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