1 /*- 2 * Copyright (c) 1982, 1986, 1991, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_ipsec.h" 34 #include "opt_inet6.h" 35 #include "opt_mac.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/mac.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/domain.h> 43 #include <sys/protosw.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/proc.h> 47 #include <sys/jail.h> 48 #include <sys/kernel.h> 49 #include <sys/sysctl.h> 50 51 #include <vm/uma.h> 52 53 #include <net/if.h> 54 #include <net/if_types.h> 55 #include <net/route.h> 56 57 #include <netinet/in.h> 58 #include <netinet/in_pcb.h> 59 #include <netinet/in_var.h> 60 #include <netinet/ip_var.h> 61 #include <netinet/tcp_var.h> 62 #include <netinet/udp.h> 63 #include <netinet/udp_var.h> 64 #ifdef INET6 65 #include <netinet/ip6.h> 66 #include <netinet6/ip6_var.h> 67 #endif /* INET6 */ 68 69 #ifdef IPSEC 70 #include <netinet6/ipsec.h> 71 #include <netkey/key.h> 72 #endif /* IPSEC */ 73 74 #ifdef FAST_IPSEC 75 #if defined(IPSEC) || defined(IPSEC_ESP) 76 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!" 77 #endif 78 79 #include <netipsec/ipsec.h> 80 #include <netipsec/key.h> 81 #endif /* FAST_IPSEC */ 82 83 /* 84 * These configure the range of local port addresses assigned to 85 * "unspecified" outgoing connections/packets/whatever. 86 */ 87 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */ 88 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */ 89 int ipport_firstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 90 int ipport_lastauto = IPPORT_HILASTAUTO; /* 65535 */ 91 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 92 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */ 93 94 /* 95 * Reserved ports accessible only to root. There are significant 96 * security considerations that must be accounted for when changing these, 97 * but the security benefits can be great. Please be careful. 98 */ 99 int ipport_reservedhigh = IPPORT_RESERVED - 1; /* 1023 */ 100 int ipport_reservedlow = 0; 101 102 /* Variables dealing with random ephemeral port allocation. */ 103 int ipport_randomized = 1; /* user controlled via sysctl */ 104 int ipport_randomcps = 10; /* user controlled via sysctl */ 105 int ipport_randomtime = 45; /* user controlled via sysctl */ 106 int ipport_stoprandom = 0; /* toggled by ipport_tick */ 107 int ipport_tcpallocs; 108 int ipport_tcplastcount; 109 110 #define RANGECHK(var, min, max) \ 111 if ((var) < (min)) { (var) = (min); } \ 112 else if ((var) > (max)) { (var) = (max); } 113 114 static int 115 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 116 { 117 int error; 118 119 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); 120 if (error == 0) { 121 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 122 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 123 RANGECHK(ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 124 RANGECHK(ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 125 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 126 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 127 } 128 return (error); 129 } 130 131 #undef RANGECHK 132 133 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); 134 135 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW, 136 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", ""); 137 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW, 138 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", ""); 139 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW, 140 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", ""); 141 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW, 142 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", ""); 143 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW, 144 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", ""); 145 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW, 146 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", ""); 147 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 148 CTLFLAG_RW|CTLFLAG_SECURE, &ipport_reservedhigh, 0, ""); 149 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 150 CTLFLAG_RW|CTLFLAG_SECURE, &ipport_reservedlow, 0, ""); 151 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_RW, 152 &ipport_randomized, 0, "Enable random port allocation"); 153 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_RW, 154 &ipport_randomcps, 0, "Maximum number of random port " 155 "allocations before switching to a sequental one"); 156 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_RW, 157 &ipport_randomtime, 0, "Minimum time to keep sequental port " 158 "allocation before switching to a random one"); 159 160 /* 161 * in_pcb.c: manage the Protocol Control Blocks. 