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 * 4. 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 static VNET_DEFINE(int, udp_require_l2_bcast) = 0; 127 #define V_udp_require_l2_bcast VNET(udp_require_l2_bcast) 128 SYSCTL_INT(_net_inet_udp, OID_AUTO, require_l2_bcast, CTLFLAG_VNET | CTLFLAG_RW, 129 &VNET_NAME(udp_require_l2_bcast), 0, 130 "Only treat packets sent to an L2 broadcast address as broadcast packets"); 131 132 u_long udp_sendspace = 9216; /* really max datagram size */ 133 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 134 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 135 136 u_long udp_recvspace = 40 * (1024 + 137 #ifdef INET6 138 sizeof(struct sockaddr_in6) 139 #else 140 sizeof(struct sockaddr_in) 141 #endif 142 ); /* 40 1K datagrams */ 143 144 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 145 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 146 147 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 148 VNET_DEFINE(struct inpcbinfo, udbinfo); 149 VNET_DEFINE(struct inpcbhead, ulitecb); 150 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 151 static VNET_DEFINE(uma_zone_t, udpcb_zone); 152 #define V_udpcb_zone VNET(udpcb_zone) 153 154 #ifndef UDBHASHSIZE 155 #define UDBHASHSIZE 128 156 #endif 157 158 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 159 VNET_PCPUSTAT_SYSINIT(udpstat); 160 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 161 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 162 163 #ifdef VIMAGE 164 VNET_PCPUSTAT_SYSUNINIT(udpstat); 165 #endif /* VIMAGE */ 166 #ifdef INET 167 static void udp_detach(struct socket *so); 168 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 169 struct mbuf *, struct thread *); 170 #endif 171 172 static void 173 udp_zone_change(void *tag) 174 { 175 176 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 177 uma_zone_set_max(V_udpcb_zone, maxsockets); 178 } 179 180 static int 181 udp_inpcb_init(void *mem, int size, int flags) 182 { 183 struct inpcb *inp; 184 185 inp = mem; 186 INP_LOCK_INIT(inp, "inp", "udpinp"); 187 return (0); 188 } 189 190 static int 191 udplite_inpcb_init(void *mem, int size, int flags) 192 { 193 struct inpcb *inp; 194 195 inp = mem; 196 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 197 return (0); 198 } 199 200 void 201 udp_init(void) 202 { 203 204 /* 205 * For now default to 2-tuple UDP hashing - until the fragment 206 * reassembly code can also update the flowid. 207 * 208 * Once we can calculate the flowid that way and re-establish 209 * a 4-tuple, flip this to 4-tuple. 210 */ 211 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 212 "udp_inpcb", udp_inpcb_init, NULL, 0, 213 IPI_HASHFIELDS_2TUPLE); 214 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 215 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 216 uma_zone_set_max(V_udpcb_zone, maxsockets); 217 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 218 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 219 EVENTHANDLER_PRI_ANY); 220 } 221 222 void 223 udplite_init(void) 224 { 225 226 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 227 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL, 228 0, IPI_HASHFIELDS_2TUPLE); 229 } 230 231 /* 232 * Kernel module interface for updating udpstat. The argument is an index 233 * into udpstat treated as an array of u_long. While this encodes the 234 * general layout of udpstat into the caller, it doesn't encode its location, 235 * so that future changes to add, for example, per-CPU stats support won't 236 * cause binary compatibility problems for kernel modules. 237 */ 238 void 239 kmod_udpstat_inc(int statnum) 240 { 241 242 counter_u64_add(VNET(udpstat)[statnum], 1); 243 } 244 245 int 246 udp_newudpcb(struct inpcb *inp) 247 { 248 struct udpcb *up; 249 250 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 251 if (up == NULL) 252 return (ENOBUFS); 253 inp->inp_ppcb = up; 254 return (0); 255 } 256 257 void 258 udp_discardcb(struct udpcb *up) 259 { 260 261 uma_zfree(V_udpcb_zone, up); 262 } 263 264 #ifdef VIMAGE 265 static void 266 udp_destroy(void *unused __unused) 267 { 268 269 in_pcbinfo_destroy(&V_udbinfo); 270 uma_zdestroy(V_udpcb_zone); 271 } 272 VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); 273 274 static void 275 udplite_destroy(void *unused __unused) 276 { 277 278 in_pcbinfo_destroy(&V_ulitecbinfo); 279 } 280 VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, 281 NULL); 282 #endif 283 284 #ifdef INET 285 /* 286 * Subroutine of udp_input(), which appends the provided mbuf chain to the 287 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 288 * contains the source address. If the socket ends up being an IPv6 socket, 289 * udp_append() will convert to a sockaddr_in6 before passing the address 290 * into the socket code. 291 * 292 * In the normal case udp_append() will return 0, indicating that you 293 * must unlock the inp. However if a tunneling protocol is in place we increment 294 * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we 295 * then decrement the reference count. If the inp_rele returns 1, indicating the 296 * inp is gone, we return that to the caller to tell them *not* to unlock 297 * the inp. In the case of multi-cast this will cause the distribution 298 * to stop (though most tunneling protocols known currently do *not* use 299 * multicast). 300 */ 301 static int 302 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 303 struct sockaddr_in *udp_in) 304 { 305 struct sockaddr *append_sa; 306 struct socket *so; 307 struct mbuf *tmpopts, *opts = NULL; 308 #ifdef INET6 309 struct sockaddr_in6 udp_in6; 310 #endif 311 struct udpcb *up; 312 313 INP_LOCK_ASSERT(inp); 314 315 /* 316 * Engage the tunneling protocol. 