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