1 /*- 2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 3 * 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 * 3. Neither the name of the project 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 PROJECT 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 PROJECT 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 * $KAME: in6_pcb.c,v 1.31 2001/05/21 05:45:10 jinmei Exp $ 30 */ 31 32 /*- 33 * Copyright (c) 1982, 1986, 1991, 1993 34 * The Regents of the University of California. All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 4. Neither the name of the University nor the names of its contributors 45 * may be used to endorse or promote products derived from this software 46 * without specific prior written permission. 47 * 48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 58 * SUCH DAMAGE. 59 * 60 * @(#)in_pcb.c 8.2 (Berkeley) 1/4/94 61 */ 62 63 #include <sys/cdefs.h> 64 __FBSDID("$FreeBSD$"); 65 66 #include "opt_inet.h" 67 #include "opt_inet6.h" 68 #include "opt_ipsec.h" 69 #include "opt_mac.h" 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/malloc.h> 74 #include <sys/mbuf.h> 75 #include <sys/domain.h> 76 #include <sys/protosw.h> 77 #include <sys/socket.h> 78 #include <sys/socketvar.h> 79 #include <sys/sockio.h> 80 #include <sys/errno.h> 81 #include <sys/time.h> 82 #include <sys/priv.h> 83 #include <sys/proc.h> 84 #include <sys/jail.h> 85 #include <sys/vimage.h> 86 87 #include <vm/uma.h> 88 89 #include <net/if.h> 90 #include <net/if_types.h> 91 #include <net/route.h> 92 93 #include <netinet/in.h> 94 #include <netinet/in_var.h> 95 #include <netinet/in_systm.h> 96 #include <netinet/tcp_var.h> 97 #include <netinet/ip6.h> 98 #include <netinet/ip_var.h> 99 100 #include <netinet6/ip6_var.h> 101 #include <netinet6/nd6.h> 102 #include <netinet/in_pcb.h> 103 #include <netinet6/in6_pcb.h> 104 #include <netinet6/scope6_var.h> 105 106 #include <security/mac/mac_framework.h> 107 108 struct in6_addr zeroin6_addr; 109 110 int 111 in6_pcbbind(register struct inpcb *inp, struct sockaddr *nam, 112 struct ucred *cred) 113 { 114 INIT_VNET_INET6(inp->inp_vnet); 115 INIT_VNET_INET(inp->inp_vnet); 116 struct socket *so = inp->inp_socket; 117 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)NULL; 118 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 119 u_short lport = 0; 120 int wild = 0, reuseport = (so->so_options & SO_REUSEPORT); 121 122 INP_INFO_WLOCK_ASSERT(pcbinfo); 123 INP_WLOCK_ASSERT(inp); 124 125 if (!V_in6_ifaddr) /* XXX broken! */ 126 return (EADDRNOTAVAIL); 127 if (inp->inp_lport || !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 128 return (EINVAL); 129 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0) 130 wild = INPLOOKUP_WILDCARD; 131 if (nam) { 132 int error; 133 134 sin6 = (struct sockaddr_in6 *)nam; 135 if (nam->sa_len != sizeof(*sin6)) 136 return (EINVAL); 137 /* 138 * family check. 