317 */ 318 up = intoudpcb(inp); 319 if (up->u_tun_func != NULL) { 320 in_pcbref(inp); 321 INP_RUNLOCK(inp); 322 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], 323 up->u_tun_ctx); 324 INP_RLOCK(inp); 325 return (in_pcbrele_rlocked(inp)); 326 } 327 328 off += sizeof(struct udphdr); 329 330 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 331 /* Check AH/ESP integrity. */ 332 if (IPSEC_ENABLED(ipv4) && 333 IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { 334 m_freem(n); 335 return (0); 336 } 337 if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ 338 if (IPSEC_ENABLED(ipv4) && 339 UDPENCAP_INPUT(n, off, AF_INET) != 0) 340 return (0); /* Consumed. */ 341 } 342 #endif /* IPSEC */ 343 #ifdef MAC 344 if (mac_inpcb_check_deliver(inp, n) != 0) { 345 m_freem(n); 346 return (0); 347 } 348 #endif /* MAC */ 349 if (inp->inp_flags & INP_CONTROLOPTS || 350 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 351 #ifdef INET6 352 if (inp->inp_vflag & INP_IPV6) 353 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 354 else 355 #endif /* INET6 */ 356 ip_savecontrol(inp, &opts, ip, n); 357 } 358 if (inp->inp_vflag & INP_IPV4 && inp->inp_flags2 & INP_ORIGDSTADDR) { 359 tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], 360 sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); 361 if (tmpopts) { 362 if (opts) { 363 tmpopts->m_next = opts; 364 opts = tmpopts; 365 } else 366 opts = tmpopts; 367 } 368 } 369 #ifdef INET6 370 if (inp->inp_vflag & INP_IPV6) { 371 bzero(&udp_in6, sizeof(udp_in6)); 372 udp_in6.sin6_len = sizeof(udp_in6); 373 udp_in6.sin6_family = AF_INET6; 374 in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); 375 append_sa = (struct sockaddr *)&udp_in6; 376 } else 377 #endif /* INET6 */ 378 append_sa = (struct sockaddr *)&udp_in[0]; 379 m_adj(n, off); 380 381 so = inp->inp_socket; 382 SOCKBUF_LOCK(&so->so_rcv); 383 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 384 SOCKBUF_UNLOCK(&so->so_rcv); 385 m_freem(n); 386 if (opts) 387 m_freem(opts); 388 UDPSTAT_INC(udps_fullsock); 389 } else 390 sorwakeup_locked(so); 391 return (0); 392 } 393 394 int 395 udp_input(struct mbuf **mp, int *offp, int proto) 396 { 397 struct ip *ip; 398 struct udphdr *uh; 399 struct ifnet *ifp; 400 struct inpcb *inp; 401 uint16_t len, ip_len; 402 struct inpcbinfo *pcbinfo; 403 struct ip save_ip; 404 struct sockaddr_in udpin[2]; 405 struct mbuf *m; 406 struct m_tag *fwd_tag; 407 int cscov_partial, iphlen; 408 409 m = *mp; 410 iphlen = *offp; 411 ifp = m->m_pkthdr.rcvif; 412 *mp = NULL; 413 UDPSTAT_INC(udps_ipackets); 414 415 /* 416 * Strip IP options, if any; should skip this, make available to 417 * user, and use on returned packets, but we don't yet have a way to 418 * check the checksum with options still present. 419 */ 420 if (iphlen > sizeof (struct ip)) { 421 ip_stripoptions(m); 422 iphlen = sizeof(struct ip); 423 } 424 425 /* 426 * Get IP and UDP header together in first mbuf. 427 */ 428 ip = mtod(m, struct ip *); 429 if (m->m_len < iphlen + sizeof(struct udphdr)) { 430 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 431 UDPSTAT_INC(udps_hdrops); 432 return (IPPROTO_DONE); 433 } 434 ip = mtod(m, struct ip *); 435 } 436 uh = (struct udphdr *)((caddr_t)ip + iphlen); 437 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 438 439 /* 440 * Destination port of 0 is illegal, based on RFC768. 441 */ 442 if (uh->uh_dport == 0) 443 goto badunlocked; 444 445 /* 446 * Construct sockaddr format source address. Stuff source address 447 * and datagram in user buffer. 448 */ 449 bzero(&udpin[0], sizeof(struct sockaddr_in)); 450 udpin[0].sin_len = sizeof(struct sockaddr_in); 451 udpin[0].sin_family = AF_INET; 452 udpin[0].sin_port = uh->uh_sport; 453 udpin[0].sin_addr = ip->ip_src; 454 bzero(&udpin[1], sizeof(struct sockaddr_in)); 455 udpin[1].sin_len = sizeof(struct sockaddr_in); 456 udpin[1].sin_family = AF_INET; 457 udpin[1].sin_port = uh->uh_dport; 458 udpin[1].sin_addr = ip->ip_dst; 459 460 /* 461 * Make mbuf data length reflect UDP length. If not enough data to 462 * reflect UDP length, drop. 463 */ 464 len = ntohs((u_short)uh->uh_ulen); 465 ip_len = ntohs(ip->ip_len) - iphlen; 466 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 467 /* Zero means checksum over the complete packet. */ 468 if (len == 0) 469 len = ip_len; 470 cscov_partial = 0; 471 } 472 if (ip_len != len) { 473 if (len > ip_len || len < sizeof(struct udphdr)) { 474 UDPSTAT_INC(udps_badlen); 475 goto badunlocked; 476 } 477 if (proto == IPPROTO_UDP) 478 m_adj(m, len - ip_len); 479 } 480 481 /* 482 * Save a copy of the IP header in case we want restore it for 483 * sending an ICMP error message in response. 484 */ 485 if (!V_udp_blackhole) 486 save_ip = *ip; 487 else 488 memset(&save_ip, 0, sizeof(save_ip)); 489 490 /* 491 * Checksum extended UDP header and data. 492 */ 493 if (uh->uh_sum) { 494 u_short uh_sum; 495 496 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 497 !cscov_partial) { 498 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 499 uh_sum = m->m_pkthdr.csum_data; 500 else 501 uh_sum = in_pseudo(ip->ip_src.s_addr, 502 ip->ip_dst.s_addr, htonl((u_short)len + 503 m->m_pkthdr.csum_data + proto)); 504 uh_sum ^= 0xffff; 505 } else { 506 char b[9]; 507 508 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 509 bzero(((struct ipovly *)ip)->ih_x1, 9); 510 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 511 uh->uh_ulen : htons(ip_len); 512 uh_sum = in_cksum(m, len + sizeof (struct ip)); 513 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 514 } 515 if (uh_sum) { 516 UDPSTAT_INC(udps_badsum); 517 m_freem(m); 518 return (IPPROTO_DONE); 519 } 520 } else { 521 if (proto == IPPROTO_UDP) { 522 UDPSTAT_INC(udps_nosum); 523 } else { 524 /* UDPLite requires a checksum */ 525 /* XXX: What is the right UDPLite MIB counter here? */ 526 m_freem(m); 527 return (IPPROTO_DONE); 528 } 529 } 530 531 pcbinfo = udp_get_inpcbinfo(proto); 532 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 533 ((!V_udp_require_l2_bcast || m->m_flags & M_BCAST) && 534 in_broadcast(ip->ip_dst, ifp))) { 535 