1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1991, 1993, 1995 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 63 * $FreeBSD: src/sys/netinet/in_pcb.c,v 1.59.2.27 2004/01/02 04:06:42 ambrisko Exp $ 64 */ 65 66 #include "opt_ipsec.h" 67 #include "opt_inet6.h" 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/malloc.h> 72 #include <sys/mbuf.h> 73 #include <sys/domain.h> 74 #include <sys/protosw.h> 75 #include <sys/socket.h> 76 #include <sys/socketvar.h> 77 #include <sys/proc.h> 78 #include <sys/priv.h> 79 #include <sys/jail.h> 80 #include <sys/kernel.h> 81 #include <sys/sysctl.h> 82 83 #include <sys/thread2.h> 84 #include <sys/socketvar2.h> 85 #include <sys/msgport2.h> 86 87 #include <machine/limits.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_pcb.h> 95 #include <netinet/in_var.h> 96 #include <netinet/ip_var.h> 97 #ifdef INET6 98 #include <netinet/ip6.h> 99 #include <netinet6/ip6_var.h> 100 #endif /* INET6 */ 101 102 #ifdef IPSEC 103 #include <netinet6/ipsec.h> 104 #include <netproto/key/key.h> 105 #include <netproto/ipsec/esp_var.h> 106 #endif 107 108 #ifdef FAST_IPSEC 109 #if defined(IPSEC) || defined(IPSEC_ESP) 110 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!" 111 #endif 112 113 #include <netproto/ipsec/ipsec.h> 114 #include <netproto/ipsec/key.h> 115 #define IPSEC 116 #endif /* FAST_IPSEC */ 117 118 #define INP_LOCALGROUP_SIZMIN 8 119 #define INP_LOCALGROUP_SIZMAX 256 120 121 struct in_addr zeroin_addr; 122 123 /* 124 * These configure the range of local port addresses assigned to 125 * "unspecified" outgoing connections/packets/whatever. 126 */ 127 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */ 128 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */ 129 130 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */ 131 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */ 132 133 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 134 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */ 135 136 #define RANGECHK(var, min, max) \ 137 if ((var) < (min)) { (var) = (min); } \ 138 else if ((var) > (max)) { (var) = (max); } 139 140 int udpencap_enable = 1; /* enabled by default */ 141 int udpencap_port = 4500; /* triggers decapsulation */ 142 143 static int 144 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 145 { 146 int error; 147 148 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); 149 if (!error) { 150 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 151 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 152 153 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX); 154 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX); 155 156 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX); 157 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX); 158 } 159 return (error); 160 } 161 162 #undef RANGECHK 163 164 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); 165 166 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW, 167 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", ""); 168 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW, 169 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", ""); 170 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW, 171 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", ""); 172 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW, 173 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", ""); 174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW, 175 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", ""); 176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW, 177 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", ""); 178 179 /* 180 * in_pcb.c: manage the Protocol Control Blocks. 181 * 182 * NOTE: It is assumed that most of these functions will be called from 183 * a critical section. XXX - There are, unfortunately, a few exceptions 184 * to this rule that should be fixed. 185 * 186 * NOTE: The caller should initialize the cpu field to the cpu running the 187 * protocol stack associated with this inpcbinfo. 188 */ 189 190 void 191 in_pcbinfo_init(struct inpcbinfo *pcbinfo) 192 { 193 LIST_INIT(&pcbinfo->pcblisthead); 194 pcbinfo->cpu = -1; 195 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB, 196 M_WAITOK | M_ZERO); 197 } 198 199 struct baddynamicports baddynamicports; 200 201 /* 202 * Check if the specified port is invalid for dynamic allocation. 203 */ 204 int 205 in_baddynamic(u_int16_t port, u_int16_t proto) 206 { 207 switch (proto) { 208 case IPPROTO_TCP: 209 return (DP_ISSET(baddynamicports.tcp, port)); 210 case IPPROTO_UDP: 211 #ifdef IPSEC 212 /* Cannot preset this as it is a sysctl */ 213 if (port == udpencap_port) 214 return (1); 215 #endif 216 return (DP_ISSET(baddynamicports.udp, port)); 217 default: 218 return (0); 219 } 220 } 221 222 223 /* 224 * Allocate a PCB and associate it with the socket. 225 */ 226 int 227 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 228 { 229 struct inpcb *inp; 230 #ifdef IPSEC 231 int error; 232 #endif 233 234 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK); 235 if (inp == NULL) 236 return (ENOMEM); 237 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 238 inp->inp_pcbinfo = inp->inp_cpcbinfo = pcbinfo; 239 inp->inp_socket = so; 240 #ifdef IPSEC 241 error = ipsec_init_policy(so, &inp->inp_sp); 242 if (error != 0) { 243 kfree(inp, M_PCB); 244 return (error); 245 } 246 #endif 247 #ifdef INET6 248 if (INP_SOCKAF(so) == AF_INET6 && ip6_v6only) 249 inp->inp_flags |= IN6P_IPV6_V6ONLY; 250 if (ip6_auto_flowlabel) 251 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 252 #endif 253 soreference(so); 254 so->so_pcb = inp; 255 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list); 256 pcbinfo->ipi_count++; 257 return (0); 258 } 259 260 /* 261 * Unlink a pcb with the intention of moving it to another cpu with a 262 * different pcbinfo. While unlinked nothing should attempt to dereference 263 * inp_pcbinfo, NULL it out so we assert if it does. 264 */ 265 void 266 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 267 { 268 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch")); 269 KASSERT(inp->inp_cpcbinfo == pcbinfo, ("cpcbinfo mismatch")); 270 KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0, 271 ("already linked")); 272 273 LIST_REMOVE(inp, inp_list); 274 pcbinfo->ipi_count--; 275 inp->inp_pcbinfo = NULL; 276 inp->inp_cpcbinfo = NULL; 277 } 278 279 /* 280 * Relink a pcb into a new pcbinfo. 