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_ipfw.h" 43 #include "opt_inet.h" 44 #include "opt_inet6.h" 45 #include "opt_ipsec.h" 46 #include "opt_rss.h" 47 48 #include <sys/param.h> 49 #include <sys/domain.h> 50 #include <sys/eventhandler.h> 51 #include <sys/jail.h> 52 #include <sys/kernel.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/protosw.h> 59 #include <sys/sdt.h> 60 #include <sys/signalvar.h> 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/sx.h> 64 #include <sys/sysctl.h> 65 #include <sys/syslog.h> 66 #include <sys/systm.h> 67 68 #include <vm/uma.h> 69 70 #include <net/if.h> 71 #include <net/if_var.h> 72 #include <net/route.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 #ifdef IPSEC 96 #include <netipsec/ipsec.h> 97 #include <netipsec/esp.h> 98 #endif 99 100 #include <machine/in_cksum.h> 101 102 #include <security/mac/mac_framework.h> 103 104 /* 105 * UDP and UDP-Lite protocols implementation. 106 * Per RFC 768, August, 1980. 107 * Per RFC 3828, July, 2004. 108 */ 109 110 /* 111 * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums 112 * removes the only data integrity mechanism for packets and malformed 113 * packets that would otherwise be discarded due to bad checksums, and may 114 * cause problems (especially for NFS data blocks). 115 */ 116 VNET_DEFINE(int, udp_cksum) = 1; 117 SYSCTL_VNET_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, 118 &VNET_NAME(udp_cksum), 0, "compute udp checksum"); 119 120 int udp_log_in_vain = 0; 121 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 122 &udp_log_in_vain, 0, "Log all incoming UDP packets"); 123 124 VNET_DEFINE(int, udp_blackhole) = 0; 125 SYSCTL_VNET_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, 126 &VNET_NAME(udp_blackhole), 0, 127 "Do not send port unreachables for refused connects"); 128 129 u_long udp_sendspace = 9216; /* really max datagram size */ 130 /* 40 1K datagrams */ 131 SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 132 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 133 134 u_long udp_recvspace = 40 * (1024 + 135 #ifdef INET6 136 sizeof(struct sockaddr_in6) 137 #else 138 sizeof(struct sockaddr_in) 139 #endif 140 ); 141 142 SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 143 &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); 144 145 VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ 146 VNET_DEFINE(struct inpcbinfo, udbinfo); 147 VNET_DEFINE(struct inpcbhead, ulitecb); 148 VNET_DEFINE(struct inpcbinfo, ulitecbinfo); 149 static VNET_DEFINE(uma_zone_t, udpcb_zone); 150 #define V_udpcb_zone VNET(udpcb_zone) 151 152 #ifndef UDBHASHSIZE 153 #define UDBHASHSIZE 128 154 #endif 155 156 VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ 157 VNET_PCPUSTAT_SYSINIT(udpstat); 158 SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, 159 udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); 160 161 #ifdef VIMAGE 162 VNET_PCPUSTAT_SYSUNINIT(udpstat); 163 #endif /* VIMAGE */ 164 #ifdef INET 165 static void udp_detach(struct socket *so); 166 static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, 167 struct mbuf *, struct thread *); 168 #endif 169 170 #ifdef IPSEC 171 #ifdef IPSEC_NAT_T 172 #define UF_ESPINUDP_ALL (UF_ESPINUDP_NON_IKE|UF_ESPINUDP) 173 #ifdef INET 174 static struct mbuf *udp4_espdecap(struct inpcb *, struct mbuf *, int); 175 #endif 176 #endif /* IPSEC_NAT_T */ 177 #endif /* IPSEC */ 178 179 static void 180 udp_zone_change(void *tag) 181 { 182 183 uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); 184 uma_zone_set_max(V_udpcb_zone, maxsockets); 185 } 186 187 static int 188 udp_inpcb_init(void *mem, int size, int flags) 189 { 190 struct inpcb *inp; 191 192 inp = mem; 193 INP_LOCK_INIT(inp, "inp", "udpinp"); 194 return (0); 195 } 196 197 static int 198 udplite_inpcb_init(void *mem, int size, int flags) 199 { 200 struct inpcb *inp; 201 202 inp = mem; 203 INP_LOCK_INIT(inp, "inp", "udpliteinp"); 204 return (0); 205 } 206 207 void 208 udp_init(void) 209 { 210 211 /* 212 * For now default to 2-tuple UDP hashing - until the fragment 213 * reassembly code can also update the flowid. 214 * 215 * Once we can calculate the flowid that way and re-establish 216 * a 4-tuple, flip this to 4-tuple. 217 */ 218 in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, 219 "udp_inpcb", udp_inpcb_init, NULL, UMA_ZONE_NOFREE, 220 IPI_HASHFIELDS_2TUPLE); 221 V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), 222 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 223 uma_zone_set_max(V_udpcb_zone, maxsockets); 224 uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); 225 EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, 226 EVENTHANDLER_PRI_ANY); 227 } 228 229 void 230 udplite_init(void) 231 { 232 233 in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, 234 UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, NULL, 235 UMA_ZONE_NOFREE, IPI_HASHFIELDS_2TUPLE); 236 } 237 238 /* 239 * Kernel module interface for updating udpstat. The argument is an index 240 * into udpstat treated as an array of u_long. While this encodes the 241 * general layout of udpstat into the caller, it doesn't encode its location, 242 * so that future changes to add, for example, per-CPU stats support won't 243 * cause binary compatibility problems for kernel modules. 244 */ 245 void 246 kmod_udpstat_inc(int statnum) 247 { 248 249 counter_u64_add(VNET(udpstat)[statnum], 1); 250 } 251 252 int 253 udp_newudpcb(struct inpcb *inp) 254 { 255 struct udpcb *up; 256 257 up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); 258 if (up == NULL) 259 return (ENOBUFS); 260 inp->inp_ppcb = up; 261 return (0); 262 } 263 264 void 265 udp_discardcb(struct udpcb *up) 266 { 267 268 uma_zfree(V_udpcb_zone, up); 269 } 270 271 #ifdef VIMAGE 272 void 273 udp_destroy(void) 274 { 275 276 in_pcbinfo_destroy(&V_udbinfo); 277 uma_zdestroy(V_udpcb_zone); 278 } 279 280 void 281 udplite_destroy(void) 282 { 283 284 in_pcbinfo_destroy(&V_ulitecbinfo); 285 } 286 #endif 287 288 #ifdef INET 289 /* 290 * Subroutine of udp_input(), which appends the provided mbuf chain to the 291 * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that 292 * contains the source address. If the socket ends up being an IPv6 socket, 293 * udp_append() will convert to a sockaddr_in6 before passing the address 294 * into the socket code. 295 */ 296 static void 297 udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, 298 struct sockaddr_in *udp_in) 299 { 300 struct sockaddr *append_sa; 301 struct socket *so; 302 struct mbuf *opts = 0; 303 #ifdef INET6 304 struct sockaddr_in6 udp_in6; 305 #endif 306 struct udpcb *up; 307 308 INP_LOCK_ASSERT(inp); 309 310 /* 311 * Engage the tunneling protocol. 