139 */ 140 if (nam->sa_family != AF_INET6) 141 return (EAFNOSUPPORT); 142 143 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) 144 return(error); 145 146 if (prison_local_ip6(cred, &sin6->sin6_addr, 147 ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) 148 return (EINVAL); 149 150 lport = sin6->sin6_port; 151 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) { 152 /* 153 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 154 * allow compepte duplication of binding if 155 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 156 * and a multicast address is bound on both 157 * new and duplicated sockets. 158 */ 159 if (so->so_options & SO_REUSEADDR) 160 reuseport = SO_REUSEADDR|SO_REUSEPORT; 161 } else if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 162 struct ifaddr *ia = NULL; 163 164 sin6->sin6_port = 0; /* yech... */ 165 if ((ia = ifa_ifwithaddr((struct sockaddr *)sin6)) == 0) 166 return (EADDRNOTAVAIL); 167 168 /* 169 * XXX: bind to an anycast address might accidentally 170 * cause sending a packet with anycast source address. 171 * We should allow to bind to a deprecated address, since 172 * the application dares to use it. 173 */ 174 if (ia && 175 ((struct in6_ifaddr *)ia)->ia6_flags & 176 (IN6_IFF_ANYCAST|IN6_IFF_NOTREADY|IN6_IFF_DETACHED)) { 177 return (EADDRNOTAVAIL); 178 } 179 } 180 if (lport) { 181 struct inpcb *t; 182 183 /* GROSS */ 184 if (ntohs(lport) <= V_ipport_reservedhigh && 185 ntohs(lport) >= V_ipport_reservedlow && 186 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 187 0)) 188 return (EACCES); 189 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr) && 190 priv_check_cred(inp->inp_cred, 191 PRIV_NETINET_REUSEPORT, 0) != 0) { 192 t = in6_pcblookup_local(pcbinfo, 193 &sin6->sin6_addr, lport, 194 INPLOOKUP_WILDCARD, cred); 195 if (t && 196 ((t->inp_vflag & INP_TIMEWAIT) == 0) && 197 (so->so_type != SOCK_STREAM || 198 IN6_IS_ADDR_UNSPECIFIED(&t->in6p_faddr)) && 199 (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) || 200 !IN6_IS_ADDR_UNSPECIFIED(&t->in6p_laddr) || 201 (t->inp_socket->so_options & SO_REUSEPORT) 202 == 0) && (inp->inp_cred->cr_uid != 203 t->inp_cred->cr_uid)) 204 return (EADDRINUSE); 205 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 && 206 IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 207 struct sockaddr_in sin; 208 209 in6_sin6_2_sin(&sin, sin6); 210 t = in_pcblookup_local(pcbinfo, 211 sin.sin_addr, lport, 212 INPLOOKUP_WILDCARD, cred); 213 if (t && 214 ((t->inp_vflag & 215 INP_TIMEWAIT) == 0) && 216 (so->so_type != SOCK_STREAM || 217 ntohl(t->inp_faddr.s_addr) == 218 INADDR_ANY) && 219 (inp->inp_cred->cr_uid != 220 t->inp_cred->cr_uid)) 221 return (EADDRINUSE); 222 } 223 } 224 if (prison_local_ip6(cred, &sin6->sin6_addr, 225 ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) 226 return (EADDRNOTAVAIL); 227 t = in6_pcblookup_local(pcbinfo, &sin6->sin6_addr, 228 lport, wild, cred); 229 if (t && (reuseport & ((t->inp_vflag & INP_TIMEWAIT) ? 230 intotw(t)->tw_so_options : 231 t->inp_socket->so_options)) == 0) 232 return (EADDRINUSE); 233 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0 && 234 IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 235 struct sockaddr_in sin; 236 237 in6_sin6_2_sin(&sin, sin6); 238 t = in_pcblookup_local(pcbinfo, sin.sin_addr, 239 lport, wild, cred); 240 if (t && t->inp_vflag & INP_TIMEWAIT) { 241 if ((reuseport & 242 intotw(t)->tw_so_options) == 0 && 243 (ntohl(t->inp_laddr.s_addr) != 244 INADDR_ANY || ((inp->inp_vflag & 245 INP_IPV6PROTO) == 246 (t->inp_vflag & INP_IPV6PROTO)))) 247 return (EADDRINUSE); 248 } 249 else if (t && 250 (reuseport & t->inp_socket->so_options) 251 == 0 && (ntohl(t->inp_laddr.s_addr) != 252 INADDR_ANY || INP_SOCKAF(so) == 253 INP_SOCKAF(t->inp_socket))) 254 return (EADDRINUSE); 255 } 256 } 257 inp->in6p_laddr = sin6->sin6_addr; 258 } 259 if (prison_local_ip6(cred, &inp->in6p_laddr, 260 ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) 261 return (EINVAL); 262 if (lport == 0) { 263 int e; 264 if ((e = in6_pcbsetport(&inp->in6p_laddr, inp, cred)) != 0) 265 return (e); 266 } else { 267 inp->inp_lport = lport; 268 if (in_pcbinshash(inp) != 0) { 269 inp->in6p_laddr = in6addr_any; 270 inp->inp_lport = 0; 271 return (EAGAIN); 272 } 273 } 274 return (0); 275 } 276 277 /* 278 * Transform old in6_pcbconnect() into an inner subroutine for new 279 * in6_pcbconnect(): Do some validity-checking on the remote 280 * address (in mbuf 'nam') and then determine local host address 281 * (i.e., which interface) to use to access that remote host. 282 * 283 * This preserves definition of in6_pcbconnect(), while supporting a 284 * slightly different version for T/TCP. (This is more than 285 * a bit of a kludge, but cleaning up the internal interfaces would 286 * have forced minor changes in every protocol). 287 */ 288 int 289 in6_pcbladdr(register struct inpcb *inp, struct sockaddr *nam, 290 struct in6_addr **plocal_addr6) 291 { 292 INIT_VNET_INET6(inp->inp_vnet); 293 register struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 294 int error = 0; 295 struct ifnet *ifp = NULL; 296 int scope_ambiguous = 0; 297 298 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 299 INP_WLOCK_ASSERT(inp); 300 301 if (nam->sa_len != sizeof (*sin6)) 302 return (EINVAL); 303 if (sin6->sin6_family != AF_INET6) 304 return (EAFNOSUPPORT); 305 if (sin6->sin6_port == 0) 306 return (EADDRNOTAVAIL); 307 308 if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone) 309 scope_ambiguous = 1; 310 if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0) 311 return(error); 312 313 if (V_in6_ifaddr) { 314 /* 315 * If the destination address is UNSPECIFIED addr, 316 * use the loopback addr, e.g ::1. 317 */ 318 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) 319 sin6->sin6_addr = in6addr_loopback; 320 } 321 if (prison_remote_ip6(inp->inp_cred, &sin6->sin6_addr) != 0) 322 return (EADDRNOTAVAIL); 323 324 /* 325 * XXX: in6_selectsrc might replace the bound local address 326 * with the address specified by setsockopt(IPV6_PKTINFO). 327 * Is it the intended behavior? 328 */ 329 *plocal_addr6 = in6_selectsrc(sin6, inp->in6p_outputopts, 330 inp, NULL, 331 inp->inp_cred, 332 &ifp, &error); 333 if (ifp && scope_ambiguous && 334 (error = in6_setscope(&sin6->sin6_addr, ifp, NULL)) != 0) { 335 return(error); 336 } 337 338 if (*plocal_addr6 == NULL) { 339 if (error == 0) 340 error = EADDRNOTAVAIL; 341 return (error); 342 } 343 /* 344 * Don't do pcblookup call here; return interface in 345 * plocal_addr6 346 * and exit to caller, that will do the lookup. 347 */ 348 349 return (0); 350 } 351 352 /* 353 * Outer subroutine: 354 * Connect from a socket to a specified address. 