281 */ 282 void 283 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 284 { 285 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo")); 286 KASSERT(inp->inp_cpcbinfo == NULL, ("has cpcbinfo")); 287 KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0, 288 ("already linked")); 289 290 inp->inp_cpcbinfo = pcbinfo; 291 inp->inp_pcbinfo = pcbinfo; 292 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list); 293 pcbinfo->ipi_count++; 294 } 295 296 static int 297 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred) 298 { 299 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 300 struct inpcbportinfo *portinfo; 301 u_short first, last, lport, step; 302 u_short *lastport; 303 int count, error; 304 int portinfo_first, portinfo_idx; 305 306 inp->inp_flags |= INP_ANONPORT; 307 308 step = pcbinfo->portinfo_mask + 1; 309 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 310 portinfo_idx = portinfo_first; 311 loop: 312 portinfo = &pcbinfo->portinfo[portinfo_idx]; 313 314 if (inp->inp_flags & INP_HIGHPORT) { 315 first = ipport_hifirstauto; /* sysctl */ 316 last = ipport_hilastauto; 317 lastport = &portinfo->lasthi; 318 } else if (inp->inp_flags & INP_LOWPORT) { 319 if (cred && 320 (error = 321 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 322 inp->inp_laddr.s_addr = INADDR_ANY; 323 return error; 324 } 325 first = ipport_lowfirstauto; /* 1023 */ 326 last = ipport_lowlastauto; /* 600 */ 327 lastport = &portinfo->lastlow; 328 } else { 329 first = ipport_firstauto; /* sysctl */ 330 last = ipport_lastauto; 331 lastport = &portinfo->lastport; 332 } 333 334 /* 335 * This has to be atomic. If the porthash is shared across multiple 336 * protocol threads (aka tcp) then the token must be held. 337 */ 338 GET_PORT_TOKEN(portinfo); 339 340 /* 341 * Simple check to ensure all ports are not used up causing 342 * a deadlock here. 343 * 344 * We split the two cases (up and down) so that the direction 345 * is not being tested on each round of the loop. 346 */ 347 if (first > last) { 348 /* 349 * counting down 350 */ 351 in_pcbportrange(&first, &last, portinfo->offset, step); 352 count = (first - last) / step; 353 354 do { 355 if (count-- < 0) { /* completely used? */ 356 error = EADDRNOTAVAIL; 357 goto done; 358 } 359 *lastport -= step; 360 if (*lastport > first || *lastport < last) 361 *lastport = first; 362 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 363 portinfo->offset); 364 lport = htons(*lastport); 365 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport, 366 wild, cred)); 367 } else { 368 /* 369 * counting up 370 */ 371 in_pcbportrange(&last, &first, portinfo->offset, step); 372 count = (last - first) / step; 373 374 do { 375 if (count-- < 0) { /* completely used? */ 376 error = EADDRNOTAVAIL; 377 goto done; 378 } 379 *lastport += step; 380 if (*lastport < first || *lastport > last) 381 *lastport = first; 382 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 383 portinfo->offset); 384 lport = htons(*lastport); 385 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport, 386 wild, cred)); 387 } 388 inp->inp_lport = lport; 389 in_pcbinsporthash(portinfo, inp); 390 error = 0; 391 done: 392 REL_PORT_TOKEN(portinfo); 393 394 if (error) { 395 /* Try next portinfo */ 396 portinfo_idx++; 397 portinfo_idx &= pcbinfo->portinfo_mask; 398 if (portinfo_idx != portinfo_first) 399 goto loop; 400 inp->inp_laddr.s_addr = INADDR_ANY; 401 } 402 return error; 403 } 404 405 int 406 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 407 { 408 struct socket *so = inp->inp_socket; 409 struct sockaddr_in jsin; 410 struct ucred *cred = NULL; 411 int wild = 0; 412 413 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 414 return (EADDRNOTAVAIL); 415 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 416 return (EINVAL); /* already bound */ 417 418 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT))) 419 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */ 420 if (td->td_proc) 421 cred = td->td_proc->p_ucred; 422 423 if (nam != NULL) { 424 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 425 struct inpcbinfo *pcbinfo; 426 struct inpcbportinfo *portinfo; 427 struct inpcb *t; 428 u_short lport, lport_ho; 429 int reuseport = (so->so_options & SO_REUSEPORT); 430 int error; 431 432 if (nam->sa_len != sizeof *sin) 433 return (EINVAL); 434 #ifdef notdef 435 /* 436 * We should check the family, but old programs 437 * incorrectly fail to initialize it. 438 */ 439 if (sin->sin_family != AF_INET) 440 return (EAFNOSUPPORT); 441 #endif 442 if (!prison_replace_wildcards(td, nam)) 443 return (EINVAL); 444 445 lport = sin->sin_port; 446 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 447 /* 448 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 449 * allow complete duplication of binding if 450 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 451 * and a multicast address is bound on both 452 * new and duplicated sockets. 