162 * 163 * NOTE: It is assumed that most of these functions will be called with 164 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 165 * functions often modify hash chains or addresses in pcbs. 166 */ 167 168 /* 169 * Allocate a PCB and associate it with the socket. 170 */ 171 int 172 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo, const char *type) 173 { 174 struct inpcb *inp; 175 int error; 176 177 INP_INFO_WLOCK_ASSERT(pcbinfo); 178 error = 0; 179 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT | M_ZERO); 180 if (inp == NULL) 181 return (ENOBUFS); 182 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 183 inp->inp_pcbinfo = pcbinfo; 184 inp->inp_socket = so; 185 #ifdef MAC 186 error = mac_init_inpcb(inp, M_NOWAIT); 187 if (error != 0) 188 goto out; 189 SOCK_LOCK(so); 190 mac_create_inpcb_from_socket(so, inp); 191 SOCK_UNLOCK(so); 192 #endif 193 #if defined(IPSEC) || defined(FAST_IPSEC) 194 #ifdef FAST_IPSEC 195 error = ipsec_init_policy(so, &inp->inp_sp); 196 #else 197 error = ipsec_init_pcbpolicy(so, &inp->inp_sp); 198 #endif 199 if (error != 0) 200 goto out; 201 #endif /*IPSEC*/ 202 #if defined(INET6) 203 if (INP_SOCKAF(so) == AF_INET6) { 204 inp->inp_vflag |= INP_IPV6PROTO; 205 if (ip6_v6only) 206 inp->inp_flags |= IN6P_IPV6_V6ONLY; 207 } 208 #endif 209 LIST_INSERT_HEAD(pcbinfo->listhead, inp, inp_list); 210 pcbinfo->ipi_count++; 211 so->so_pcb = (caddr_t)inp; 212 INP_LOCK_INIT(inp, "inp", type); 213 #ifdef INET6 214 if (ip6_auto_flowlabel) 215 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 216 #endif 217 #if defined(IPSEC) || defined(FAST_IPSEC) || defined(MAC) 218 out: 219 if (error != 0) 220 uma_zfree(pcbinfo->ipi_zone, inp); 221 #endif 222 return (error); 223 } 224 225 int 226 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 227 { 228 int anonport, error; 229 230 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 231 INP_LOCK_ASSERT(inp); 232 233 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 234 return (EINVAL); 235 anonport = inp->inp_lport == 0 && (nam == NULL || 236 ((struct sockaddr_in *)nam)->sin_port == 0); 237 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 238 &inp->inp_lport, cred); 239 if (error) 240 return (error); 241 if (in_pcbinshash(inp) != 0) { 242 inp->inp_laddr.s_addr = INADDR_ANY; 243 inp->inp_lport = 0; 244 return (EAGAIN); 245 } 246 if (anonport) 247 inp->inp_flags |= INP_ANONPORT; 248 return (0); 249 } 250 251 /* 252 * Set up a bind operation on a PCB, performing port allocation 253 * as required, but do not actually modify the PCB. Callers can 254 * either complete the bind by setting inp_laddr/inp_lport and 255 * calling in_pcbinshash(), or they can just use the resulting 256 * port and address to authorise the sending of a once-off packet. 257 * 258 * On error, the values of *laddrp and *lportp are not changed. 259 */ 260 int 261 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 262 u_short *lportp, struct ucred *cred) 263 { 264 struct socket *so = inp->inp_socket; 265 unsigned short *lastport; 266 struct sockaddr_in *sin; 267 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 268 struct in_addr laddr; 269 u_short lport = 0; 270 int wild = 0, reuseport = (so->so_options & SO_REUSEPORT); 271 int error, prison = 0; 272 int dorandom; 273 274 INP_INFO_WLOCK_ASSERT(pcbinfo); 275 INP_LOCK_ASSERT(inp); 276 277 if (TAILQ_EMPTY(&in_ifaddrhead)) /* XXX broken! */ 278 return (EADDRNOTAVAIL); 279 laddr.s_addr = *laddrp; 280 if (nam != NULL && laddr.s_addr != INADDR_ANY) 281 return (EINVAL); 282 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0) 283 wild = 1; 284 if (nam) { 285 sin = (struct sockaddr_in *)nam; 286 if (nam->sa_len != sizeof (*sin)) 287 return (EINVAL); 288 #ifdef notdef 289 /* 290 * We should check the family, but old programs 291 * incorrectly fail to initialize it. 292 */ 293 if (sin->sin_family != AF_INET) 294 return (EAFNOSUPPORT); 295 #endif 296 if (sin->sin_addr.s_addr != INADDR_ANY) 297 if (prison_ip(cred, 0, &sin->sin_addr.s_addr)) 298 return(EINVAL); 299 if (sin->sin_port != *lportp) { 300 /* Don't allow the port to change. */ 301 if (*lportp != 0) 302 return (EINVAL); 303 lport = sin->sin_port; 304 } 305 /* NB: lport is left as 0 if the port isn't being changed. */ 306 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 307 /* 308 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 309 * allow complete duplication of binding if 310 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 311 * and a multicast address is bound on both 312 * new and duplicated sockets. 