in_pcb.c revision 309108
1/*- 2 * Copyright (c) 1982, 1986, 1991, 1993, 1995 3 * The Regents of the University of California. 4 * Copyright (c) 2007-2009 Robert N. M. Watson 5 * Copyright (c) 2010-2011 Juniper Networks, Inc. 6 * All rights reserved. 7 * 8 * Portions of this software were developed by Robert N. M. Watson under 9 * contract to Juniper Networks, Inc. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 4. Neither the name of the University nor the names of its contributors 20 * may be used to endorse or promote products derived from this software 21 * without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 36 */ 37 38#include <sys/cdefs.h> 39__FBSDID("$FreeBSD: stable/10/sys/netinet/in_pcb.c 309108 2016-11-24 14:48:46Z jch $"); 40 41#include "opt_ddb.h" 42#include "opt_ipsec.h" 43#include "opt_inet.h" 44#include "opt_inet6.h" 45#include "opt_pcbgroup.h" 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/malloc.h> 50#include <sys/mbuf.h> 51#include <sys/callout.h> 52#include <sys/domain.h> 53#include <sys/protosw.h> 54#include <sys/socket.h> 55#include <sys/socketvar.h> 56#include <sys/priv.h> 57#include <sys/proc.h> 58#include <sys/refcount.h> 59#include <sys/jail.h> 60#include <sys/kernel.h> 61#include <sys/sysctl.h> 62 63#ifdef DDB 64#include <ddb/ddb.h> 65#endif 66 67#include <vm/uma.h> 68 69#include <net/if.h> 70#include <net/if_types.h> 71#include <net/route.h> 72#include <net/vnet.h> 73 74#if defined(INET) || defined(INET6) 75#include <netinet/in.h> 76#include <netinet/in_pcb.h> 77#include <netinet/ip_var.h> 78#include <netinet/tcp_var.h> 79#include <netinet/udp.h> 80#include <netinet/udp_var.h> 81#endif 82#ifdef INET 83#include <netinet/in_var.h> 84#endif 85#ifdef INET6 86#include <netinet/ip6.h> 87#include <netinet6/in6_pcb.h> 88#include <netinet6/in6_var.h> 89#include <netinet6/ip6_var.h> 90#endif /* INET6 */ 91 92 93#ifdef IPSEC 94#include <netipsec/ipsec.h> 95#include <netipsec/key.h> 96#endif /* IPSEC */ 97 98#include <security/mac/mac_framework.h> 99 100static struct callout ipport_tick_callout; 101 102/* 103 * These configure the range of local port addresses assigned to 104 * "unspecified" outgoing connections/packets/whatever. 105 */ 106VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 107VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 108VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 109VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 110VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 111VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 112 113/* 114 * Reserved ports accessible only to root. There are significant 115 * security considerations that must be accounted for when changing these, 116 * but the security benefits can be great. Please be careful. 117 */ 118VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 119VNET_DEFINE(int, ipport_reservedlow); 120 121/* Variables dealing with random ephemeral port allocation. */ 122VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ 123VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ 124VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ 125VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ 126VNET_DEFINE(int, ipport_tcpallocs); 127static VNET_DEFINE(int, ipport_tcplastcount); 128 129#define V_ipport_tcplastcount VNET(ipport_tcplastcount) 130 131static void in_pcbremlists(struct inpcb *inp); 132#ifdef INET 133static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 134 struct in_addr faddr, u_int fport_arg, 135 struct in_addr laddr, u_int lport_arg, 136 int lookupflags, struct ifnet *ifp); 137 138#define RANGECHK(var, min, max) \ 139 if ((var) < (min)) { (var) = (min); } \ 140 else if ((var) > (max)) { (var) = (max); } 141 142static int 143sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 144{ 145 int error; 146 147 error = sysctl_handle_int(oidp, arg1, arg2, req); 148 if (error == 0) { 149 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 150 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 151 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 152 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 153 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 154 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 155 } 156 return (error); 157} 158 159#undef RANGECHK 160 161static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, 162 "IP Ports"); 163 164SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 165 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowfirstauto), 0, 166 &sysctl_net_ipport_check, "I", ""); 167SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 168 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lowlastauto), 0, 169 &sysctl_net_ipport_check, "I", ""); 170SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, first, 171 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_firstauto), 0, 172 &sysctl_net_ipport_check, "I", ""); 173SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, last, 174 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_lastauto), 0, 175 &sysctl_net_ipport_check, "I", ""); 176SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 177 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hifirstauto), 0, 178 &sysctl_net_ipport_check, "I", ""); 179SYSCTL_VNET_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 180 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(ipport_hilastauto), 0, 181 &sysctl_net_ipport_check, "I", ""); 182SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 183 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedhigh), 0, ""); 184SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 185 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 186SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_RW, 187 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 188SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_RW, 189 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " 190 "allocations before switching to a sequental one"); 191SYSCTL_VNET_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_RW, 192 &VNET_NAME(ipport_randomtime), 0, 193 "Minimum time to keep sequental port " 194 "allocation before switching to a random one"); 195#endif /* INET */ 196 197/* 198 * in_pcb.c: manage the Protocol Control Blocks. 199 * 200 * NOTE: It is assumed that most of these functions will be called with 201 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 202 * functions often modify hash chains or addresses in pcbs. 203 */ 204 205/* 206 * Initialize an inpcbinfo -- we should be able to reduce the number of 207 * arguments in time. 208 */ 209void 210in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 211 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 212 char *inpcbzone_name, uma_init inpcbzone_init, uma_fini inpcbzone_fini, 213 uint32_t inpcbzone_flags, u_int hashfields) 214{ 215 216 INP_INFO_LOCK_INIT(pcbinfo, name); 217 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 218 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 219#ifdef VIMAGE 220 pcbinfo->ipi_vnet = curvnet; 221#endif 222 pcbinfo->ipi_listhead = listhead; 223 LIST_INIT(pcbinfo->ipi_listhead); 224 pcbinfo->ipi_count = 0; 225 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 226 &pcbinfo->ipi_hashmask); 227 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 228 &pcbinfo->ipi_porthashmask); 229#ifdef PCBGROUP 230 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 231#endif 232 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 233 NULL, NULL, inpcbzone_init, inpcbzone_fini, UMA_ALIGN_PTR, 234 inpcbzone_flags); 235 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 236 uma_zone_set_warning(pcbinfo->ipi_zone, 237 "kern.ipc.maxsockets limit reached"); 238} 239 240/* 241 * Destroy an inpcbinfo. 242 */ 243void 244in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 245{ 246 247 KASSERT(pcbinfo->ipi_count == 0, 248 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 249 250 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 251 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 252 pcbinfo->ipi_porthashmask); 253#ifdef PCBGROUP 254 in_pcbgroup_destroy(pcbinfo); 255#endif 256 uma_zdestroy(pcbinfo->ipi_zone); 257 INP_LIST_LOCK_DESTROY(pcbinfo); 258 INP_HASH_LOCK_DESTROY(pcbinfo); 259 INP_INFO_LOCK_DESTROY(pcbinfo); 260} 261 262/* 263 * Allocate a PCB and associate it with the socket. 264 * On success return with the PCB locked. 