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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_ipsec.h" 38 #include "opt_inet6.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/domain.h> 45 #include <sys/protosw.h> 46 #include <sys/socket.h> 47 #include <sys/socketvar.h> 48 #include <sys/proc.h> 49 #include <sys/jail.h> 50 #include <sys/kernel.h> 51 #include <sys/sysctl.h> 52 53 #include <machine/limits.h> 54 55 #include <vm/vm_zone.h> 56 57 #include <net/if.h> 58 #include <net/if_types.h> 59 #include <net/route.h> 60 61 #include <netinet/in.h> 62 #include <netinet/in_pcb.h> 63 #include <netinet/in_var.h> 64 #include <netinet/ip_var.h> 65 #ifdef INET6 66 #include <netinet/ip6.h> 67 #include <netinet6/ip6_var.h> 68 #endif /* INET6 */ 69 70 #ifdef IPSEC 71 #include <netinet6/ipsec.h> 72 #include <netkey/key.h> 73 #endif /* IPSEC */ 74 75 struct in_addr zeroin_addr; 76 77 /* 78 * These configure the range of local port addresses assigned to 79 * "unspecified" outgoing connections/packets/whatever. 80 */ 81 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */ 82 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */ 83 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */ 84 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */ 85 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 86 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */ 87 88 #define RANGECHK(var, min, max) \ 89 if ((var) < (min)) { (var) = (min); } \ 90 else if ((var) > (max)) { (var) = (max); } 91 92 static int 93 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 94 { 95 int error = sysctl_handle_int(oidp, 96 oidp->oid_arg1, oidp->oid_arg2, req); 97 if (!error) { 98 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 99 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 100 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX); 101 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX); 102 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX); 103 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX); 104 } 105 return error; 106 } 107 108 #undef RANGECHK 109 110 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); 111 112 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW, 113 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", ""); 114 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW, 115 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", ""); 116 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW, 117 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", ""); 118 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW, 119 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", ""); 120 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW, 121 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", ""); 122 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW, 123 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", ""); 124 125 /* 126 * in_pcb.c: manage the Protocol Control Blocks. 127 * 128 * NOTE: It is assumed that most of these functions will be called at 129 * splnet(). XXX - There are, unfortunately, a few exceptions to this 130 * rule that should be fixed. 131 */ 132 133 /* 134 * Allocate a PCB and associate it with the socket. 