313 */ 314 if (so->so_options & SO_REUSEADDR) 315 reuseport = SO_REUSEADDR|SO_REUSEPORT; 316 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 317 sin->sin_port = 0; /* yech... */ 318 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 319 if (ifa_ifwithaddr((struct sockaddr *)sin) == 0) 320 return (EADDRNOTAVAIL); 321 } 322 laddr = sin->sin_addr; 323 if (lport) { 324 struct inpcb *t; 325 struct tcptw *tw; 326 327 /* GROSS */ 328 if (ntohs(lport) <= ipport_reservedhigh && 329 ntohs(lport) >= ipport_reservedlow && 330 suser_cred(cred, SUSER_ALLOWJAIL)) 331 return (EACCES); 332 if (jailed(cred)) 333 prison = 1; 334 if (so->so_cred->cr_uid != 0 && 335 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 336 t = in_pcblookup_local(inp->inp_pcbinfo, 337 sin->sin_addr, lport, 338 prison ? 0 : INPLOOKUP_WILDCARD); 339 /* 340 * XXX 341 * This entire block sorely needs a rewrite. 342 */ 343 if (t && 344 ((t->inp_vflag & INP_TIMEWAIT) == 0) && 345 (so->so_type != SOCK_STREAM || 346 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 347 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 348 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 349 (t->inp_socket->so_options & 350 SO_REUSEPORT) == 0) && 351 (so->so_cred->cr_uid != 352 t->inp_socket->so_cred->cr_uid)) 353 return (EADDRINUSE); 354 } 355 if (prison && prison_ip(cred, 0, &sin->sin_addr.s_addr)) 356 return (EADDRNOTAVAIL); 357 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 358 lport, prison ? 0 : wild); 359 if (t && (t->inp_vflag & INP_TIMEWAIT)) { 360 /* 361 * XXXRW: If an incpb has had its timewait 362 * state recycled, we treat the address as 363 * being in use (for now). This is better 364 * than a panic, but not desirable. 365 */ 366 tw = intotw(inp); 367 if (tw == NULL || 368 (reuseport & tw->tw_so_options) == 0) 369 return (EADDRINUSE); 370 } else if (t && 371 (reuseport & t->inp_socket->so_options) == 0) { 372 #if defined(INET6) 373 if (ntohl(sin->sin_addr.s_addr) != 374 INADDR_ANY || 375 ntohl(t->inp_laddr.s_addr) != 376 INADDR_ANY || 377 INP_SOCKAF(so) == 378 INP_SOCKAF(t->inp_socket)) 379 #endif /* defined(INET6) */ 380 return (EADDRINUSE); 381 } 382 } 383 } 384 if (*lportp != 0) 385 lport = *lportp; 386 if (lport == 0) { 387 u_short first, last; 388 int count; 389 390 if (laddr.s_addr != INADDR_ANY) 391 if (prison_ip(cred, 0, &laddr.s_addr)) 392 return (EINVAL); 393 394 if (inp->inp_flags & INP_HIGHPORT) { 395 first = ipport_hifirstauto; /* sysctl */ 396 last = ipport_hilastauto; 397 lastport = &pcbinfo->lasthi; 398 } else if (inp->inp_flags & INP_LOWPORT) { 399 if ((error = suser_cred(cred, SUSER_ALLOWJAIL)) != 0) 400 return error; 401 first = ipport_lowfirstauto; /* 1023 */ 402 last = ipport_lowlastauto; /* 600 */ 403 lastport = &pcbinfo->lastlow; 404 } else { 405 first = ipport_firstauto; /* sysctl */ 406 last = ipport_lastauto; 407 lastport = &pcbinfo->lastport; 408 } 409 /* 410 * For UDP, use random port allocation as long as the user 411 * allows it. For TCP (and as of yet unknown) connections, 412 * use random port allocation only if the user allows it AND 413 * ipport_tick() allows it. 414 */ 415 if (ipport_randomized && 416 (!ipport_stoprandom || pcbinfo == &udbinfo)) 417 dorandom = 1; 418 else 419 dorandom = 0; 420 /* 421 * It makes no sense to do random port allocation if 422 * we have the only port available. 