265 */ 266int 267in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 268{ 269 struct inpcb *inp; 270 int error; 271 272#ifdef INVARIANTS 273 if (pcbinfo == &V_tcbinfo) { 274 INP_INFO_RLOCK_ASSERT(pcbinfo); 275 } else { 276 INP_INFO_WLOCK_ASSERT(pcbinfo); 277 } 278#endif 279 280 error = 0; 281 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 282 if (inp == NULL) 283 return (ENOBUFS); 284 bzero(inp, inp_zero_size); 285 inp->inp_pcbinfo = pcbinfo; 286 inp->inp_socket = so; 287 inp->inp_cred = crhold(so->so_cred); 288 inp->inp_inc.inc_fibnum = so->so_fibnum; 289#ifdef MAC 290 error = mac_inpcb_init(inp, M_NOWAIT); 291 if (error != 0) 292 goto out; 293 mac_inpcb_create(so, inp); 294#endif 295#ifdef IPSEC 296 error = ipsec_init_policy(so, &inp->inp_sp); 297 if (error != 0) { 298#ifdef MAC 299 mac_inpcb_destroy(inp); 300#endif 301 goto out; 302 } 303#endif /*IPSEC*/ 304#ifdef INET6 305 if (INP_SOCKAF(so) == AF_INET6) { 306 inp->inp_vflag |= INP_IPV6PROTO; 307 if (V_ip6_v6only) 308 inp->inp_flags |= IN6P_IPV6_V6ONLY; 309 } 310#endif 311 INP_WLOCK(inp); 312 INP_LIST_WLOCK(pcbinfo); 313 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 314 pcbinfo->ipi_count++; 315 so->so_pcb = (caddr_t)inp; 316#ifdef INET6 317 if (V_ip6_auto_flowlabel) 318 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 319#endif 320 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 321 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 322 INP_LIST_WUNLOCK(pcbinfo); 323#if defined(IPSEC) || defined(MAC) 324out: 325 if (error != 0) { 326 crfree(inp->inp_cred); 327 uma_zfree(pcbinfo->ipi_zone, inp); 328 } 329#endif 330 return (error); 331} 332 333#ifdef INET 334int 335in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 336{ 337 int anonport, error; 338 339 INP_WLOCK_ASSERT(inp); 340 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 341 342 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 343 return (EINVAL); 344 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 345 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 346 &inp->inp_lport, cred); 347 if (error) 348 return (error); 349 if (in_pcbinshash(inp) != 0) { 350 inp->inp_laddr.s_addr = INADDR_ANY; 351 inp->inp_lport = 0; 352 return (EAGAIN); 353 } 354 if (anonport) 355 inp->inp_flags |= INP_ANONPORT; 356 return (0); 357} 358#endif 359 360#if defined(INET) || defined(INET6) 361int 362in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 363 struct ucred *cred, int lookupflags) 364{ 365 struct inpcbinfo *pcbinfo; 366 struct inpcb *tmpinp; 367 unsigned short *lastport; 368 int count, dorandom, error; 369 u_short aux, first, last, lport; 370#ifdef INET 371 struct in_addr laddr; 372#endif 373 374 pcbinfo = inp->inp_pcbinfo; 375 376 /* 377 * Because no actual state changes occur here, a global write lock on 378 * the pcbinfo isn't required. 379 */ 380 INP_LOCK_ASSERT(inp); 381 INP_HASH_LOCK_ASSERT(pcbinfo); 382 383 if (inp->inp_flags & INP_HIGHPORT) { 384 first = V_ipport_hifirstauto; /* sysctl */ 385 last = V_ipport_hilastauto; 386 lastport = &pcbinfo->ipi_lasthi; 387 } else if (inp->inp_flags & INP_LOWPORT) { 388 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0); 389 if (error) 390 return (error); 391 first = V_ipport_lowfirstauto; /* 1023 */ 392 last = V_ipport_lowlastauto; /* 600 */ 393 lastport = &pcbinfo->ipi_lastlow; 394 } else { 395 first = V_ipport_firstauto; /* sysctl */ 396 last = V_ipport_lastauto; 397 lastport = &pcbinfo->ipi_lastport; 398 } 399 /* 400 * For UDP(-Lite), use random port allocation as long as the user 401 * allows it. For TCP (and as of yet unknown) connections, 402 * use random port allocation only if the user allows it AND 403 * ipport_tick() allows it. 404 */ 405 if (V_ipport_randomized && 406 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 407 pcbinfo == &V_ulitecbinfo)) 408 dorandom = 1; 409 else 410 dorandom = 0; 411 /* 412 * It makes no sense to do random port allocation if 413 * we have the only port available. 414 */ 415 if (first == last) 416 dorandom = 0; 417 /* Make sure to not include UDP(-Lite) packets in the count. */ 418 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 419 V_ipport_tcpallocs++; 420 /* 421 * Instead of having two loops further down counting up or down 422 * make sure that first is always <= last and go with only one 423 * code path implementing all logic. 424 */ 425 if (first > last) { 426 aux = first; 427 first = last; 428 last = aux; 429 } 430 431#ifdef INET 432 /* Make the compiler happy. */ 433 laddr.s_addr = 0; 434 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 435 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", 436 __func__, inp)); 437 laddr = *laddrp; 438 } 439#endif 440 tmpinp = NULL; /* Make compiler happy. */ 441 lport = *lportp; 442 443 if (dorandom) 444 *lastport = first + (arc4random() % (last - first)); 445 446 count = last - first; 447 448 do { 449 if (count-- < 0) /* completely used? */ 450 return (EADDRNOTAVAIL); 451 ++*lastport; 452 if (*lastport < first || *lastport > last) 453 *lastport = first; 454 lport = htons(*lastport); 455 456#ifdef INET6 457 if ((inp->inp_vflag & INP_IPV6) != 0) 458 tmpinp = in6_pcblookup_local(pcbinfo, 459 &inp->in6p_laddr, lport, lookupflags, cred); 460#endif 461#if defined(INET) && defined(INET6) 462 else 463#endif 464#ifdef INET 465 tmpinp = in_pcblookup_local(pcbinfo, laddr, 466 lport, lookupflags, cred); 467#endif 468 } while (tmpinp != NULL); 469 470#ifdef INET 471 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) 472 laddrp->s_addr = laddr.s_addr; 473#endif 474 *lportp = lport; 475 476 return (0); 477} 478 479/* 480 * Return cached socket options. 481 */ 482short 483inp_so_options(const struct inpcb *inp) 484{ 485 short so_options; 486 487 so_options = 0; 488 489 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 490 so_options |= SO_REUSEPORT; 491 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 492 so_options |= SO_REUSEADDR; 493 return (so_options); 494} 495#endif /* INET || INET6 */ 496 497#ifdef INET 498/* 499 * Set up a bind operation on a PCB, performing port allocation 500 * as required, but do not actually modify the PCB. Callers can 501 * either complete the bind by setting inp_laddr/inp_lport and 502 * calling in_pcbinshash(), or they can just use the resulting 503 * port and address to authorise the sending of a once-off packet. 504 * 505 * On error, the values of *laddrp and *lportp are not changed. 506 */ 507int 508in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 509 u_short *lportp, struct ucred *cred) 510{ 511 struct socket *so = inp->inp_socket; 512 struct sockaddr_in *sin; 513 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 514 struct in_addr laddr; 515 u_short lport = 0; 516 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 517 int error; 518 519 /* 520 * No state changes, so read locks are sufficient here. 521 */ 522 INP_LOCK_ASSERT(inp); 523 INP_HASH_LOCK_ASSERT(pcbinfo); 524 525 if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 526 return (EADDRNOTAVAIL); 527 laddr.s_addr = *laddrp; 528 if (nam != NULL && laddr.s_addr != INADDR_ANY) 529 return (EINVAL); 530 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0) 531 lookupflags = INPLOOKUP_WILDCARD; 532 if (nam == NULL) { 533 if ((error = prison_local_ip4(cred, &laddr)) != 0) 534 return (error); 535 } else { 536 sin = (struct sockaddr_in *)nam; 537 if (nam->sa_len != sizeof (*sin)) 538 return (EINVAL); 539#ifdef notdef 540 /* 541 * We should check the family, but old programs 542 * incorrectly fail to initialize it. 543 */ 544 if (sin->sin_family != AF_INET) 545 return (EAFNOSUPPORT); 546#endif 547 error = prison_local_ip4(cred, &sin->sin_addr); 548 if (error) 549 return (error); 550 if (sin->sin_port != *lportp) { 551 /* Don't allow the port to change. */ 552 if (*lportp != 0) 553 return (EINVAL); 554 lport = sin->sin_port; 555 } 556 /* NB: lport is left as 0 if the port isn't being changed. */ 557 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 558 /* 559 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 560 * allow complete duplication of binding if 561 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 562 * and a multicast address is bound on both 563 * new and duplicated sockets. 564 */ 565 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 566 reuseport = SO_REUSEADDR|SO_REUSEPORT; 567 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 568 sin->sin_port = 0; /* yech... */ 569 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 570 /* 571 * Is the address a local IP address? 572 * If INP_BINDANY is set, then the socket may be bound 573 * to any endpoint address, local or not. 574 */ 575 if ((inp->inp_flags & INP_BINDANY) == 0 && 576 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 577 return (EADDRNOTAVAIL); 578 } 579 laddr = sin->sin_addr; 580 if (lport) { 581 struct inpcb *t; 582 struct tcptw *tw; 583 584 /* GROSS */ 585 if (ntohs(lport) <= V_ipport_reservedhigh && 586 ntohs(lport) >= V_ipport_reservedlow && 587 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 588 0)) 589 return (EACCES); 590 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 591 priv_check_cred(inp->inp_cred, 592 PRIV_NETINET_REUSEPORT, 0) != 0) { 593 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 594 lport, INPLOOKUP_WILDCARD, cred); 595 /* 596 * XXX 597 * This entire block sorely needs a rewrite. 