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