423 */ 424 if (first == last) 425 dorandom = 0; 426 /* Make sure to not include UDP packets in the count. */ 427 if (pcbinfo != &udbinfo) 428 ipport_tcpallocs++; 429 /* 430 * Simple check to ensure all ports are not used up causing 431 * a deadlock here. 432 * 433 * We split the two cases (up and down) so that the direction 434 * is not being tested on each round of the loop. 435 */ 436 if (first > last) { 437 /* 438 * counting down 439 */ 440 if (dorandom) 441 *lastport = first - 442 (arc4random() % (first - last)); 443 count = first - last; 444 445 do { 446 if (count-- < 0) /* completely used? */ 447 return (EADDRNOTAVAIL); 448 --*lastport; 449 if (*lastport > first || *lastport < last) 450 *lastport = first; 451 lport = htons(*lastport); 452 } while (in_pcblookup_local(pcbinfo, laddr, lport, 453 wild)); 454 } else { 455 /* 456 * counting up 457 */ 458 if (dorandom) 459 *lastport = first + 460 (arc4random() % (last - first)); 461 count = last - first; 462 463 do { 464 if (count-- < 0) /* completely used? */ 465 return (EADDRNOTAVAIL); 466 ++*lastport; 467 if (*lastport < first || *lastport > last) 468 *lastport = first; 469 lport = htons(*lastport); 470 } while (in_pcblookup_local(pcbinfo, laddr, lport, 471 wild)); 472 } 473 } 474 if (prison_ip(cred, 0, &laddr.s_addr)) 475 return (EINVAL); 476 *laddrp = laddr.s_addr; 477 *lportp = lport; 478 return (0); 479 } 480 481 /* 482 * Connect from a socket to a specified address. 483 * Both address and port must be specified in argument sin. 484 * If don't have a local address for this socket yet, 485 * then pick one. 486 */ 487 int 488 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 489 { 490 u_short lport, fport; 491 in_addr_t laddr, faddr; 492 int anonport, error; 493 494 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 495 INP_LOCK_ASSERT(inp); 496 497 lport = inp->inp_lport; 498 laddr = inp->inp_laddr.s_addr; 499 anonport = (lport == 0); 500 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 501 NULL, cred); 502 if (error) 503 return (error); 504 505 /* Do the initial binding of the local address if required. */ 506 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 507 inp->inp_lport = lport; 508 inp->inp_laddr.s_addr = laddr; 509 if (in_pcbinshash(inp) != 0) { 510 inp->inp_laddr.s_addr = INADDR_ANY; 511 inp->inp_lport = 0; 512 return (EAGAIN); 513 } 514 } 515 516 /* Commit the remaining changes. */ 517 inp->inp_lport = lport; 518 inp->inp_laddr.s_addr = laddr; 519 inp->inp_faddr.s_addr = faddr; 520 inp->inp_fport = fport; 521 in_pcbrehash(inp); 522 #ifdef IPSEC 523 if (inp->inp_socket->so_type == SOCK_STREAM) 524 ipsec_pcbconn(inp->inp_sp); 525 #endif 526 if (anonport) 527 inp->inp_flags |= INP_ANONPORT; 528 return (0); 529 } 530 531 /* 532 * Set up for a connect from a socket to the specified address. 533 * On entry, *laddrp and *lportp should contain the current local 534 * address and port for the PCB; these are updated to the values 535 * that should be placed in inp_laddr and inp_lport to complete 536 * the connect. 537 * 538 * On success, *faddrp and *fportp will be set to the remote address 539 * and port. These are not updated in the error case. 540 * 541 * If the operation fails because the connection already exists, 542 * *oinpp will be set to the PCB of that connection so that the 543 * caller can decide to override it. In all other cases, *oinpp 544 * is set to NULL. 545 */ 546 int 547 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 548 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 549 struct inpcb **oinpp, struct ucred *cred) 550 { 551 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 552 struct in_ifaddr *ia; 553 struct sockaddr_in sa; 554 struct ucred *socred; 555 struct inpcb *oinp; 556 struct in_addr laddr, faddr; 557 u_short lport, fport; 558 int error; 559 560 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 561 INP_LOCK_ASSERT(inp); 562 563 if (oinpp != NULL) 564 *oinpp = NULL; 565 if (nam->sa_len != sizeof (*sin)) 566 return (EINVAL); 567 if (sin->sin_family != AF_INET) 568 return (EAFNOSUPPORT); 569 if (sin->sin_port == 0) 570 return (EADDRNOTAVAIL); 571 laddr.s_addr = *laddrp; 572 lport = *lportp; 573 faddr = sin->sin_addr; 574 fport = sin->sin_port; 575 socred = inp->inp_socket->so_cred; 576 if (laddr.s_addr == INADDR_ANY && jailed(socred)) { 577 bzero(&sa, sizeof(sa)); 578 sa.sin_addr.s_addr = htonl(prison_getip(socred)); 579 sa.sin_len = sizeof(sa); 580 sa.sin_family = AF_INET; 581 error = in_pcbbind_setup(inp, (struct sockaddr *)&sa, 582 &laddr.s_addr, &lport, cred); 583 if (error) 584 return (error); 585 } 586 if (!TAILQ_EMPTY(&in_ifaddrhead)) { 587 /* 588 * If the destination address is INADDR_ANY, 589 * use the primary local address. 590 * If the supplied address is INADDR_BROADCAST, 591 * and the primary interface supports broadcast, 592 * choose the broadcast address for that interface. 