598 */ 599 if (t && 600 ((t->inp_flags & INP_TIMEWAIT) == 0) && 601 (so->so_type != SOCK_STREAM || 602 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 603 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 604 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 605 (t->inp_flags2 & INP_REUSEPORT) == 0) && 606 (inp->inp_cred->cr_uid != 607 t->inp_cred->cr_uid)) 608 return (EADDRINUSE); 609 } 610 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 611 lport, lookupflags, cred); 612 if (t && (t->inp_flags & INP_TIMEWAIT)) { 613 /* 614 * XXXRW: If an incpb has had its timewait 615 * state recycled, we treat the address as 616 * being in use (for now). This is better 617 * than a panic, but not desirable. 618 */ 619 tw = intotw(t); 620 if (tw == NULL || 621 (reuseport & tw->tw_so_options) == 0) 622 return (EADDRINUSE); 623 } else if (t && (reuseport & inp_so_options(t)) == 0) { 624#ifdef INET6 625 if (ntohl(sin->sin_addr.s_addr) != 626 INADDR_ANY || 627 ntohl(t->inp_laddr.s_addr) != 628 INADDR_ANY || 629 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 630 (t->inp_vflag & INP_IPV6PROTO) == 0) 631#endif 632 return (EADDRINUSE); 633 } 634 } 635 } 636 if (*lportp != 0) 637 lport = *lportp; 638 if (lport == 0) { 639 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 640 if (error != 0) 641 return (error); 642 643 } 644 *laddrp = laddr.s_addr; 645 *lportp = lport; 646 return (0); 647} 648 649/* 650 * Connect from a socket to a specified address. 651 * Both address and port must be specified in argument sin. 652 * If don't have a local address for this socket yet, 653 * then pick one. 654 */ 655int 656in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 657 struct ucred *cred, struct mbuf *m) 658{ 659 u_short lport, fport; 660 in_addr_t laddr, faddr; 661 int anonport, error; 662 663 INP_WLOCK_ASSERT(inp); 664 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 665 666 lport = inp->inp_lport; 667 laddr = inp->inp_laddr.s_addr; 668 anonport = (lport == 0); 669 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 670 NULL, cred); 671 if (error) 672 return (error); 673 674 /* Do the initial binding of the local address if required. */ 675 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 676 inp->inp_lport = lport; 677 inp->inp_laddr.s_addr = laddr; 678 if (in_pcbinshash(inp) != 0) { 679 inp->inp_laddr.s_addr = INADDR_ANY; 680 inp->inp_lport = 0; 681 return (EAGAIN); 682 } 683 } 684 685 /* Commit the remaining changes. */ 686 inp->inp_lport = lport; 687 inp->inp_laddr.s_addr = laddr; 688 inp->inp_faddr.s_addr = faddr; 689 inp->inp_fport = fport; 690 in_pcbrehash_mbuf(inp, m); 691 692 if (anonport) 693 inp->inp_flags |= INP_ANONPORT; 694 return (0); 695} 696 697int 698in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 699{ 700 701 return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); 702} 703 704/* 705 * Do proper source address selection on an unbound socket in case 706 * of connect. Take jails into account as well. 707 */ 708int 709in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 710 struct ucred *cred) 711{ 712 struct ifaddr *ifa; 713 struct sockaddr *sa; 714 struct sockaddr_in *sin; 715 struct route sro; 716 int error; 717 718 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 719 720 /* 721 * Bypass source address selection and use the primary jail IP 722 * if requested. 723 */ 724 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 725 return (0); 726 727 error = 0; 728 bzero(&sro, sizeof(sro)); 729 730 sin = (struct sockaddr_in *)&sro.ro_dst; 731 sin->sin_family = AF_INET; 732 sin->sin_len = sizeof(struct sockaddr_in); 733 sin->sin_addr.s_addr = faddr->s_addr; 734 735 /* 736 * If route is known our src addr is taken from the i/f, 737 * else punt. 738 * 739 * Find out route to destination. 740 */ 741 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 742 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 743 744 /* 745 * If we found a route, use the address corresponding to 746 * the outgoing interface. 747 * 748 * Otherwise assume faddr is reachable on a directly connected 749 * network and try to find a corresponding interface to take 750 * the source address from. 751 */ 752 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 753 struct in_ifaddr *ia; 754 struct ifnet *ifp; 755 756 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin)); 757 if (ia == NULL) 758 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0)); 759 if (ia == NULL) { 760 error = ENETUNREACH; 761 goto done; 762 } 763 764 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 765 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 766 ifa_free(&ia->ia_ifa); 767 goto done; 768 } 769 770 ifp = ia->ia_ifp; 771 ifa_free(&ia->ia_ifa); 772 ia = NULL; 773 IF_ADDR_RLOCK(ifp); 774 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 775 776 sa = ifa->ifa_addr; 777 if (sa->sa_family != AF_INET) 778 continue; 779 sin = (struct sockaddr_in *)sa; 780 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 781 ia = (struct in_ifaddr *)ifa; 782 break; 783 } 784 } 785 if (ia != NULL) { 786 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 787 IF_ADDR_RUNLOCK(ifp); 788 goto done; 789 } 790 IF_ADDR_RUNLOCK(ifp); 791 792 /* 3. As a last resort return the 'default' jail address. */ 793 error = prison_get_ip4(cred, laddr); 794 goto done; 795 } 796 797 /* 798 * If the outgoing interface on the route found is not 799 * a loopback interface, use the address from that interface. 800 * In case of jails do those three steps: 801 * 1. check if the interface address belongs to the jail. If so use it. 802 * 2. check if we have any address on the outgoing interface 803 * belonging to this jail. If so use it. 804 * 3. as a last resort return the 'default' jail address. 805 */ 806 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 807 struct in_ifaddr *ia; 808 struct ifnet *ifp; 809 810 /* If not jailed, use the default returned. */ 811 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 812 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 813 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 814 goto done; 815 } 816 817 /* Jailed. */ 818 /* 1. Check if the iface address belongs to the jail. */ 819 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 820 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 821 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 822 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 823 goto done; 824 } 825 826 /* 827 * 2. Check if we have any address on the outgoing interface 828 * belonging to this jail. 829 */ 830 ia = NULL; 831 ifp = sro.ro_rt->rt_ifp; 832 IF_ADDR_RLOCK(ifp); 833 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 834 sa = ifa->ifa_addr; 835 if (sa->sa_family != AF_INET) 836 continue; 837 sin = (struct sockaddr_in *)sa; 838 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 839 ia = (struct in_ifaddr *)ifa; 840 break; 841 } 842 } 843 if (ia != NULL) { 844 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 845 IF_ADDR_RUNLOCK(ifp); 846 goto done; 847 } 848 IF_ADDR_RUNLOCK(ifp); 849 850 /* 3. As a last resort return the 'default' jail address. */ 851 error = prison_get_ip4(cred, laddr); 852 goto done; 853 } 854 855 /* 856 * The outgoing interface is marked with 'loopback net', so a route 857 * to ourselves is here. 858 * Try to find the interface of the destination address and then 859 * take the address from there. That interface is not necessarily 860 * a loopback interface. 861 * In case of jails, check that it is an address of the jail 862 * and if we cannot find, fall back to the 'default' jail address. 863 */ 864 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 865 struct sockaddr_in sain; 866 struct in_ifaddr *ia; 867 868 bzero(&sain, sizeof(struct sockaddr_in)); 869 sain.sin_family = AF_INET; 870 sain.sin_len = sizeof(struct sockaddr_in); 871 sain.sin_addr.s_addr = faddr->s_addr; 872 873 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain))); 874 if (ia == NULL) 875 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0)); 876 if (ia == NULL) 877 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 878 879 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 880 if (ia == NULL) { 881 error = ENETUNREACH; 882 goto done; 883 } 884 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 885 ifa_free(&ia->ia_ifa); 886 goto done; 887 } 888 889 /* Jailed. */ 890 if (ia != NULL) { 891 struct ifnet *ifp; 892 893 ifp = ia->ia_ifp; 894 ifa_free(&ia->ia_ifa); 895 ia = NULL; 896 IF_ADDR_RLOCK(ifp); 897 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 898 899 sa = ifa->ifa_addr; 900 if (sa->sa_family != AF_INET) 901 continue; 902 sin = (struct sockaddr_in *)sa; 903 if (prison_check_ip4(cred, 904 &sin->sin_addr) == 0) { 905 ia = (struct in_ifaddr *)ifa; 906 break; 907 } 908 } 909 if (ia != NULL) { 910 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 911 IF_ADDR_RUNLOCK(ifp); 912 goto done; 913 } 914 IF_ADDR_RUNLOCK(ifp); 915 } 916 917 /* 3. As a last resort return the 'default' jail address. */ 918 error = prison_get_ip4(cred, laddr); 919 goto done; 920 } 921 922done: 923 if (sro.ro_rt != NULL) 924 RTFREE(sro.ro_rt); 925 return (error); 926} 927 928/* 929 * Set up for a connect from a socket to the specified address. 