355 * Both address and port must be specified in argument sin. 356 * If don't have a local address for this socket yet, 357 * then pick one. 358 */ 359 int 360 in6_pcbconnect(register struct inpcb *inp, struct sockaddr *nam, 361 struct ucred *cred) 362 { 363 struct in6_addr *addr6; 364 register struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam; 365 int error; 366 367 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 368 INP_WLOCK_ASSERT(inp); 369 370 /* 371 * Call inner routine, to assign local interface address. 372 * in6_pcbladdr() may automatically fill in sin6_scope_id. 373 */ 374 if ((error = in6_pcbladdr(inp, nam, &addr6)) != 0) 375 return (error); 376 377 if (in6_pcblookup_hash(inp->inp_pcbinfo, &sin6->sin6_addr, 378 sin6->sin6_port, 379 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) 380 ? addr6 : &inp->in6p_laddr, 381 inp->inp_lport, 0, NULL) != NULL) { 382 return (EADDRINUSE); 383 } 384 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { 385 if (inp->inp_lport == 0) { 386 error = in6_pcbbind(inp, (struct sockaddr *)0, cred); 387 if (error) 388 return (error); 389 } 390 inp->in6p_laddr = *addr6; 391 } 392 inp->in6p_faddr = sin6->sin6_addr; 393 inp->inp_fport = sin6->sin6_port; 394 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 395 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 396 if (inp->in6p_flags & IN6P_AUTOFLOWLABEL) 397 inp->in6p_flowinfo |= 398 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK); 399 400 in_pcbrehash(inp); 401 402 return (0); 403 } 404 405 void 406 in6_pcbdisconnect(struct inpcb *inp) 407 { 408 409 INP_INFO_WLOCK_ASSERT(inp->inp_pcbinfo); 410 INP_WLOCK_ASSERT(inp); 411 412 bzero((caddr_t)&inp->in6p_faddr, sizeof(inp->in6p_faddr)); 413 inp->inp_fport = 0; 414 /* clear flowinfo - draft-itojun-ipv6-flowlabel-api-00 */ 415 inp->in6p_flowinfo &= ~IPV6_FLOWLABEL_MASK; 416 in_pcbrehash(inp); 417 } 418 419 struct sockaddr * 420 in6_sockaddr(in_port_t port, struct in6_addr *addr_p) 421 { 422 struct sockaddr_in6 *sin6; 423 424 sin6 = malloc(sizeof *sin6, M_SONAME, M_WAITOK); 425 bzero(sin6, sizeof *sin6); 426 sin6->sin6_family = AF_INET6; 427 sin6->sin6_len = sizeof(*sin6); 428 sin6->sin6_port = port; 429 sin6->sin6_addr = *addr_p; 430 (void)sa6_recoverscope(sin6); /* XXX: should catch errors */ 431 432 return (struct sockaddr *)sin6; 433 } 434 435 struct sockaddr * 436 in6_v4mapsin6_sockaddr(in_port_t port, struct in_addr *addr_p) 437 { 438 struct sockaddr_in sin; 439 struct sockaddr_in6 *sin6_p; 440 441 bzero(&sin, sizeof sin); 442 sin.sin_family = AF_INET; 443 sin.sin_len = sizeof(sin); 444 sin.sin_port = port; 445 sin.sin_addr = *addr_p; 446 447 sin6_p = malloc(sizeof *sin6_p, M_SONAME, 448 M_WAITOK); 449 in6_sin_2_v4mapsin6(&sin, sin6_p); 450 451 return (struct sockaddr *)sin6_p; 452 } 453 454 int 455 in6_getsockaddr(struct socket *so, struct sockaddr **nam) 456 { 457 register struct inpcb *inp; 458 struct in6_addr addr; 459 in_port_t port; 460 461 inp = sotoinpcb(so); 462 KASSERT(inp != NULL, ("in6_getsockaddr: inp == NULL")); 463 464 INP_RLOCK(inp); 465 port = inp->inp_lport; 466 addr = inp->in6p_laddr; 467 INP_RUNLOCK(inp); 468 469 *nam = in6_sockaddr(port, &addr); 470 return 0; 471 } 472 473 int 474 in6_getpeeraddr(struct socket *so, struct sockaddr **nam) 475 { 476 struct inpcb *inp; 477 struct in6_addr addr; 478 in_port_t port; 479 480 inp = sotoinpcb(so); 481 KASSERT(inp != NULL, ("in6_getpeeraddr: inp == NULL")); 482 483 INP_RLOCK(inp); 484 port = inp->inp_fport; 485 addr = inp->in6p_faddr; 486 INP_RUNLOCK(inp); 487 488 *nam = in6_sockaddr(port, &addr); 489 return 0; 490 } 491 492 int 493 in6_mapped_sockaddr(struct socket *so, struct sockaddr **nam) 494 { 495 struct inpcb *inp; 496 int error; 497 498 inp = sotoinpcb(so); 499 KASSERT(inp != NULL, ("in6_mapped_sockaddr: inp == NULL")); 500 501 if ((inp->inp_vflag & (INP_IPV4 | INP_IPV6)) == INP_IPV4) { 502 error = in_getsockaddr(so, nam); 503 if (error == 0) 504 in6_sin_2_v4mapsin6_in_sock(nam); 505 } else { 506 /* scope issues will be handled in in6_getsockaddr(). */ 507 error = in6_getsockaddr(so, nam); 508 } 509 510 return error; 511 } 512 513 int 514 in6_mapped_peeraddr(struct socket *so, struct sockaddr **nam) 515 { 516 struct inpcb *inp; 517 int error; 518 519 inp = sotoinpcb(so); 520 KASSERT(inp != NULL, ("in6_mapped_peeraddr: inp == NULL")); 521 522 if ((inp->inp_vflag & (INP_IPV4 | INP_IPV6)) == INP_IPV4) { 523 error = in_getpeeraddr(so, nam); 524 if (error == 0) 525 in6_sin_2_v4mapsin6_in_sock(nam); 526 } else 527 /* scope issues will be handled in in6_getpeeraddr(). */ 528 error = in6_getpeeraddr(so, nam); 529 530 return error; 531 } 532 533 /* 534 * Pass some notification to all connections of a protocol 535 * associated with address dst. The local address and/or port numbers 536 * may be specified to limit the search. The "usual action" will be 537 * taken, depending on the ctlinput cmd. The caller must filter any 538 * cmds that are uninteresting (e.g., no error in the map). 539 * Call the protocol specific routine (if any) to report 540 * any errors for each matching socket. 541 */ 542 void 543 in6_pcbnotify(struct inpcbinfo *pcbinfo, struct sockaddr *dst, 544 u_int fport_arg, const struct sockaddr *src, u_int lport_arg, 545 int cmd, void *cmdarg, 546 struct inpcb *(*notify)(struct inpcb *, int)) 547 { 548 struct inpcb *inp, *inp_temp; 549 struct sockaddr_in6 sa6_src, *sa6_dst; 550 u_short fport = fport_arg, lport = lport_arg; 551 u_int32_t flowinfo; 552 int errno; 553 554 if ((unsigned)cmd >= PRC_NCMDS || dst->sa_family != AF_INET6) 555 return; 556 557 sa6_dst = (struct sockaddr_in6 *)dst; 558 if (IN6_IS_ADDR_UNSPECIFIED(&sa6_dst->sin6_addr)) 559 return; 560 561 /* 562 * note that src can be NULL when we get notify by local fragmentation. 563 */ 564 sa6_src = (src == NULL) ? sa6_any : *(const struct sockaddr_in6 *)src; 565 flowinfo = sa6_src.sin6_flowinfo; 566 567 /* 568 * Redirects go to all references to the destination, 569 * and use in6_rtchange to invalidate the route cache. 570 * Dead host indications: also use in6_rtchange to invalidate 571 * the cache, and deliver the error to all the sockets. 572 * Otherwise, if we have knowledge of the local port and address, 573 * deliver only to that socket. 