593 */ 594 if (faddr.s_addr == INADDR_ANY) 595 faddr = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr; 596 else if (faddr.s_addr == (u_long)INADDR_BROADCAST && 597 (TAILQ_FIRST(&in_ifaddrhead)->ia_ifp->if_flags & 598 IFF_BROADCAST)) 599 faddr = satosin(&TAILQ_FIRST( 600 &in_ifaddrhead)->ia_broadaddr)->sin_addr; 601 } 602 if (laddr.s_addr == INADDR_ANY) { 603 ia = (struct in_ifaddr *)0; 604 /* 605 * If route is known our src addr is taken from the i/f, 606 * else punt. 607 * 608 * Find out route to destination 609 */ 610 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 611 ia = ip_rtaddr(faddr); 612 /* 613 * If we found a route, use the address corresponding to 614 * the outgoing interface. 615 * 616 * Otherwise assume faddr is reachable on a directly connected 617 * network and try to find a corresponding interface to take 618 * the source address from. 619 */ 620 if (ia == 0) { 621 bzero(&sa, sizeof(sa)); 622 sa.sin_addr = faddr; 623 sa.sin_len = sizeof(sa); 624 sa.sin_family = AF_INET; 625 626 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sa))); 627 if (ia == 0) 628 ia = ifatoia(ifa_ifwithnet(sintosa(&sa))); 629 if (ia == 0) 630 return (ENETUNREACH); 631 } 632 /* 633 * If the destination address is multicast and an outgoing 634 * interface has been set as a multicast option, use the 635 * address of that interface as our source address. 636 */ 637 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 638 inp->inp_moptions != NULL) { 639 struct ip_moptions *imo; 640 struct ifnet *ifp; 641 642 imo = inp->inp_moptions; 643 if (imo->imo_multicast_ifp != NULL) { 644 ifp = imo->imo_multicast_ifp; 645 TAILQ_FOREACH(ia, &in_ifaddrhead, ia_link) 646 if (ia->ia_ifp == ifp) 647 break; 648 if (ia == 0) 649 return (EADDRNOTAVAIL); 650 } 651 } 652 laddr = ia->ia_addr.sin_addr; 653 } 654 655 oinp = in_pcblookup_hash(inp->inp_pcbinfo, faddr, fport, laddr, lport, 656 0, NULL); 657 if (oinp != NULL) { 658 if (oinpp != NULL) 659 *oinpp = oinp; 660 return (EADDRINUSE); 661 } 662 if (lport == 0) { 663 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 664 cred); 665 if (error) 666 return (error); 667 } 668 *laddrp = laddr.s_addr; 669 *lportp = lport; 670 *faddrp = faddr.s_addr; 671 *fportp = fport; 672 return (0); 673 } 674 675 void 676 in_pcbdisconnect(struct inpcb *inp) 677 { 678 679 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 680 INP_LOCK_ASSERT(inp); 681 682 inp->inp_faddr.s_addr = INADDR_ANY; 683 inp->inp_fport = 0; 684 in_pcbrehash(inp); 685 #ifdef IPSEC 686 ipsec_pcbdisconn(inp->inp_sp); 687 #endif 688 } 689 690 /* 691 * In the old world order, in_pcbdetach() served two functions: to detach the 692 * pcb from the socket/potentially free the socket, and to free the pcb 693 * itself. In the new world order, the protocol code is responsible for 694 * managing the relationship with the socket, and this code simply frees the 695 * pcb. 696 */ 697 void 698 in_pcbdetach(struct inpcb *inp) 699 { 700 701 KASSERT(inp->inp_socket != NULL, ("in_pcbdetach: inp_socket == NULL")); 702 inp->inp_socket->so_pcb = NULL; 703 inp->inp_socket = NULL; 704 } 705 706 void 707 in_pcbfree(struct inpcb *inp) 708 { 709 struct inpcbinfo *ipi = inp->inp_pcbinfo; 710 711 KASSERT(inp->inp_socket == NULL, ("in_pcbfree: inp_socket != NULL")); 712 INP_INFO_WLOCK_ASSERT(ipi); 713 INP_LOCK_ASSERT(inp); 714 715 #if defined(IPSEC) || defined(FAST_IPSEC) 716 ipsec4_delete_pcbpolicy(inp); 717 #endif /*IPSEC*/ 718 inp->inp_gencnt = ++ipi->ipi_gencnt; 719 in_pcbremlists(inp); 720 if (inp->inp_options) 721 (void)m_free(inp->inp_options); 722 ip_freemoptions(inp->inp_moptions); 723 inp->inp_vflag = 0; 724 INP_LOCK_DESTROY(inp); 725 #ifdef MAC 726 mac_destroy_inpcb(inp); 727 #endif 728 uma_zfree(ipi->ipi_zone, inp); 729 } 730 731 /* 732 * TCP needs to maintain its inpcb structure after the TCP connection has 733 * been torn down. However, it must be disconnected from the inpcb hashes as 734 * it must not prevent binding of future connections to the same port/ip 735 * combination by other inpcbs. 