930 * On entry, *laddrp and *lportp should contain the current local 931 * address and port for the PCB; these are updated to the values 932 * that should be placed in inp_laddr and inp_lport to complete 933 * the connect. 934 * 935 * On success, *faddrp and *fportp will be set to the remote address 936 * and port. These are not updated in the error case. 937 * 938 * If the operation fails because the connection already exists, 939 * *oinpp will be set to the PCB of that connection so that the 940 * caller can decide to override it. In all other cases, *oinpp 941 * is set to NULL. 942 */ 943int 944in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 945 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 946 struct inpcb **oinpp, struct ucred *cred) 947{ 948 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 949 struct in_ifaddr *ia; 950 struct inpcb *oinp; 951 struct in_addr laddr, faddr; 952 u_short lport, fport; 953 int error; 954 955 /* 956 * Because a global state change doesn't actually occur here, a read 957 * lock is sufficient. 958 */ 959 INP_LOCK_ASSERT(inp); 960 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 961 962 if (oinpp != NULL) 963 *oinpp = NULL; 964 if (nam->sa_len != sizeof (*sin)) 965 return (EINVAL); 966 if (sin->sin_family != AF_INET) 967 return (EAFNOSUPPORT); 968 if (sin->sin_port == 0) 969 return (EADDRNOTAVAIL); 970 laddr.s_addr = *laddrp; 971 lport = *lportp; 972 faddr = sin->sin_addr; 973 fport = sin->sin_port; 974 975 if (!TAILQ_EMPTY(&V_in_ifaddrhead)) { 976 /* 977 * If the destination address is INADDR_ANY, 978 * use the primary local address. 979 * If the supplied address is INADDR_BROADCAST, 980 * and the primary interface supports broadcast, 981 * choose the broadcast address for that interface. 982 */ 983 if (faddr.s_addr == INADDR_ANY) { 984 IN_IFADDR_RLOCK(); 985 faddr = 986 IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 987 IN_IFADDR_RUNLOCK(); 988 if (cred != NULL && 989 (error = prison_get_ip4(cred, &faddr)) != 0) 990 return (error); 991 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 992 IN_IFADDR_RLOCK(); 993 if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 994 IFF_BROADCAST) 995 faddr = satosin(&TAILQ_FIRST( 996 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 997 IN_IFADDR_RUNLOCK(); 998 } 999 } 1000 if (laddr.s_addr == INADDR_ANY) { 1001 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1002 /* 1003 * If the destination address is multicast and an outgoing 1004 * interface has been set as a multicast option, prefer the 1005 * address of that interface as our source address. 1006 */ 1007 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1008 inp->inp_moptions != NULL) { 1009 struct ip_moptions *imo; 1010 struct ifnet *ifp; 1011 1012 imo = inp->inp_moptions; 1013 if (imo->imo_multicast_ifp != NULL) { 1014 ifp = imo->imo_multicast_ifp; 1015 IN_IFADDR_RLOCK(); 1016 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1017 if ((ia->ia_ifp == ifp) && 1018 (cred == NULL || 1019 prison_check_ip4(cred, 1020 &ia->ia_addr.sin_addr) == 0)) 1021 break; 1022 } 1023 if (ia == NULL) 1024 error = EADDRNOTAVAIL; 1025 else { 1026 laddr = ia->ia_addr.sin_addr; 1027 error = 0; 1028 } 1029 IN_IFADDR_RUNLOCK(); 1030 } 1031 } 1032 if (error) 1033 return (error); 1034 } 1035 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1036 laddr, lport, 0, NULL); 1037 if (oinp != NULL) { 1038 if (oinpp != NULL) 1039 *oinpp = oinp; 1040 return (EADDRINUSE); 1041 } 1042 if (lport == 0) { 1043 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1044 cred); 1045 if (error) 1046 return (error); 1047 } 1048 *laddrp = laddr.s_addr; 1049 *lportp = lport; 1050 *faddrp = faddr.s_addr; 1051 *fportp = fport; 1052 return (0); 1053} 1054 1055void 1056in_pcbdisconnect(struct inpcb *inp) 1057{ 1058 1059 INP_WLOCK_ASSERT(inp); 1060 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1061 1062 inp->inp_faddr.s_addr = INADDR_ANY; 1063 inp->inp_fport = 0; 1064 in_pcbrehash(inp); 1065} 1066#endif /* INET */ 1067 1068/* 1069 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1070 * For most protocols, this will be invoked immediately prior to calling 1071 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1072 * socket, in which case in_pcbfree() is deferred. 1073 */ 1074void 1075in_pcbdetach(struct inpcb *inp) 1076{ 1077 1078 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1079 1080 inp->inp_socket->so_pcb = NULL; 1081 inp->inp_socket = NULL; 1082} 1083 1084/* 1085 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1086 * stability of an inpcb pointer despite the inpcb lock being released. This 1087 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1088 * but where the inpcb lock may already held, or when acquiring a reference 1089 * via a pcbgroup. 1090 * 1091 * in_pcbref() should be used only to provide brief memory stability, and 1092 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1093 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1094 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1095 * lock and rele are the *only* safe operations that may be performed on the 1096 * inpcb. 1097 * 1098 * While the inpcb will not be freed, releasing the inpcb lock means that the 1099 * connection's state may change, so the caller should be careful to 1100 * revalidate any cached state on reacquiring the lock. Drop the reference 1101 * using in_pcbrele(). 1102 */ 1103void 1104in_pcbref(struct inpcb *inp) 1105{ 1106 1107 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1108 1109 refcount_acquire(&inp->inp_refcount); 1110} 1111 1112/* 1113 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1114 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1115 * return a flag indicating whether or not the inpcb remains valid. If it is 1116 * valid, we return with the inpcb lock held. 1117 * 1118 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1119 * reference on an inpcb. Historically more work was done here (actually, in 1120 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1121 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1122 * about memory stability (and continued use of the write lock). 1123 */ 1124int 1125in_pcbrele_rlocked(struct inpcb *inp) 1126{ 1127 struct inpcbinfo *pcbinfo; 1128 1129 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1130 1131 INP_RLOCK_ASSERT(inp); 1132 1133 if (refcount_release(&inp->inp_refcount) == 0) { 1134 /* 1135 * If the inpcb has been freed, let the caller know, even if 1136 * this isn't the last reference. 1137 */ 1138 if (inp->inp_flags2 & INP_FREED) { 1139 INP_RUNLOCK(inp); 1140 return (1); 1141 } 1142 return (0); 1143 } 1144 1145 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1146 1147 INP_RUNLOCK(inp); 1148 pcbinfo = inp->inp_pcbinfo; 1149 uma_zfree(pcbinfo->ipi_zone, inp); 1150 return (1); 1151} 1152 1153int 1154in_pcbrele_wlocked(struct inpcb *inp) 1155{ 1156 struct inpcbinfo *pcbinfo; 1157 1158 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1159 1160 INP_WLOCK_ASSERT(inp); 1161 1162 if (refcount_release(&inp->inp_refcount) == 0) { 1163 /* 1164 * If the inpcb has been freed, let the caller know, even if 1165 * this isn't the last reference. 1166 */ 1167 if (inp->inp_flags2 & INP_FREED) { 1168 INP_WUNLOCK(inp); 1169 return (1); 1170 } 1171 return (0); 1172 } 1173 1174 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1175 1176 INP_WUNLOCK(inp); 1177 pcbinfo = inp->inp_pcbinfo; 1178 uma_zfree(pcbinfo->ipi_zone, inp); 1179 return (1); 1180} 1181 1182/* 1183 * Temporary wrapper. 