453 */ 454 if (so->so_options & SO_REUSEADDR) 455 reuseport = SO_REUSEADDR | SO_REUSEPORT; 456 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 457 sin->sin_port = 0; /* yech... */ 458 bzero(&sin->sin_zero, sizeof sin->sin_zero); 459 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL) 460 return (EADDRNOTAVAIL); 461 } 462 463 inp->inp_laddr = sin->sin_addr; 464 465 jsin.sin_family = AF_INET; 466 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 467 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 468 inp->inp_laddr.s_addr = INADDR_ANY; 469 return (EINVAL); 470 } 471 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 472 473 if (lport == 0) { 474 /* Auto-select local port */ 475 return in_pcbsetlport(inp, wild, cred); 476 } 477 lport_ho = ntohs(lport); 478 479 /* GROSS */ 480 if (lport_ho < IPPORT_RESERVED && cred && 481 (error = 482 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 483 inp->inp_laddr.s_addr = INADDR_ANY; 484 return (error); 485 } 486 487 /* 488 * Locate the proper portinfo based on lport 489 */ 490 pcbinfo = inp->inp_pcbinfo; 491 portinfo = 492 &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 493 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == 494 portinfo->offset); 495 496 /* 497 * This has to be atomic. If the porthash is shared across 498 * multiple protocol threads (aka tcp) then the token must 499 * be held. 500 */ 501 GET_PORT_TOKEN(portinfo); 502 503 if (so->so_cred->cr_uid != 0 && 504 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 505 t = in_pcblookup_local(portinfo, sin->sin_addr, lport, 506 INPLOOKUP_WILDCARD, cred); 507 if (t && 508 (!in_nullhost(sin->sin_addr) || 509 !in_nullhost(t->inp_laddr) || 510 (t->inp_socket->so_options & SO_REUSEPORT) == 0) && 511 (so->so_cred->cr_uid != 512 t->inp_socket->so_cred->cr_uid)) { 513 #ifdef INET6 514 if (!in_nullhost(sin->sin_addr) || 515 !in_nullhost(t->inp_laddr) || 516 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket)) 517 #endif 518 { 519 inp->inp_laddr.s_addr = INADDR_ANY; 520 error = EADDRINUSE; 521 goto done; 522 } 523 } 524 } 525 if (cred && !prison_replace_wildcards(td, nam)) { 526 inp->inp_laddr.s_addr = INADDR_ANY; 527 error = EADDRNOTAVAIL; 528 goto done; 529 } 530 t = in_pcblookup_local(portinfo, sin->sin_addr, lport, 531 wild, cred); 532 if (t && !(reuseport & t->inp_socket->so_options)) { 533 #ifdef INET6 534 if (!in_nullhost(sin->sin_addr) || 535 !in_nullhost(t->inp_laddr) || 536 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket)) 537 #endif 538 { 539 inp->inp_laddr.s_addr = INADDR_ANY; 540 error = EADDRINUSE; 541 goto done; 542 } 543 } 544 inp->inp_lport = lport; 545 in_pcbinsporthash(portinfo, inp); 546 error = 0; 547 done: 548 REL_PORT_TOKEN(portinfo); 549 return (error); 550 } else { 551 jsin.sin_family = AF_INET; 552 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 553 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 554 inp->inp_laddr.s_addr = INADDR_ANY; 555 return (EINVAL); 556 } 557 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 558 559 return in_pcbsetlport(inp, wild, cred); 560 } 561 } 562 563 static struct inpcb * 564 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr, 565 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred) 566 { 567 struct inpcb *inp; 568 struct inpcbporthead *porthash; 569 struct inpcbport *phd; 570 struct inpcb *match = NULL; 571 572 /* 573 * If the porthashbase is shared across several cpus, it must 574 * have been locked. 575 */ 576 ASSERT_PORT_TOKEN_HELD(portinfo); 577 578 /* 579 * Best fit PCB lookup. 580 * 581 * First see if this local port is in use by looking on the 582 * port hash list. 583 */ 584 porthash = &portinfo->porthashbase[ 585 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 586 LIST_FOREACH(phd, porthash, phd_hash) { 587 if (phd->phd_port == lport) 588 break; 589 } 590 if (phd != NULL) { 591 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 592 #ifdef INET6 593 if ((inp->inp_vflag & INP_IPV4) == 0) 594 continue; 595 #endif 596 if (inp->inp_laddr.s_addr != INADDR_ANY && 597 inp->inp_laddr.s_addr != laddr.s_addr) 598 continue; 599 600 if (inp->inp_faddr.s_addr != INADDR_ANY && 601 inp->inp_faddr.s_addr != faddr.s_addr) 602 continue; 603 604 if (inp->inp_fport != 0 && inp->inp_fport != fport) 605 continue; 606 607 if (cred == NULL || 608 cred->cr_prison == 609 inp->inp_socket->so_cred->cr_prison) { 610 match = inp; 611 break; 612 } 613 } 614 } 615 return (match); 616 } 617 618 int 619 in_pcbsetlport_remote(struct inpcb *inp, const struct sockaddr *remote, 620 struct thread *td) 621 { 622 struct proc *p = td->td_proc; 623 unsigned short *lastport; 624 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote; 625 struct sockaddr_in jsin; 626 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 627 struct inpcbportinfo *portinfo; 628 struct ucred *cred = NULL; 629 u_short first, last, lport, step; 630 int count, error, dup; 631 int portinfo_first, portinfo_idx; 632 633 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 634 return (EADDRNOTAVAIL); 635 636 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY); 637 if (inp->inp_lport != 0) 638 return (EINVAL); /* already bound */ 639 640 KKASSERT(p); 641 cred = p->p_ucred; 642 643 jsin.sin_family = AF_INET; 644 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 645 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 646 inp->inp_laddr.s_addr = INADDR_ANY; 647 return (EINVAL); 648 } 649 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 650 651 inp->inp_flags |= INP_ANONPORT; 652 653 step = pcbinfo->portinfo_mask + 1; 654 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 655 portinfo_idx = portinfo_first; 656 loop: 657 portinfo = &pcbinfo->portinfo[portinfo_idx]; 658 dup = 0; 659 660 if (inp->inp_flags & INP_HIGHPORT) { 661 first = ipport_hifirstauto; /* sysctl */ 662 last = ipport_hilastauto; 663 lastport = &portinfo->lasthi; 664 } else if (inp->inp_flags & INP_LOWPORT) { 665 if (cred && 666 (error = 667 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 668 inp->inp_laddr.s_addr = INADDR_ANY; 669 return (error); 670 } 671 first = ipport_lowfirstauto; /* 1023 */ 672 last = ipport_lowlastauto; /* 600 */ 673 lastport = &portinfo->lastlow; 674 } else { 675 first = ipport_firstauto; /* sysctl */ 676 last = ipport_lastauto; 677 lastport = &portinfo->lastport; 678 } 679 680 /* 681 * This has to be atomic. If the porthash is shared across multiple 682 * protocol threads (aka tcp) then the token must be held. 683 */ 684 GET_PORT_TOKEN(portinfo); 685 686 again: 687 /* 688 * Simple check to ensure all ports are not used up causing 689 * a deadlock here. 690 * 691 * We split the two cases (up and down) so that the direction 692 * is not being tested on each round of the loop. 693 */ 694 if (first > last) { 695 /* 696 * counting down 697 */ 698 in_pcbportrange(&first, &last, portinfo->offset, step); 699 count = (first - last) / step; 700 701 do { 