312 */ 313 up = intoudpcb(inp); 314 if (up->u_tun_func != NULL) { 315 (*up->u_tun_func)(n, off, inp, (struct sockaddr *)udp_in, 316 up->u_tun_ctx); 317 return; 318 } 319 320 off += sizeof(struct udphdr); 321 322 #ifdef IPSEC 323 /* Check AH/ESP integrity. */ 324 if (ipsec4_in_reject(n, inp)) { 325 m_freem(n); 326 IPSECSTAT_INC(ips_in_polvio); 327 return; 328 } 329 #ifdef IPSEC_NAT_T 330 up = intoudpcb(inp); 331 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 332 if (up->u_flags & UF_ESPINUDP_ALL) { /* IPSec UDP encaps. */ 333 n = udp4_espdecap(inp, n, off); 334 if (n == NULL) /* Consumed. */ 335 return; 336 } 337 #endif /* IPSEC_NAT_T */ 338 #endif /* IPSEC */ 339 #ifdef MAC 340 if (mac_inpcb_check_deliver(inp, n) != 0) { 341 m_freem(n); 342 return; 343 } 344 #endif /* MAC */ 345 if (inp->inp_flags & INP_CONTROLOPTS || 346 inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { 347 #ifdef INET6 348 if (inp->inp_vflag & INP_IPV6) 349 (void)ip6_savecontrol_v4(inp, n, &opts, NULL); 350 else 351 #endif /* INET6 */ 352 ip_savecontrol(inp, &opts, ip, n); 353 } 354 #ifdef INET6 355 if (inp->inp_vflag & INP_IPV6) { 356 bzero(&udp_in6, sizeof(udp_in6)); 357 udp_in6.sin6_len = sizeof(udp_in6); 358 udp_in6.sin6_family = AF_INET6; 359 in6_sin_2_v4mapsin6(udp_in, &udp_in6); 360 append_sa = (struct sockaddr *)&udp_in6; 361 } else 362 #endif /* INET6 */ 363 append_sa = (struct sockaddr *)udp_in; 364 m_adj(n, off); 365 366 so = inp->inp_socket; 367 SOCKBUF_LOCK(&so->so_rcv); 368 if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { 369 SOCKBUF_UNLOCK(&so->so_rcv); 370 m_freem(n); 371 if (opts) 372 m_freem(opts); 373 UDPSTAT_INC(udps_fullsock); 374 } else 375 sorwakeup_locked(so); 376 } 377 378 int 379 udp_input(struct mbuf **mp, int *offp, int proto) 380 { 381 struct ip *ip; 382 struct udphdr *uh; 383 struct ifnet *ifp; 384 struct inpcb *inp; 385 uint16_t len, ip_len; 386 struct inpcbinfo *pcbinfo; 387 struct ip save_ip; 388 struct sockaddr_in udp_in; 389 struct mbuf *m; 390 struct m_tag *fwd_tag; 391 int cscov_partial, iphlen; 392 393 m = *mp; 394 iphlen = *offp; 395 ifp = m->m_pkthdr.rcvif; 396 *mp = NULL; 397 UDPSTAT_INC(udps_ipackets); 398 399 /* 400 * Strip IP options, if any; should skip this, make available to 401 * user, and use on returned packets, but we don't yet have a way to 402 * check the checksum with options still present. 403 */ 404 if (iphlen > sizeof (struct ip)) { 405 ip_stripoptions(m); 406 iphlen = sizeof(struct ip); 407 } 408 409 /* 410 * Get IP and UDP header together in first mbuf. 411 */ 412 ip = mtod(m, struct ip *); 413 if (m->m_len < iphlen + sizeof(struct udphdr)) { 414 if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { 415 UDPSTAT_INC(udps_hdrops); 416 return (IPPROTO_DONE); 417 } 418 ip = mtod(m, struct ip *); 419 } 420 uh = (struct udphdr *)((caddr_t)ip + iphlen); 421 cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; 422 423 /* 424 * Destination port of 0 is illegal, based on RFC768. 425 */ 426 if (uh->uh_dport == 0) 427 goto badunlocked; 428 429 /* 430 * Construct sockaddr format source address. Stuff source address 431 * and datagram in user buffer. 432 */ 433 bzero(&udp_in, sizeof(udp_in)); 434 udp_in.sin_len = sizeof(udp_in); 435 udp_in.sin_family = AF_INET; 436 udp_in.sin_port = uh->uh_sport; 437 udp_in.sin_addr = ip->ip_src; 438 439 /* 440 * Make mbuf data length reflect UDP length. If not enough data to 441 * reflect UDP length, drop. 442 */ 443 len = ntohs((u_short)uh->uh_ulen); 444 ip_len = ntohs(ip->ip_len) - iphlen; 445 if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { 446 /* Zero means checksum over the complete packet. */ 447 if (len == 0) 448 len = ip_len; 449 cscov_partial = 0; 450 } 451 if (ip_len != len) { 452 if (len > ip_len || len < sizeof(struct udphdr)) { 453 UDPSTAT_INC(udps_badlen); 454 goto badunlocked; 455 } 456 if (proto == IPPROTO_UDP) 457 m_adj(m, len - ip_len); 458 } 459 460 /* 461 * Save a copy of the IP header in case we want restore it for 462 * sending an ICMP error message in response. 463 */ 464 if (!V_udp_blackhole) 465 save_ip = *ip; 466 else 467 memset(&save_ip, 0, sizeof(save_ip)); 468 469 /* 470 * Checksum extended UDP header and data. 471 */ 472 if (uh->uh_sum) { 473 u_short uh_sum; 474 475 if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && 476 !cscov_partial) { 477 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 478 uh_sum = m->m_pkthdr.csum_data; 479 else 480 uh_sum = in_pseudo(ip->ip_src.s_addr, 481 ip->ip_dst.s_addr, htonl((u_short)len + 482 m->m_pkthdr.csum_data + proto)); 483 uh_sum ^= 0xffff; 484 } else { 485 char b[9]; 486 487 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 488 bzero(((struct ipovly *)ip)->ih_x1, 9); 489 ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? 490 uh->uh_ulen : htons(ip_len); 491 uh_sum = in_cksum(m, len + sizeof (struct ip)); 492 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 493 } 494 if (uh_sum) { 495 UDPSTAT_INC(udps_badsum); 496 m_freem(m); 497 return (IPPROTO_DONE); 498 } 499 } else { 500 if (proto == IPPROTO_UDP) { 501 UDPSTAT_INC(udps_nosum); 502 } else { 503 /* UDPLite requires a checksum */ 504 /* XXX: What is the right UDPLite MIB counter here? */ 505 m_freem(m); 506 return (IPPROTO_DONE); 507 } 508 } 509 510 pcbinfo = get_inpcbinfo(proto); 511 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 512 in_broadcast(ip->ip_dst, ifp)) { 513 struct inpcb *last; 514 struct inpcbhead *pcblist; 515 struct ip_moptions *imo; 516 517 INP_INFO_RLOCK(pcbinfo); 518 pcblist = get_pcblist(proto); 519 last = NULL; 520 LIST_FOREACH(inp, pcblist, inp_list) { 521 if (inp->inp_lport != uh->uh_dport) 522 continue; 523 #ifdef INET6 524 if ((inp->inp_vflag & INP_IPV4) == 0) 525 continue; 526 #endif 527 if (inp->inp_laddr.s_addr != INADDR_ANY && 528 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 529 continue; 530 if (inp->inp_faddr.s_addr != INADDR_ANY && 531 inp->inp_faddr.s_addr != ip->ip_src.s_addr) 532 continue; 533 if (inp->inp_fport != 0 && 534 inp->inp_fport != uh->uh_sport) 535 continue; 536 537 INP_RLOCK(inp); 