1184 */ 1185int 1186in_pcbrele(struct inpcb *inp) 1187{ 1188 1189 return (in_pcbrele_wlocked(inp)); 1190} 1191 1192/* 1193 * Unconditionally schedule an inpcb to be freed by decrementing its 1194 * reference count, which should occur only after the inpcb has been detached 1195 * from its socket. If another thread holds a temporary reference (acquired 1196 * using in_pcbref()) then the free is deferred until that reference is 1197 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1198 * work, including removal from global lists, is done in this context, where 1199 * the pcbinfo lock is held. 1200 */ 1201void 1202in_pcbfree(struct inpcb *inp) 1203{ 1204 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1205 1206 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1207 1208#ifdef INVARIANTS 1209 if (pcbinfo == &V_tcbinfo) { 1210 INP_INFO_LOCK_ASSERT(pcbinfo); 1211 } else { 1212 INP_INFO_WLOCK_ASSERT(pcbinfo); 1213 } 1214#endif 1215 INP_WLOCK_ASSERT(inp); 1216 1217 /* XXXRW: Do as much as possible here. */ 1218#ifdef IPSEC 1219 if (inp->inp_sp != NULL) 1220 ipsec_delete_pcbpolicy(inp); 1221#endif 1222 INP_LIST_WLOCK(pcbinfo); 1223 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1224 in_pcbremlists(inp); 1225 INP_LIST_WUNLOCK(pcbinfo); 1226#ifdef INET6 1227 if (inp->inp_vflag & INP_IPV6PROTO) { 1228 ip6_freepcbopts(inp->in6p_outputopts); 1229 if (inp->in6p_moptions != NULL) 1230 ip6_freemoptions(inp->in6p_moptions); 1231 } 1232#endif 1233 if (inp->inp_options) 1234 (void)m_free(inp->inp_options); 1235#ifdef INET 1236 if (inp->inp_moptions != NULL) 1237 inp_freemoptions(inp->inp_moptions); 1238#endif 1239 inp->inp_vflag = 0; 1240 inp->inp_flags2 |= INP_FREED; 1241 crfree(inp->inp_cred); 1242#ifdef MAC 1243 mac_inpcb_destroy(inp); 1244#endif 1245 if (!in_pcbrele_wlocked(inp)) 1246 INP_WUNLOCK(inp); 1247} 1248 1249/* 1250 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1251 * port reservation, and preventing it from being returned by inpcb lookups. 1252 * 1253 * It is used by TCP to mark an inpcb as unused and avoid future packet 1254 * delivery or event notification when a socket remains open but TCP has 1255 * closed. This might occur as a result of a shutdown()-initiated TCP close 1256 * or a RST on the wire, and allows the port binding to be reused while still 1257 * maintaining the invariant that so_pcb always points to a valid inpcb until 1258 * in_pcbdetach(). 1259 * 1260 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1261 * in_pcbnotifyall() and in_pcbpurgeif0()? 1262 */ 1263void 1264in_pcbdrop(struct inpcb *inp) 1265{ 1266 1267 INP_WLOCK_ASSERT(inp); 1268 1269 /* 1270 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1271 * the hash lock...? 1272 */ 1273 inp->inp_flags |= INP_DROPPED; 1274 if (inp->inp_flags & INP_INHASHLIST) { 1275 struct inpcbport *phd = inp->inp_phd; 1276 1277 INP_HASH_WLOCK(inp->inp_pcbinfo); 1278 LIST_REMOVE(inp, inp_hash); 1279 LIST_REMOVE(inp, inp_portlist); 1280 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1281 LIST_REMOVE(phd, phd_hash); 1282 free(phd, M_PCB); 1283 } 1284 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1285 inp->inp_flags &= ~INP_INHASHLIST; 1286#ifdef PCBGROUP 1287 in_pcbgroup_remove(inp); 1288#endif 1289 } 1290} 1291 1292#ifdef INET 1293/* 1294 * Common routines to return the socket addresses associated with inpcbs. 1295 */ 1296struct sockaddr * 1297in_sockaddr(in_port_t port, struct in_addr *addr_p) 1298{ 1299 struct sockaddr_in *sin; 1300 1301 sin = malloc(sizeof *sin, M_SONAME, 1302 M_WAITOK | M_ZERO); 1303 sin->sin_family = AF_INET; 1304 sin->sin_len = sizeof(*sin); 1305 sin->sin_addr = *addr_p; 1306 sin->sin_port = port; 1307 1308 return (struct sockaddr *)sin; 1309} 1310 1311int 1312in_getsockaddr(struct socket *so, struct sockaddr **nam) 1313{ 1314 struct inpcb *inp; 1315 struct in_addr addr; 1316 in_port_t port; 1317 1318 inp = sotoinpcb(so); 1319 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1320 1321 INP_RLOCK(inp); 1322 port = inp->inp_lport; 1323 addr = inp->inp_laddr; 1324 INP_RUNLOCK(inp); 1325 1326 *nam = in_sockaddr(port, &addr); 1327 return 0; 1328} 1329 1330int 1331in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1332{ 1333 struct inpcb *inp; 1334 struct in_addr addr; 1335 in_port_t port; 1336 1337 inp = sotoinpcb(so); 1338 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1339 1340 INP_RLOCK(inp); 1341 port = inp->inp_fport; 1342 addr = inp->inp_faddr; 1343 INP_RUNLOCK(inp); 1344 1345 *nam = in_sockaddr(port, &addr); 1346 return 0; 1347} 1348 1349void 1350in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1351 struct inpcb *(*notify)(struct inpcb *, int)) 1352{ 1353 struct inpcb *inp, *inp_temp; 1354 1355 INP_INFO_WLOCK(pcbinfo); 1356 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1357 INP_WLOCK(inp); 1358#ifdef INET6 1359 if ((inp->inp_vflag & INP_IPV4) == 0) { 1360 INP_WUNLOCK(inp); 1361 continue; 1362 } 1363#endif 1364 if (inp->inp_faddr.s_addr != faddr.s_addr || 1365 inp->inp_socket == NULL) { 1366 INP_WUNLOCK(inp); 1367 continue; 1368 } 1369 if ((*notify)(inp, errno)) 1370 INP_WUNLOCK(inp); 1371 } 1372 INP_INFO_WUNLOCK(pcbinfo); 1373} 1374 1375void 1376in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1377{ 1378 struct inpcb *inp; 1379 struct ip_moptions *imo; 1380 int i, gap; 1381 1382 INP_INFO_WLOCK(pcbinfo); 1383 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1384 INP_WLOCK(inp); 1385 imo = inp->inp_moptions; 1386 if ((inp->inp_vflag & INP_IPV4) && 1387 imo != NULL) { 1388 /* 1389 * Unselect the outgoing interface if it is being 1390 * detached. 1391 */ 1392 if (imo->imo_multicast_ifp == ifp) 1393 imo->imo_multicast_ifp = NULL; 1394 1395 /* 1396 * Drop multicast group membership if we joined 1397 * through the interface being detached. 1398 */ 1399 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1400 i++) { 1401 if (imo->imo_membership[i]->inm_ifp == ifp) { 1402 in_delmulti(imo->imo_membership[i]); 1403 gap++; 1404 } else if (gap != 0) 1405 imo->imo_membership[i - gap] = 1406 imo->imo_membership[i]; 1407 } 1408 imo->imo_num_memberships -= gap; 1409 } 1410 INP_WUNLOCK(inp); 1411 } 1412 INP_INFO_WUNLOCK(pcbinfo); 1413} 1414 1415/* 1416 * Lookup a PCB based on the local address and port. Caller must hold the 1417 * hash lock. No inpcb locks or references are acquired. 1418 */ 1419#define INP_LOOKUP_MAPPED_PCB_COST 3 1420struct inpcb * 1421in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1422 u_short lport, int lookupflags, struct ucred *cred) 1423{ 1424 struct inpcb *inp; 1425#ifdef INET6 1426 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1427#else 1428 int matchwild = 3; 1429#endif 1430 int wildcard; 1431 1432 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1433 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1434 1435 INP_HASH_LOCK_ASSERT(pcbinfo); 1436 1437 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1438 struct inpcbhead *head; 1439 /* 1440 * Look for an unconnected (wildcard foreign addr) PCB that 1441 * matches the local address and port we're looking for. 1442 */ 1443 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1444 0, pcbinfo->ipi_hashmask)]; 1445 LIST_FOREACH(inp, head, inp_hash) { 1446#ifdef INET6 1447 /* XXX inp locking */ 1448 if ((inp->inp_vflag & INP_IPV4) == 0) 1449 continue; 1450#endif 1451 if (inp->inp_faddr.s_addr == INADDR_ANY && 1452 inp->inp_laddr.s_addr == laddr.s_addr && 1453 inp->inp_lport == lport) { 1454 /* 1455 * Found? 1456 */ 1457 if (cred == NULL || 1458 prison_equal_ip4(cred->cr_prison, 1459 inp->inp_cred->cr_prison)) 1460 return (inp); 1461 } 1462 } 1463 /* 1464 * Not found. 1465 */ 1466 return (NULL); 1467 } else { 1468 struct inpcbporthead *porthash; 1469 struct inpcbport *phd; 1470 struct inpcb *match = NULL; 1471 /* 1472 * Best fit PCB lookup. 1473 * 1474 * First see if this local port is in use by looking on the 1475 * port hash list. 1476 */ 1477 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1478 pcbinfo->ipi_porthashmask)]; 1479 LIST_FOREACH(phd, porthash, phd_hash) { 1480 if (phd->phd_port == lport) 1481 break; 1482 } 1483 if (phd != NULL) { 1484 /* 1485 * Port is in use by one or more PCBs. Look for best 1486 * fit. 1487 */ 1488 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1489 wildcard = 0; 1490 if (cred != NULL && 1491 !prison_equal_ip4(inp->inp_cred->cr_prison, 1492 cred->cr_prison)) 1493 continue; 1494#ifdef INET6 1495 /* XXX inp locking */ 1496 if ((inp->inp_vflag & INP_IPV4) == 0) 1497 continue; 1498 /* 1499 * We never select the PCB that has 1500 * INP_IPV6 flag and is bound to :: if 1501 * we have another PCB which is bound 1502 * to 0.0.0.0. If a PCB has the 1503 * INP_IPV6 flag, then we set its cost 1504 * higher than IPv4 only PCBs. 1505 * 1506 * Note that the case only happens 1507 * when a socket is bound to ::, under 1508 * the condition that the use of the 1509 * mapped address is allowed. 