702 if (count-- < 0) { /* completely used? */ 703 error = EADDRNOTAVAIL; 704 goto done; 705 } 706 *lastport -= step; 707 if (*lastport > first || *lastport < last) 708 *lastport = first; 709 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 710 portinfo->offset); 711 lport = htons(*lastport); 712 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr, 713 lport, sin->sin_addr, sin->sin_port, cred)); 714 } else { 715 /* 716 * counting up 717 */ 718 in_pcbportrange(&last, &first, portinfo->offset, step); 719 count = (last - first) / step; 720 721 do { 722 if (count-- < 0) { /* completely used? */ 723 error = EADDRNOTAVAIL; 724 goto done; 725 } 726 *lastport += step; 727 if (*lastport < first || *lastport > last) 728 *lastport = first; 729 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 730 portinfo->offset); 731 lport = htons(*lastport); 732 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr, 733 lport, sin->sin_addr, sin->sin_port, cred)); 734 } 735 736 /* This could happen on loopback interface */ 737 if (sin->sin_port == lport && 738 sin->sin_addr.s_addr == inp->inp_laddr.s_addr) { 739 if (dup) { 740 /* 741 * Duplicate again; give up 742 */ 743 error = EADDRNOTAVAIL; 744 goto done; 745 } 746 dup = 1; 747 goto again; 748 } 749 inp->inp_lport = lport; 750 in_pcbinsporthash(portinfo, inp); 751 error = 0; 752 done: 753 REL_PORT_TOKEN(portinfo); 754 755 if (error) { 756 /* Try next portinfo */ 757 portinfo_idx++; 758 portinfo_idx &= pcbinfo->portinfo_mask; 759 if (portinfo_idx != portinfo_first) 760 goto loop; 761 inp->inp_laddr.s_addr = INADDR_ANY; 762 } 763 return error; 764 } 765 766 /* 767 * Transform old in_pcbconnect() into an inner subroutine for new 768 * in_pcbconnect(): Do some validity-checking on the remote 769 * address (in mbuf 'nam') and then determine local host address 770 * (i.e., which interface) to use to access that remote host. 771 * 772 * This preserves definition of in_pcbconnect(), while supporting a 773 * slightly different version for T/TCP. (This is more than 774 * a bit of a kludge, but cleaning up the internal interfaces would 775 * have forced minor changes in every protocol). 776 */ 777 int 778 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam, 779 struct sockaddr_in **plocal_sin, struct thread *td, int find) 780 { 781 struct in_ifaddr *ia; 782 struct ucred *cred = NULL; 783 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 784 struct sockaddr *jsin; 785 int jailed = 0, alloc_route = 0; 786 787 if (nam->sa_len != sizeof *sin) 788 return (EINVAL); 789 if (sin->sin_family != AF_INET) 790 return (EAFNOSUPPORT); 791 if (sin->sin_port == 0) 792 return (EADDRNOTAVAIL); 793 if (td && td->td_proc && td->td_proc->p_ucred) 794 cred = td->td_proc->p_ucred; 795 if (cred && cred->cr_prison) 796 jailed = 1; 797 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) { 798 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 799 /* 800 * If the destination address is INADDR_ANY, 801 * use the primary local address. 802 * If the supplied address is INADDR_BROADCAST, 803 * and the primary interface supports broadcast, 804 * choose the broadcast address for that interface. 805 */ 806 if (sin->sin_addr.s_addr == INADDR_ANY) 807 sin->sin_addr = IA_SIN(ia)->sin_addr; 808 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST && 809 (ia->ia_ifp->if_flags & IFF_BROADCAST)) 810 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr; 811 } 812 if (find) { 813 struct route *ro; 814 815 ia = NULL; 816 /* 817 * If route is known or can be allocated now, 818 * our src addr is taken from the i/f, else punt. 819 * Note that we should check the address family of the cached 820 * destination, in case of sharing the cache with IPv6. 821 */ 822 ro = &inp->inp_route; 823 if (ro->ro_rt && 824 (!(ro->ro_rt->rt_flags & RTF_UP) || 825 ro->ro_dst.sa_family != AF_INET || 826 satosin(&ro->ro_dst)->sin_addr.s_addr != 827 sin->sin_addr.s_addr || 828 inp->inp_socket->so_options & SO_DONTROUTE)) { 829 RTFREE(ro->ro_rt); 830 ro->ro_rt = NULL; 831 } 832 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 833 (ro->ro_rt == NULL || 834 ro->ro_rt->rt_ifp == NULL)) { 835 /* No route yet, so try to acquire one */ 836 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 837 ro->ro_dst.sa_family = AF_INET; 838 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 839 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 840 sin->sin_addr; 841 rtalloc(ro); 842 alloc_route = 1; 843 } 844 /* 845 * If we found a route, use the address 846 * corresponding to the outgoing interface 847 * unless it is the loopback (in case a route 848 * to our address on another net goes to loopback). 849 */ 850 if (ro->ro_rt && 851 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) { 852 if (jailed) { 853 if (jailed_ip(cred->cr_prison, 854 ro->ro_rt->rt_ifa->ifa_addr)) { 855 ia = ifatoia(ro->ro_rt->rt_ifa); 856 } 857 } else { 858 ia = ifatoia(ro->ro_rt->rt_ifa); 859 } 860 } 861 if (ia == NULL) { 862 u_short fport = sin->sin_port; 863 864 sin->sin_port = 0; 865 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 866 if (ia && jailed && !jailed_ip(cred->cr_prison, 867 sintosa(&ia->ia_addr))) 868 ia = NULL; 869 if (ia == NULL) 870 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 871 if (ia && jailed && !jailed_ip(cred->cr_prison, 872 sintosa(&ia->ia_addr))) 873 ia = NULL; 874 sin->sin_port = fport; 875 if (ia == NULL && 876 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 877 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 878 if (ia && jailed && !jailed_ip(cred->cr_prison, 879 sintosa(&ia->ia_addr))) 880 ia = NULL; 881 882 if (!jailed && ia == NULL) 883 goto fail; 884 } 885 /* 886 * If the destination address is multicast and an outgoing 887 * interface has been set as a multicast option, use the 888 * address of that interface as our source address. 