574 */ 575 if (PRC_IS_REDIRECT(cmd) || cmd == PRC_HOSTDEAD) { 576 fport = 0; 577 lport = 0; 578 bzero((caddr_t)&sa6_src.sin6_addr, sizeof(sa6_src.sin6_addr)); 579 580 if (cmd != PRC_HOSTDEAD) 581 notify = in6_rtchange; 582 } 583 errno = inet6ctlerrmap[cmd]; 584 INP_INFO_WLOCK(pcbinfo); 585 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 586 INP_WLOCK(inp); 587 if ((inp->inp_vflag & INP_IPV6) == 0) { 588 INP_WUNLOCK(inp); 589 continue; 590 } 591 592 /* 593 * If the error designates a new path MTU for a destination 594 * and the application (associated with this socket) wanted to 595 * know the value, notify. Note that we notify for all 596 * disconnected sockets if the corresponding application 597 * wanted. This is because some UDP applications keep sending 598 * sockets disconnected. 599 * XXX: should we avoid to notify the value to TCP sockets? 600 */ 601 if (cmd == PRC_MSGSIZE && (inp->inp_flags & IN6P_MTU) != 0 && 602 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) || 603 IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, &sa6_dst->sin6_addr))) { 604 ip6_notify_pmtu(inp, (struct sockaddr_in6 *)dst, 605 (u_int32_t *)cmdarg); 606 } 607 608 /* 609 * Detect if we should notify the error. If no source and 610 * destination ports are specifed, but non-zero flowinfo and 611 * local address match, notify the error. This is the case 612 * when the error is delivered with an encrypted buffer 613 * by ESP. Otherwise, just compare addresses and ports 614 * as usual. 615 */ 616 if (lport == 0 && fport == 0 && flowinfo && 617 inp->inp_socket != NULL && 618 flowinfo == (inp->in6p_flowinfo & IPV6_FLOWLABEL_MASK) && 619 IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, &sa6_src.sin6_addr)) 620 goto do_notify; 621 else if (!IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, 622 &sa6_dst->sin6_addr) || 623 inp->inp_socket == 0 || 624 (lport && inp->inp_lport != lport) || 625 (!IN6_IS_ADDR_UNSPECIFIED(&sa6_src.sin6_addr) && 626 !IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, 627 &sa6_src.sin6_addr)) || 628 (fport && inp->inp_fport != fport)) { 629 INP_WUNLOCK(inp); 630 continue; 631 } 632 633 do_notify: 634 if (notify) { 635 if ((*notify)(inp, errno)) 636 INP_WUNLOCK(inp); 637 } else 638 INP_WUNLOCK(inp); 639 } 640 INP_INFO_WUNLOCK(pcbinfo); 641 } 642 643 /* 644 * Lookup a PCB based on the local address and port. 645 */ 646 struct inpcb * 647 in6_pcblookup_local(struct inpcbinfo *pcbinfo, struct in6_addr *laddr, 648 u_short lport, int wild_okay, struct ucred *cred) 649 { 650 register struct inpcb *inp; 651 int matchwild = 3, wildcard; 652 653 INP_INFO_WLOCK_ASSERT(pcbinfo); 654 655 if (!wild_okay) { 656 struct inpcbhead *head; 657 /* 658 * Look for an unconnected (wildcard foreign addr) PCB that 659 * matches the local address and port we're looking for. 660 */ 661 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 662 0, pcbinfo->ipi_hashmask)]; 663 LIST_FOREACH(inp, head, inp_hash) { 664 /* XXX inp locking */ 665 if ((inp->inp_vflag & INP_IPV6) == 0) 666 continue; 667 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) && 668 IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) && 669 inp->inp_lport == lport) { 670 /* Found. */ 671 if (cred == NULL || 672 inp->inp_cred->cr_prison == cred->cr_prison) 673 return (inp); 674 } 675 } 676 /* 677 * Not found. 