1510 */ 1511 if ((inp->inp_vflag & INP_IPV6) != 0) 1512 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1513#endif 1514 if (inp->inp_faddr.s_addr != INADDR_ANY) 1515 wildcard++; 1516 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1517 if (laddr.s_addr == INADDR_ANY) 1518 wildcard++; 1519 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1520 continue; 1521 } else { 1522 if (laddr.s_addr != INADDR_ANY) 1523 wildcard++; 1524 } 1525 if (wildcard < matchwild) { 1526 match = inp; 1527 matchwild = wildcard; 1528 if (matchwild == 0) 1529 break; 1530 } 1531 } 1532 } 1533 return (match); 1534 } 1535} 1536#undef INP_LOOKUP_MAPPED_PCB_COST 1537 1538#ifdef PCBGROUP 1539/* 1540 * Lookup PCB in hash list, using pcbgroup tables. 1541 */ 1542static struct inpcb * 1543in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 1544 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 1545 u_int lport_arg, int lookupflags, struct ifnet *ifp) 1546{ 1547 struct inpcbhead *head; 1548 struct inpcb *inp, *tmpinp; 1549 u_short fport = fport_arg, lport = lport_arg; 1550 1551 /* 1552 * First look for an exact match. 1553 */ 1554 tmpinp = NULL; 1555 INP_GROUP_LOCK(pcbgroup); 1556 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 1557 pcbgroup->ipg_hashmask)]; 1558 LIST_FOREACH(inp, head, inp_pcbgrouphash) { 1559#ifdef INET6 1560 /* XXX inp locking */ 1561 if ((inp->inp_vflag & INP_IPV4) == 0) 1562 continue; 1563#endif 1564 if (inp->inp_faddr.s_addr == faddr.s_addr && 1565 inp->inp_laddr.s_addr == laddr.s_addr && 1566 inp->inp_fport == fport && 1567 inp->inp_lport == lport) { 1568 /* 1569 * XXX We should be able to directly return 1570 * the inp here, without any checks. 1571 * Well unless both bound with SO_REUSEPORT? 1572 */ 1573 if (prison_flag(inp->inp_cred, PR_IP4)) 1574 goto found; 1575 if (tmpinp == NULL) 1576 tmpinp = inp; 1577 } 1578 } 1579 if (tmpinp != NULL) { 1580 inp = tmpinp; 1581 goto found; 1582 } 1583 1584 /* 1585 * Then look for a wildcard match, if requested. 1586 */ 1587 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 1588 struct inpcb *local_wild = NULL, *local_exact = NULL; 1589#ifdef INET6 1590 struct inpcb *local_wild_mapped = NULL; 1591#endif 1592 struct inpcb *jail_wild = NULL; 1593 struct inpcbhead *head; 1594 int injail; 1595 1596 /* 1597 * Order of socket selection - we always prefer jails. 1598 * 1. jailed, non-wild. 1599 * 2. jailed, wild. 1600 * 3. non-jailed, non-wild. 1601 * 4. non-jailed, wild. 1602 */ 1603 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 1604 0, pcbinfo->ipi_wildmask)]; 1605 LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 1606#ifdef INET6 1607 /* XXX inp locking */ 1608 if ((inp->inp_vflag & INP_IPV4) == 0) 1609 continue; 1610#endif 1611 if (inp->inp_faddr.s_addr != INADDR_ANY || 1612 inp->inp_lport != lport) 1613 continue; 1614 1615 /* XXX inp locking */ 1616 if (ifp && ifp->if_type == IFT_FAITH && 1617 (inp->inp_flags & INP_FAITH) == 0) 1618 continue; 1619 1620 injail = prison_flag(inp->inp_cred, PR_IP4); 1621 if (injail) { 1622 if (prison_check_ip4(inp->inp_cred, 1623 &laddr) != 0) 1624 continue; 1625 } else { 1626 if (local_exact != NULL) 1627 continue; 1628 } 1629 1630 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1631 if (injail) 1632 goto found; 1633 else 1634 local_exact = inp; 1635 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 1636#ifdef INET6 1637 /* XXX inp locking, NULL check */ 1638 if (inp->inp_vflag & INP_IPV6PROTO) 1639 local_wild_mapped = inp; 1640 else 1641#endif 1642 if (injail) 1643 jail_wild = inp; 1644 else 1645 local_wild = inp; 1646 } 1647 } /* LIST_FOREACH */ 1648 inp = jail_wild; 1649 if (inp == NULL) 1650 inp = local_exact; 1651 if (inp == NULL) 1652 inp = local_wild; 1653#ifdef INET6 1654 if (inp == NULL) 1655 inp = local_wild_mapped; 1656#endif 1657 if (inp != NULL) 1658 goto found; 1659 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 1660 INP_GROUP_UNLOCK(pcbgroup); 1661 return (NULL); 1662 1663found: 1664 in_pcbref(inp); 1665 INP_GROUP_UNLOCK(pcbgroup); 1666 if (lookupflags & INPLOOKUP_WLOCKPCB) { 1667 INP_WLOCK(inp); 1668 if (in_pcbrele_wlocked(inp)) 1669 return (NULL); 1670 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 1671 INP_RLOCK(inp); 1672 if (in_pcbrele_rlocked(inp)) 1673 return (NULL); 1674 } else 1675 panic("%s: locking bug", __func__); 1676 return (inp); 1677} 1678#endif /* PCBGROUP */ 1679 1680/* 1681 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 1682 * that the caller has locked the hash list, and will not perform any further 1683 * locking or reference operations on either the hash list or the connection. 1684 */ 1685static struct inpcb * 1686in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1687 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 1688 struct ifnet *ifp) 1689{ 1690 struct inpcbhead *head; 1691 struct inpcb *inp, *tmpinp; 1692 u_short fport = fport_arg, lport = lport_arg; 1693 1694 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1695 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1696 1697 INP_HASH_LOCK_ASSERT(pcbinfo); 1698 1699 /* 1700 * First look for an exact match. 1701 */ 1702 tmpinp = NULL; 1703 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 1704 pcbinfo->ipi_hashmask)]; 1705 LIST_FOREACH(inp, head, inp_hash) { 1706#ifdef INET6 1707 /* XXX inp locking */ 1708 if ((inp->inp_vflag & INP_IPV4) == 0) 1709 continue; 1710#endif 1711 if (inp->inp_faddr.s_addr == faddr.s_addr && 1712 inp->inp_laddr.s_addr == laddr.s_addr && 1713 inp->inp_fport == fport && 1714 inp->inp_lport == lport) { 1715 /* 1716 * XXX We should be able to directly return 1717 * the inp here, without any checks. 1718 * Well unless both bound with SO_REUSEPORT? 1719 */ 1720 if (prison_flag(inp->inp_cred, PR_IP4)) 1721 return (inp); 1722 if (tmpinp == NULL) 1723 tmpinp = inp; 1724 } 1725 } 1726 if (tmpinp != NULL) 1727 return (tmpinp); 1728 1729 /* 1730 * Then look for a wildcard match, if requested. 1731 */ 1732 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 1733 struct inpcb *local_wild = NULL, *local_exact = NULL; 1734#ifdef INET6 1735 struct inpcb *local_wild_mapped = NULL; 1736#endif 1737 struct inpcb *jail_wild = NULL; 1738 int injail; 1739 1740 /* 1741 * Order of socket selection - we always prefer jails. 1742 * 1. jailed, non-wild. 1743 * 2. jailed, wild. 1744 * 3. non-jailed, non-wild. 1745 * 4. non-jailed, wild. 1746 */ 1747 1748 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1749 0, pcbinfo->ipi_hashmask)]; 1750 LIST_FOREACH(inp, head, inp_hash) { 1751#ifdef INET6 1752 /* XXX inp locking */ 1753 if ((inp->inp_vflag & INP_IPV4) == 0) 1754 continue; 1755#endif 1756 if (inp->inp_faddr.s_addr != INADDR_ANY || 1757 inp->inp_lport != lport) 1758 continue; 1759 1760 /* XXX inp locking */ 1761 if (ifp && ifp->if_type == IFT_FAITH && 1762 (inp->inp_flags & INP_FAITH) == 0) 1763 continue; 1764 1765 injail = prison_flag(inp->inp_cred, PR_IP4); 1766 if (injail) { 1767 if (prison_check_ip4(inp->inp_cred, 1768 &laddr) != 0) 1769 continue; 1770 } else { 1771 if (local_exact != NULL) 1772 continue; 1773 } 1774 1775 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1776 if (injail) 1777 return (inp); 1778 else 1779 local_exact = inp; 1780 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 1781#ifdef INET6 1782 /* XXX inp locking, NULL check */ 1783 if (inp->inp_vflag & INP_IPV6PROTO) 1784 local_wild_mapped = inp; 1785 else 1786#endif 1787 if (injail) 1788 jail_wild = inp; 1789 else 1790 local_wild = inp; 1791 } 1792 } /* LIST_FOREACH */ 1793 if (jail_wild != NULL) 1794 return (jail_wild); 1795 if (local_exact != NULL) 1796 return (local_exact); 1797 if (local_wild != NULL) 1798 return (local_wild); 1799#ifdef INET6 1800 if (local_wild_mapped != NULL) 1801 return (local_wild_mapped); 1802#endif 1803 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 1804 1805 return (NULL); 1806} 1807 1808/* 1809 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 1810 * hash list lock, and will return the inpcb locked (i.e., requires 1811 * INPLOOKUP_LOCKPCB). 1812 */ 1813static struct inpcb * 1814in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1815 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 1816 struct ifnet *ifp) 1817{ 1818 struct inpcb *inp; 1819 1820 INP_HASH_RLOCK(pcbinfo); 1821 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 1822 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 1823 if (inp != NULL) { 1824 in_pcbref(inp); 1825 INP_HASH_RUNLOCK(pcbinfo); 1826 if (lookupflags & INPLOOKUP_WLOCKPCB) { 1827 INP_WLOCK(inp); 1828 if (in_pcbrele_wlocked(inp)) 1829 return (NULL); 1830 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 1831 INP_RLOCK(inp); 1832 if (in_pcbrele_rlocked(inp)) 1833 return (NULL); 1834 } else 1835 panic("%s: locking bug", __func__); 1836 } else 1837 INP_HASH_RUNLOCK(pcbinfo); 1838 return (inp); 1839} 1840 1841/* 1842 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 1843 * from which a pre-calculated hash value may be extracted. 