135 */ 136 int 137 in_pcballoc(so, pcbinfo, td) 138 struct socket *so; 139 struct inpcbinfo *pcbinfo; 140 struct thread *td; 141 { 142 register struct inpcb *inp; 143 #ifdef IPSEC 144 int error; 145 #endif 146 147 inp = zalloc(pcbinfo->ipi_zone); 148 if (inp == NULL) 149 return (ENOBUFS); 150 bzero((caddr_t)inp, sizeof(*inp)); 151 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 152 inp->inp_pcbinfo = pcbinfo; 153 inp->inp_socket = so; 154 #ifdef IPSEC 155 error = ipsec_init_policy(so, &inp->inp_sp); 156 if (error != 0) { 157 zfree(pcbinfo->ipi_zone, inp); 158 return error; 159 } 160 #endif /*IPSEC*/ 161 #if defined(INET6) 162 if (INP_SOCKAF(so) == AF_INET6 && !ip6_mapped_addr_on) 163 inp->inp_flags |= IN6P_IPV6_V6ONLY; 164 #endif 165 LIST_INSERT_HEAD(pcbinfo->listhead, inp, inp_list); 166 pcbinfo->ipi_count++; 167 so->so_pcb = (caddr_t)inp; 168 #ifdef INET6 169 if (ip6_auto_flowlabel) 170 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 171 #endif 172 return (0); 173 } 174 175 int 176 in_pcbbind(inp, nam, td) 177 register struct inpcb *inp; 178 struct sockaddr *nam; 179 struct thread *td; 180 { 181 struct proc *p = td->td_proc; 182 register struct socket *so = inp->inp_socket; 183 unsigned short *lastport; 184 struct sockaddr_in *sin; 185 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 186 u_short lport = 0; 187 int wild = 0, reuseport = (so->so_options & SO_REUSEPORT); 188 int error, prison = 0; 189 190 if (TAILQ_EMPTY(&in_ifaddrhead)) /* XXX broken! */ 191 return (EADDRNOTAVAIL); 192 if (inp->inp_lport || inp->inp_laddr.s_addr != INADDR_ANY) 193 return (EINVAL); 194 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0) 195 wild = 1; 196 if (nam) { 197 sin = (struct sockaddr_in *)nam; 198 if (nam->sa_len != sizeof (*sin)) 199 return (EINVAL); 200 #ifdef notdef 201 /* 202 * We should check the family, but old programs 203 * incorrectly fail to initialize it. 204 */ 205 if (sin->sin_family != AF_INET) 206 return (EAFNOSUPPORT); 207 #endif 208 if (sin->sin_addr.s_addr != INADDR_ANY) 209 if (prison_ip(p->p_ucred, 0, &sin->sin_addr.s_addr)) 210 return(EINVAL); 211 lport = sin->sin_port; 212 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 213 /* 214 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 215 * allow complete duplication of binding if 216 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 217 * and a multicast address is bound on both 218 * new and duplicated sockets. 219 */ 220 if (so->so_options & SO_REUSEADDR) 221 reuseport = SO_REUSEADDR|SO_REUSEPORT; 222 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 223 sin->sin_port = 0; /* yech... */ 224 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 225 if (ifa_ifwithaddr((struct sockaddr *)sin) == 0) 226 return (EADDRNOTAVAIL); 227 } 228 if (lport) { 229 struct inpcb *t; 230 /* GROSS */ 231 if (ntohs(lport) < IPPORT_RESERVED && p && 232 suser_xxx(0, p, PRISON_ROOT)) 233 return (EACCES); 234 if (p && jailed(p->p_ucred)) 235 prison = 1; 236 if (so->so_cred->cr_uid != 0 && 237 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 238 t = in_pcblookup_local(inp->inp_pcbinfo, 239 sin->sin_addr, lport, 240 prison ? 0 : INPLOOKUP_WILDCARD); 241 if (t && 242 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 243 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 244 (t->inp_socket->so_options & 245 SO_REUSEPORT) == 0) && 246 (so->so_cred->cr_uid != 247 t->inp_socket->so_cred->cr_uid)) { 248 #if defined(INET6) 249 if (ntohl(sin->sin_addr.s_addr) != 250 INADDR_ANY || 251 ntohl(t->inp_laddr.s_addr) != 252 INADDR_ANY || 253 INP_SOCKAF(so) == 254 INP_SOCKAF(t->inp_socket)) 255 #endif /* defined(INET6) */ 256 return (EADDRINUSE); 257 } 258 } 259 if (prison && 260 prison_ip(p->p_ucred, 0, &sin->sin_addr.s_addr)) 261 return (EADDRNOTAVAIL); 262 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 263 lport, prison ? 