736 */ 737 void 738 in_pcbdrop(struct inpcb *inp) 739 { 740 741 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 742 INP_LOCK_ASSERT(inp); 743 744 inp->inp_vflag |= INP_DROPPED; 745 if (inp->inp_lport) { 746 struct inpcbport *phd = inp->inp_phd; 747 748 LIST_REMOVE(inp, inp_hash); 749 LIST_REMOVE(inp, inp_portlist); 750 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 751 LIST_REMOVE(phd, phd_hash); 752 free(phd, M_PCB); 753 } 754 inp->inp_lport = 0; 755 } 756 } 757 758 struct sockaddr * 759 in_sockaddr(in_port_t port, struct in_addr *addr_p) 760 { 761 struct sockaddr_in *sin; 762 763 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 764 M_WAITOK | M_ZERO); 765 sin->sin_family = AF_INET; 766 sin->sin_len = sizeof(*sin); 767 sin->sin_addr = *addr_p; 768 sin->sin_port = port; 769 770 return (struct sockaddr *)sin; 771 } 772 773 /* 774 * The wrapper function will pass down the pcbinfo for this function to lock. 775 * The socket must have a valid 776 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one 777 * except through a kernel programming error, so it is acceptable to panic 778 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap 779 * because there actually /is/ a programming error somewhere... XXX) 780 */ 781 int 782 in_setsockaddr(struct socket *so, struct sockaddr **nam, 783 struct inpcbinfo *pcbinfo) 784 { 785 struct inpcb *inp; 786 struct in_addr addr; 787 in_port_t port; 788 789 inp = sotoinpcb(so); 790 KASSERT(inp != NULL, ("in_setsockaddr: inp == NULL")); 791 792 INP_LOCK(inp); 793 port = inp->inp_lport; 794 addr = inp->inp_laddr; 795 INP_UNLOCK(inp); 796 797 *nam = in_sockaddr(port, &addr); 798 return 0; 799 } 800 801 /* 802 * The wrapper function will pass down the pcbinfo for this function to lock. 803 */ 804 int 805 in_setpeeraddr(struct socket *so, struct sockaddr **nam, 806 struct inpcbinfo *pcbinfo) 807 { 808 struct inpcb *inp; 809 struct in_addr addr; 810 in_port_t port; 811 812 inp = sotoinpcb(so); 813 KASSERT(inp != NULL, ("in_setpeeraddr: inp == NULL")); 814 815 INP_LOCK(inp); 816 port = inp->inp_fport; 817 addr = inp->inp_faddr; 818 INP_UNLOCK(inp); 819 820 *nam = in_sockaddr(port, &addr); 821 return 0; 822 } 823 824 void 825 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 826 struct inpcb *(*notify)(struct inpcb *, int)) 827 { 828 struct inpcb *inp, *ninp; 829 struct inpcbhead *head; 830 831 INP_INFO_WLOCK(pcbinfo); 832 head = pcbinfo->listhead; 833 for (inp = LIST_FIRST(head); inp != NULL; inp = ninp) { 834 INP_LOCK(inp); 835 ninp = LIST_NEXT(inp, inp_list); 836 #ifdef INET6 837 if ((inp->inp_vflag & INP_IPV4) == 0) { 838 INP_UNLOCK(inp); 839 continue; 840 } 841 #endif 842 if (inp->inp_faddr.s_addr != faddr.s_addr || 843 inp->inp_socket == NULL) { 844 INP_UNLOCK(inp); 845 continue; 846 } 847 if ((*notify)(inp, errno)) 848 INP_UNLOCK(inp); 849 } 850 INP_INFO_WUNLOCK(pcbinfo); 851 } 852 853 void 854 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 855 { 856 struct inpcb *inp; 857 struct ip_moptions *imo; 858 int i, gap; 859 860 INP_INFO_RLOCK(pcbinfo); 861 LIST_FOREACH(inp, pcbinfo->listhead, inp_list) { 862 INP_LOCK(inp); 863 imo = inp->inp_moptions; 864 if ((inp->inp_vflag & INP_IPV4) && 865 imo != NULL) { 866 /* 867 * Unselect the outgoing interface if it is being 868 * detached. 869 */ 870 if (imo->imo_multicast_ifp == ifp) 871 imo->imo_multicast_ifp = NULL; 872 873 /* 874 * Drop multicast group membership if we joined 875 * through the interface being detached. 876 */ 877 for (i = 0, gap = 0; i < imo->imo_num_memberships; 878 i++) { 879 if (imo->imo_membership[i]->inm_ifp == ifp) { 880 in_delmulti(imo->imo_membership[i]); 881 gap++; 882 } else if (gap != 0) 883 imo->imo_membership[i - gap] = 884 imo->imo_membership[i]; 885 } 886 imo->imo_num_memberships -= gap; 887 } 888 INP_UNLOCK(inp); 889 } 890 INP_INFO_RUNLOCK(pcbinfo); 891 } 892 893 /* 894 * Lookup a PCB based on the local address and port. 895 */ 896 #define INP_LOOKUP_MAPPED_PCB_COST 3 897 struct inpcb * 898 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 899 u_int lport_arg, int wild_okay) 900 { 901 struct inpcb *inp; 902 struct tcptw *tw; 903 #ifdef INET6 904 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 905 #else 906 int matchwild = 3; 907 #endif 908 int wildcard; 909 u_short lport = lport_arg; 910 911 INP_INFO_WLOCK_ASSERT(pcbinfo); 912 913 if (!wild_okay) { 914 struct inpcbhead *head; 915 /* 916 * Look for an unconnected (wildcard foreign addr) PCB that 917 * matches the local address and port we're looking for. 918 */ 919 head = &pcbinfo->hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->hashmask)]; 920 LIST_FOREACH(inp, head, inp_hash) { 921 #ifdef INET6 922 if ((inp->inp_vflag & INP_IPV4) == 0) 923 continue; 924 #endif 925 if (inp->inp_faddr.s_addr == INADDR_ANY && 926 inp->inp_laddr.s_addr == laddr.s_addr && 927 inp->inp_lport == lport) { 928 /* 929 * Found. 