1844 * 1845 * Possibly more of this logic should be in in_pcbgroup.c. 1846 */ 1847struct inpcb * 1848in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 1849 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 1850{ 1851#if defined(PCBGROUP) 1852 struct inpcbgroup *pcbgroup; 1853#endif 1854 1855 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 1856 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1857 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 1858 ("%s: LOCKPCB not set", __func__)); 1859 1860#if defined(PCBGROUP) 1861 if (in_pcbgroup_enabled(pcbinfo)) { 1862 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 1863 fport); 1864 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 1865 laddr, lport, lookupflags, ifp)); 1866 } 1867#endif 1868 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 1869 lookupflags, ifp)); 1870} 1871 1872struct inpcb * 1873in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1874 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 1875 struct ifnet *ifp, struct mbuf *m) 1876{ 1877#ifdef PCBGROUP 1878 struct inpcbgroup *pcbgroup; 1879#endif 1880 1881 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 1882 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1883 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 1884 ("%s: LOCKPCB not set", __func__)); 1885 1886#ifdef PCBGROUP 1887 if (in_pcbgroup_enabled(pcbinfo)) { 1888 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 1889 m->m_pkthdr.flowid); 1890 if (pcbgroup != NULL) 1891 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 1892 fport, laddr, lport, lookupflags, ifp)); 1893 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 1894 fport); 1895 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 1896 laddr, lport, lookupflags, ifp)); 1897 } 1898#endif 1899 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 1900 lookupflags, ifp)); 1901} 1902#endif /* INET */ 1903 1904/* 1905 * Insert PCB onto various hash lists. 1906 */ 1907static int 1908in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 1909{ 1910 struct inpcbhead *pcbhash; 1911 struct inpcbporthead *pcbporthash; 1912 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1913 struct inpcbport *phd; 1914 u_int32_t hashkey_faddr; 1915 1916 INP_WLOCK_ASSERT(inp); 1917 INP_HASH_WLOCK_ASSERT(pcbinfo); 1918 1919 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 1920 ("in_pcbinshash: INP_INHASHLIST")); 1921 1922#ifdef INET6 1923 if (inp->inp_vflag & INP_IPV6) 1924 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */; 1925 else 1926#endif 1927 hashkey_faddr = inp->inp_faddr.s_addr; 1928 1929 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 1930 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 1931 1932 pcbporthash = &pcbinfo->ipi_porthashbase[ 1933 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 1934 1935 /* 1936 * Go through port list and look for a head for this lport. 1937 */ 1938 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1939 if (phd->phd_port == inp->inp_lport) 1940 break; 1941 } 1942 /* 1943 * If none exists, malloc one and tack it on. 1944 */ 1945 if (phd == NULL) { 1946 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 1947 if (phd == NULL) { 1948 return (ENOBUFS); /* XXX */ 1949 } 1950 phd->phd_port = inp->inp_lport; 1951 LIST_INIT(&phd->phd_pcblist); 1952 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1953 } 1954 inp->inp_phd = phd; 1955 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1956 LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 1957 inp->inp_flags |= INP_INHASHLIST; 1958#ifdef PCBGROUP 1959 if (do_pcbgroup_update) 1960 in_pcbgroup_update(inp); 1961#endif 1962 return (0); 1963} 1964 1965/* 1966 * For now, there are two public interfaces to insert an inpcb into the hash 1967 * lists -- one that does update pcbgroups, and one that doesn't. The latter 1968 * is used only in the TCP syncache, where in_pcbinshash is called before the 1969 * full 4-tuple is set for the inpcb, and we don't want to install in the 1970 * pcbgroup until later. 1971 * 1972 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 1973 * connection groups, and partially initialised inpcbs should not be exposed 1974 * to either reservation hash tables or pcbgroups. 1975 */ 1976int 1977in_pcbinshash(struct inpcb *inp) 1978{ 1979 1980 return (in_pcbinshash_internal(inp, 1)); 1981} 1982 1983int 1984in_pcbinshash_nopcbgroup(struct inpcb *inp) 1985{ 1986 1987 return (in_pcbinshash_internal(inp, 0)); 1988} 1989 1990/* 1991 * Move PCB to the proper hash bucket when { faddr, fport } have been 1992 * changed. NOTE: This does not handle the case of the lport changing (the 1993 * hashed port list would have to be updated as well), so the lport must 1994 * not change after in_pcbinshash() has been called. 1995 */ 1996void 1997in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 1998{ 1999 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2000 struct inpcbhead *head; 2001 u_int32_t hashkey_faddr; 2002 2003 INP_WLOCK_ASSERT(inp); 2004 INP_HASH_WLOCK_ASSERT(pcbinfo); 2005 2006 KASSERT(inp->inp_flags & INP_INHASHLIST, 2007 ("in_pcbrehash: !INP_INHASHLIST")); 2008 2009#ifdef INET6 2010 if (inp->inp_vflag & INP_IPV6) 2011 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */; 2012 else 2013#endif 2014 hashkey_faddr = inp->inp_faddr.s_addr; 2015 2016 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2017 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2018 2019 LIST_REMOVE(inp, inp_hash); 2020 LIST_INSERT_HEAD(head, inp, inp_hash); 2021 2022#ifdef PCBGROUP 2023 if (m != NULL) 2024 in_pcbgroup_update_mbuf(inp, m); 2025 else 2026 in_pcbgroup_update(inp); 2027#endif 2028} 2029 2030void 2031in_pcbrehash(struct inpcb *inp) 2032{ 2033 2034 in_pcbrehash_mbuf(inp, NULL); 2035} 2036 2037/* 2038 * Remove PCB from various lists. 2039 */ 2040static void 2041in_pcbremlists(struct inpcb *inp) 2042{ 2043 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2044 2045#ifdef INVARIANTS 2046 if (pcbinfo == &V_tcbinfo) { 2047 INP_INFO_RLOCK_ASSERT(pcbinfo); 2048 } else { 2049 INP_INFO_WLOCK_ASSERT(pcbinfo); 2050 } 2051#endif 2052 2053 INP_WLOCK_ASSERT(inp); 2054 INP_LIST_WLOCK_ASSERT(pcbinfo); 2055 2056 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2057 if (inp->inp_flags & INP_INHASHLIST) { 2058 struct inpcbport *phd = inp->inp_phd; 2059 2060 INP_HASH_WLOCK(pcbinfo); 2061 LIST_REMOVE(inp, inp_hash); 2062 LIST_REMOVE(inp, inp_portlist); 2063 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 2064 LIST_REMOVE(phd, phd_hash); 2065 free(phd, M_PCB); 2066 } 2067 INP_HASH_WUNLOCK(pcbinfo); 2068 inp->inp_flags &= ~INP_INHASHLIST; 2069 } 2070 LIST_REMOVE(inp, inp_list); 2071 pcbinfo->ipi_count--; 2072#ifdef PCBGROUP 2073 in_pcbgroup_remove(inp); 2074#endif 2075} 2076 2077/* 2078 * A set label operation has occurred at the socket layer, propagate the 2079 * label change into the in_pcb for the socket. 2080 */ 2081void 2082in_pcbsosetlabel(struct socket *so) 2083{ 2084#ifdef MAC 2085 struct inpcb *inp; 2086 2087 inp = sotoinpcb(so); 2088 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2089 2090 INP_WLOCK(inp); 2091 SOCK_LOCK(so); 2092 mac_inpcb_sosetlabel(so, inp); 2093 SOCK_UNLOCK(so); 2094 INP_WUNLOCK(inp); 2095#endif 2096} 2097 2098/* 2099 * ipport_tick runs once per second, determining if random port allocation 2100 * should be continued. If more than ipport_randomcps ports have been 2101 * allocated in the last second, then we return to sequential port 2102 * allocation. We return to random allocation only once we drop below 2103 * ipport_randomcps for at least ipport_randomtime seconds. 2104 */ 2105static void 2106ipport_tick(void *xtp) 2107{ 2108 VNET_ITERATOR_DECL(vnet_iter); 2109 2110 VNET_LIST_RLOCK_NOSLEEP(); 2111 VNET_FOREACH(vnet_iter) { 2112 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2113 if (V_ipport_tcpallocs <= 2114 V_ipport_tcplastcount + V_ipport_randomcps) { 2115 if (V_ipport_stoprandom > 0) 2116 V_ipport_stoprandom--; 2117 } else 2118 V_ipport_stoprandom = V_ipport_randomtime; 2119 V_ipport_tcplastcount = V_ipport_tcpallocs; 2120 CURVNET_RESTORE(); 2121 } 2122 VNET_LIST_RUNLOCK_NOSLEEP(); 2123 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2124} 2125 2126static void 2127ip_fini(void *xtp) 2128{ 2129 2130 callout_stop(&ipport_tick_callout); 2131} 2132 2133/* 2134 * The ipport_callout should start running at about the time we attach the 2135 * inet or inet6 domains. 