 1407 ui->ui_len = htons(plen); 1408 ui->ui_ulen = htons(cscov); 1409 /* 1410 * For UDP-Lite, checksum coverage length of zero means 1411 * the entire UDPLite packet is covered by the checksum. 1412 */ 1413 cscov_partial = (cscov == 0) ? 0 : 1; 1414 } else 1415 ui->ui_v = IPVERSION << 4; 1416 1417 /* 1418 * Set the Don't Fragment bit in the IP header. 1419 */ 1420 if (inp->inp_flags & INP_DONTFRAG) { 1421 struct ip *ip; 1422 1423 ip = (struct ip *)&ui->ui_i; 1424 ip->ip_off |= htons(IP_DF); 1425 } 1426 1427 ipflags = 0; 1428 if (inp->inp_socket->so_options & SO_DONTROUTE) 1429 ipflags |= IP_ROUTETOIF; 1430 if (inp->inp_socket->so_options & SO_BROADCAST) 1431 ipflags |= IP_ALLOWBROADCAST; 1432 if (inp->inp_flags & INP_ONESBCAST) 1433 ipflags |= IP_SENDONES; 1434 1435 #ifdef MAC 1436 mac_inpcb_create_mbuf(inp, m); 1437 #endif 1438 1439 /* 1440 * Set up checksum and output datagram. 1441 */ 1442 ui->ui_sum = 0; 1443 if (pr == IPPROTO_UDPLITE) { 1444 if (inp->inp_flags & INP_ONESBCAST) 1445 faddr.s_addr = INADDR_BROADCAST; 1446 if (cscov_partial) { 1447 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1448 ui->ui_sum = 0xffff; 1449 } else { 1450 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1451 ui->ui_sum = 0xffff; 1452 } 1453 } else if (V_udp_cksum) { 1454 if (inp->inp_flags & INP_ONESBCAST) 1455 faddr.s_addr = INADDR_BROADCAST; 1456 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1457 htons((u_short)len + sizeof(struct udphdr) + pr)); 1458 m->m_pkthdr.csum_flags = CSUM_UDP; 1459 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1460 } 1461 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1462 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1463 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1464 UDPSTAT_INC(udps_opackets); 1465 1466 /* 1467 * Setup flowid / RSS information for outbound socket. 1468 * 1469 * Once the UDP code decides to set a flowid some other way, 1470 * this allows the flowid to be overridden by userland. 1471 */ 1472 if (flowtype != M_HASHTYPE_NONE) { 1473 m->m_pkthdr.flowid = flowid; 1474 M_HASHTYPE_SET(m, flowtype); 1475 #ifdef RSS 1476 } else { 1477 uint32_t hash_val, hash_type; 1478 /* 1479 * Calculate an appropriate RSS hash for UDP and 1480 * UDP Lite. 1481 * 1482 * The called function will take care of figuring out 1483 * whether a 2-tuple or 4-tuple hash is required based 1484 * on the currently configured scheme. 1485 * 1486 * Later later on connected socket values should be 1487 * cached in the inpcb and reused, rather than constantly 1488 * re-calculating it. 1489 * 1490 * UDP Lite is a different protocol number and will 1491 * likely end up being hashed as a 2-tuple until 1492 * RSS / NICs grow UDP Lite protocol awareness. 1493 */ 1494 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1495 pr, &hash_val, &hash_type) == 0) { 1496 m->m_pkthdr.flowid = hash_val; 1497 M_HASHTYPE_SET(m, hash_type); 1498 } 1499 #endif 1500 } 1501 1502 #ifdef RSS 1503 /* 1504 * Don't override with the inp cached flowid value. 1505 * 1506 * Depending upon the kind of send being done, the inp 1507 * flowid/flowtype values may actually not be appropriate 1508 * for this particular socket send. 1509 * 1510 * We should either leave the flowid at zero (which is what is 1511 * currently done) or set it to some software generated 1512 * hash value based on the packet contents. 