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