678 */ 679 return (NULL); 680 } else { 681 struct inpcbporthead *porthash; 682 struct inpcbport *phd; 683 struct inpcb *match = NULL; 684 /* 685 * Best fit PCB lookup. 686 * 687 * First see if this local port is in use by looking on the 688 * port hash list. 689 */ 690 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 691 pcbinfo->ipi_porthashmask)]; 692 LIST_FOREACH(phd, porthash, phd_hash) { 693 if (phd->phd_port == lport) 694 break; 695 } 696 if (phd != NULL) { 697 /* 698 * Port is in use by one or more PCBs. Look for best 699 * fit. 700 */ 701 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 702 wildcard = 0; 703 if (cred != NULL && 704 inp->inp_cred->cr_prison != cred->cr_prison) 705 continue; 706 /* XXX inp locking */ 707 if ((inp->inp_vflag & INP_IPV6) == 0) 708 continue; 709 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) 710 wildcard++; 711 if (!IN6_IS_ADDR_UNSPECIFIED( 712 &inp->in6p_laddr)) { 713 if (IN6_IS_ADDR_UNSPECIFIED(laddr)) 714 wildcard++; 715 else if (!IN6_ARE_ADDR_EQUAL( 716 &inp->in6p_laddr, laddr)) 717 continue; 718 } else { 719 if (!IN6_IS_ADDR_UNSPECIFIED(laddr)) 720 wildcard++; 721 } 722 if (wildcard < matchwild) { 723 match = inp; 724 matchwild = wildcard; 725 if (matchwild == 0) 726 break; 727 } 728 } 729 } 730 return (match); 731 } 732 } 733 734 void 735 in6_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 736 { 737 struct in6pcb *in6p; 738 struct ip6_moptions *im6o; 739 struct in6_multi_mship *imm, *nimm; 740 741 INP_INFO_RLOCK(pcbinfo); 742 LIST_FOREACH(in6p, pcbinfo->ipi_listhead, inp_list) { 743 INP_WLOCK(in6p); 744 im6o = in6p->in6p_moptions; 745 if ((in6p->inp_vflag & INP_IPV6) && 746 im6o) { 747 /* 748 * Unselect the outgoing interface if it is being 749 * detached. 750 */ 751 if (im6o->im6o_multicast_ifp == ifp) 752 im6o->im6o_multicast_ifp = NULL; 753 754 /* 755 * Drop multicast group membership if we joined 756 * through the interface being detached. 757 * XXX controversial - is it really legal for kernel 758 * to force this? 759 */ 760 for (imm = im6o->im6o_memberships.lh_first; 761 imm != NULL; imm = nimm) { 762 nimm = imm->i6mm_chain.le_next; 763 if (imm->i6mm_maddr->in6m_ifp == ifp) { 764 LIST_REMOVE(imm, i6mm_chain); 765 in6_delmulti(imm->i6mm_maddr); 766 free(imm, M_IP6MADDR); 767 } 768 } 769 } 770 INP_WUNLOCK(in6p); 771 } 772 INP_INFO_RUNLOCK(pcbinfo); 773 } 774 775 /* 776 * Check for alternatives when higher level complains 777 * about service problems. For now, invalidate cached 778 * routing information. If the route was created dynamically 779 * (by a redirect), time to try a default gateway again. 780 */ 781 void 782 in6_losing(struct inpcb *in6p) 783 { 784 785 /* 786 * We don't store route pointers in the routing table anymore 787 */ 788 return; 789 } 790 791 /* 792 * After a routing change, flush old routing 793 * and allocate a (hopefully) better one. 794 */ 795 struct inpcb * 796 in6_rtchange(struct inpcb *inp, int errno) 797 { 798 /* 799 * We don't store route pointers in the routing table anymore 800 */ 801 return inp; 802 } 803 804 /* 805 * Lookup PCB in hash list. 806 */ 807 struct inpcb * 808 