struct inpcb *last; 536 struct inpcbhead *pcblist; 537 struct ip_moptions *imo; 538 539 INP_INFO_RLOCK(pcbinfo); 540 pcblist = udp_get_pcblist(proto); 541 last = NULL; 542 LIST_FOREACH(inp, pcblist, inp_list) { 543 if (inp->inp_lport != uh->uh_dport) 544 continue; 545 #ifdef INET6 546 if ((inp->inp_vflag & INP_IPV4) == 0) 547 continue; 548 #endif 549 if (inp->inp_laddr.s_addr != INADDR_ANY && 550 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 551 continue; 552 if (inp->inp_faddr.s_addr != INADDR_ANY && 553 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 554 continue; 555 if (inp->inp_fport != 0 && 556 inp->inp_fport != uh->uh_sport) 557 continue; 558 559 INP_RLOCK(inp); 560 561 /* 562 * XXXRW: Because we weren't holding either the inpcb 563 * or the hash lock when we checked for a match 564 * before, we should probably recheck now that the 565 * inpcb lock is held. 566 */ 567 568 /* 569 * Handle socket delivery policy for any-source 570 * and source-specific multicast. [RFC3678] 571 */ 572 imo = inp->inp_moptions; 573 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 574 struct sockaddr_in group; 575 int blocked; 576 if (imo == NULL) { 577 INP_RUNLOCK(inp); 578 continue; 579 } 580 bzero(&group, sizeof(struct sockaddr_in)); 581 group.sin_len = sizeof(struct sockaddr_in); 582 group.sin_family = AF_INET; 583 group.sin_addr = ip->ip_dst; 584 585 blocked = imo_multi_filter(imo, ifp, 586 (struct sockaddr *)&group, 587 (struct sockaddr *)&udpin[0]); 588 if (blocked != MCAST_PASS) { 589 if (blocked == MCAST_NOTGMEMBER) 590 IPSTAT_INC(ips_notmember); 591 if (blocked == MCAST_NOTSMEMBER || 592 blocked == MCAST_MUTED) 593 UDPSTAT_INC(udps_filtermcast); 594 INP_RUNLOCK(inp); 595 continue; 596 } 597 } 598 if (last != NULL) { 599 struct mbuf *n; 600 601 if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != 602 NULL) { 603 UDP_PROBE(receive, NULL, last, ip, 604 last, uh); 605 if (udp_append(last, ip, n, iphlen, 606 udpin)) { 607 goto inp_lost; 608 } 609 } 610 INP_RUNLOCK(last); 611 } 612 last = inp; 613 /* 614 * Don't look for additional matches if this one does 615 * not have either the SO_REUSEPORT or SO_REUSEADDR 616 * socket options set. This heuristic avoids 617 * searching through all pcbs in the common case of a 618 * non-shared port. It assumes that an application 619 * will never clear these options after setting them. 620 */ 621 if ((last->inp_socket->so_options & 622 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 623 break; 624 } 625 626 if (last == NULL) { 627 /* 628 * No matching pcb found; discard datagram. (No need 629 * to send an ICMP Port Unreachable for a broadcast 630 * or multicast datgram.) 631 */ 632 UDPSTAT_INC(udps_noportbcast); 633 if (inp) 634 INP_RUNLOCK(inp); 635 INP_INFO_RUNLOCK(pcbinfo); 636 goto badunlocked; 637 } 638 UDP_PROBE(receive, NULL, last, ip, last, uh); 639 if (udp_append(last, ip, m, iphlen, udpin) == 0) 640 INP_RUNLOCK(last); 641 inp_lost: 642 INP_INFO_RUNLOCK(pcbinfo); 643 return (IPPROTO_DONE); 644 } 645 646 /* 647 * Locate pcb for datagram. 648 */ 649 650 /* 651 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 652 */ 653 if ((m->m_flags & M_IP_NEXTHOP) && 654 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 655 struct sockaddr_in *next_hop; 656 657 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 658 659 /* 660 * Transparently forwarded. Pretend to be the destination. 661 * Already got one like this? 662 */ 663 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 664 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 665 if (!inp) { 666 /* 667 * It's new. Try to find the ambushing socket. 668 * Because we've rewritten the destination address, 669 * any hardware-generated hash is ignored. 670 */ 671 inp = in_pcblookup(pcbinfo, ip->ip_src, 672 uh->uh_sport, next_hop->sin_addr, 673 next_hop->sin_port ? htons(next_hop->sin_port) : 674 uh->uh_dport, INPLOOKUP_WILDCARD | 675 INPLOOKUP_RLOCKPCB, ifp); 676 } 677 /* Remove the tag from the packet. We don't need it anymore. */ 678 m_tag_delete(m, fwd_tag); 679 m->m_flags &= ~M_IP_NEXTHOP; 680 } else 681 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 682 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 683 INPLOOKUP_RLOCKPCB, ifp, m); 684 if (inp == NULL) { 685 if (udp_log_in_vain) { 686 char src[INET_ADDRSTRLEN]; 687 char dst[INET_ADDRSTRLEN]; 688 689 log(LOG_INFO, 690 "Connection attempt to UDP %s:%d from %s:%d\n", 691 inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), 692 inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); 693 } 694 UDPSTAT_INC(udps_noport); 695 if (m->m_flags & (M_BCAST | M_MCAST)) { 696 UDPSTAT_INC(udps_noportbcast); 697 goto badunlocked; 698 } 699 if (V_udp_blackhole) 700 goto badunlocked; 701 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 702 goto badunlocked; 703 *ip = save_ip; 704 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 705 return (IPPROTO_DONE); 706 } 707 708 /* 709 * Check the minimum TTL for socket. 710 */ 711 INP_RLOCK_ASSERT(inp); 712 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 713 INP_RUNLOCK(inp); 714 m_freem(m); 715 return (IPPROTO_DONE); 716 } 717 if (cscov_partial) { 718 struct udpcb *up; 719 720 up = intoudpcb(inp); 721 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 722 INP_RUNLOCK(inp); 723 m_freem(m); 724 return (IPPROTO_DONE); 725 } 726 } 727 728 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 729 if (udp_append(inp, ip, m, iphlen, udpin) == 0) 730 INP_RUNLOCK(inp); 731 return (IPPROTO_DONE); 732 733 badunlocked: 734 m_freem(m); 735 return (IPPROTO_DONE); 736 } 737 #endif /* INET */ 738 739 /* 740 * Notify a udp user of an asynchronous error; just wake up so that they can 741 * collect error status. 742 */ 743 struct inpcb * 744 udp_notify(struct inpcb *inp, int errno) 745 { 746 747 /* 748 * While udp_ctlinput() always calls udp_notify() with a read lock 749 * when invoking it directly, in_pcbnotifyall() currently uses write 750 * locks due to sharing code with TCP. For now, accept either a read 751 * or a write lock, but a read lock is sufficient. 