538 539 /* 540 * XXXRW: Because we weren't holding either the inpcb 541 * or the hash lock when we checked for a match 542 * before, we should probably recheck now that the 543 * inpcb lock is held. 544 */ 545 546 /* 547 * Handle socket delivery policy for any-source 548 * and source-specific multicast. [RFC3678] 549 */ 550 imo = inp->inp_moptions; 551 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 552 struct sockaddr_in group; 553 int blocked; 554 if (imo == NULL) { 555 INP_RUNLOCK(inp); 556 continue; 557 } 558 bzero(&group, sizeof(struct sockaddr_in)); 559 group.sin_len = sizeof(struct sockaddr_in); 560 group.sin_family = AF_INET; 561 group.sin_addr = ip->ip_dst; 562 563 blocked = imo_multi_filter(imo, ifp, 564 (struct sockaddr *)&group, 565 (struct sockaddr *)&udp_in); 566 if (blocked != MCAST_PASS) { 567 if (blocked == MCAST_NOTGMEMBER) 568 IPSTAT_INC(ips_notmember); 569 if (blocked == MCAST_NOTSMEMBER || 570 blocked == MCAST_MUTED) 571 UDPSTAT_INC(udps_filtermcast); 572 INP_RUNLOCK(inp); 573 continue; 574 } 575 } 576 if (last != NULL) { 577 struct mbuf *n; 578 579 if ((n = m_copy(m, 0, M_COPYALL)) != NULL) { 580 UDP_PROBE(receive, NULL, last, ip, 581 last, uh); 582 udp_append(last, ip, n, iphlen, 583 &udp_in); 584 } 585 INP_RUNLOCK(last); 586 } 587 last = inp; 588 /* 589 * Don't look for additional matches if this one does 590 * not have either the SO_REUSEPORT or SO_REUSEADDR 591 * socket options set. This heuristic avoids 592 * searching through all pcbs in the common case of a 593 * non-shared port. It assumes that an application 594 * will never clear these options after setting them. 595 */ 596 if ((last->inp_socket->so_options & 597 (SO_REUSEPORT|SO_REUSEADDR)) == 0) 598 break; 599 } 600 601 if (last == NULL) { 602 /* 603 * No matching pcb found; discard datagram. (No need 604 * to send an ICMP Port Unreachable for a broadcast 605 * or multicast datgram.) 606 */ 607 UDPSTAT_INC(udps_noportbcast); 608 if (inp) 609 INP_RUNLOCK(inp); 610 INP_INFO_RUNLOCK(pcbinfo); 611 goto badunlocked; 612 } 613 UDP_PROBE(receive, NULL, last, ip, last, uh); 614 udp_append(last, ip, m, iphlen, &udp_in); 615 INP_RUNLOCK(last); 616 INP_INFO_RUNLOCK(pcbinfo); 617 return (IPPROTO_DONE); 618 } 619 620 /* 621 * Locate pcb for datagram. 622 */ 623 624 /* 625 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 626 */ 627 if ((m->m_flags & M_IP_NEXTHOP) && 628 (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { 629 struct sockaddr_in *next_hop; 630 631 next_hop = (struct sockaddr_in *)(fwd_tag + 1); 632 633 /* 634 * Transparently forwarded. Pretend to be the destination. 635 * Already got one like this? 636 */ 637 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 638 ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); 639 if (!inp) { 640 /* 641 * It's new. Try to find the ambushing socket. 642 * Because we've rewritten the destination address, 643 * any hardware-generated hash is ignored. 644 */ 645 inp = in_pcblookup(pcbinfo, ip->ip_src, 646 uh->uh_sport, next_hop->sin_addr, 647 next_hop->sin_port ? htons(next_hop->sin_port) : 648 uh->uh_dport, INPLOOKUP_WILDCARD | 649 INPLOOKUP_RLOCKPCB, ifp); 650 } 651 /* Remove the tag from the packet. We don't need it anymore. */ 652 m_tag_delete(m, fwd_tag); 653 m->m_flags &= ~M_IP_NEXTHOP; 654 } else 655 inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, 656 ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | 657 INPLOOKUP_RLOCKPCB, ifp, m); 658 if (inp == NULL) { 659 if (udp_log_in_vain) { 660 char buf[4*sizeof "123"]; 661 662 strcpy(buf, inet_ntoa(ip->ip_dst)); 663 log(LOG_INFO, 664 "Connection attempt to UDP %s:%d from %s:%d\n", 665 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 666 ntohs(uh->uh_sport)); 667 } 668 UDPSTAT_INC(udps_noport); 669 if (m->m_flags & (M_BCAST | M_MCAST)) { 670 UDPSTAT_INC(udps_noportbcast); 671 goto badunlocked; 672 } 673 if (V_udp_blackhole) 674 goto badunlocked; 675 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 676 goto badunlocked; 677 *ip = save_ip; 678 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 679 return (IPPROTO_DONE); 680 } 681 682 /* 683 * Check the minimum TTL for socket. 684 */ 685 INP_RLOCK_ASSERT(inp); 686 if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { 687 INP_RUNLOCK(inp); 688 m_freem(m); 689 return (IPPROTO_DONE); 690 } 691 if (cscov_partial) { 692 struct udpcb *up; 693 694 up = intoudpcb(inp); 695 if (up->u_rxcslen == 0 || up->u_rxcslen > len) { 696 INP_RUNLOCK(inp); 697 m_freem(m); 698 return (IPPROTO_DONE); 699 } 700 } 701 702 UDP_PROBE(receive, NULL, inp, ip, inp, uh); 703 udp_append(inp, ip, m, iphlen, &udp_in); 704 INP_RUNLOCK(inp); 705 return (IPPROTO_DONE); 706 707 badunlocked: 708 m_freem(m); 709 return (IPPROTO_DONE); 710 } 711 #endif /* INET */ 712 713 /* 714 * Notify a udp user of an asynchronous error; just wake up so that they can 715 * collect error status. 716 */ 717 struct inpcb * 718 udp_notify(struct inpcb *inp, int errno) 719 { 720 721 /* 722 * While udp_ctlinput() always calls udp_notify() with a read lock 723 * when invoking it directly, in_pcbnotifyall() currently uses write 724 * locks due to sharing code with TCP. For now, accept either a read 725 * or a write lock, but a read lock is sufficient. 726 */ 727 INP_LOCK_ASSERT(inp); 728 729 inp->inp_socket->so_error = errno; 730 sorwakeup(inp->inp_socket); 731 sowwakeup(inp->inp_socket); 732 return (inp); 733 } 734 735 #ifdef INET 736 static void 737 udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, 738 struct inpcbinfo *pcbinfo) 739 { 740 struct ip *ip = vip; 741 struct udphdr *uh; 742 struct in_addr faddr; 743 struct inpcb *inp; 744 745 faddr = ((struct sockaddr_in *)sa)->sin_addr; 746 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 747 return; 748 749 /* 750 * Redirects don't need to be handled up here. 751 */ 752 if (PRC_IS_REDIRECT(cmd)) 753 return; 754 755 /* 756 * Hostdead is ugly because it goes linearly through all PCBs. 757 * 758 * XXX: We never get this from ICMP, otherwise it makes an excellent 759 * DoS attack on machines with many connections. 