889 */ 890 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 891 inp->inp_moptions != NULL) { 892 struct ip_moptions *imo; 893 struct ifnet *ifp; 894 895 imo = inp->inp_moptions; 896 if (imo->imo_multicast_ifp != NULL) { 897 struct in_ifaddr_container *iac; 898 899 ifp = imo->imo_multicast_ifp; 900 ia = NULL; 901 TAILQ_FOREACH(iac, 902 &in_ifaddrheads[mycpuid], ia_link) { 903 if (iac->ia->ia_ifp == ifp) { 904 ia = iac->ia; 905 break; 906 } 907 } 908 if (ia == NULL) 909 goto fail; 910 } 911 } 912 /* 913 * Don't do pcblookup call here; return interface in plocal_sin 914 * and exit to caller, that will do the lookup. 915 */ 916 if (ia == NULL && jailed) { 917 if ((jsin = prison_get_nonlocal( 918 cred->cr_prison, AF_INET, NULL)) != NULL || 919 (jsin = prison_get_local( 920 cred->cr_prison, AF_INET, NULL)) != NULL) { 921 *plocal_sin = satosin(jsin); 922 } else { 923 /* IPv6 only Jail */ 924 goto fail; 925 } 926 } else { 927 *plocal_sin = &ia->ia_addr; 928 } 929 } 930 return (0); 931 fail: 932 if (alloc_route) { 933 struct route *ro = &inp->inp_route; 934 935 if (ro->ro_rt != NULL) 936 RTFREE(ro->ro_rt); 937 bzero(ro, sizeof(*ro)); 938 } 939 return (EADDRNOTAVAIL); 940 } 941 942 int 943 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam, 944 struct sockaddr_in **plocal_sin, struct thread *td) 945 { 946 return in_pcbladdr_find(inp, nam, plocal_sin, td, 947 (inp->inp_laddr.s_addr == INADDR_ANY)); 948 } 949 950 /* 951 * Outer subroutine: 952 * Connect from a socket to a specified address. 953 * Both address and port must be specified in argument sin. 954 * If don't have a local address for this socket yet, 955 * then pick one. 956 */ 957 int 958 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 959 { 960 struct sockaddr_in *if_sin; 961 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 962 int error; 963 964 /* Call inner routine to assign local interface address. */ 965 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0) 966 return (error); 967 968 if (in_pcblookup_hash(inp->inp_cpcbinfo, sin->sin_addr, sin->sin_port, 969 inp->inp_laddr.s_addr ? 970 inp->inp_laddr : if_sin->sin_addr, 971 inp->inp_lport, FALSE, NULL) != NULL) { 972 return (EADDRINUSE); 973 } 974 if (inp->inp_laddr.s_addr == INADDR_ANY) { 975 if (inp->inp_lport == 0) { 976 error = in_pcbbind(inp, NULL, td); 977 if (error) 978 return (error); 979 } 980 inp->inp_laddr = if_sin->sin_addr; 981 } 982 inp->inp_faddr = sin->sin_addr; 983 inp->inp_fport = sin->sin_port; 984 in_pcbinsconnhash(inp); 985 return (0); 986 } 987 988 void 989 in_pcbdisconnect(struct inpcb *inp) 990 { 991 992 inp->inp_faddr.s_addr = INADDR_ANY; 993 inp->inp_fport = 0; 994 in_pcbremconnhash(inp); 995 if (inp->inp_socket->so_state & SS_NOFDREF) 996 in_pcbdetach(inp); 997 } 998 999 void 1000 in_pcbdetach(struct inpcb *inp) 1001 { 1002 struct socket *so = inp->inp_socket; 1003 struct inpcbinfo *ipi = inp->inp_pcbinfo; 1004 1005 #ifdef IPSEC 1006 ipsec4_delete_pcbpolicy(inp); 1007 #endif /*IPSEC*/ 1008 inp->inp_gencnt = ++ipi->ipi_gencnt; 1009 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 1010 in_pcbremlists(inp); 1011 so->so_pcb = NULL; 1012 sofree(so); /* remove pcb ref */ 1013 if (inp->inp_options) 1014 m_free(inp->inp_options); 1015 if (inp->inp_route.ro_rt) 1016 rtfree(inp->inp_route.ro_rt); 1017 ip_freemoptions(inp->inp_moptions); 1018 inp->inp_vflag = 0; 1019 kfree(inp, M_PCB); 1020 } 1021 1022 /* 1023 * The calling convention of in_setsockaddr() and in_setpeeraddr() was 1024 * modified to match the pru_sockaddr() and pru_peeraddr() entry points 1025 * in struct pr_usrreqs, so that protocols can just reference then directly 1026 * without the need for a wrapper function. The socket must have a valid 1027 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one 1028 * except through a kernel programming error, so it is acceptable to panic 1029 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap 1030 * because there actually /is/ a programming error somewhere... XXX) 1031 */ 1032 int 1033 in_setsockaddr(struct socket *so, struct sockaddr **nam) 1034 { 1035 struct inpcb *inp; 1036 struct sockaddr_in *sin; 1037 1038 /* 1039 * Do the malloc first in case it blocks. 1040 */ 1041 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1042 sin->sin_family = AF_INET; 1043 sin->sin_len = sizeof *sin; 1044 1045 crit_enter(); 1046 inp = so->so_pcb; 1047 if (!inp) { 1048 crit_exit(); 1049 kfree(sin, M_SONAME); 1050 return (ECONNRESET); 1051 } 1052 sin->sin_port = inp->inp_lport; 1053 sin->sin_addr = inp->inp_laddr; 1054 crit_exit(); 1055 1056 *nam = (struct sockaddr *)sin; 1057 return (0); 1058 } 1059 1060 void 1061 in_setsockaddr_dispatch(netmsg_t msg) 1062 { 1063 int error; 1064 1065 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1066 lwkt_replymsg(&msg->lmsg, error); 1067 } 1068 1069 int 1070 in_setpeeraddr(struct socket *so, struct sockaddr **nam) 1071 { 1072 struct inpcb *inp; 1073 struct sockaddr_in *sin; 1074 1075 /* 1076 * Do the malloc first in case it blocks. 1077 */ 1078 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1079 sin->sin_family = AF_INET; 1080 sin->sin_len = sizeof *sin; 1081 1082 crit_enter(); 1083 inp = so->so_pcb; 1084 if (!inp) { 1085 crit_exit(); 1086 kfree(sin, M_SONAME); 1087 return (ECONNRESET); 1088 } 1089 sin->sin_port = inp->inp_fport; 1090 sin->sin_addr = inp->inp_faddr; 1091 crit_exit(); 1092 1093 *nam = (struct sockaddr *)sin; 1094 return (0); 1095 } 1096 1097 void 1098 in_setpeeraddr_dispatch(netmsg_t msg) 1099 { 1100 int error; 1101 1102 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1103 lwkt_replymsg(&msg->lmsg, error); 1104 } 1105 1106 void 1107 in_pcbnotifyall(struct inpcbhead *head, struct in_addr faddr, int err, 1108 void (*notify)(struct inpcb *, int)) 1109 { 1110 struct inpcb *inp, *ninp; 1111 1112 /* 1113 * note: if INP_PLACEMARKER is set we must ignore the rest of 1114 * the structure and skip it. 1115 */ 1116 crit_enter(); 1117 LIST_FOREACH_MUTABLE(inp, head, inp_list, ninp) { 1118 if (inp->inp_flags & INP_PLACEMARKER) 1119 continue; 1120 #ifdef INET6 1121 if (!