752 */ 753 INP_LOCK_ASSERT(inp); 754 if ((errno == EHOSTUNREACH || errno == ENETUNREACH || 755 errno == EHOSTDOWN) && inp->inp_route.ro_rt) { 756 RTFREE(inp->inp_route.ro_rt); 757 inp->inp_route.ro_rt = (struct rtentry *)NULL; 758 } 759 760 inp->inp_socket->so_error = errno; 761 sorwakeup(inp->inp_socket); 762 sowwakeup(inp->inp_socket); 763 return (inp); 764 } 765 766 #ifdef INET 767 static void 768 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 769 struct inpcbinfo *pcbinfo) 770 { 771 struct ip *ip = vip; 772 struct udphdr *uh; 773 struct in_addr faddr; 774 struct inpcb *inp; 775 776 faddr = ((struct sockaddr_in *)sa)->sin_addr; 777 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 778 return; 779 780 if (PRC_IS_REDIRECT(cmd)) { 781 /* signal EHOSTDOWN, as it flushes the cached route */ 782 in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); 783 return; 784 } 785 786 /* 787 * Hostdead is ugly because it goes linearly through all PCBs. 788 * 789 * XXX: We never get this from ICMP, otherwise it makes an excellent 790 * DoS attack on machines with many connections. 791 */ 792 if (cmd == PRC_HOSTDEAD) 793 ip = NULL; 794 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 795 return; 796 if (ip != NULL) { 797 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 798 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 799 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 800 if (inp != NULL) { 801 INP_RLOCK_ASSERT(inp); 802 if (inp->inp_socket != NULL) { 803 udp_notify(inp, inetctlerrmap[cmd]); 804 } 805 INP_RUNLOCK(inp); 806 } else { 807 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 808 ip->ip_src, uh->uh_sport, 809 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 810 if (inp != NULL) { 811 struct udpcb *up; 812 813 up = intoudpcb(inp); 814 if (up->u_icmp_func != NULL) { 815 INP_RUNLOCK(inp); 816 (*up->u_icmp_func)(cmd, sa, vip, up->u_tun_ctx); 817 } else { 818 INP_RUNLOCK(inp); 819 } 820 } 821 } 822 } else 823 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 824 udp_notify); 825 } 826 void 827 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 828 { 829 830 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 831 } 832 833 void 834 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 835 { 836 837 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 838 } 839 #endif /* INET */ 840 841 static int 842 udp_pcblist(SYSCTL_HANDLER_ARGS) 843 { 844 int error, i, n; 845 struct inpcb *inp, **inp_list; 846 inp_gen_t gencnt; 847 struct xinpgen xig; 848 849 /* 850 * The process of preparing the PCB list is too time-consuming and 851 * resource-intensive to repeat twice on every request. 852 */ 853 if (req->oldptr == 0) { 854 n = V_udbinfo.ipi_count; 855 n += imax(n / 8, 10); 856 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 857 return (0); 858 } 859 860 if (req->newptr != 0) 861 return (EPERM); 862 863 /* 864 * OK, now we're committed to doing something. 865 */ 866 INP_INFO_RLOCK(&V_udbinfo); 867 gencnt = V_udbinfo.ipi_gencnt; 868 n = V_udbinfo.ipi_count; 869 INP_INFO_RUNLOCK(&V_udbinfo); 870 871 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 872 + n * sizeof(struct xinpcb)); 873 if (error != 0) 874 return (error); 875 876 xig.xig_len = sizeof xig; 877 xig.xig_count = n; 878 xig.xig_gen = gencnt; 879 xig.xig_sogen = so_gencnt; 880 error = SYSCTL_OUT(req, &xig, sizeof xig); 881 if (error) 882 return (error); 883 884 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 885 if (inp_list == NULL) 886 return (ENOMEM); 887 888 INP_INFO_RLOCK(&V_udbinfo); 889 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 890 inp = LIST_NEXT(inp, inp_list)) { 891 INP_WLOCK(inp); 892 if (inp->inp_gencnt <= gencnt && 893 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 894 in_pcbref(inp); 895 inp_list[i++] = inp; 896 } 897 INP_WUNLOCK(inp); 898 } 899 INP_INFO_RUNLOCK(&V_udbinfo); 900 n = i; 901 902 error = 0; 903 for (i = 0; i < n; i++) { 904 inp = inp_list[i]; 905 INP_RLOCK(inp); 906 if (inp->inp_gencnt <= gencnt) { 907 struct xinpcb xi; 908 909 bzero(&xi, sizeof(xi)); 910 xi.xi_len = sizeof xi; 911 /* XXX should avoid extra copy */ 912 bcopy(inp, &xi.xi_inp, sizeof *inp); 913 if (inp->inp_socket) 914 sotoxsocket(inp->inp_socket, &xi.xi_socket); 915 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 916 INP_RUNLOCK(inp); 917 error = SYSCTL_OUT(req, &xi, sizeof xi); 918 } else 919 INP_RUNLOCK(inp); 920 } 921 INP_INFO_WLOCK(&V_udbinfo); 922 for (i = 0; i < n; i++) { 923 inp = inp_list[i]; 924 INP_RLOCK(inp); 925 if (!in_pcbrele_rlocked(inp)) 926 INP_RUNLOCK(inp); 927 } 928 INP_INFO_WUNLOCK(&V_udbinfo); 929 930 if (!error) { 931 /* 932 * Give the user an updated idea of our state. If the 933 * generation differs from what we told her before, she knows 934 * that something happened while we were processing this 935 * request, and it might be necessary to retry. 