760 */ 761 if (cmd == PRC_HOSTDEAD) 762 ip = NULL; 763 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 764 return; 765 if (ip != NULL) { 766 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 767 inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, 768 ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); 769 if (inp != NULL) { 770 INP_RLOCK_ASSERT(inp); 771 if (inp->inp_socket != NULL) { 772 udp_notify(inp, inetctlerrmap[cmd]); 773 } 774 INP_RUNLOCK(inp); 775 } 776 } else 777 in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], 778 udp_notify); 779 } 780 void 781 udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 782 { 783 784 return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); 785 } 786 787 void 788 udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) 789 { 790 791 return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); 792 } 793 #endif /* INET */ 794 795 static int 796 udp_pcblist(SYSCTL_HANDLER_ARGS) 797 { 798 int error, i, n; 799 struct inpcb *inp, **inp_list; 800 inp_gen_t gencnt; 801 struct xinpgen xig; 802 803 /* 804 * The process of preparing the PCB list is too time-consuming and 805 * resource-intensive to repeat twice on every request. 806 */ 807 if (req->oldptr == 0) { 808 n = V_udbinfo.ipi_count; 809 n += imax(n / 8, 10); 810 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 811 return (0); 812 } 813 814 if (req->newptr != 0) 815 return (EPERM); 816 817 /* 818 * OK, now we're committed to doing something. 819 */ 820 INP_INFO_RLOCK(&V_udbinfo); 821 gencnt = V_udbinfo.ipi_gencnt; 822 n = V_udbinfo.ipi_count; 823 INP_INFO_RUNLOCK(&V_udbinfo); 824 825 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 826 + n * sizeof(struct xinpcb)); 827 if (error != 0) 828 return (error); 829 830 xig.xig_len = sizeof xig; 831 xig.xig_count = n; 832 xig.xig_gen = gencnt; 833 xig.xig_sogen = so_gencnt; 834 error = SYSCTL_OUT(req, &xig, sizeof xig); 835 if (error) 836 return (error); 837 838 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 839 if (inp_list == 0) 840 return (ENOMEM); 841 842 INP_INFO_RLOCK(&V_udbinfo); 843 for (inp = LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; 844 inp = LIST_NEXT(inp, inp_list)) { 845 INP_WLOCK(inp); 846 if (inp->inp_gencnt <= gencnt && 847 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 848 in_pcbref(inp); 849 inp_list[i++] = inp; 850 } 851 INP_WUNLOCK(inp); 852 } 853 INP_INFO_RUNLOCK(&V_udbinfo); 854 n = i; 855 856 error = 0; 857 for (i = 0; i < n; i++) { 858 inp = inp_list[i]; 859 INP_RLOCK(inp); 860 if (inp->inp_gencnt <= gencnt) { 861 struct xinpcb xi; 862 863 bzero(&xi, sizeof(xi)); 864 xi.xi_len = sizeof xi; 865 /* XXX should avoid extra copy */ 866 bcopy(inp, &xi.xi_inp, sizeof *inp); 867 if (inp->inp_socket) 868 sotoxsocket(inp->inp_socket, &xi.xi_socket); 869 xi.xi_inp.inp_gencnt = inp->inp_gencnt; 870 INP_RUNLOCK(inp); 871 error = SYSCTL_OUT(req, &xi, sizeof xi); 872 } else 873 INP_RUNLOCK(inp); 874 } 875 INP_INFO_WLOCK(&V_udbinfo); 876 for (i = 0; i < n; i++) { 877 inp = inp_list[i]; 878 INP_RLOCK(inp); 879 if (!in_pcbrele_rlocked(inp)) 880 INP_RUNLOCK(inp); 881 } 882 INP_INFO_WUNLOCK(&V_udbinfo); 883 884 if (!error) { 885 /* 886 * Give the user an updated idea of our state. If the 887 * generation differs from what we told her before, she knows 888 * that something happened while we were processing this 889 * request, and it might be necessary to retry. 890 */ 891 INP_INFO_RLOCK(&V_udbinfo); 892 xig.xig_gen = V_udbinfo.ipi_gencnt; 893 xig.xig_sogen = so_gencnt; 894 xig.xig_count = V_udbinfo.ipi_count; 895 INP_INFO_RUNLOCK(&V_udbinfo); 896 error = SYSCTL_OUT(req, &xig, sizeof xig); 897 } 898 free(inp_list, M_TEMP); 899 return (error); 900 } 901 902 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, 903 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 904 udp_pcblist, "S,xinpcb", "List of active UDP sockets"); 905 906 #ifdef INET 907 static int 908 udp_getcred(SYSCTL_HANDLER_ARGS) 909 { 910 struct xucred xuc; 911 struct sockaddr_in addrs[2]; 912 struct inpcb *inp; 913 int error; 914 915 error = priv_check(req->td, PRIV_NETINET_GETCRED); 916 if (error) 917 return (error); 918 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 919 if (error) 920 return (error); 921 inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, 922 addrs[0].sin_addr, addrs[0].sin_port, 923 INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); 924 if (inp != NULL) { 925 INP_RLOCK_ASSERT(inp); 926 if (inp->inp_socket == NULL) 927 error = ENOENT; 928 if (error == 0) 929 error = cr_canseeinpcb(req->td->td_ucred, inp); 930 if (error == 0) 931 cru2x(inp->inp_cred, &xuc); 932 INP_RUNLOCK(inp); 933 } else 934 error = ENOENT; 935 if (error == 0) 936 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 937 return (error); 938 } 939 940 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, 941 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 942 udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); 943 #endif /* INET */ 944 945 int 946 udp_ctloutput(struct socket *so, struct sockopt *sopt) 947 { 948 struct inpcb *inp; 949 struct udpcb *up; 950 int isudplite, error, optval; 951 952 error = 0; 953 isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; 954 inp = sotoinpcb(so); 955 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 956 INP_WLOCK(inp); 957 if (sopt->sopt_level != so->so_proto->pr_protocol) { 958 #ifdef INET6 959 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 960 INP_WUNLOCK(inp); 961 error = ip6_ctloutput(so, sopt); 962 } 963 #endif 964 #if defined(INET) && defined(INET6) 965 else 966 #endif 967 #ifdef INET 968 { 969 INP_WUNLOCK(inp); 970 error = ip_ctloutput(so, sopt); 971 } 972 #endif 973 return (error); 974 } 975 976 switch (sopt->sopt_dir) { 977 case SOPT_SET: 978 switch (sopt->sopt_name) { 979 case UDP_ENCAP: 980 INP_WUNLOCK(inp); 981 error = sooptcopyin(sopt, &optval, sizeof optval, 982 sizeof optval); 983 if (error) 984 break; 985 inp = sotoinpcb(so); 986 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 987 INP_WLOCK(inp); 988 #ifdef IPSEC_NAT_T 989 up = intoudpcb(inp); 990 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 991 #endif 992 switch (optval) { 993 case 0: 994 /* Clear all UDP encap. */ 995 #ifdef IPSEC_NAT_T 996 up->u_flags &= ~UF_ESPINUDP_ALL; 997 #endif 998 break; 999 #ifdef IPSEC_NAT_T 1000 case UDP_ENCAP_ESPINUDP: 1001 case UDP_ENCAP_ESPINUDP_NON_IKE: 1002 up->u_flags &= ~UF_ESPINUDP_ALL; 1003 if (optval == UDP_ENCAP_ESPINUDP) 1004 up->u_flags |= UF_ESPINUDP; 1005 else if (optval == UDP_ENCAP_ESPINUDP_NON_IKE) 1006 up->u_flags |= UF_ESPINUDP_NON_IKE; 1007 break; 1008 #endif 1009 default: 1010 error = EINVAL; 1011 