(inp->inp_vflag & INP_IPV4)) 1122 continue; 1123 #endif 1124 if (inp->inp_faddr.s_addr != faddr.s_addr || 1125 inp->inp_socket == NULL) 1126 continue; 1127 (*notify)(inp, err); /* can remove inp from list! */ 1128 } 1129 crit_exit(); 1130 } 1131 1132 void 1133 in_pcbpurgeif0(struct inpcb *head, struct ifnet *ifp) 1134 { 1135 struct inpcb *inp; 1136 struct ip_moptions *imo; 1137 int i, gap; 1138 1139 for (inp = head; inp != NULL; inp = LIST_NEXT(inp, inp_list)) { 1140 if (inp->inp_flags & INP_PLACEMARKER) 1141 continue; 1142 imo = inp->inp_moptions; 1143 if ((inp->inp_vflag & INP_IPV4) && imo != NULL) { 1144 /* 1145 * Unselect the outgoing interface if it is being 1146 * detached. 1147 */ 1148 if (imo->imo_multicast_ifp == ifp) 1149 imo->imo_multicast_ifp = NULL; 1150 1151 /* 1152 * Drop multicast group membership if we joined 1153 * through the interface being detached. 1154 */ 1155 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1156 i++) { 1157 if (imo->imo_membership[i]->inm_ifp == ifp) { 1158 in_delmulti(imo->imo_membership[i]); 1159 gap++; 1160 } else if (gap != 0) 1161 imo->imo_membership[i - gap] = 1162 imo->imo_membership[i]; 1163 } 1164 imo->imo_num_memberships -= gap; 1165 } 1166 } 1167 } 1168 1169 /* 1170 * Check for alternatives when higher level complains 1171 * about service problems. For now, invalidate cached 1172 * routing information. If the route was created dynamically 1173 * (by a redirect), time to try a default gateway again. 1174 */ 1175 void 1176 in_losing(struct inpcb *inp) 1177 { 1178 struct rtentry *rt; 1179 struct rt_addrinfo rtinfo; 1180 1181 if ((rt = inp->inp_route.ro_rt)) { 1182 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1183 rtinfo.rti_info[RTAX_DST] = rt_key(rt); 1184 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1185 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt); 1186 rtinfo.rti_flags = rt->rt_flags; 1187 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0); 1188 if (rt->rt_flags & RTF_DYNAMIC) { 1189 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1190 rt_mask(rt), rt->rt_flags, NULL); 1191 } 1192 inp->inp_route.ro_rt = NULL; 1193 rtfree(rt); 1194 /* 1195 * A new route can be allocated 1196 * the next time output is attempted. 1197 */ 1198 } 1199 } 1200 1201 /* 1202 * After a routing change, flush old routing 1203 * and allocate a (hopefully) better one. 1204 */ 1205 void 1206 in_rtchange(struct inpcb *inp, int err) 1207 { 1208 if (inp->inp_route.ro_rt) { 1209 rtfree(inp->inp_route.ro_rt); 1210 inp->inp_route.ro_rt = NULL; 1211 /* 1212 * A new route can be allocated the next time 1213 * output is attempted. 1214 */ 1215 } 1216 } 1217 1218 /* 1219 * Lookup a PCB based on the local address and port. 1220 */ 1221 struct inpcb * 1222 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr, 1223 u_int lport_arg, int wild_okay, struct ucred *cred) 1224 { 1225 struct inpcb *inp; 1226 int matchwild = 3, wildcard; 1227 u_short lport = lport_arg; 1228 struct inpcbporthead *porthash; 1229 struct inpcbport *phd; 1230 struct inpcb *match = NULL; 1231 1232 /* 1233 * If the porthashbase is shared across several cpus, it must 1234 * have been locked. 1235 */ 1236 ASSERT_PORT_TOKEN_HELD(portinfo); 1237 1238 /* 1239 * Best fit PCB lookup. 1240 * 1241 * First see if this local port is in use by looking on the 1242 * port hash list. 1243 */ 1244 porthash = &portinfo->porthashbase[ 1245 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 1246 LIST_FOREACH(phd, porthash, phd_hash) { 1247 if (phd->phd_port == lport) 1248 break; 1249 } 1250 if (phd != NULL) { 1251 /* 1252 * Port is in use by one or more PCBs. Look for best 1253 * fit. 1254 */ 1255 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1256 wildcard = 0; 1257 #ifdef INET6 1258 if ((inp->inp_vflag & INP_IPV4) == 0) 1259 continue; 1260 #endif 1261 if (inp->inp_faddr.s_addr != INADDR_ANY) 1262 wildcard++; 1263 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1264 if (laddr.s_addr == INADDR_ANY) 1265 wildcard++; 1266 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1267 continue; 1268 } else { 1269 if (laddr.s_addr != INADDR_ANY) 1270 wildcard++; 1271 } 1272 if (wildcard && !wild_okay) 1273 continue; 1274 if (wildcard < matchwild && 1275 (cred == NULL || 1276 cred->cr_prison == 1277 inp->inp_socket->so_cred->cr_prison)) { 1278 match = inp; 1279 matchwild = wildcard; 1280 if (matchwild == 0) { 1281 break; 1282 } 1283 } 1284 } 1285 } 1286 return (match); 1287 } 1288 1289 struct inpcb * 1290 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo, 1291 const struct inpcb *inp) 1292 { 1293 const struct inp_localgrphead *hdr; 1294 const struct inp_localgroup *grp; 1295 int i; 1296 1297 if (pcbinfo->localgrphashbase == NULL) 1298 return NULL; 1299 1300 hdr = &pcbinfo->localgrphashbase[ 1301 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1302 1303 LIST_FOREACH(grp, hdr, il_list) { 1304 if (grp->il_vflag == inp->inp_vflag && 1305 grp->il_lport == inp->inp_lport && 1306 memcmp(&grp->il_dependladdr, 1307 &inp->inp_inc.inc_ie.ie_dependladdr, 1308 sizeof(grp->il_dependladdr)) == 0) { 1309 break; 1310 } 1311 } 1312 if (grp == NULL || grp->il_inpcnt == 1) 1313 return NULL; 1314 1315 KASSERT(grp->il_inpcnt >= 2, 1316 ("invalid localgroup inp count %d", grp->il_inpcnt)); 1317 for (i = 0; i < grp->il_inpcnt; ++i) { 1318 if (grp->il_inp[i] == inp) { 1319 int last = grp->il_inpcnt - 1; 1320 1321 if (i == last) 1322 last = grp->il_inpcnt - 2; 1323 return grp->il_inp[last]; 1324 } 1325 } 1326 return NULL; 1327 } 1328 1329 static struct inpcb * 1330 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo, 1331 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash) 1332 { 1333 struct inpcb *local_wild = NULL; 1334 const struct inp_localgrphead *hdr; 1335 const struct inp_localgroup *grp; 1336 1337 hdr = &pcbinfo->localgrphashbase[ 1338 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)]; 1339 #ifdef INP_LOCALGROUP_HASHTHR 1340 pkt_hash >>= ncpus2_shift; 1341 #endif 1342 1343 /* 1344 * Order of socket selection: 1345 * 1. non-wild. 1346 * 2. wild. 1347 * 1348 * NOTE: 1349 * - Local group does not contain jailed sockets 1350 * - Local group does not contain IPv4 mapped INET6 wild sockets 1351 */ 1352 LIST_FOREACH(grp, hdr, il_list) { 1353 #ifdef INET6 1354 if (!