936 */ 937 INP_INFO_RLOCK(&V_udbinfo); 938 xig.xig_gen = V_udbinfo.ipi_gencnt; 939 xig.xig_sogen = so_gencnt; 940 xig.xig_count = V_udbinfo.ipi_count; 941 INP_INFO_RUNLOCK(&V_udbinfo); 942 error = SYSCTL_OUT(req, &xig, sizeof xig); 943 } 944 free(inp_list, M_TEMP); 945 return (error); 946 } 947 948 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 949 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 950 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 951 952 #ifdef INET 953 static int 954 udp_getcred(SYSCTL_HANDLER_ARGS) 955 { 956 struct xucred xuc; 957 struct sockaddr_in addrs[2]; 958 struct inpcb *inp; 959 int error; 960 961 error = priv_check(req->td, PRIV_NETINET_GETCRED); 962 if (error) 963 return (error); 964 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 965 if (error) 966 return (error); 967 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 968 addrs[0].sin_addr, addrs[0].sin_port, 969 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 970 if (inp != NULL) { 971 INP_RLOCK_ASSERT(inp); 972 if (inp->inp_socket == NULL) 973 error = ENOENT; 974 if (error == 0) 975 error = cr_canseeinpcb(req->td->td_ucred, inp); 976 if (error == 0) 977 cru2x(inp->inp_cred, &xuc); 978 INP_RUNLOCK(inp); 979 } else 980 error = ENOENT; 981 if (error == 0) 982 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 983 return (error); 984 } 985 986 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 987 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 988 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 989 #endif /* INET */ 990 991 int 992 udp_ctloutput(struct socket *so, struct sockopt *sopt) 993 { 994 struct inpcb *inp; 995 struct udpcb *up; 996 int isudplite, error, optval; 997 998 error = 0; 999 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 1000 inp = sotoinpcb(so); 1001 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1002 INP_WLOCK(inp); 1003 if (sopt->sopt_level != so->so_proto->pr_protocol) { 1004 #ifdef INET6 1005 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 1006 INP_WUNLOCK(inp); 1007 error = ip6_ctloutput(so, sopt); 1008 } 1009 #endif 1010 #if defined(INET) && defined(INET6) 1011 else 1012 #endif 1013 #ifdef INET 1014 { 1015 INP_WUNLOCK(inp); 1016 error = ip_ctloutput(so, sopt); 1017 } 1018 #endif 1019 return (error); 1020 } 1021 1022 switch (sopt->sopt_dir) { 1023 case SOPT_SET: 1024 switch (sopt->sopt_name) { 1025 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1026 #ifdef INET 1027 case UDP_ENCAP: 1028 if (!IPSEC_ENABLED(ipv4)) { 1029 INP_WUNLOCK(inp); 1030 return (ENOPROTOOPT); 1031 } 1032 error = UDPENCAP_PCBCTL(inp, sopt); 1033 break; 1034 #endif /* INET */ 1035 #endif /* IPSEC */ 1036 case UDPLITE_SEND_CSCOV: 1037 case UDPLITE_RECV_CSCOV: 1038 if (!isudplite) { 1039 INP_WUNLOCK(inp); 1040 error = ENOPROTOOPT; 1041 break; 1042 } 1043 INP_WUNLOCK(inp); 1044 error = sooptcopyin(sopt, &optval, sizeof(optval), 1045 sizeof(optval)); 1046 if (error != 0) 1047 break; 1048 inp = sotoinpcb(so); 1049 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1050 INP_WLOCK(inp); 1051 up = intoudpcb(inp); 1052 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1053 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1054 INP_WUNLOCK(inp); 1055 error = EINVAL; 1056 break; 1057 } 1058 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1059 up->u_txcslen = optval; 1060 else 1061 up->u_rxcslen = optval; 1062 INP_WUNLOCK(inp); 1063 break; 1064 default: 1065 INP_WUNLOCK(inp); 1066 error = ENOPROTOOPT; 1067 break; 1068 } 1069 break; 1070 case SOPT_GET: 1071 switch (sopt->sopt_name) { 1072 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1073 #ifdef INET 1074 case UDP_ENCAP: 1075 if (!IPSEC_ENABLED(ipv4)) { 1076 INP_WUNLOCK(inp); 1077 return (ENOPROTOOPT); 1078 } 1079 error = UDPENCAP_PCBCTL(inp, sopt); 1080 break; 1081 #endif /* INET */ 1082 #endif /* IPSEC */ 1083 case UDPLITE_SEND_CSCOV: 1084 case UDPLITE_RECV_CSCOV: 1085 if (!isudplite) { 1086 INP_WUNLOCK(inp); 1087 error = ENOPROTOOPT; 1088 break; 1089 } 1090 up = intoudpcb(inp); 1091 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1092 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1093 optval = up->u_txcslen; 1094 else 1095 optval = up->u_rxcslen; 1096 INP_WUNLOCK(inp); 1097 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1098 break; 1099 default: 1100 INP_WUNLOCK(inp); 1101 error = ENOPROTOOPT; 1102 break; 1103 } 1104 break; 1105 } 1106 return (error); 1107 } 1108 1109 #ifdef INET 1110 #define UH_WLOCKED 2 1111 #define UH_RLOCKED 1 1112 #define UH_UNLOCKED 0 1113 static int 1114 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1115 struct mbuf *control, struct thread *td) 1116 { 1117 struct udpiphdr *ui; 1118 int len = m->m_pkthdr.len; 1119 struct in_addr faddr, laddr; 1120 struct cmsghdr *cm; 1121 struct inpcbinfo *pcbinfo; 1122 struct sockaddr_in *sin, src; 1123 int cscov_partial = 0; 1124 int error = 0; 1125 int ipflags; 1126 u_short fport, lport; 1127 int unlock_udbinfo, unlock_inp; 1128 u_char tos; 1129 uint8_t pr; 1130 uint16_t cscov = 0; 1131 uint32_t flowid = 0; 1132 uint8_t flowtype = M_HASHTYPE_NONE; 1133 1134 /* 1135 * udp_output() may need to temporarily bind or connect the current 1136 * inpcb. As such, we don't know up front whether we will need the 1137 * pcbinfo lock or not. Do any work to decide what is needed up 1138 * front before acquiring any locks. 1139 */ 1140 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1141 if (control) 1142 m_freem(control); 1143 m_freem(m); 1144 return (EMSGSIZE); 1145 } 1146 1147 src.sin_family = 0; 1148 sin = (struct sockaddr_in *)addr; 1149 if (sin == NULL || 1150 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1151 INP_WLOCK(inp); 1152 unlock_inp = UH_WLOCKED; 1153 } else { 1154 INP_RLOCK(inp); 1155 unlock_inp = UH_RLOCKED; 1156 } 1157 tos = inp->inp_ip_tos; 1158 if (control != NULL) { 1159 /* 1160 * XXX: Currently, we assume all the optional information is 1161 * stored in a single mbuf. 1162 */ 1163 if (control->m_next) { 1164 if (unlock_inp == UH_WLOCKED) 1165 INP_WUNLOCK(inp); 1166 else 1167 INP_RUNLOCK(inp); 1168 m_freem(control); 1169 m_freem(m); 1170 return (EINVAL); 1171 } 1172 for (; control->m_len > 0; 1173 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1174 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1175 cm = mtod(control, struct cmsghdr *); 1176 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1177 || cm->cmsg_len > control->m_len) { 1178 error = EINVAL; 1179 break; 1180 } 1181 if (cm->cmsg_level != IPPROTO_IP) 1182 continue; 1183 1184 switch (cm->cmsg_type) { 1185 case IP_SENDSRCADDR: 1186 if (cm->cmsg_len != 1187 CMSG_LEN(sizeof(struct in_addr))) { 1188 error = EINVAL; 1189 break; 1190 } 1191 bzero(&src, sizeof(src)); 1192 