break; 1012 } 1013 INP_WUNLOCK(inp); 1014 break; 1015 case UDPLITE_SEND_CSCOV: 1016 case UDPLITE_RECV_CSCOV: 1017 if (!isudplite) { 1018 INP_WUNLOCK(inp); 1019 error = ENOPROTOOPT; 1020 break; 1021 } 1022 INP_WUNLOCK(inp); 1023 error = sooptcopyin(sopt, &optval, sizeof(optval), 1024 sizeof(optval)); 1025 if (error != 0) 1026 break; 1027 inp = sotoinpcb(so); 1028 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 1029 INP_WLOCK(inp); 1030 up = intoudpcb(inp); 1031 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1032 if ((optval != 0 && optval < 8) || (optval > 65535)) { 1033 INP_WUNLOCK(inp); 1034 error = EINVAL; 1035 break; 1036 } 1037 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1038 up->u_txcslen = optval; 1039 else 1040 up->u_rxcslen = optval; 1041 INP_WUNLOCK(inp); 1042 break; 1043 default: 1044 INP_WUNLOCK(inp); 1045 error = ENOPROTOOPT; 1046 break; 1047 } 1048 break; 1049 case SOPT_GET: 1050 switch (sopt->sopt_name) { 1051 #ifdef IPSEC_NAT_T 1052 case UDP_ENCAP: 1053 up = intoudpcb(inp); 1054 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1055 optval = up->u_flags & UF_ESPINUDP_ALL; 1056 INP_WUNLOCK(inp); 1057 error = sooptcopyout(sopt, &optval, sizeof optval); 1058 break; 1059 #endif 1060 case UDPLITE_SEND_CSCOV: 1061 case UDPLITE_RECV_CSCOV: 1062 if (!isudplite) { 1063 INP_WUNLOCK(inp); 1064 error = ENOPROTOOPT; 1065 break; 1066 } 1067 up = intoudpcb(inp); 1068 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1069 if (sopt->sopt_name == UDPLITE_SEND_CSCOV) 1070 optval = up->u_txcslen; 1071 else 1072 optval = up->u_rxcslen; 1073 INP_WUNLOCK(inp); 1074 error = sooptcopyout(sopt, &optval, sizeof(optval)); 1075 break; 1076 default: 1077 INP_WUNLOCK(inp); 1078 error = ENOPROTOOPT; 1079 break; 1080 } 1081 break; 1082 } 1083 return (error); 1084 } 1085 1086 #ifdef INET 1087 #define UH_WLOCKED 2 1088 #define UH_RLOCKED 1 1089 #define UH_UNLOCKED 0 1090 static int 1091 udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, 1092 struct mbuf *control, struct thread *td) 1093 { 1094 struct udpiphdr *ui; 1095 int len = m->m_pkthdr.len; 1096 struct in_addr faddr, laddr; 1097 struct cmsghdr *cm; 1098 struct inpcbinfo *pcbinfo; 1099 struct sockaddr_in *sin, src; 1100 int cscov_partial = 0; 1101 int error = 0; 1102 int ipflags; 1103 u_short fport, lport; 1104 int unlock_udbinfo; 1105 u_char tos; 1106 uint8_t pr; 1107 uint16_t cscov = 0; 1108 uint32_t flowid = 0; 1109 int flowid_type = 0; 1110 int use_flowid = 0; 1111 1112 /* 1113 * udp_output() may need to temporarily bind or connect the current 1114 * inpcb. As such, we don't know up front whether we will need the 1115 * pcbinfo lock or not. Do any work to decide what is needed up 1116 * front before acquiring any locks. 1117 */ 1118 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 1119 if (control) 1120 m_freem(control); 1121 m_freem(m); 1122 return (EMSGSIZE); 1123 } 1124 1125 src.sin_family = 0; 1126 INP_RLOCK(inp); 1127 tos = inp->inp_ip_tos; 1128 if (control != NULL) { 1129 /* 1130 * XXX: Currently, we assume all the optional information is 1131 * stored in a single mbuf. 1132 */ 1133 if (control->m_next) { 1134 INP_RUNLOCK(inp); 1135 m_freem(control); 1136 m_freem(m); 1137 return (EINVAL); 1138 } 1139 for (; control->m_len > 0; 1140 control->m_data += CMSG_ALIGN(cm->cmsg_len), 1141 control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { 1142 cm = mtod(control, struct cmsghdr *); 1143 if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 1144 || cm->cmsg_len > control->m_len) { 1145 error = EINVAL; 1146 break; 1147 } 1148 if (cm->cmsg_level != IPPROTO_IP) 1149 continue; 1150 1151 switch (cm->cmsg_type) { 1152 case IP_SENDSRCADDR: 1153 if (cm->cmsg_len != 1154 CMSG_LEN(sizeof(struct in_addr))) { 1155 error = EINVAL; 1156 break; 1157 } 1158 bzero(&src, sizeof(src)); 1159 src.sin_family = AF_INET; 1160 src.sin_len = sizeof(src); 1161 src.sin_port = inp->inp_lport; 1162 src.sin_addr = 1163 *(struct in_addr *)CMSG_DATA(cm); 1164 break; 1165 1166 case IP_TOS: 1167 if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { 1168 error = EINVAL; 1169 break; 1170 } 1171 tos = *(u_char *)CMSG_DATA(cm); 1172 break; 1173 1174 case IP_FLOWID: 1175 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1176 error = EINVAL; 1177 break; 1178 } 1179 flowid = *(uint32_t *) CMSG_DATA(cm); 1180 break; 1181 1182 case IP_FLOWTYPE: 1183 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1184 error = EINVAL; 1185 break; 1186 } 1187 flowid_type = *(uint32_t *) CMSG_DATA(cm); 1188 use_flowid = 1; 1189 break; 1190 1191 #ifdef RSS 1192 case IP_RSSBUCKETID: 1193 if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { 1194 error = EINVAL; 1195 break; 1196 } 1197 /* This is just a placeholder for now */ 1198 break; 1199 #endif /* RSS */ 1200 default: 1201 error = ENOPROTOOPT; 1202 break; 1203 } 1204 if (error) 1205 break; 1206 } 1207 m_freem(control); 1208 } 1209 if (error) { 1210 INP_RUNLOCK(inp); 1211 m_freem(m); 1212 return (error); 1213 } 1214 1215 /* 1216 * Depending on whether or not the application has bound or connected 1217 * the socket, we may have to do varying levels of work. The optimal 1218 * case is for a connected UDP socket, as a global lock isn't 1219 * required at all. 1220 * 1221 * In order to decide which we need, we require stability of the 1222 * inpcb binding, which we ensure by acquiring a read lock on the 1223 * inpcb. This doesn't strictly follow the lock order, so we play 1224 * the trylock and retry game; note that we may end up with more 1225 * conservative locks than required the second time around, so later 1226 * assertions have to accept that. Further analysis of the number of 1227 * misses under contention is required. 1228 * 1229 * XXXRW: Check that hash locking update here is correct. 1230 */ 1231 pr = inp->inp_socket->so_proto->pr_protocol; 1232 pcbinfo = get_inpcbinfo(pr); 1233 sin = (struct sockaddr_in *)addr; 1234 if (sin != NULL && 1235 (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { 1236 INP_RUNLOCK(inp); 1237 INP_WLOCK(inp); 1238 INP_HASH_WLOCK(pcbinfo); 1239 unlock_udbinfo = UH_WLOCKED; 1240 } else if ((sin != NULL && ( 1241 (sin->sin_addr.s_addr == INADDR_ANY) || 1242 (sin->sin_addr.s_addr == INADDR_BROADCAST) || 1243 (inp->inp_laddr.s_addr == INADDR_ANY) || 1244 (inp->inp_lport == 0))) || 1245 (src.sin_family == AF_INET)) { 1246 INP_HASH_RLOCK(pcbinfo); 1247 unlock_udbinfo = UH_RLOCKED; 1248 } else 1249 unlock_udbinfo = UH_UNLOCKED; 1250 1251 /* 1252 * If the IP_SENDSRCADDR control message was specified, override the 1253 * source address for this datagram. Its use is invalidated if the 1254 * address thus specified is incomplete or clobbers other inpcbs. 