0 : wild); 264 if (t && 265 (reuseport & t->inp_socket->so_options) == 0) { 266 #if defined(INET6) 267 if (ntohl(sin->sin_addr.s_addr) != 268 INADDR_ANY || 269 ntohl(t->inp_laddr.s_addr) != 270 INADDR_ANY || 271 INP_SOCKAF(so) == 272 INP_SOCKAF(t->inp_socket)) 273 #endif /* defined(INET6) */ 274 return (EADDRINUSE); 275 } 276 } 277 inp->inp_laddr = sin->sin_addr; 278 } 279 if (lport == 0) { 280 ushort first, last; 281 int count; 282 283 if (inp->inp_laddr.s_addr != INADDR_ANY) 284 if (prison_ip(p->p_ucred, 0, &inp->inp_laddr.s_addr )) { 285 inp->inp_laddr.s_addr = INADDR_ANY; 286 return (EINVAL); 287 } 288 inp->inp_flags |= INP_ANONPORT; 289 290 if (inp->inp_flags & INP_HIGHPORT) { 291 first = ipport_hifirstauto; /* sysctl */ 292 last = ipport_hilastauto; 293 lastport = &pcbinfo->lasthi; 294 } else if (inp->inp_flags & INP_LOWPORT) { 295 if (p && (error = suser_xxx(0, p, PRISON_ROOT))) { 296 inp->inp_laddr.s_addr = INADDR_ANY; 297 return error; 298 } 299 first = ipport_lowfirstauto; /* 1023 */ 300 last = ipport_lowlastauto; /* 600 */ 301 lastport = &pcbinfo->lastlow; 302 } else { 303 first = ipport_firstauto; /* sysctl */ 304 last = ipport_lastauto; 305 lastport = &pcbinfo->lastport; 306 } 307 /* 308 * Simple check to ensure all ports are not used up causing 309 * a deadlock here. 310 * 311 * We split the two cases (up and down) so that the direction 312 * is not being tested on each round of the loop. 313 */ 314 if (first > last) { 315 /* 316 * counting down 317 */ 318 count = first - last; 319 320 do { 321 if (count-- < 0) { /* completely used? */ 322 inp->inp_laddr.s_addr = INADDR_ANY; 323 return (EADDRNOTAVAIL); 324 } 325 --*lastport; 326 if (*lastport > first || *lastport < last) 327 *lastport = first; 328 lport = htons(*lastport); 329 } while (in_pcblookup_local(pcbinfo, 330 inp->inp_laddr, lport, wild)); 331 } else { 332 /* 333 * counting up 334 */ 335 count = last - first; 336 337 do { 338 if (count-- < 0) { /* completely used? */ 339 /* 340 * Undo any address bind that may have 341 * occurred above. 342 */ 343 inp->inp_laddr.s_addr = INADDR_ANY; 344 return (EADDRNOTAVAIL); 345 } 346 ++*lastport; 347 if (*lastport < first || *lastport > last) 348 *lastport = first; 349 lport = htons(*lastport); 350 } while (in_pcblookup_local(pcbinfo, 351 inp->inp_laddr, lport, wild)); 352 } 353 } 354 inp->inp_lport = lport; 355 if (prison_ip(p->p_ucred, 0, &inp->inp_laddr.s_addr)) { 356 inp->inp_laddr.s_addr = INADDR_ANY; 357 inp->inp_lport = 0; 358 return (EINVAL); 359 } 360 if (in_pcbinshash(inp) != 0) { 361 inp->inp_laddr.s_addr = INADDR_ANY; 362 inp->inp_lport = 0; 363 return (EAGAIN); 364 } 365 return (0); 366 } 367 368 /* 369 * Transform old in_pcbconnect() into an inner subroutine for new 370 * in_pcbconnect(): Do some validity-checking on the remote 371 * address (in mbuf 'nam') and then determine local host address 372 * (i.e., which interface) to use to access that remote host. 373 * 374 * This preserves definition of in_pcbconnect(), while supporting a 375 * slightly different version for T/TCP. (This is more than 376 * a bit of a kludge, but cleaning up the internal interfaces would 377 * have forced minor changes in every protocol). 