(grp->il_vflag & INP_IPV4)) 1355 continue; 1356 #endif 1357 if (grp->il_lport == lport) { 1358 int idx; 1359 1360 #ifdef INP_LOCALGROUP_HASHTHR 1361 idx = pkt_hash / grp->il_factor; 1362 KASSERT(idx < grp->il_inpcnt && idx >= 0, 1363 ("invalid hash %04x, cnt %d or fact %d", 1364 pkt_hash, grp->il_inpcnt, grp->il_factor)); 1365 #else 1366 /* 1367 * Modulo-N is used here, which greatly reduces 1368 * completion queue token contention, thus more 1369 * cpu time is saved. 1370 */ 1371 idx = pkt_hash % grp->il_inpcnt; 1372 #endif 1373 1374 if (grp->il_laddr.s_addr == laddr.s_addr) 1375 return grp->il_inp[idx]; 1376 else if (grp->il_laddr.s_addr == INADDR_ANY) 1377 local_wild = grp->il_inp[idx]; 1378 } 1379 } 1380 if (local_wild != NULL) 1381 return local_wild; 1382 return NULL; 1383 } 1384 1385 /* 1386 * Lookup PCB in hash list. 1387 */ 1388 struct inpcb * 1389 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1390 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1391 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m) 1392 { 1393 struct inpcbhead *head; 1394 struct inpcb *inp, *jinp=NULL; 1395 u_short fport = fport_arg, lport = lport_arg; 1396 1397 /* 1398 * First look for an exact match. 1399 */ 1400 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport, 1401 laddr.s_addr, lport, pcbinfo->hashmask)]; 1402 LIST_FOREACH(inp, head, inp_hash) { 1403 #ifdef INET6 1404 if (!(inp->inp_vflag & INP_IPV4)) 1405 continue; 1406 #endif 1407 if (in_hosteq(inp->inp_faddr, faddr) && 1408 in_hosteq(inp->inp_laddr, laddr) && 1409 inp->inp_fport == fport && inp->inp_lport == lport) { 1410 /* found */ 1411 if (inp->inp_socket == NULL || 1412 inp->inp_socket->so_cred->cr_prison == NULL) { 1413 return (inp); 1414 } else { 1415 if (jinp == NULL) 1416 jinp = inp; 1417 } 1418 } 1419 } 1420 if (jinp != NULL) 1421 return (jinp); 1422 if (wildcard) { 1423 struct inpcb *local_wild = NULL; 1424 struct inpcb *jinp_wild = NULL; 1425 #ifdef INET6 1426 struct inpcb *local_wild_mapped = NULL; 1427 #endif 1428 struct inpcontainer *ic; 1429 struct inpcontainerhead *chead; 1430 struct sockaddr_in jsin; 1431 struct ucred *cred; 1432 1433 /* 1434 * Check local group first 1435 */ 1436 if (pcbinfo->localgrphashbase != NULL && 1437 m != NULL && (m->m_flags & M_HASH) && 1438 !(ifp && ifp->if_type == IFT_FAITH)) { 1439 inp = inp_localgroup_lookup(pcbinfo, 1440 laddr, lport, m->m_pkthdr.hash); 1441 if (inp != NULL) 1442 return inp; 1443 } 1444 1445 /* 1446 * Order of socket selection: 1447 * 1. non-jailed, non-wild. 1448 * 2. non-jailed, wild. 1449 * 3. jailed, non-wild. 1450 * 4. jailed, wild. 1451 */ 1452 jsin.sin_family = AF_INET; 1453 chead = &pcbinfo->wildcardhashbase[ 1454 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)]; 1455 LIST_FOREACH(ic, chead, ic_list) { 1456 inp = ic->ic_inp; 1457 jsin.sin_addr.s_addr = laddr.s_addr; 1458 #ifdef INET6 1459 if (!(inp->inp_vflag & INP_IPV4)) 1460 continue; 1461 #endif 1462 if (inp->inp_socket != NULL) 1463 cred = inp->inp_socket->so_cred; 1464 else 1465 cred = NULL; 1466 if (cred != NULL && jailed(cred)) { 1467 if (jinp != NULL) 1468 continue; 1469 else 1470 if (!jailed_ip(cred->cr_prison, 1471 (struct sockaddr *)&jsin)) 1472 continue; 1473 } 1474 if (inp->inp_lport == lport) { 1475 if (ifp && ifp->if_type == IFT_FAITH && 1476 !(inp->inp_flags & INP_FAITH)) 1477 continue; 1478 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1479 if (cred != NULL && jailed(cred)) 1480 jinp = inp; 1481 else 1482 return (inp); 1483 } 1484 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1485 #ifdef INET6 1486 if (INP_CHECK_SOCKAF(inp->inp_socket, 1487 AF_INET6)) 1488 local_wild_mapped = inp; 1489 else 1490 #endif 1491 if (cred != NULL && 1492 jailed(cred)) 1493 jinp_wild = inp; 1494 else 1495 local_wild = inp; 1496 } 1497 } 1498 } 1499 if (local_wild != NULL) 1500 return (local_wild); 1501 #ifdef INET6 1502 if (local_wild_mapped != NULL) 1503 return (local_wild_mapped); 1504 #endif 1505 if (jinp != NULL) 1506 return (jinp); 1507 return (jinp_wild); 1508 } 1509 1510 /* 1511 * Not found. 1512 */ 1513 return (NULL); 1514 } 1515 1516 struct inpcb * 1517 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1518 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1519 boolean_t wildcard, struct ifnet *ifp) 1520 { 1521 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg, 1522 laddr, lport_arg, wildcard, ifp, NULL); 1523 } 1524 1525 /* 1526 * Insert PCB into connection hash table. 1527 */ 1528 void 1529 in_pcbinsconnhash(struct inpcb *inp) 1530 { 1531 struct inpcbinfo *pcbinfo = inp->inp_cpcbinfo; 1532 struct inpcbhead *bucket; 1533 u_int32_t hashkey_faddr, hashkey_laddr; 1534 1535 #ifdef INET6 1536 if (inp->inp_vflag & INP_IPV6) { 1537 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */; 1538 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */; 1539 } else { 1540 #endif 1541 hashkey_faddr = inp->inp_faddr.s_addr; 1542 hashkey_laddr = inp->inp_laddr.s_addr; 1543 #ifdef INET6 1544 } 1545 #endif 1546 1547 KASSERT(!(inp->inp_flags & INP_WILDCARD), 1548 ("already on wildcardhash")); 1549 KASSERT(!(inp->inp_flags & INP_CONNECTED), 1550 ("already on connhash")); 1551 inp->inp_flags |= INP_CONNECTED; 1552 1553 /* 1554 * Insert into the connection hash table. 1555 */ 1556 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr, 1557 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)]; 1558 LIST_INSERT_HEAD(bucket, inp, inp_hash); 1559 } 1560 1561 /* 1562 * Remove PCB from connection hash table. 1563 */ 1564 void 1565 in_pcbremconnhash(struct inpcb *inp) 1566 { 1567 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected")); 1568 LIST_REMOVE(inp, inp_hash); 1569 inp->inp_flags &= ~INP_CONNECTED; 1570 } 1571 1572 /* 1573 * Insert PCB into port hash table. 1574 */ 1575 void 1576 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp) 1577 { 1578 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1579 struct inpcbporthead *pcbporthash; 1580 struct inpcbport *phd; 1581 1582 /* 1583 * If the porthashbase is shared across several cpus, it must 1584 * have been locked. 1585 */ 1586 ASSERT_PORT_TOKEN_HELD(portinfo); 1587 1588 /* 1589 * Insert into the port hash table. 