1255 */ 1256 laddr = inp->inp_laddr; 1257 lport = inp->inp_lport; 1258 if (src.sin_family == AF_INET) { 1259 INP_HASH_LOCK_ASSERT(pcbinfo); 1260 if ((lport == 0) || 1261 (laddr.s_addr == INADDR_ANY && 1262 src.sin_addr.s_addr == INADDR_ANY)) { 1263 error = EINVAL; 1264 goto release; 1265 } 1266 error = in_pcbbind_setup(inp, (struct sockaddr *)&src, 1267 &laddr.s_addr, &lport, td->td_ucred); 1268 if (error) 1269 goto release; 1270 } 1271 1272 /* 1273 * If a UDP socket has been connected, then a local address/port will 1274 * have been selected and bound. 1275 * 1276 * If a UDP socket has not been connected to, then an explicit 1277 * destination address must be used, in which case a local 1278 * address/port may not have been selected and bound. 1279 */ 1280 if (sin != NULL) { 1281 INP_LOCK_ASSERT(inp); 1282 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1283 error = EISCONN; 1284 goto release; 1285 } 1286 1287 /* 1288 * Jail may rewrite the destination address, so let it do 1289 * that before we use it. 1290 */ 1291 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1292 if (error) 1293 goto release; 1294 1295 /* 1296 * If a local address or port hasn't yet been selected, or if 1297 * the destination address needs to be rewritten due to using 1298 * a special INADDR_ constant, invoke in_pcbconnect_setup() 1299 * to do the heavy lifting. Once a port is selected, we 1300 * commit the binding back to the socket; we also commit the 1301 * binding of the address if in jail. 1302 * 1303 * If we already have a valid binding and we're not 1304 * requesting a destination address rewrite, use a fast path. 1305 */ 1306 if (inp->inp_laddr.s_addr == INADDR_ANY || 1307 inp->inp_lport == 0 || 1308 sin->sin_addr.s_addr == INADDR_ANY || 1309 sin->sin_addr.s_addr == INADDR_BROADCAST) { 1310 INP_HASH_LOCK_ASSERT(pcbinfo); 1311 error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, 1312 &lport, &faddr.s_addr, &fport, NULL, 1313 td->td_ucred); 1314 if (error) 1315 goto release; 1316 1317 /* 1318 * XXXRW: Why not commit the port if the address is 1319 * !INADDR_ANY? 1320 */ 1321 /* Commit the local port if newly assigned. */ 1322 if (inp->inp_laddr.s_addr == INADDR_ANY && 1323 inp->inp_lport == 0) { 1324 INP_WLOCK_ASSERT(inp); 1325 INP_HASH_WLOCK_ASSERT(pcbinfo); 1326 /* 1327 * Remember addr if jailed, to prevent 1328 * rebinding. 1329 */ 1330 if (prison_flag(td->td_ucred, PR_IP4)) 1331 inp->inp_laddr = laddr; 1332 inp->inp_lport = lport; 1333 if (in_pcbinshash(inp) != 0) { 1334 inp->inp_lport = 0; 1335 error = EAGAIN; 1336 goto release; 1337 } 1338 inp->inp_flags |= INP_ANONPORT; 1339 } 1340 } else { 1341 faddr = sin->sin_addr; 1342 fport = sin->sin_port; 1343 } 1344 } else { 1345 INP_LOCK_ASSERT(inp); 1346 faddr = inp->inp_faddr; 1347 fport = inp->inp_fport; 1348 if (faddr.s_addr == INADDR_ANY) { 1349 error = ENOTCONN; 1350 goto release; 1351 } 1352 } 1353 1354 /* 1355 * Calculate data length and get a mbuf for UDP, IP, and possible 1356 * link-layer headers. Immediate slide the data pointer back forward 1357 * since we won't use that space at this layer. 1358 */ 1359 M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); 1360 if (m == NULL) { 1361 error = ENOBUFS; 1362 goto release; 1363 } 1364 m->m_data += max_linkhdr; 1365 m->m_len -= max_linkhdr; 1366 m->m_pkthdr.len -= max_linkhdr; 1367 1368 /* 1369 * Fill in mbuf with extended UDP header and addresses and length put 1370 * into network format. 1371 */ 1372 ui = mtod(m, struct udpiphdr *); 1373 bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ 1374 ui->ui_pr = pr; 1375 ui->ui_src = laddr; 1376 ui->ui_dst = faddr; 1377 ui->ui_sport = lport; 1378 ui->ui_dport = fport; 1379 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1380 if (pr == IPPROTO_UDPLITE) { 1381 struct udpcb *up; 1382 uint16_t plen; 1383 1384 up = intoudpcb(inp); 1385 cscov = up->u_txcslen; 1386 plen = (u_short)len + sizeof(struct udphdr); 1387 if (cscov >= plen) 1388 cscov = 0; 1389 ui->ui_len = htons(plen); 1390 ui->ui_ulen = htons(cscov); 1391 /* 1392 * For UDP-Lite, checksum coverage length of zero means 1393 * the entire UDPLite packet is covered by the checksum. 1394 */ 1395 cscov_partial = (cscov == 0) ? 0 : 1; 1396 } else 1397 ui->ui_v = IPVERSION << 4; 1398 1399 /* 1400 * Set the Don't Fragment bit in the IP header. 1401 */ 1402 if (inp->inp_flags & INP_DONTFRAG) { 1403 struct ip *ip; 1404 1405 ip = (struct ip *)&ui->ui_i; 1406 ip->ip_off |= htons(IP_DF); 1407 } 1408 1409 ipflags = 0; 1410 if (inp->inp_socket->so_options & SO_DONTROUTE) 1411 ipflags |= IP_ROUTETOIF; 1412 if (inp->inp_socket->so_options & SO_BROADCAST) 1413 ipflags |= IP_ALLOWBROADCAST; 1414 if (inp->inp_flags & INP_ONESBCAST) 1415 ipflags |= IP_SENDONES; 1416 1417 #ifdef MAC 1418 mac_inpcb_create_mbuf(inp, m); 1419 #endif 1420 1421 /* 1422 * Set up checksum and output datagram. 1423 */ 1424 ui->ui_sum = 0; 1425 if (pr == IPPROTO_UDPLITE) { 1426 if (inp->inp_flags & INP_ONESBCAST) 1427 faddr.s_addr = INADDR_BROADCAST; 1428 if (cscov_partial) { 1429 if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) 1430 ui->ui_sum = 0xffff; 1431 } else { 1432 if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) 1433 ui->ui_sum = 0xffff; 1434 } 1435 } else if (V_udp_cksum) { 1436 if (inp->inp_flags & INP_ONESBCAST) 1437 faddr.s_addr = INADDR_BROADCAST; 1438 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, 1439 htons((u_short)len + sizeof(struct udphdr) + pr)); 1440 m->m_pkthdr.csum_flags = CSUM_UDP; 1441 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1442 } 1443 ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); 1444 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1445 ((struct ip *)ui)->ip_tos = tos; /* XXX */ 1446 UDPSTAT_INC(udps_opackets); 1447 1448 /* 1449 * Setup flowid / RSS information for outbound socket. 1450 * 1451 * Once the UDP code decides to set a flowid some other way, 1452 * this allows the flowid to be overridden by userland. 