378 */ 379 380 int 381 in_pcbladdr(inp, nam, plocal_sin) 382 register struct inpcb *inp; 383 struct sockaddr *nam; 384 struct sockaddr_in **plocal_sin; 385 { 386 struct in_ifaddr *ia; 387 register struct sockaddr_in *sin = (struct sockaddr_in *)nam; 388 389 if (nam->sa_len != sizeof (*sin)) 390 return (EINVAL); 391 if (sin->sin_family != AF_INET) 392 return (EAFNOSUPPORT); 393 if (sin->sin_port == 0) 394 return (EADDRNOTAVAIL); 395 if (!TAILQ_EMPTY(&in_ifaddrhead)) { 396 /* 397 * If the destination address is INADDR_ANY, 398 * use the primary local address. 399 * If the supplied address is INADDR_BROADCAST, 400 * and the primary interface supports broadcast, 401 * choose the broadcast address for that interface. 402 */ 403 if (sin->sin_addr.s_addr == INADDR_ANY) 404 sin->sin_addr = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr; 405 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST && 406 (TAILQ_FIRST(&in_ifaddrhead)->ia_ifp->if_flags & IFF_BROADCAST)) 407 sin->sin_addr = satosin(&TAILQ_FIRST(&in_ifaddrhead)->ia_broadaddr)->sin_addr; 408 } 409 if (inp->inp_laddr.s_addr == INADDR_ANY) { 410 register struct route *ro; 411 412 ia = (struct in_ifaddr *)0; 413 /* 414 * If route is known or can be allocated now, 415 * our src addr is taken from the i/f, else punt. 416 * Note that we should check the address family of the cached 417 * destination, in case of sharing the cache with IPv6. 418 */ 419 ro = &inp->inp_route; 420 if (ro->ro_rt && 421 (ro->ro_dst.sa_family != AF_INET || 422 satosin(&ro->ro_dst)->sin_addr.s_addr != 423 sin->sin_addr.s_addr || 424 inp->inp_socket->so_options & SO_DONTROUTE)) { 425 RTFREE(ro->ro_rt); 426 ro->ro_rt = (struct rtentry *)0; 427 } 428 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0 && /*XXX*/ 429 (ro->ro_rt == (struct rtentry *)0 || 430 ro->ro_rt->rt_ifp == (struct ifnet *)0)) { 431 /* No route yet, so try to acquire one */ 432 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 433 ro->ro_dst.sa_family = AF_INET; 434 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 435 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 436 sin->sin_addr; 437 rtalloc(ro); 438 } 439 /* 440 * If we found a route, use the address 441 * corresponding to the outgoing interface 442 * unless it is the loopback (in case a route 443 * to our address on another net goes to loopback). 444 */ 445 if (ro->ro_rt && !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) 446 ia = ifatoia(ro->ro_rt->rt_ifa); 447 if (ia == 0) { 448 u_short fport = sin->sin_port; 449 450 sin->sin_port = 0; 451 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 452 if (ia == 0) 453 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 454 sin->sin_port = fport; 455 if (ia == 0) 456 ia = TAILQ_FIRST(&in_ifaddrhead); 457 if (ia == 0) 458 return (EADDRNOTAVAIL); 459 } 460 /* 461 * If the destination address is multicast and an outgoing 462 * interface has been set as a multicast option, use the 463 * address of that interface as our source address. 464 */ 465 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 466 inp->inp_moptions != NULL) { 467 struct ip_moptions *imo; 468 struct ifnet *ifp; 469 470 imo = inp->inp_moptions; 471 if (imo->imo_multicast_ifp != NULL) { 472 ifp = imo->imo_multicast_ifp; 473 TAILQ_FOREACH(ia, &in_ifaddrhead, ia_link) 474 if (ia->ia_ifp == ifp) 475 break; 476 if (ia == 0) 477 return (EADDRNOTAVAIL); 478 } 479 } 480 /* 481 * Don't do pcblookup call here; return interface in plocal_sin 482 * and exit to caller, that will do the lookup. 483 */ 484 *plocal_sin = &ia->ia_addr; 485 486 } 487 return(0); 488 } 489 490 /* 491 * Outer subroutine: 492 * Connect from a socket to a specified address. 