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