930 */ 931 return (inp); 932 } 933 } 934 /* 935 * Not found. 936 */ 937 return (NULL); 938 } else { 939 struct inpcbporthead *porthash; 940 struct inpcbport *phd; 941 struct inpcb *match = NULL; 942 /* 943 * Best fit PCB lookup. 944 * 945 * First see if this local port is in use by looking on the 946 * port hash list. 947 */ 948 retrylookup: 949 porthash = &pcbinfo->porthashbase[INP_PCBPORTHASH(lport, 950 pcbinfo->porthashmask)]; 951 LIST_FOREACH(phd, porthash, phd_hash) { 952 if (phd->phd_port == lport) 953 break; 954 } 955 if (phd != NULL) { 956 /* 957 * Port is in use by one or more PCBs. Look for best 958 * fit. 959 */ 960 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 961 wildcard = 0; 962 #ifdef INET6 963 if ((inp->inp_vflag & INP_IPV4) == 0) 964 continue; 965 /* 966 * We never select the PCB that has 967 * INP_IPV6 flag and is bound to :: if 968 * we have another PCB which is bound 969 * to 0.0.0.0. If a PCB has the 970 * INP_IPV6 flag, then we set its cost 971 * higher than IPv4 only PCBs. 972 * 973 * Note that the case only happens 974 * when a socket is bound to ::, under 975 * the condition that the use of the 976 * mapped address is allowed. 977 */ 978 if ((inp->inp_vflag & INP_IPV6) != 0) 979 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 980 #endif 981 /* 982 * Clean out old time_wait sockets if they 983 * are clogging up needed local ports. 984 */ 985 if ((inp->inp_vflag & INP_TIMEWAIT) != 0) { 986 tw = intotw(inp); 987 if (tw != NULL && 988 tcp_twrecycleable(tw)) { 989 INP_LOCK(inp); 990 tcp_twclose(tw, 0); 991 match = NULL; 992 goto retrylookup; 993 } 994 } 995 if (inp->inp_faddr.s_addr != INADDR_ANY) 996 wildcard++; 997 if (inp->inp_laddr.s_addr != INADDR_ANY) { 998 if (laddr.s_addr == INADDR_ANY) 999 wildcard++; 1000 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1001 continue; 1002 } else { 1003 if (laddr.s_addr != INADDR_ANY) 1004 wildcard++; 1005 } 1006 if (wildcard < matchwild) { 1007 match = inp; 1008 matchwild = wildcard; 1009 if (matchwild == 0) { 1010 break; 1011 } 1012 } 1013 } 1014 } 1015 return (match); 1016 } 1017 } 1018 #undef INP_LOOKUP_MAPPED_PCB_COST 1019 1020 /* 1021 * Lookup PCB in hash list. 1022 */ 1023 struct inpcb * 1024 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1025 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int wildcard, 1026 struct ifnet *ifp) 1027 { 1028 struct inpcbhead *head; 1029 struct inpcb *inp; 1030 u_short fport = fport_arg, lport = lport_arg; 1031 1032 INP_INFO_RLOCK_ASSERT(pcbinfo); 1033 1034 /* 1035 * First look for an exact match. 1036 */ 1037 head = &pcbinfo->hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, pcbinfo->hashmask)]; 1038 LIST_FOREACH(inp, head, inp_hash) { 1039 #ifdef INET6 1040 if ((inp->inp_vflag & INP_IPV4) == 0) 1041 continue; 1042 #endif 1043 if (inp->inp_faddr.s_addr == faddr.s_addr && 1044 inp->inp_laddr.s_addr == laddr.s_addr && 1045 inp->inp_fport == fport && 1046 inp->inp_lport == lport) { 1047 /* 1048 * Found. 1049 */ 1050 return (inp); 1051 } 1052 } 1053 if (wildcard) { 1054 struct inpcb *local_wild = NULL; 1055 #if defined(INET6) 1056 struct inpcb *local_wild_mapped = NULL; 1057 #endif /* defined(INET6) */ 1058 1059 head = &pcbinfo->hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->hashmask)]; 1060 LIST_FOREACH(inp, head, inp_hash) { 1061 #ifdef INET6 1062 if ((inp->inp_vflag & INP_IPV4) == 0) 1063 continue; 1064 #endif 1065 if (inp->inp_faddr.s_addr == INADDR_ANY && 1066 inp->inp_lport == lport) { 1067 if (ifp && ifp->if_type == IFT_FAITH && 1068 (inp->inp_flags & INP_FAITH) == 0) 1069 continue; 1070 if (inp->inp_laddr.s_addr == laddr.s_addr) 1071 return (inp); 1072 else if (inp->inp_laddr.s_addr == INADDR_ANY) { 1073 #if defined(INET6) 1074 if (INP_CHECK_SOCKAF(inp->inp_socket, 1075 AF_INET6)) 1076 local_wild_mapped = inp; 1077 else 1078 #endif /* defined(INET6) */ 1079 local_wild = inp; 1080 } 1081 } 1082 } 1083 #if defined(INET6) 1084 if (local_wild == NULL) 1085 return (local_wild_mapped); 1086 #endif /* defined(INET6) */ 1087 return (local_wild); 1088 } 1089 1090 /* 1091 * Not found. 