493 * Both address and port must be specified in argument sin. 494 * If don't have a local address for this socket yet, 495 * then pick one. 496 */ 497 int 498 in_pcbconnect(inp, nam, td) 499 register struct inpcb *inp; 500 struct sockaddr *nam; 501 struct thread *td; 502 { 503 struct sockaddr_in *ifaddr; 504 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 505 struct sockaddr_in sa; 506 struct ucred *cred; 507 int error; 508 509 cred = inp->inp_socket->so_cred; 510 if (inp->inp_laddr.s_addr == INADDR_ANY && jailed(cred)) { 511 bzero(&sa, sizeof (sa)); 512 sa.sin_addr.s_addr = htonl(prison_getip(cred)); 513 sa.sin_len=sizeof (sa); 514 sa.sin_family = AF_INET; 515 error = in_pcbbind(inp, (struct sockaddr *)&sa, td); 516 if (error) 517 return (error); 518 } 519 /* 520 * Call inner routine, to assign local interface address. 521 */ 522 if ((error = in_pcbladdr(inp, nam, &ifaddr)) != 0) 523 return(error); 524 525 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port, 526 inp->inp_laddr.s_addr ? inp->inp_laddr : ifaddr->sin_addr, 527 inp->inp_lport, 0, NULL) != NULL) { 528 return (EADDRINUSE); 529 } 530 if (inp->inp_laddr.s_addr == INADDR_ANY) { 531 if (inp->inp_lport == 0) { 532 error = in_pcbbind(inp, (struct sockaddr *)0, td); 533 if (error) 534 return (error); 535 } 536 inp->inp_laddr = ifaddr->sin_addr; 537 } 538 inp->inp_faddr = sin->sin_addr; 539 inp->inp_fport = sin->sin_port; 540 in_pcbrehash(inp); 541 return (0); 542 } 543 544 void 545 in_pcbdisconnect(inp) 546 struct inpcb *inp; 547 { 548 549 inp->inp_faddr.s_addr = INADDR_ANY; 550 inp->inp_fport = 0; 551 in_pcbrehash(inp); 552 if (inp->inp_socket->so_state & SS_NOFDREF) 553 in_pcbdetach(inp); 554 } 555 556 void 557 in_pcbdetach(inp) 558 struct inpcb *inp; 559 { 560 struct socket *so = inp->inp_socket; 561 struct inpcbinfo *ipi = inp->inp_pcbinfo; 562 563 #ifdef IPSEC 564 ipsec4_delete_pcbpolicy(inp); 565 #endif /*IPSEC*/ 566 inp->inp_gencnt = ++ipi->ipi_gencnt; 567 in_pcbremlists(inp); 568 so->so_pcb = 0; 569 sotryfree(so); 570 if (inp->inp_options) 571 (void)m_free(inp->inp_options); 572 if (inp->inp_route.ro_rt) 573 rtfree(inp->inp_route.ro_rt); 574 ip_freemoptions(inp->inp_moptions); 575 inp->inp_vflag = 0; 576 zfree(ipi->ipi_zone, inp); 577 } 578 579 /* 580 * The calling convention of in_setsockaddr() and in_setpeeraddr() was 581 * modified to match the pru_sockaddr() and pru_peeraddr() entry points 582 * in struct pr_usrreqs, so that protocols can just reference then directly 583 * without the need for a wrapper function. The socket must have a valid 584 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one 585 * except through a kernel programming error, so it is acceptable to panic 586 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap 587 * because there actually /is/ a programming error somewhere... XXX) 588 */ 589 int 590 in_setsockaddr(so, nam) 591 struct socket *so; 592 struct sockaddr **nam; 593 { 594 int s; 595 register struct inpcb *inp; 596 register struct sockaddr_in *sin; 597 598 /* 599 * Do the malloc first in case it blocks. 600 */ 601 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 602 M_WAITOK | M_ZERO); 603 sin->sin_family = AF_INET; 604 sin->sin_len = sizeof(*sin); 605 606 s = splnet(); 607 inp = sotoinpcb(so); 608 if (!inp) { 609 splx(s); 610 free(sin, M_SONAME); 611 return ECONNRESET; 612 } 613 sin->sin_port = inp->inp_lport; 614 sin->sin_addr = inp->inp_laddr; 615 splx(s); 616 617 *nam = (struct sockaddr *)sin; 618 return 0; 619 } 620 621 int 622 in_setpeeraddr(so, nam) 623 struct socket *so; 624 struct sockaddr **nam; 625 { 626 int s; 627 struct inpcb *inp; 628 register struct sockaddr_in *sin; 629 630 /* 631 * Do the malloc first in case it blocks. 