1513 */ 1514 ipflags |= IP_NODEFAULTFLOWID; 1515 #endif /* RSS */ 1516 1517 if (unlock_udbinfo == UH_WLOCKED) 1518 INP_HASH_WUNLOCK(pcbinfo); 1519 else if (unlock_udbinfo == UH_RLOCKED) 1520 INP_HASH_RUNLOCK(pcbinfo); 1521 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1522 error = ip_output(m, inp->inp_options, 1523 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1524 inp->inp_moptions, inp); 1525 if (unlock_inp == UH_WLOCKED) 1526 INP_WUNLOCK(inp); 1527 else 1528 INP_RUNLOCK(inp); 1529 return (error); 1530 1531 release: 1532 if (unlock_udbinfo == UH_WLOCKED) { 1533 KASSERT(unlock_inp == UH_WLOCKED, 1534 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1535 INP_HASH_WUNLOCK(pcbinfo); 1536 INP_WUNLOCK(inp); 1537 } else if (unlock_udbinfo == UH_RLOCKED) { 1538 KASSERT(unlock_inp == UH_RLOCKED, 1539 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1540 INP_HASH_RUNLOCK(pcbinfo); 1541 INP_RUNLOCK(inp); 1542 } else if (unlock_inp == UH_WLOCKED) 1543 INP_WUNLOCK(inp); 1544 else 1545 INP_RUNLOCK(inp); 1546 m_freem(m); 1547 return (error); 1548 } 1549 1550 static void 1551 udp_abort(struct socket *so) 1552 { 1553 struct inpcb *inp; 1554 struct inpcbinfo *pcbinfo; 1555 1556 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1557 inp = sotoinpcb(so); 1558 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1559 INP_WLOCK(inp); 1560 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1561 INP_HASH_WLOCK(pcbinfo); 1562 in_pcbdisconnect(inp); 1563 inp->inp_laddr.s_addr = INADDR_ANY; 1564 INP_HASH_WUNLOCK(pcbinfo); 1565 soisdisconnected(so); 1566 } 1567 INP_WUNLOCK(inp); 1568 } 1569 1570 static int 1571 udp_attach(struct socket *so, int proto, struct thread *td) 1572 { 1573 struct inpcb *inp; 1574 struct inpcbinfo *pcbinfo; 1575 int error; 1576 1577 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1578 inp = sotoinpcb(so); 1579 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1580 error = soreserve(so, udp_sendspace, udp_recvspace); 1581 if (error) 1582 return (error); 1583 INP_INFO_WLOCK(pcbinfo); 1584 error = in_pcballoc(so, pcbinfo); 1585 if (error) { 1586 INP_INFO_WUNLOCK(pcbinfo); 1587 return (error); 1588 } 1589 1590 inp = sotoinpcb(so); 1591 inp->inp_vflag |= INP_IPV4; 1592 inp->inp_ip_ttl = V_ip_defttl; 1593 1594 error = udp_newudpcb(inp); 1595 if (error) { 1596 in_pcbdetach(inp); 1597 in_pcbfree(inp); 1598 INP_INFO_WUNLOCK(pcbinfo); 1599 return (error); 1600 } 1601 1602 INP_WUNLOCK(inp); 1603 INP_INFO_WUNLOCK(pcbinfo); 1604 return (0); 1605 } 1606 #endif /* INET */ 1607 1608 int 1609 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1610 { 1611 struct inpcb *inp; 1612 struct udpcb *up; 1613 1614 KASSERT(so->so_type == SOCK_DGRAM, 1615 ("udp_set_kernel_tunneling: !dgram")); 1616 inp = sotoinpcb(so); 1617 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1618 INP_WLOCK(inp); 1619 up = intoudpcb(inp); 1620 if ((up->u_tun_func != NULL) || 1621 (up->u_icmp_func != NULL)) { 1622 INP_WUNLOCK(inp); 1623 return (EBUSY); 1624 } 1625 up->u_tun_func = f; 1626 up->u_icmp_func = i; 1627 up->u_tun_ctx = ctx; 1628 INP_WUNLOCK(inp); 1629 return (0); 1630 } 1631 1632 #ifdef INET 1633 static int 1634 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1635 { 1636 struct inpcb *inp; 1637 struct inpcbinfo *pcbinfo; 1638 int error; 1639 1640 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1641 inp = sotoinpcb(so); 1642 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1643 INP_WLOCK(inp); 1644 INP_HASH_WLOCK(pcbinfo); 1645 error = in_pcbbind(inp, nam, td->td_ucred); 1646 INP_HASH_WUNLOCK(pcbinfo); 1647 INP_WUNLOCK(inp); 1648 return (error); 1649 } 1650 1651 static void 1652 udp_close(struct socket *so) 1653 { 1654 struct inpcb *inp; 1655 struct inpcbinfo *pcbinfo; 1656 1657 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1658 inp = sotoinpcb(so); 1659 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1660 