src.sin_family = AF_INET; 1193 src.sin_len = sizeof(src); 1194 src.sin_port = inp->inp_lport; 1195 src.sin_addr = 1196 *(struct in_addr *)CMSG_DATA(cm); 1197 break; 1198 1199 case IP_TOS: 1200 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1201 error = EINVAL; 1202 break; 1203 } 1204 tos = *(u_char *)CMSG_DATA(cm); 1205 break; 1206 1207 case IP_FLOWID: 1208 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1209 error = EINVAL; 1210 break; 1211 } 1212 flowid = *(uint32_t *) CMSG_DATA(cm); 1213 break; 1214 1215 case IP_FLOWTYPE: 1216 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1217 error = EINVAL; 1218 break; 1219 } 1220 flowtype = *(uint32_t *) CMSG_DATA(cm); 1221 break; 1222 1223 #ifdef RSS 1224 case IP_RSSBUCKETID: 1225 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1226 error = EINVAL; 1227 break; 1228 } 1229 /* This is just a placeholder for now */ 1230 break; 1231 #endif /* RSS */ 1232 default: 1233 error = ENOPROTOOPT; 1234 break; 1235 } 1236 if (error) 1237 break; 1238 } 1239 m_freem(control); 1240 } 1241 if (error) { 1242 if (unlock_inp == UH_WLOCKED) 1243 INP_WUNLOCK(inp); 1244 else 1245 INP_RUNLOCK(inp); 1246 m_freem(m); 1247 return (error); 1248 } 1249 1250 /* 1251 * Depending on whether or not the application has bound or connected 1252 * the socket, we may have to do varying levels of work. The optimal 1253 * case is for a connected UDP socket, as a global lock isn't 1254 * required at all. 1255 * 1256 * In order to decide which we need, we require stability of the 1257 * inpcb binding, which we ensure by acquiring a read lock on the 1258 * inpcb. This doesn't strictly follow the lock order, so we play 1259 * the trylock and retry game; note that we may end up with more 1260 * conservative locks than required the second time around, so later 1261 * assertions have to accept that. Further analysis of the number of 1262 * misses under contention is required. 1263 * 1264 * XXXRW: Check that hash locking update here is correct. 1265 */ 1266 pr = inp->inp_socket->so_proto->pr_protocol; 1267 pcbinfo = udp_get_inpcbinfo(pr); 1268 sin = (struct sockaddr_in *)addr; 1269 if (sin != NULL && 1270 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1271 INP_HASH_WLOCK(pcbinfo); 1272 unlock_udbinfo = UH_WLOCKED; 1273 } else if ((sin != NULL && ( 1274 (sin->sin_addr.s_addr == INADDR_ANY) || 1275 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1276 (inp->inp_laddr.s_addr == INADDR_ANY) || 1277 (inp->inp_lport == 0))) || 1278 (src.sin_family == AF_INET)) { 1279 INP_HASH_RLOCK(pcbinfo); 1280 unlock_udbinfo = UH_RLOCKED; 1281 } else 1282 unlock_udbinfo = UH_UNLOCKED; 1283 1284 /* 1285 * If the IP_SENDSRCADDR control message was specified, override the 1286 * source address for this datagram. Its use is invalidated if the 1287 * address thus specified is incomplete or clobbers other inpcbs. 1288 */ 1289 laddr = inp->inp_laddr; 1290 lport = inp->inp_lport; 1291 if (src.sin_family == AF_INET) { 1292 INP_HASH_LOCK_ASSERT(pcbinfo); 1293 if ((lport == 0) || 1294 (laddr.s_addr == INADDR_ANY && 1295 src.sin_addr.s_addr == INADDR_ANY)) { 1296 error = EINVAL; 1297 goto release; 1298 } 1299 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1300 &laddr.s_addr, &lport, td->td_ucred); 1301 if (error) 1302 goto release; 1303 } 1304 1305 /* 1306 * If a UDP socket has been connected, then a local address/port will 1307 * have been selected and bound. 1308 * 1309 * If a UDP socket has not been connected to, then an explicit 1310 * destination address must be used, in which case a local 1311 * address/port may not have been selected and bound. 1312 */ 1313 if (sin != NULL) { 1314 INP_LOCK_ASSERT(inp); 1315 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1316 error = EISCONN; 1317 goto release; 1318 } 1319 1320 /* 1321 * Jail may rewrite the destination address, so let it do 1322 * that before we use it. 1323 */ 1324 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1325 if (error) 1326 goto release; 1327 1328 /* 1329 * If a local address or port hasn't yet been selected, or if 1330 * the destination address needs to be rewritten due to using 1331 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1332 * to do the heavy lifting. Once a port is selected, we 1333 * commit the binding back to the socket; we also commit the 1334 * binding of the address if in jail. 1335 * 1336 * If we already have a valid binding and we're not 1337 * requesting a destination address rewrite, use a fast path. 1338 */ 1339 if (inp->inp_laddr.s_addr == INADDR_ANY || 1340 inp->inp_lport == 0 || 1341 sin->sin_addr.s_addr == INADDR_ANY || 1342 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1343 INP_HASH_LOCK_ASSERT(pcbinfo); 1344 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1345 &lport, &faddr.s_addr, &fport, NULL, 1346 td->td_ucred); 1347 if (error) 1348 goto release; 1349 1350 /* 1351 * XXXRW: Why not commit the port if the address is 1352 * !INADDR_ANY? 1353 */ 1354 /* Commit the local port if newly assigned. */ 1355 if (inp->inp_laddr.s_addr == INADDR_ANY && 1356 inp->inp_lport == 0) { 1357 INP_WLOCK_ASSERT(inp); 1358 INP_HASH_WLOCK_ASSERT(pcbinfo); 1359 /* 1360 * Remember addr if jailed, to prevent 1361 * rebinding. 1362 */ 1363 if (prison_flag(td->td_ucred, PR_IP4)) 1364 inp->inp_laddr = laddr; 1365 inp->inp_lport = lport; 1366 if (in_pcbinshash(inp) != 0) { 1367 inp->inp_lport = 0; 1368 error = EAGAIN; 1369 goto release; 1370 } 1371 inp->inp_flags |= INP_ANONPORT; 1372 } 1373 } else { 1374 faddr = sin->sin_addr; 1375 fport = sin->sin_port; 1376 } 1377 } else { 1378 INP_LOCK_ASSERT(inp); 1379 faddr = inp->inp_faddr; 1380 fport = inp->inp_fport; 1381 if (faddr.s_addr == INADDR_ANY) { 1382 error = ENOTCONN; 1383 goto release; 1384 } 1385 } 1386 1387 /* 1388 * Calculate data length and get a mbuf for UDP, IP, and possible 1389 * link-layer headers. Immediate slide the data pointer back forward 1390 * since we won't use that space at this layer. 