632 */ 633 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 634 M_WAITOK | M_ZERO); 635 sin->sin_family = AF_INET; 636 sin->sin_len = sizeof(*sin); 637 638 s = splnet(); 639 inp = sotoinpcb(so); 640 if (!inp) { 641 splx(s); 642 free(sin, M_SONAME); 643 return ECONNRESET; 644 } 645 sin->sin_port = inp->inp_fport; 646 sin->sin_addr = inp->inp_faddr; 647 splx(s); 648 649 *nam = (struct sockaddr *)sin; 650 return 0; 651 } 652 653 void 654 in_pcbnotifyall(head, faddr, errno, notify) 655 struct inpcbhead *head; 656 struct in_addr faddr; 657 int errno; 658 void (*notify) __P((struct inpcb *, int)); 659 { 660 struct inpcb *inp, *ninp; 661 int s; 662 663 s = splnet(); 664 for (inp = LIST_FIRST(head); inp != NULL; inp = ninp) { 665 ninp = LIST_NEXT(inp, inp_list); 666 #ifdef INET6 667 if ((inp->inp_vflag & INP_IPV4) == 0) 668 continue; 669 #endif 670 if (inp->inp_faddr.s_addr != faddr.s_addr || 671 inp->inp_socket == NULL) 672 continue; 673 (*notify)(inp, errno); 674 } 675 splx(s); 676 } 677 678 void 679 in_pcbpurgeif0(head, ifp) 680 struct inpcb *head; 681 struct ifnet *ifp; 682 { 683 struct inpcb *inp; 684 struct ip_moptions *imo; 685 int i, gap; 686 687 for (inp = head; inp != NULL; inp = LIST_NEXT(inp, inp_list)) { 688 imo = inp->inp_moptions; 689 if ((inp->inp_vflag & INP_IPV4) && 690 imo != NULL) { 691 /* 692 * Unselect the outgoing interface if it is being 693 * detached. 694 */ 695 if (imo->imo_multicast_ifp == ifp) 696 imo->imo_multicast_ifp = NULL; 697 698 /* 699 * Drop multicast group membership if we joined 700 * through the interface being detached. 701 */ 702 for (i = 0, gap = 0; i < imo->imo_num_memberships; 703 i++) { 704 if (imo->imo_membership[i]->inm_ifp == ifp) { 705 in_delmulti(imo->imo_membership[i]); 706 gap++; 707 } else if (gap != 0) 708 imo->imo_membership[i - gap] = 709 imo->imo_membership[i]; 710 } 711 imo->imo_num_memberships -= gap; 712 } 713 } 714 } 715 716 /* 717 * Check for alternatives when higher level complains 718 * about service problems. For now, invalidate cached 719 * routing information. If the route was created dynamically 720 * (by a redirect), time to try a default gateway again. 721 */ 722 void 723 in_losing(inp) 724 struct inpcb *inp; 725 { 726 register struct rtentry *rt; 727 struct rt_addrinfo info; 728 729 if ((rt = inp->inp_route.ro_rt)) { 730 bzero((caddr_t)&info, sizeof(info)); 731 info.rti_flags = rt->rt_flags; 732 info.rti_info[RTAX_DST] = rt_key(rt); 733 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 734 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 735 rt_missmsg(RTM_LOSING, &info, rt->rt_flags, 0); 736 if (rt->rt_flags & RTF_DYNAMIC) 737 (void) rtrequest1(RTM_DELETE, &info, NULL); 738 inp->inp_route.ro_rt = NULL; 739 rtfree(rt); 740 /* 741 * A new route can be allocated 742 * the next time output is attempted. 743 */ 744 } 745 } 746 747 /* 748 * After a routing change, flush old routing 749 * and allocate a (hopefully) better one. 750 */ 751 void 752 in_rtchange(inp, errno) 753 register struct inpcb *inp; 754 int errno; 755 { 756 if (inp->inp_route.ro_rt) { 757 rtfree(inp->inp_route.ro_rt); 758 inp->inp_route.ro_rt = 0; 759 /* 760 * A new route can be allocated the next time 761 * output is attempted. 