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