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