762 */ 763 } 764 } 765 766 /* 767 * Lookup a PCB based on the local address and port. 768 */ 769 struct inpcb * 770 in_pcblookup_local(pcbinfo, laddr, lport_arg, wild_okay) 771 struct inpcbinfo *pcbinfo; 772 struct in_addr laddr; 773 u_int lport_arg; 774 int wild_okay; 775 { 776 register struct inpcb *inp; 777 int matchwild = 3, wildcard; 778 u_short lport = lport_arg; 779 780 if (!wild_okay) { 781 struct inpcbhead *head; 782 /* 783 * Look for an unconnected (wildcard foreign addr) PCB that 784 * matches the local address and port we're looking for. 785 */ 786 head = &pcbinfo->hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->hashmask)]; 787 LIST_FOREACH(inp, head, inp_hash) { 788 #ifdef INET6 789 if ((inp->inp_vflag & INP_IPV4) == 0) 790 continue; 791 #endif 792 if (inp->inp_faddr.s_addr == INADDR_ANY && 793 inp->inp_laddr.s_addr == laddr.s_addr && 794 inp->inp_lport == lport) { 795 /* 796 * Found. 797 */ 798 return (inp); 799 } 800 } 801 /* 802 * Not found. 803 */ 804 return (NULL); 805 } else { 806 struct inpcbporthead *porthash; 807 struct inpcbport *phd; 808 struct inpcb *match = NULL; 809 /* 810 * Best fit PCB lookup. 811 * 812 * First see if this local port is in use by looking on the 813 * port hash list. 814 */ 815 porthash = &pcbinfo->porthashbase[INP_PCBPORTHASH(lport, 816 pcbinfo->porthashmask)]; 817 LIST_FOREACH(phd, porthash, phd_hash) { 818 if (phd->phd_port == lport) 819 break; 820 } 821 if (phd != NULL) { 822 /* 823 * Port is in use by one or more PCBs. Look for best 824 * fit. 825 */ 826 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 827 wildcard = 0; 828 #ifdef INET6 829 if ((inp->inp_vflag & INP_IPV4) == 0) 830 continue; 831 #endif 832 if (inp->inp_faddr.s_addr != INADDR_ANY) 833 wildcard++; 834 if (inp->inp_laddr.s_addr != INADDR_ANY) { 835 if (laddr.s_addr == INADDR_ANY) 836 wildcard++; 837 else if (inp->inp_laddr.s_addr != laddr.s_addr) 838 continue; 839 } else { 840 if (laddr.s_addr != INADDR_ANY) 841 wildcard++; 842 } 843 if (wildcard < matchwild) { 844 match = inp; 845 matchwild = wildcard; 846 if (matchwild == 0) { 847 break; 848 } 849 } 850 } 851 } 852 return (match); 853 } 854 } 855 856 /* 857 * Lookup PCB in hash list. 858 */ 859 struct inpcb * 860 in_pcblookup_hash(pcbinfo, faddr, fport_arg, laddr, lport_arg, wildcard, 861 ifp) 862 struct inpcbinfo *pcbinfo; 863 struct in_addr faddr, laddr; 864 u_int fport_arg, lport_arg; 865 int wildcard; 866 struct ifnet *ifp; 867 { 868 struct inpcbhead *head; 869 register struct inpcb *inp; 870 u_short fport = fport_arg, lport = lport_arg; 871 872 /* 873 * First look for an exact match. 874 */ 875 head = &pcbinfo->hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, pcbinfo->hashmask)]; 876 LIST_FOREACH(inp, head, inp_hash) { 877 #ifdef INET6 878 if ((inp->inp_vflag & INP_IPV4) == 0) 879 continue; 880 #endif 881 if (inp->inp_faddr.s_addr == faddr.s_addr && 882 inp->inp_laddr.s_addr == laddr.s_addr && 883 inp->inp_fport == fport && 884 inp->inp_lport == lport) { 885 /* 886 * Found. 887 */ 888 return (inp); 889 } 890 } 891 if (wildcard) { 892 struct inpcb *local_wild = NULL; 893 #if defined(INET6) 894 struct inpcb *local_wild_mapped = NULL; 895 #endif /* defined(INET6) */ 896 897 head = &pcbinfo->hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->hashmask)]; 898 LIST_FOREACH(inp, head, inp_hash) { 899 #ifdef INET6 900 if ((inp->inp_vflag & INP_IPV4) == 0) 901 continue; 902 #endif 903 if (inp->inp_faddr.s_addr == INADDR_ANY && 904 inp->inp_lport == lport) { 905 if (ifp && ifp->if_type == IFT_FAITH && 906 (inp->inp_flags & INP_FAITH) == 0) 907 continue; 908 if (inp->inp_laddr.s_addr == laddr.s_addr) 909 return (inp); 910 else if (inp->inp_laddr.s_addr == INADDR_ANY) { 911 #if defined(INET6) 912 if (INP_CHECK_SOCKAF(inp->inp_socket, 913 AF_INET6)) 914 local_wild_mapped = inp; 915 else 916 #endif /* defined(INET6) */ 917 local_wild = inp; 918 } 919 } 920 } 921 #if defined(INET6) 922 if (local_wild == NULL) 923 return (local_wild_mapped); 924 #endif /* defined(INET6) */ 925 return (local_wild); 926 } 927 928 /* 929 * Not found. 