1453 */ 1454 if (use_flowid) { 1455 m->m_flags |= M_FLOWID; 1456 m->m_pkthdr.flowid = flowid; 1457 M_HASHTYPE_SET(m, flowid_type); 1458 #ifdef RSS 1459 } else { 1460 uint32_t hash_val, hash_type; 1461 /* 1462 * Calculate an appropriate RSS hash for UDP and 1463 * UDP Lite. 1464 * 1465 * The called function will take care of figuring out 1466 * whether a 2-tuple or 4-tuple hash is required based 1467 * on the currently configured scheme. 1468 * 1469 * Later later on connected socket values should be 1470 * cached in the inpcb and reused, rather than constantly 1471 * re-calculating it. 1472 * 1473 * UDP Lite is a different protocol number and will 1474 * likely end up being hashed as a 2-tuple until 1475 * RSS / NICs grow UDP Lite protocol awareness. 1476 */ 1477 if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, 1478 pr, &hash_val, &hash_type) == 0) { 1479 m->m_pkthdr.flowid = hash_val; 1480 m->m_flags |= M_FLOWID; 1481 M_HASHTYPE_SET(m, hash_type); 1482 } 1483 #endif 1484 } 1485 1486 #ifdef RSS 1487 /* 1488 * Don't override with the inp cached flowid value. 1489 * 1490 * Depending upon the kind of send being done, the inp 1491 * flowid/flowtype values may actually not be appropriate 1492 * for this particular socket send. 1493 * 1494 * We should either leave the flowid at zero (which is what is 1495 * currently done) or set it to some software generated 1496 * hash value based on the packet contents. 1497 */ 1498 ipflags |= IP_NODEFAULTFLOWID; 1499 #endif /* RSS */ 1500 1501 if (unlock_udbinfo == UH_WLOCKED) 1502 INP_HASH_WUNLOCK(pcbinfo); 1503 else if (unlock_udbinfo == UH_RLOCKED) 1504 INP_HASH_RUNLOCK(pcbinfo); 1505 UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); 1506 error = ip_output(m, inp->inp_options, NULL, ipflags, 1507 inp->inp_moptions, inp); 1508 if (unlock_udbinfo == UH_WLOCKED) 1509 INP_WUNLOCK(inp); 1510 else 1511 INP_RUNLOCK(inp); 1512 return (error); 1513 1514 release: 1515 if (unlock_udbinfo == UH_WLOCKED) { 1516 INP_HASH_WUNLOCK(pcbinfo); 1517 INP_WUNLOCK(inp); 1518 } else if (unlock_udbinfo == UH_RLOCKED) { 1519 INP_HASH_RUNLOCK(pcbinfo); 1520 INP_RUNLOCK(inp); 1521 } else 1522 INP_RUNLOCK(inp); 1523 m_freem(m); 1524 return (error); 1525 } 1526 1527 1528 #if defined(IPSEC) && defined(IPSEC_NAT_T) 1529 /* 1530 * Potentially decap ESP in UDP frame. Check for an ESP header 1531 * and optional marker; if present, strip the UDP header and 1532 * push the result through IPSec. 1533 * 1534 * Returns mbuf to be processed (potentially re-allocated) or 1535 * NULL if consumed and/or processed. 1536 */ 1537 static struct mbuf * 1538 udp4_espdecap(struct inpcb *inp, struct mbuf *m, int off) 1539 { 1540 size_t minlen, payload, skip, iphlen; 1541 caddr_t data; 1542 struct udpcb *up; 1543 struct m_tag *tag; 1544 struct udphdr *udphdr; 1545 struct ip *ip; 1546 1547 INP_RLOCK_ASSERT(inp); 1548 1549 /* 1550 * Pull up data so the longest case is contiguous: 1551 * IP/UDP hdr + non ESP marker + ESP hdr. 1552 */ 1553 minlen = off + sizeof(uint64_t) + sizeof(struct esp); 1554 if (minlen > m->m_pkthdr.len) 1555 minlen = m->m_pkthdr.len; 1556 if ((m = m_pullup(m, minlen)) == NULL) { 1557 IPSECSTAT_INC(ips_in_inval); 1558 return (NULL); /* Bypass caller processing. */ 1559 } 1560 data = mtod(m, caddr_t); /* Points to ip header. */ 1561 payload = m->m_len - off; /* Size of payload. */ 1562 1563 if (payload == 1 && data[off] == '\xff') 1564 return (m); /* NB: keepalive packet, no decap. */ 1565 1566 up = intoudpcb(inp); 1567 KASSERT(up != NULL, ("%s: udpcb NULL", __func__)); 1568 KASSERT((up->u_flags & UF_ESPINUDP_ALL) != 0, 1569 ("u_flags 0x%x", up->u_flags)); 1570 1571 /* 1572 * Check that the payload is large enough to hold an 1573 * ESP header and compute the amount of data to remove. 1574 * 1575 * NB: the caller has already done a pullup for us. 1576 * XXX can we assume alignment and eliminate bcopys? 1577 */ 1578 if (up->u_flags & UF_ESPINUDP_NON_IKE) { 1579 /* 1580 * draft-ietf-ipsec-nat-t-ike-0[01].txt and 1581 * draft-ietf-ipsec-udp-encaps-(00/)01.txt, ignoring 1582 * possible AH mode non-IKE marker+non-ESP marker 1583 * from draft-ietf-ipsec-udp-encaps-00.txt. 1584 */ 1585 uint64_t marker; 1586 1587 if (payload <= sizeof(uint64_t) + sizeof(struct esp)) 1588 return (m); /* NB: no decap. */ 1589 bcopy(data + off, &marker, sizeof(uint64_t)); 1590 if (marker != 0) /* Non-IKE marker. */ 1591 return (m); /* NB: no decap. */ 1592 skip = sizeof(uint64_t) + sizeof(struct udphdr); 1593 } else { 1594 uint32_t spi; 1595 1596 if (payload <= sizeof(struct esp)) { 1597 IPSECSTAT_INC(ips_in_inval); 1598 m_freem(m); 1599 return (NULL); /* Discard. */ 1600 } 1601 bcopy(data + off, &spi, sizeof(uint32_t)); 1602 if (spi == 0) /* Non-ESP marker. */ 1603 return (m); /* NB: no decap. */ 1604 skip = sizeof(struct udphdr); 1605 } 1606 1607 /* 1608 * Setup a PACKET_TAG_IPSEC_NAT_T_PORT tag to remember 1609 * the UDP ports. This is required if we want to select 1610 * the right SPD for multiple hosts behind same NAT. 1611 * 1612 * NB: ports are maintained in network byte order everywhere 1613 * in the NAT-T code. 1614 */ 1615 tag = m_tag_get(PACKET_TAG_IPSEC_NAT_T_PORTS, 1616 2 * sizeof(uint16_t), M_NOWAIT); 1617 if (tag == NULL) { 1618 IPSECSTAT_INC(ips_in_nomem); 1619 m_freem(m); 1620 return (NULL); /* Discard. */ 1621 } 1622 iphlen = off - sizeof(struct udphdr); 1623 udphdr = (struct udphdr *)(data + iphlen); 1624 ((uint16_t *)(tag + 1))[0] = udphdr->uh_sport; 1625 ((uint16_t *)(tag + 1))[1] = udphdr->uh_dport; 1626 m_tag_prepend(m, tag); 1627 1628 /* 1629 * Remove the UDP header (and possibly the non ESP marker) 1630 * IP header length is iphlen 1631 * Before: 1632 * <--- off ---> 1633 * +----+------+-----+ 1634 * | IP | UDP | ESP | 1635 * +----+------+-----+ 1636 * <-skip-> 1637 * After: 1638 * +----+-----+ 1639 * | IP | ESP | 1640 * +----+-----+ 1641 * <-skip-> 1642 */ 1643 ovbcopy(data, data + skip, iphlen); 1644 m_adj(m, skip); 1645 1646 ip = mtod(m, struct ip *); 1647 ip->ip_len = htons(ntohs(ip->ip_len) - skip); 1648 ip->ip_p = IPPROTO_ESP; 1649 1650 /* 1651 * We cannot yet update the cksums so clear any 1652 * h/w cksum flags as they are no longer valid. 