in6_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in6_addr *faddr, 809 u_int fport_arg, struct in6_addr *laddr, u_int lport_arg, int wildcard, 810 struct ifnet *ifp) 811 { 812 struct inpcbhead *head; 813 struct inpcb *inp, *tmpinp; 814 u_short fport = fport_arg, lport = lport_arg; 815 int faith; 816 817 INP_INFO_LOCK_ASSERT(pcbinfo); 818 819 if (faithprefix_p != NULL) 820 faith = (*faithprefix_p)(laddr); 821 else 822 faith = 0; 823 824 /* 825 * First look for an exact match. 826 */ 827 tmpinp = NULL; 828 head = &pcbinfo->ipi_hashbase[ 829 INP_PCBHASH(faddr->s6_addr32[3] /* XXX */, lport, fport, 830 pcbinfo->ipi_hashmask)]; 831 LIST_FOREACH(inp, head, inp_hash) { 832 /* XXX inp locking */ 833 if ((inp->inp_vflag & INP_IPV6) == 0) 834 continue; 835 if (IN6_ARE_ADDR_EQUAL(&inp->in6p_faddr, faddr) && 836 IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr) && 837 inp->inp_fport == fport && 838 inp->inp_lport == lport) { 839 /* 840 * XXX We should be able to directly return 841 * the inp here, without any checks. 842 * Well unless both bound with SO_REUSEPORT? 843 */ 844 if (jailed(inp->inp_cred)) 845 return (inp); 846 if (tmpinp == NULL) 847 tmpinp = inp; 848 } 849 } 850 if (tmpinp != NULL) 851 return (tmpinp); 852 853 /* 854 * Then look for a wildcard match, if requested. 855 */ 856 if (wildcard == INPLOOKUP_WILDCARD) { 857 struct inpcb *local_wild = NULL, *local_exact = NULL; 858 struct inpcb *jail_wild = NULL; 859 int injail; 860 861 /* 862 * Order of socket selection - we always prefer jails. 863 * 1. jailed, non-wild. 864 * 2. jailed, wild. 865 * 3. non-jailed, non-wild. 866 * 4. non-jailed, wild. 867 */ 868 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 869 0, pcbinfo->ipi_hashmask)]; 870 LIST_FOREACH(inp, head, inp_hash) { 871 /* XXX inp locking */ 872 if ((inp->inp_vflag & INP_IPV6) == 0) 873 continue; 874 875 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) || 876 inp->inp_lport != lport) { 877 continue; 878 } 879 880 /* XXX inp locking */ 881 if (faith && (inp->inp_flags & INP_FAITH) == 0) 882 continue; 883 884 injail = jailed(inp->inp_cred); 885 if (injail) { 886 if (!prison_check_ip6(inp->inp_cred, laddr)) 887 continue; 888 } else { 889 if (local_exact != NULL) 890 continue; 891 } 892 893 if (IN6_ARE_ADDR_EQUAL(&inp->in6p_laddr, laddr)) { 894 if (injail) 895 return (inp); 896 else 897 local_exact = inp; 898 } else if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { 899 if (injail) 900 jail_wild = inp; 901 else 902 local_wild = inp; 903 } 904 } /* LIST_FOREACH */ 905 906 if (jail_wild != NULL) 907 return (jail_wild); 908 if (local_exact != NULL) 909 return (local_exact); 910 if (local_wild != NULL) 911 return (local_wild); 912 } /* if (wildcard == INPLOOKUP_WILDCARD) */ 913 914 /* 915 * Not found. 916 */ 917 return (NULL); 918 } 919 920 void 921 init_sin6(struct sockaddr_in6 *sin6, struct mbuf *m) 922 { 923 struct ip6_hdr *ip; 924 925 ip = mtod(m, struct ip6_hdr *); 926 bzero(sin6, sizeof(*sin6)); 927 sin6->sin6_len = sizeof(*sin6); 928 sin6->sin6_family = AF_INET6; 929 sin6->sin6_addr = ip->ip6_src; 930 931 (void)sa6_recoverscope(sin6); /* XXX: should catch errors... */ 932 933 return; 934 } 935