930 */ 931 return (NULL); 932 } 933 934 /* 935 * Insert PCB onto various hash lists. 936 */ 937 int 938 in_pcbinshash(inp) 939 struct inpcb *inp; 940 { 941 struct inpcbhead *pcbhash; 942 struct inpcbporthead *pcbporthash; 943 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 944 struct inpcbport *phd; 945 u_int32_t hashkey_faddr; 946 947 #ifdef INET6 948 if (inp->inp_vflag & INP_IPV6) 949 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */; 950 else 951 #endif /* INET6 */ 952 hashkey_faddr = inp->inp_faddr.s_addr; 953 954 pcbhash = &pcbinfo->hashbase[INP_PCBHASH(hashkey_faddr, 955 inp->inp_lport, inp->inp_fport, pcbinfo->hashmask)]; 956 957 pcbporthash = &pcbinfo->porthashbase[INP_PCBPORTHASH(inp->inp_lport, 958 pcbinfo->porthashmask)]; 959 960 /* 961 * Go through port list and look for a head for this lport. 962 */ 963 LIST_FOREACH(phd, pcbporthash, phd_hash) { 964 if (phd->phd_port == inp->inp_lport) 965 break; 966 } 967 /* 968 * If none exists, malloc one and tack it on. 969 */ 970 if (phd == NULL) { 971 MALLOC(phd, struct inpcbport *, sizeof(struct inpcbport), M_PCB, M_NOWAIT); 972 if (phd == NULL) { 973 return (ENOBUFS); /* XXX */ 974 } 975 phd->phd_port = inp->inp_lport; 976 LIST_INIT(&phd->phd_pcblist); 977 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 978 } 979 inp->inp_phd = phd; 980 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 981 LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 982 return (0); 983 } 984 985 /* 986 * Move PCB to the proper hash bucket when { faddr, fport } have been 987 * changed. NOTE: This does not handle the case of the lport changing (the 988 * hashed port list would have to be updated as well), so the lport must 989 * not change after in_pcbinshash() has been called. 990 */ 991 void 992 in_pcbrehash(inp) 993 struct inpcb *inp; 994 { 995 struct inpcbhead *head; 996 u_int32_t hashkey_faddr; 997 998 #ifdef INET6 999 if (inp->inp_vflag & INP_IPV6) 1000 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX */; 1001 else 1002 #endif /* INET6 */ 1003 hashkey_faddr = inp->inp_faddr.s_addr; 1004 1005 head = &inp->inp_pcbinfo->hashbase[INP_PCBHASH(hashkey_faddr, 1006 inp->inp_lport, inp->inp_fport, inp->inp_pcbinfo->hashmask)]; 1007 1008 LIST_REMOVE(inp, inp_hash); 1009 LIST_INSERT_HEAD(head, inp, inp_hash); 1010 } 1011 1012 /* 1013 * Remove PCB from various lists. 1014 */ 1015 void 1016 in_pcbremlists(inp) 1017 struct inpcb *inp; 1018 { 1019 inp->inp_gencnt = ++inp->inp_pcbinfo->ipi_gencnt; 1020 if (inp->inp_lport) { 1021 struct inpcbport *phd = inp->inp_phd; 1022 1023 LIST_REMOVE(inp, inp_hash); 1024 LIST_REMOVE(inp, inp_portlist); 1025 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1026 LIST_REMOVE(phd, phd_hash); 1027 free(phd, M_PCB); 1028 } 1029 } 1030 LIST_REMOVE(inp, inp_list); 1031 inp->inp_pcbinfo->ipi_count--; 1032 } 1033 1034 int 1035 prison_xinpcb(struct proc *p, struct inpcb *inp) 1036 { 1037 if (!jailed(p->p_ucred)) 1038 return (0); 1039 if (ntohl(inp->inp_laddr.s_addr) == prison_getip(p->p_ucred)) 1040 return (0); 1041 return (1); 1042 } 1043