1391 */ 1392 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1393 if (m == NULL) { 1394 error = ENOBUFS; 1395 goto release; 1396 } 1397 m->m_data += max_linkhdr; 1398 m->m_len -= max_linkhdr; 1399 m->m_pkthdr.len -= max_linkhdr; 1400 1401 /* 1402 * Fill in mbuf with extended UDP header and addresses and length put 1403 * into network format. 1404 */ 1405 ui = mtod(m, struct udpiphdr *); 1406 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1407 ui->ui_pr = pr; 1408 ui->ui_src = laddr; 1409 ui->ui_dst = faddr; 1410 ui->ui_sport = lport; 1411 ui->ui_dport = fport; 1412 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1413 if (pr == IPPROTO_UDPLITE) { 1414 struct udpcb *up; 1415 uint16_t plen; 1416 1417 up = intoudpcb(inp); 1418 cscov = up->u_txcslen; 1419 plen = (u_short)len + sizeof(struct udphdr); 1420 if (cscov >= plen) 1421 cscov = 0; 1422 ui->ui_len = htons(plen); 1423 ui->ui_ulen = htons(cscov); 1424 /* 1425 * For UDP-Lite, checksum coverage length of zero means 1426 * the entire UDPLite packet is covered by the checksum. 1427 */ 1428 cscov_partial = (cscov == 0) ? 0 : 1; 1429 } else 1430 ui->ui_v = IPVERSION << 4; 1431 1432 /* 1433 * Set the Don't Fragment bit in the IP header. 1434 */ 1435 if (inp->inp_flags & INP_DONTFRAG) { 1436 struct ip *ip; 1437 1438 ip = (struct ip *)&ui->ui_i; 1439 ip->ip_off |= htons(IP_DF); 1440 } 1441 1442 ipflags = 0; 1443 if (inp->inp_socket->so_options & SO_DONTROUTE) 1444 ipflags |= IP_ROUTETOIF; 1445 if (inp->inp_socket->so_options & SO_BROADCAST) 1446 ipflags |= IP_ALLOWBROADCAST; 1447 if (inp->inp_flags & INP_ONESBCAST) 1448 ipflags |= IP_SENDONES; 1449 1450 #ifdef MAC 1451 mac_inpcb_create_mbuf(inp, m); 1452 #endif 1453 1454 /* 1455 * Set up checksum and output datagram. 1456 */ 1457 ui->ui_sum = 0; 1458 if (pr == IPPROTO_UDPLITE) { 1459 if (inp->inp_flags & INP_ONESBCAST) 1460 faddr.s_addr = INADDR_BROADCAST; 1461 if (cscov_partial) { 1462 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1463 ui->ui_sum = 0xffff; 1464 } else { 1465 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1466 ui->ui_sum = 0xffff; 1467 } 1468 } else if (V_udp_cksum) { 1469 if (inp->inp_flags & INP_ONESBCAST) 1470 faddr.s_addr = INADDR_BROADCAST; 1471 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1472 htons((u_short)len + sizeof(struct udphdr) + pr)); 1473 m->m_pkthdr.csum_flags = CSUM_UDP; 1474 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1475 } 1476 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1477 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1478 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1479 UDPSTAT_INC(udps_opackets); 1480 1481 /* 1482 * Setup flowid / RSS information for outbound socket. 1483 * 1484 * Once the UDP code decides to set a flowid some other way, 1485 * this allows the flowid to be overridden by userland. 1486 */ 1487 if (flowtype != M_HASHTYPE_NONE) { 1488 m->m_pkthdr.flowid = flowid; 1489 M_HASHTYPE_SET(m, flowtype); 1490 #ifdef RSS 1491 } else { 1492 uint32_t hash_val, hash_type; 1493 /* 1494 * Calculate an appropriate RSS hash for UDP and 1495 * UDP Lite. 1496 * 1497 * The called function will take care of figuring out 1498 * whether a 2-tuple or 4-tuple hash is required based 1499 * on the currently configured scheme. 1500 * 1501 * Later later on connected socket values should be 1502 * cached in the inpcb and reused, rather than constantly 1503 * re-calculating it. 1504 * 1505 * UDP Lite is a different protocol number and will 1506 * likely end up being hashed as a 2-tuple until 1507 * RSS / NICs grow UDP Lite protocol awareness. 1508 */ 1509 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1510 pr, &hash_val, &hash_type) == 0) { 1511 m->m_pkthdr.flowid = hash_val; 1512 M_HASHTYPE_SET(m, hash_type); 1513 } 1514 #endif 1515 } 1516 1517 #ifdef RSS 1518 /* 1519 * Don't override with the inp cached flowid value. 1520 * 1521 * Depending upon the kind of send being done, the inp 1522 * flowid/flowtype values may actually not be appropriate 1523 * for this particular socket send. 1524 * 1525 * We should either leave the flowid at zero (which is what is 1526 * currently done) or set it to some software generated 1527 * hash value based on the packet contents. 1528 */ 1529 ipflags |= IP_NODEFAULTFLOWID; 1530 #endif /* RSS */ 1531 1532 if (unlock_udbinfo == UH_WLOCKED) 1533 INP_HASH_WUNLOCK(pcbinfo); 1534 else if (unlock_udbinfo == UH_RLOCKED) 1535 INP_HASH_RUNLOCK(pcbinfo); 1536 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1537 error = ip_output(m, inp->inp_options, 1538 (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, 1539 inp->inp_moptions, inp); 1540 if (unlock_inp == UH_WLOCKED) 1541 INP_WUNLOCK(inp); 1542 else 1543 INP_RUNLOCK(inp); 1544 return (error); 1545 1546 release: 1547 if (unlock_udbinfo == UH_WLOCKED) { 1548 KASSERT(unlock_inp == UH_WLOCKED, 1549 ("%s: excl udbinfo lock, shared inp lock", __func__)); 1550 INP_HASH_WUNLOCK(pcbinfo); 1551 INP_WUNLOCK(inp); 1552 } else if (unlock_udbinfo == UH_RLOCKED) { 1553 KASSERT(unlock_inp == UH_RLOCKED, 1554 ("%s: shared udbinfo lock, excl inp lock", __func__)); 1555 INP_HASH_RUNLOCK(pcbinfo); 1556 INP_RUNLOCK(inp); 1557 } else if (unlock_inp == UH_WLOCKED) 1558 INP_WUNLOCK(inp); 1559 else 1560 INP_RUNLOCK(inp); 1561 m_freem(m); 1562 return (error); 1563 } 1564 1565 static void 1566 udp_abort(struct socket *so) 1567 { 1568 struct inpcb *inp; 1569 struct inpcbinfo *pcbinfo; 1570 1571 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1572 inp = sotoinpcb(so); 1573 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1574 INP_WLOCK(inp); 1575 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1576 INP_HASH_WLOCK(pcbinfo); 1577 in_pcbdisconnect(inp); 1578 inp->inp_laddr.s_addr = INADDR_ANY; 1579 INP_HASH_WUNLOCK(pcbinfo); 1580 soisdisconnected(so); 1581 } 1582 INP_WUNLOCK(inp); 1583 } 1584 1585 static int 