2136 */ 2137static void 2138ipport_tick_init(const void *unused __unused) 2139{ 2140 2141 /* Start ipport_tick. */ 2142 callout_init(&ipport_tick_callout, CALLOUT_MPSAFE); 2143 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2144 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2145 SHUTDOWN_PRI_DEFAULT); 2146} 2147SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2148 ipport_tick_init, NULL); 2149 2150void 2151inp_wlock(struct inpcb *inp) 2152{ 2153 2154 INP_WLOCK(inp); 2155} 2156 2157void 2158inp_wunlock(struct inpcb *inp) 2159{ 2160 2161 INP_WUNLOCK(inp); 2162} 2163 2164void 2165inp_rlock(struct inpcb *inp) 2166{ 2167 2168 INP_RLOCK(inp); 2169} 2170 2171void 2172inp_runlock(struct inpcb *inp) 2173{ 2174 2175 INP_RUNLOCK(inp); 2176} 2177 2178#ifdef INVARIANTS 2179void 2180inp_lock_assert(struct inpcb *inp) 2181{ 2182 2183 INP_WLOCK_ASSERT(inp); 2184} 2185 2186void 2187inp_unlock_assert(struct inpcb *inp) 2188{ 2189 2190 INP_UNLOCK_ASSERT(inp); 2191} 2192#endif 2193 2194void 2195inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2196{ 2197 struct inpcb *inp; 2198 2199 INP_INFO_WLOCK(&V_tcbinfo); 2200 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2201 INP_WLOCK(inp); 2202 func(inp, arg); 2203 INP_WUNLOCK(inp); 2204 } 2205 INP_INFO_WUNLOCK(&V_tcbinfo); 2206} 2207 2208struct socket * 2209inp_inpcbtosocket(struct inpcb *inp) 2210{ 2211 2212 INP_WLOCK_ASSERT(inp); 2213 return (inp->inp_socket); 2214} 2215 2216struct tcpcb * 2217inp_inpcbtotcpcb(struct inpcb *inp) 2218{ 2219 2220 INP_WLOCK_ASSERT(inp); 2221 return ((struct tcpcb *)inp->inp_ppcb); 2222} 2223 2224int 2225inp_ip_tos_get(const struct inpcb *inp) 2226{ 2227 2228 return (inp->inp_ip_tos); 2229} 2230 2231void 2232inp_ip_tos_set(struct inpcb *inp, int val) 2233{ 2234 2235 inp->inp_ip_tos = val; 2236} 2237 2238void 2239inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2240 uint32_t *faddr, uint16_t *fp) 2241{ 2242 2243 INP_LOCK_ASSERT(inp); 2244 *laddr = inp->inp_laddr.s_addr; 2245 *faddr = inp->inp_faddr.s_addr; 2246 *lp = inp->inp_lport; 2247 *fp = inp->inp_fport; 2248} 2249 2250struct inpcb * 2251so_sotoinpcb(struct socket *so) 2252{ 2253 2254 return (sotoinpcb(so)); 2255} 2256 2257struct tcpcb * 2258so_sototcpcb(struct socket *so) 2259{ 2260 2261 return (sototcpcb(so)); 2262} 2263 2264#ifdef DDB 2265static void 2266db_print_indent(int indent) 2267{ 2268 int i; 2269 2270 for (i = 0; i < indent; i++) 2271 db_printf(" "); 2272} 2273 2274static void 2275db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2276{ 2277 char faddr_str[48], laddr_str[48]; 2278 2279 db_print_indent(indent); 2280 db_printf("%s at %p\n", name, inc); 2281 2282 indent += 2; 2283 2284#ifdef INET6 2285 if (inc->inc_flags & INC_ISIPV6) { 2286 /* IPv6. */ 2287 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2288 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2289 } else 2290#endif 2291 { 2292 /* IPv4. */ 2293 inet_ntoa_r(inc->inc_laddr, laddr_str); 2294 inet_ntoa_r(inc->inc_faddr, faddr_str); 2295 } 2296 db_print_indent(indent); 2297 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2298 ntohs(inc->inc_lport)); 2299 db_print_indent(indent); 2300 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2301 ntohs(inc->inc_fport)); 2302} 2303 2304static void 2305db_print_inpflags(int inp_flags) 2306{ 2307 int comma; 2308 2309 comma = 0; 2310 if (inp_flags & INP_RECVOPTS) { 2311 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2312 comma = 1; 2313 } 2314 if (inp_flags & INP_RECVRETOPTS) { 2315 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2316 comma = 1; 2317 } 2318 if (inp_flags & INP_RECVDSTADDR) { 2319 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2320 comma = 1; 2321 } 2322 if (inp_flags & INP_HDRINCL) { 2323 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2324 comma = 1; 2325 } 2326 if (inp_flags & INP_HIGHPORT) { 2327 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2328 comma = 1; 2329 } 2330 if (inp_flags & INP_LOWPORT) { 2331 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 2332 comma = 1; 2333 } 2334 if (inp_flags & INP_ANONPORT) { 2335 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 2336 comma = 1; 2337 } 2338 if (inp_flags & INP_RECVIF) { 2339 db_printf("%sINP_RECVIF", comma ? ", " : ""); 2340 comma = 1; 2341 } 2342 if (inp_flags & INP_MTUDISC) { 2343 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 2344 comma = 1; 2345 } 2346 if (inp_flags & INP_FAITH) { 2347 db_printf("%sINP_FAITH", comma ? ", " : ""); 2348 comma = 1; 2349 } 2350 if (inp_flags & INP_RECVTTL) { 2351 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 2352 comma = 1; 2353 } 2354 if (inp_flags & INP_DONTFRAG) { 2355 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 2356 comma = 1; 2357 } 2358 if (inp_flags & INP_RECVTOS) { 2359 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 2360 comma = 1; 2361 } 2362 if (inp_flags & IN6P_IPV6_V6ONLY) { 2363 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 2364 comma = 1; 2365 } 2366 if (inp_flags & IN6P_PKTINFO) { 2367 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 2368 comma = 1; 2369 } 2370 if (inp_flags & IN6P_HOPLIMIT) { 2371 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 2372 comma = 1; 2373 } 2374 if (inp_flags & IN6P_HOPOPTS) { 2375 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 2376 comma = 1; 2377 } 2378 if (inp_flags & IN6P_DSTOPTS) { 2379 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 2380 comma = 1; 2381 } 2382 if (inp_flags & IN6P_RTHDR) { 2383 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 2384 comma = 1; 2385 } 2386 if (inp_flags & IN6P_RTHDRDSTOPTS) { 2387 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 2388 comma = 1; 2389 } 2390 if (inp_flags & IN6P_TCLASS) { 2391 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 2392 comma = 1; 2393 } 2394 if (inp_flags & IN6P_AUTOFLOWLABEL) { 2395 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 2396 comma = 1; 2397 } 2398 if (inp_flags & INP_TIMEWAIT) { 2399 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 2400 comma = 1; 2401 } 2402 if (inp_flags & INP_ONESBCAST) { 2403 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 2404 comma = 1; 2405 } 2406 if (inp_flags & INP_DROPPED) { 2407 db_printf("%sINP_DROPPED", comma ? ", " : ""); 2408 comma = 1; 2409 } 2410 if (inp_flags & INP_SOCKREF) { 2411 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 2412 comma = 1; 2413 } 2414 if (inp_flags & IN6P_RFC2292) { 2415 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 2416 comma = 1; 2417 } 2418 if (inp_flags & IN6P_MTU) { 2419 db_printf("IN6P_MTU%s", comma ? ", " : ""); 2420 comma = 1; 2421 } 2422} 2423 2424static void 2425db_print_inpvflag(u_char inp_vflag) 2426{ 2427 int comma; 2428 2429 comma = 0; 2430 if (inp_vflag & INP_IPV4) { 2431 db_printf("%sINP_IPV4", comma ? ", " : ""); 2432 comma = 1; 2433 } 2434 if (inp_vflag & INP_IPV6) { 2435 db_printf("%sINP_IPV6", comma ? ", " : ""); 2436 comma = 1; 2437 } 2438 if (inp_vflag & INP_IPV6PROTO) { 2439 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 2440 comma = 1; 2441 } 2442} 2443 2444static void 2445db_print_inpcb(struct inpcb *inp, const char *name, int indent) 2446{ 2447 2448 db_print_indent(indent); 2449 db_printf("%s at %p\n", name, inp); 2450 2451 indent += 2; 2452 2453 db_print_indent(indent); 2454 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 2455 2456 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 2457 2458 db_print_indent(indent); 2459 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 2460 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 2461 2462 db_print_indent(indent); 2463 db_printf("inp_label: %p inp_flags: 0x%x (", 2464 inp->inp_label, inp->inp_flags); 2465 db_print_inpflags(inp->inp_flags); 2466 db_printf(")\n"); 2467 2468 db_print_indent(indent); 2469 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 2470 inp->inp_vflag); 2471 db_print_inpvflag(inp->inp_vflag); 2472 db_printf(")\n"); 2473 2474 db_print_indent(indent); 2475 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 2476 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 2477 2478 db_print_indent(indent); 2479#ifdef INET6 2480 if (inp->inp_vflag & INP_IPV6) { 2481 db_printf("in6p_options: %p in6p_outputopts: %p " 2482 "in6p_moptions: %p\n", inp->in6p_options, 2483 inp->in6p_outputopts, inp->in6p_moptions); 2484 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 2485 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 2486 inp->in6p_hops); 2487 } else 2488#endif 2489 { 2490 db_printf("inp_ip_tos: %d inp_ip_options: %p " 2491 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 2492 inp->inp_options, inp->inp_moptions); 2493 } 2494 2495 db_print_indent(indent); 2496 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 2497 (uintmax_t)inp->inp_gencnt); 2498} 2499 2500DB_SHOW_COMMAND(inpcb, db_show_inpcb) 2501{ 2502 struct inpcb *inp; 2503 2504 if (!have_addr) { 2505 db_printf("usage: show inpcb <addr>\n"); 2506 return; 2507 } 2508 inp = (struct inpcb *)addr; 2509 2510 db_print_inpcb(inp, "inpcb", 0); 2511} 2512#endif /* DDB */ 2513