INP_WLOCK(inp); 1661 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1662 INP_HASH_WLOCK(pcbinfo); 1663 in_pcbdisconnect(inp); 1664 inp->inp_laddr.s_addr = INADDR_ANY; 1665 INP_HASH_WUNLOCK(pcbinfo); 1666 soisdisconnected(so); 1667 } 1668 INP_WUNLOCK(inp); 1669 } 1670 1671 static int 1672 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1673 { 1674 struct inpcb *inp; 1675 struct inpcbinfo *pcbinfo; 1676 struct sockaddr_in *sin; 1677 int error; 1678 1679 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1680 inp = sotoinpcb(so); 1681 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1682 INP_WLOCK(inp); 1683 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1684 INP_WUNLOCK(inp); 1685 return (EISCONN); 1686 } 1687 sin = (struct sockaddr_in *)nam; 1688 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1689 if (error != 0) { 1690 INP_WUNLOCK(inp); 1691 return (error); 1692 } 1693 INP_HASH_WLOCK(pcbinfo); 1694 error = in_pcbconnect(inp, nam, td->td_ucred); 1695 INP_HASH_WUNLOCK(pcbinfo); 1696 if (error == 0) 1697 soisconnected(so); 1698 INP_WUNLOCK(inp); 1699 return (error); 1700 } 1701 1702 static void 1703 udp_detach(struct socket *so) 1704 { 1705 struct inpcb *inp; 1706 struct inpcbinfo *pcbinfo; 1707 struct udpcb *up; 1708 1709 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1710 inp = sotoinpcb(so); 1711 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1712 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1713 ("udp_detach: not disconnected")); 1714 INP_INFO_WLOCK(pcbinfo); 1715 INP_WLOCK(inp); 1716 up = intoudpcb(inp); 1717 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1718 inp->inp_ppcb = NULL; 1719 in_pcbdetach(inp); 1720 in_pcbfree(inp); 1721 INP_INFO_WUNLOCK(pcbinfo); 1722 udp_discardcb(up); 1723 } 1724 1725 static int 1726 udp_disconnect(struct socket *so) 1727 { 1728 struct inpcb *inp; 1729 struct inpcbinfo *pcbinfo; 1730 1731 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1732 inp = sotoinpcb(so); 1733 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1734 INP_WLOCK(inp); 1735 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1736 INP_WUNLOCK(inp); 1737 return (ENOTCONN); 1738 } 1739 INP_HASH_WLOCK(pcbinfo); 1740 in_pcbdisconnect(inp); 1741 inp->inp_laddr.s_addr = INADDR_ANY; 1742 INP_HASH_WUNLOCK(pcbinfo); 1743 SOCK_LOCK(so); 1744 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1745 SOCK_UNLOCK(so); 1746 INP_WUNLOCK(inp); 1747 return (0); 1748 } 1749 1750 static int 1751 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1752 struct mbuf *control, struct thread *td) 1753 { 1754 struct inpcb *inp; 1755 1756 inp = sotoinpcb(so); 1757 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1758 return (udp_output(inp, m, addr, control, td)); 1759 } 1760 #endif /* INET */ 1761 1762 int 1763 udp_shutdown(struct socket *so) 1764 { 1765 struct inpcb *inp; 1766 1767 inp = sotoinpcb(so); 1768 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1769 INP_WLOCK(inp); 1770 socantsendmore(so); 1771 INP_WUNLOCK(inp); 1772 return (0); 1773 } 1774 1775 #ifdef INET 1776 struct pr_usrreqs udp_usrreqs = { 1777 .pru_abort = udp_abort, 1778 .pru_attach = udp_attach, 1779 .pru_bind = udp_bind, 1780 .pru_connect = udp_connect, 1781 .pru_control = in_control, 1782 .pru_detach = udp_detach, 1783 .pru_disconnect = udp_disconnect, 1784 .pru_peeraddr = in_getpeeraddr, 1785 .pru_send = udp_send, 1786 .pru_soreceive = soreceive_dgram, 1787 .pru_sosend = sosend_dgram, 1788 .pru_shutdown = udp_shutdown, 1789 .pru_sockaddr = in_getsockaddr, 1790 .pru_sosetlabel = in_pcbsosetlabel, 1791 .pru_close = udp_close, 1792 }; 1793 #endif /* INET */ 1794