1092 */ 1093 return (NULL); 1094 } 1095 1096 /* 1097 * Insert PCB onto various hash lists. 1098 */ 1099 int 1100 in_pcbinshash(struct inpcb *inp) 1101 { 1102 struct inpcbhead *pcbhash; 1103 struct inpcbporthead *pcbporthash; 1104 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1105 struct inpcbport *phd; 1106 u_int32_t hashkey_faddr; 1107 1108 INP_INFO_WLOCK_ASSERT(pcbinfo); 1109 INP_LOCK_ASSERT(inp); 1110 1111 #ifdef INET6 1112 if (inp->inp_vflag & INP_IPV6) 1113 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */; 1114 else 1115 #endif /* INET6 */ 1116 hashkey_faddr = inp->inp_faddr.s_addr; 1117 1118 pcbhash = &pcbinfo->hashbase[INP_PCBHASH(hashkey_faddr, 1119 inp->inp_lport, inp->inp_fport, pcbinfo->hashmask)]; 1120 1121 pcbporthash = &pcbinfo->porthashbase[INP_PCBPORTHASH(inp->inp_lport, 1122 pcbinfo->porthashmask)]; 1123 1124 /* 1125 * Go through port list and look for a head for this lport. 1126 */ 1127 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1128 if (phd->phd_port == inp->inp_lport) 1129 break; 1130 } 1131 /* 1132 * If none exists, malloc one and tack it on. 1133 */ 1134 if (phd == NULL) { 1135 MALLOC(phd, struct inpcbport *, sizeof(struct inpcbport), M_PCB, M_NOWAIT); 1136 if (phd == NULL) { 1137 return (ENOBUFS); /* XXX */ 1138 } 1139 phd->phd_port = inp->inp_lport; 1140 LIST_INIT(&phd->phd_pcblist); 1141 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1142 } 1143 inp->inp_phd = phd; 1144 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1145 LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 1146 return (0); 1147 } 1148 1149 /* 1150 * Move PCB to the proper hash bucket when { faddr, fport } have been 1151 * changed. NOTE: This does not handle the case of the lport changing (the 1152 * hashed port list would have to be updated as well), so the lport must 1153 * not change after in_pcbinshash() has been called. 1154 */ 1155 void 1156 in_pcbrehash(struct inpcb *inp) 1157 { 1158 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1159 struct inpcbhead *head; 1160 u_int32_t hashkey_faddr; 1161 1162 INP_INFO_WLOCK_ASSERT(pcbinfo); 1163 INP_LOCK_ASSERT(inp); 1164 1165 #ifdef INET6 1166 if (inp->inp_vflag & INP_IPV6) 1167 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */; 1168 else 1169 #endif /* INET6 */ 1170 hashkey_faddr = inp->inp_faddr.s_addr; 1171 1172 head = &pcbinfo->hashbase[INP_PCBHASH(hashkey_faddr, 1173 inp->inp_lport, inp->inp_fport, pcbinfo->hashmask)]; 1174 1175 LIST_REMOVE(inp, inp_hash); 1176 LIST_INSERT_HEAD(head, inp, inp_hash); 1177 } 1178 1179 /* 1180 * Remove PCB from various lists. 1181 */ 1182 void 1183 in_pcbremlists(struct inpcb *inp) 1184 { 1185 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1186 1187 INP_INFO_WLOCK_ASSERT(pcbinfo); 1188 INP_LOCK_ASSERT(inp); 1189 1190 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1191 if (inp->inp_lport) { 1192 struct inpcbport *phd = inp->inp_phd; 1193 1194 LIST_REMOVE(inp, inp_hash); 1195 LIST_REMOVE(inp, inp_portlist); 1196 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1197 LIST_REMOVE(phd, phd_hash); 1198 free(phd, M_PCB); 1199 } 1200 } 1201 LIST_REMOVE(inp, inp_list); 1202 pcbinfo->ipi_count--; 1203 } 1204 1205 /* 1206 * A set label operation has occurred at the socket layer, propagate the 1207 * label change into the in_pcb for the socket. 1208 */ 1209 void 1210 in_pcbsosetlabel(struct socket *so) 1211 { 1212 #ifdef MAC 1213 struct inpcb *inp; 1214 1215 inp = sotoinpcb(so); 1216 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 1217 1218 INP_LOCK(inp); 1219 SOCK_LOCK(so); 1220 mac_inpcb_sosetlabel(so, inp); 1221 SOCK_UNLOCK(so); 1222 INP_UNLOCK(inp); 1223 #endif 1224 } 1225 1226 /* 1227 * ipport_tick runs once per second, determining if random port allocation 1228 * should be continued. If more than ipport_randomcps ports have been 1229 * allocated in the last second, then we return to sequential port 1230 * allocation. We return to random allocation only once we drop below 1231 * ipport_randomcps for at least ipport_randomtime seconds. 1232 */ 1233 void 1234 ipport_tick(void *xtp) 1235 { 1236 1237 if (ipport_tcpallocs <= ipport_tcplastcount + ipport_randomcps) { 1238 if (ipport_stoprandom > 0) 1239 ipport_stoprandom--; 1240 } else 1241 ipport_stoprandom = ipport_randomtime; 1242 ipport_tcplastcount = ipport_tcpallocs; 1243 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 1244 } 1245