1586 udp_attach(struct socket *so, int proto, struct thread *td) 1587 { 1588 struct inpcb *inp; 1589 struct inpcbinfo *pcbinfo; 1590 int error; 1591 1592 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1593 inp = sotoinpcb(so); 1594 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1595 error = soreserve(so, udp_sendspace, udp_recvspace); 1596 if (error) 1597 return (error); 1598 INP_INFO_WLOCK(pcbinfo); 1599 error = in_pcballoc(so, pcbinfo); 1600 if (error) { 1601 INP_INFO_WUNLOCK(pcbinfo); 1602 return (error); 1603 } 1604 1605 inp = sotoinpcb(so); 1606 inp->inp_vflag |= INP_IPV4; 1607 inp->inp_ip_ttl = V_ip_defttl; 1608 1609 error = udp_newudpcb(inp); 1610 if (error) { 1611 in_pcbdetach(inp); 1612 in_pcbfree(inp); 1613 INP_INFO_WUNLOCK(pcbinfo); 1614 return (error); 1615 } 1616 1617 INP_WUNLOCK(inp); 1618 INP_INFO_WUNLOCK(pcbinfo); 1619 return (0); 1620 } 1621 #endif /* INET */ 1622 1623 int 1624 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) 1625 { 1626 struct inpcb *inp; 1627 struct udpcb *up; 1628 1629 KASSERT(so->so_type == SOCK_DGRAM, 1630 ("udp_set_kernel_tunneling: !dgram")); 1631 inp = sotoinpcb(so); 1632 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1633 INP_WLOCK(inp); 1634 up = intoudpcb(inp); 1635 if ((up->u_tun_func != NULL) || 1636 (up->u_icmp_func != NULL)) { 1637 INP_WUNLOCK(inp); 1638 return (EBUSY); 1639 } 1640 up->u_tun_func = f; 1641 up->u_icmp_func = i; 1642 up->u_tun_ctx = ctx; 1643 INP_WUNLOCK(inp); 1644 return (0); 1645 } 1646 1647 #ifdef INET 1648 static int 1649 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1650 { 1651 struct inpcb *inp; 1652 struct inpcbinfo *pcbinfo; 1653 int error; 1654 1655 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1656 inp = sotoinpcb(so); 1657 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1658 INP_WLOCK(inp); 1659 INP_HASH_WLOCK(pcbinfo); 1660 error = in_pcbbind(inp, nam, td->td_ucred); 1661 INP_HASH_WUNLOCK(pcbinfo); 1662 INP_WUNLOCK(inp); 1663 return (error); 1664 } 1665 1666 static void 1667 udp_close(struct socket *so) 1668 { 1669 struct inpcb *inp; 1670 struct inpcbinfo *pcbinfo; 1671 1672 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1673 inp = sotoinpcb(so); 1674 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1675 INP_WLOCK(inp); 1676 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1677 INP_HASH_WLOCK(pcbinfo); 1678 in_pcbdisconnect(inp); 1679 inp->inp_laddr.s_addr = INADDR_ANY; 1680 INP_HASH_WUNLOCK(pcbinfo); 1681 soisdisconnected(so); 1682 } 1683 INP_WUNLOCK(inp); 1684 } 1685 1686 static int 1687 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1688 { 1689 struct inpcb *inp; 1690 struct inpcbinfo *pcbinfo; 1691 struct sockaddr_in *sin; 1692 int error; 1693 1694 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1695 inp = sotoinpcb(so); 1696 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1697 INP_WLOCK(inp); 1698 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1699 INP_WUNLOCK(inp); 1700 return (EISCONN); 1701 } 1702 sin = (struct sockaddr_in *)nam; 1703 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1704 if (error != 0) { 1705 INP_WUNLOCK(inp); 1706 return (error); 1707 } 1708 INP_HASH_WLOCK(pcbinfo); 1709 error = in_pcbconnect(inp, nam, td->td_ucred); 1710 INP_HASH_WUNLOCK(pcbinfo); 1711 if (error == 0) 1712 soisconnected(so); 1713 INP_WUNLOCK(inp); 1714 return (error); 1715 } 1716 1717 static void 1718 udp_detach(struct socket *so) 1719 { 1720 struct inpcb *inp; 1721 struct inpcbinfo *pcbinfo; 1722 struct udpcb *up; 1723 1724 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1725 inp = sotoinpcb(so); 1726 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1727 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1728 ("udp_detach: not disconnected")); 1729 INP_INFO_WLOCK(pcbinfo); 1730 INP_WLOCK(inp); 1731 up = intoudpcb(inp); 1732 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1733 inp->inp_ppcb = NULL; 1734 in_pcbdetach(inp); 1735 in_pcbfree(inp); 1736 INP_INFO_WUNLOCK(pcbinfo); 1737 udp_discardcb(up); 1738 } 1739 1740 static int 1741 udp_disconnect(struct socket *so) 1742 { 1743 struct inpcb *inp; 1744 struct inpcbinfo *pcbinfo; 1745 1746 pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); 1747 inp = sotoinpcb(so); 1748 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1749 INP_WLOCK(inp); 1750 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1751 INP_WUNLOCK(inp); 1752 return (ENOTCONN); 1753 } 1754 INP_HASH_WLOCK(pcbinfo); 1755 in_pcbdisconnect(inp); 1756 inp->inp_laddr.s_addr = INADDR_ANY; 1757 INP_HASH_WUNLOCK(pcbinfo); 1758 SOCK_LOCK(so); 1759 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1760 SOCK_UNLOCK(so); 1761 INP_WUNLOCK(inp); 1762 return (0); 1763 } 1764 1765 static int 1766 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1767 struct mbuf *control, struct thread *td) 1768 { 1769 struct inpcb *inp; 1770 1771 inp = sotoinpcb(so); 1772 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1773 return (udp_output(inp, m, addr, control, td)); 1774 } 1775 #endif /* INET */ 1776 1777 int 1778 udp_shutdown(struct socket *so) 1779 { 1780 struct inpcb *inp; 1781 1782 inp = sotoinpcb(so); 1783 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1784 INP_WLOCK(inp); 1785 socantsendmore(so); 1786 INP_WUNLOCK(inp); 1787 return (0); 1788 } 1789 1790 #ifdef INET 1791 struct pr_usrreqs udp_usrreqs = { 1792 .pru_abort = udp_abort, 1793 .pru_attach = udp_attach, 1794 .pru_bind = udp_bind, 1795 .pru_connect = udp_connect, 1796 .pru_control = in_control, 1797 .pru_detach = udp_detach, 1798 .pru_disconnect = udp_disconnect, 1799 .pru_peeraddr = in_getpeeraddr, 1800 .pru_send = udp_send, 1801 .pru_soreceive = soreceive_dgram, 1802 .pru_sosend = sosend_dgram, 1803 .pru_shutdown = udp_shutdown, 1804 .pru_sockaddr = in_getsockaddr, 1805 .pru_sosetlabel = in_pcbsosetlabel, 1806 .pru_close = udp_close, 1807 }; 1808 #endif /* INET */ 1809