1653 */ 1654 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) 1655 m->m_pkthdr.csum_flags &= ~(CSUM_DATA_VALID|CSUM_PSEUDO_HDR); 1656 1657 (void) ipsec4_common_input(m, iphlen, ip->ip_p); 1658 return (NULL); /* NB: consumed, bypass processing. */ 1659 } 1660 #endif /* defined(IPSEC) && defined(IPSEC_NAT_T) */ 1661 1662 static void 1663 udp_abort(struct socket *so) 1664 { 1665 struct inpcb *inp; 1666 struct inpcbinfo *pcbinfo; 1667 1668 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1669 inp = sotoinpcb(so); 1670 KASSERT(inp != NULL, ("udp_abort: inp == NULL")); 1671 INP_WLOCK(inp); 1672 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1673 INP_HASH_WLOCK(pcbinfo); 1674 in_pcbdisconnect(inp); 1675 inp->inp_laddr.s_addr = INADDR_ANY; 1676 INP_HASH_WUNLOCK(pcbinfo); 1677 soisdisconnected(so); 1678 } 1679 INP_WUNLOCK(inp); 1680 } 1681 1682 static int 1683 udp_attach(struct socket *so, int proto, struct thread *td) 1684 { 1685 struct inpcb *inp; 1686 struct inpcbinfo *pcbinfo; 1687 int error; 1688 1689 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1690 inp = sotoinpcb(so); 1691 KASSERT(inp == NULL, ("udp_attach: inp != NULL")); 1692 error = soreserve(so, udp_sendspace, udp_recvspace); 1693 if (error) 1694 return (error); 1695 INP_INFO_WLOCK(pcbinfo); 1696 error = in_pcballoc(so, pcbinfo); 1697 if (error) { 1698 INP_INFO_WUNLOCK(pcbinfo); 1699 return (error); 1700 } 1701 1702 inp = sotoinpcb(so); 1703 inp->inp_vflag |= INP_IPV4; 1704 inp->inp_ip_ttl = V_ip_defttl; 1705 1706 error = udp_newudpcb(inp); 1707 if (error) { 1708 in_pcbdetach(inp); 1709 in_pcbfree(inp); 1710 INP_INFO_WUNLOCK(pcbinfo); 1711 return (error); 1712 } 1713 1714 INP_WUNLOCK(inp); 1715 INP_INFO_WUNLOCK(pcbinfo); 1716 return (0); 1717 } 1718 #endif /* INET */ 1719 1720 int 1721 udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, void *ctx) 1722 { 1723 struct inpcb *inp; 1724 struct udpcb *up; 1725 1726 KASSERT(so->so_type == SOCK_DGRAM, 1727 ("udp_set_kernel_tunneling: !dgram")); 1728 inp = sotoinpcb(so); 1729 KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); 1730 INP_WLOCK(inp); 1731 up = intoudpcb(inp); 1732 if (up->u_tun_func != NULL) { 1733 INP_WUNLOCK(inp); 1734 return (EBUSY); 1735 } 1736 up->u_tun_func = f; 1737 up->u_tun_ctx = ctx; 1738 INP_WUNLOCK(inp); 1739 return (0); 1740 } 1741 1742 #ifdef INET 1743 static int 1744 udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 1745 { 1746 struct inpcb *inp; 1747 struct inpcbinfo *pcbinfo; 1748 int error; 1749 1750 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1751 inp = sotoinpcb(so); 1752 KASSERT(inp != NULL, ("udp_bind: inp == NULL")); 1753 INP_WLOCK(inp); 1754 INP_HASH_WLOCK(pcbinfo); 1755 error = in_pcbbind(inp, nam, td->td_ucred); 1756 INP_HASH_WUNLOCK(pcbinfo); 1757 INP_WUNLOCK(inp); 1758 return (error); 1759 } 1760 1761 static void 1762 udp_close(struct socket *so) 1763 { 1764 struct inpcb *inp; 1765 struct inpcbinfo *pcbinfo; 1766 1767 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1768 inp = sotoinpcb(so); 1769 KASSERT(inp != NULL, ("udp_close: inp == NULL")); 1770 INP_WLOCK(inp); 1771 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1772 INP_HASH_WLOCK(pcbinfo); 1773 in_pcbdisconnect(inp); 1774 inp->inp_laddr.s_addr = INADDR_ANY; 1775 INP_HASH_WUNLOCK(pcbinfo); 1776 soisdisconnected(so); 1777 } 1778 INP_WUNLOCK(inp); 1779 } 1780 1781 static int 1782 udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 1783 { 1784 struct inpcb *inp; 1785 struct inpcbinfo *pcbinfo; 1786 struct sockaddr_in *sin; 1787 int error; 1788 1789 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1790 inp = sotoinpcb(so); 1791 KASSERT(inp != NULL, ("udp_connect: inp == NULL")); 1792 INP_WLOCK(inp); 1793 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1794 INP_WUNLOCK(inp); 1795 return (EISCONN); 1796 } 1797 sin = (struct sockaddr_in *)nam; 1798 error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); 1799 if (error != 0) { 1800 INP_WUNLOCK(inp); 1801 return (error); 1802 } 1803 INP_HASH_WLOCK(pcbinfo); 1804 error = in_pcbconnect(inp, nam, td->td_ucred); 1805 INP_HASH_WUNLOCK(pcbinfo); 1806 if (error == 0) 1807 soisconnected(so); 1808 INP_WUNLOCK(inp); 1809 return (error); 1810 } 1811 1812 static void 1813 udp_detach(struct socket *so) 1814 { 1815 struct inpcb *inp; 1816 struct inpcbinfo *pcbinfo; 1817 struct udpcb *up; 1818 1819 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1820 inp = sotoinpcb(so); 1821 KASSERT(inp != NULL, ("udp_detach: inp == NULL")); 1822 KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, 1823 ("udp_detach: not disconnected")); 1824 INP_INFO_WLOCK(pcbinfo); 1825 INP_WLOCK(inp); 1826 up = intoudpcb(inp); 1827 KASSERT(up != NULL, ("%s: up == NULL", __func__)); 1828 inp->inp_ppcb = NULL; 1829 in_pcbdetach(inp); 1830 in_pcbfree(inp); 1831 INP_INFO_WUNLOCK(pcbinfo); 1832 udp_discardcb(up); 1833 } 1834 1835 static int 1836 udp_disconnect(struct socket *so) 1837 { 1838 struct inpcb *inp; 1839 struct inpcbinfo *pcbinfo; 1840 1841 pcbinfo = get_inpcbinfo(so->so_proto->pr_protocol); 1842 inp = sotoinpcb(so); 1843 KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); 1844 INP_WLOCK(inp); 1845 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1846 INP_WUNLOCK(inp); 1847 return (ENOTCONN); 1848 } 1849 INP_HASH_WLOCK(pcbinfo); 1850 in_pcbdisconnect(inp); 1851 inp->inp_laddr.s_addr = INADDR_ANY; 1852 INP_HASH_WUNLOCK(pcbinfo); 1853 SOCK_LOCK(so); 1854 so->so_state &= ~SS_ISCONNECTED; /* XXX */ 1855 SOCK_UNLOCK(so); 1856 INP_WUNLOCK(inp); 1857 return (0); 1858 } 1859 1860 static int 1861 udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, 1862 struct mbuf *control, struct thread *td) 1863 { 1864 struct inpcb *inp; 1865 1866 inp = sotoinpcb(so); 1867 KASSERT(inp != NULL, ("udp_send: inp == NULL")); 1868 return (udp_output(inp, m, addr, control, td)); 1869 } 1870 #endif /* INET */ 1871 1872 int 1873 udp_shutdown(struct socket *so) 1874 { 1875 struct inpcb *inp; 1876 1877 inp = sotoinpcb(so); 1878 KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); 1879 INP_WLOCK(inp); 1880 socantsendmore(so); 1881 INP_WUNLOCK(inp); 1882 return (0); 1883 } 1884 1885 #ifdef INET 1886 struct pr_usrreqs udp_usrreqs = { 1887 .pru_abort = udp_abort, 1888 .pru_attach = udp_attach, 1889 .pru_bind = udp_bind, 1890 .pru_connect = udp_connect, 1891 .pru_control = in_control, 1892 .pru_detach = udp_detach, 1893 .pru_disconnect = udp_disconnect, 1894 .pru_peeraddr = in_getpeeraddr, 1895 .pru_send = udp_send, 1896 .pru_soreceive = soreceive_dgram, 1897 .pru_sosend = sosend_dgram, 1898 .pru_shutdown = udp_shutdown, 1899 .pru_sockaddr = in_getsockaddr, 1900 .pru_sosetlabel = in_pcbsosetlabel, 1901 .pru_close = udp_close, 1902 }; 1903 #endif /* INET */ 1904