1590 */ 1591 pcbporthash = &portinfo->porthashbase[ 1592 INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)]; 1593 1594 /* Go through port list and look for a head for this lport. */ 1595 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1596 if (phd->phd_port == inp->inp_lport) 1597 break; 1598 } 1599 1600 /* If none exists, use saved one and tack it on. */ 1601 if (phd == NULL) { 1602 KKASSERT(pcbinfo->portsave != NULL); 1603 phd = pcbinfo->portsave; 1604 pcbinfo->portsave = NULL; 1605 phd->phd_port = inp->inp_lport; 1606 LIST_INIT(&phd->phd_pcblist); 1607 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1608 } 1609 1610 inp->inp_portinfo = portinfo; 1611 inp->inp_phd = phd; 1612 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1613 1614 /* 1615 * Malloc one inpcbport for later use. It is safe to use 1616 * "wait" malloc here (port token would be released, if 1617 * malloc ever blocked), since all changes to the porthash 1618 * are done. 1619 */ 1620 if (pcbinfo->portsave == NULL) { 1621 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), 1622 M_PCB, M_INTWAIT | M_ZERO); 1623 } 1624 } 1625 1626 void 1627 in_pcbinsporthash_lport(struct inpcb *inp) 1628 { 1629 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1630 struct inpcbportinfo *portinfo; 1631 u_short lport_ho; 1632 1633 /* Locate the proper portinfo based on lport */ 1634 lport_ho = ntohs(inp->inp_lport); 1635 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 1636 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset); 1637 1638 GET_PORT_TOKEN(portinfo); 1639 in_pcbinsporthash(portinfo, inp); 1640 REL_PORT_TOKEN(portinfo); 1641 } 1642 1643 static struct inp_localgroup * 1644 inp_localgroup_alloc(struct inp_localgrphead *hdr, u_char vflag, 1645 uint16_t port, const union in_dependaddr *addr, int size) 1646 { 1647 struct inp_localgroup *grp; 1648 1649 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]), 1650 M_TEMP, M_INTWAIT | M_ZERO); 1651 grp->il_vflag = vflag; 1652 grp->il_lport = port; 1653 grp->il_dependladdr = *addr; 1654 grp->il_inpsiz = size; 1655 1656 LIST_INSERT_HEAD(hdr, grp, il_list); 1657 1658 return grp; 1659 } 1660 1661 static void 1662 inp_localgroup_free(struct inp_localgroup *grp) 1663 { 1664 LIST_REMOVE(grp, il_list); 1665 kfree(grp, M_TEMP); 1666 } 1667 1668 static struct inp_localgroup * 1669 inp_localgroup_resize(struct inp_localgrphead *hdr, 1670 struct inp_localgroup *old_grp, int size) 1671 { 1672 struct inp_localgroup *grp; 1673 int i; 1674 1675 grp = inp_localgroup_alloc(hdr, old_grp->il_vflag, 1676 old_grp->il_lport, &old_grp->il_dependladdr, size); 1677 1678 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 1679 ("invalid new local group size %d and old local group count %d", 1680 grp->il_inpsiz, old_grp->il_inpcnt)); 1681 for (i = 0; i < old_grp->il_inpcnt; ++i) 1682 grp->il_inp[i] = old_grp->il_inp[i]; 1683 grp->il_inpcnt = old_grp->il_inpcnt; 1684 grp->il_factor = old_grp->il_factor; 1685 1686 inp_localgroup_free(old_grp); 1687 1688 return grp; 1689 } 1690 1691 static void 1692 inp_localgroup_factor(struct inp_localgroup *grp) 1693 { 1694 grp->il_factor = 1695 ((uint32_t)(0xffff >> ncpus2_shift) / grp->il_inpcnt) + 1; 1696 KASSERT(grp->il_factor != 0, ("invalid local group factor, " 1697 "ncpus2_shift %d, inpcnt %d", ncpus2_shift, grp->il_inpcnt)); 1698 } 1699 1700 static void 1701 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1702 { 1703 struct inp_localgrphead *hdr; 1704 struct inp_localgroup *grp; 1705 struct ucred *cred; 1706 1707 if (pcbinfo->localgrphashbase == NULL) 1708 return; 1709 1710 /* 1711 * XXX don't allow jailed socket to join local group 1712 */ 1713 if (inp->inp_socket != NULL) 1714 cred = inp->inp_socket->so_cred; 1715 else 1716 cred = NULL; 1717 if (cred != NULL && jailed(cred)) 1718 return; 1719 1720 #ifdef INET6 1721 /* 1722 * XXX don't allow IPv4 mapped INET6 wild socket 1723 */ 1724 if ((inp->inp_vflag & INP_IPV4) && 1725 inp->inp_laddr.s_addr == INADDR_ANY && 1726 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) 1727 return; 1728 #endif 1729 1730 hdr = &pcbinfo->localgrphashbase[ 1731 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1732 1733 LIST_FOREACH(grp, hdr, il_list) { 1734 if (grp->il_vflag == inp->inp_vflag && 1735 grp->il_lport == inp->inp_lport && 1736 memcmp(&grp->il_dependladdr, 1737 &inp->inp_inc.inc_ie.ie_dependladdr, 1738 sizeof(grp->il_dependladdr)) == 0) { 1739 break; 1740 } 1741 } 1742 if (grp == NULL) { 1743 /* Create new local group */ 1744 grp = inp_localgroup_alloc(hdr, inp->inp_vflag, 1745 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 1746 INP_LOCALGROUP_SIZMIN); 1747 } else if (grp->il_inpcnt == grp->il_inpsiz) { 1748 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) { 1749 static int limit_logged = 0; 1750 1751 if (!limit_logged) { 1752 limit_logged = 1; 1753 kprintf("local group port %d, " 1754 "limit reached\n", ntohs(grp->il_lport)); 1755 } 1756 return; 1757 } 1758 1759 /* Expand this local group */ 1760 grp = inp_localgroup_resize(hdr, grp, grp->il_inpsiz * 2); 1761 } 1762 1763 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 1764 ("invalid local group size %d and count %d", 1765 grp->il_inpsiz, grp->il_inpcnt)); 1766 grp->il_inp[grp->il_inpcnt] = inp; 1767 grp->il_inpcnt++; 1768 inp_localgroup_factor(grp); 1769 } 1770 1771 void 1772 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1773 { 1774 struct inpcontainer *ic; 1775 struct inpcontainerhead *bucket; 1776 1777 in_pcbinslocalgrphash_oncpu(inp, pcbinfo); 1778 1779 bucket = &pcbinfo->wildcardhashbase[ 1780 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1781 1782 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT); 1783 ic->ic_inp = inp; 1784 LIST_INSERT_HEAD(bucket, ic, ic_list); 1785 } 1786 1787 /* 1788 * Insert PCB into wildcard hash table. 1789 */ 1790 void 1791 in_pcbinswildcardhash(struct inpcb *inp) 1792 { 1793 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1794 1795 KASSERT(!(inp->inp_flags & INP_CONNECTED), 1796 ("already on connhash")); 1797 KASSERT(!(inp->inp_flags & INP_WILDCARD), 1798 ("already on wildcardhash")); 1799 inp->inp_flags |= INP_WILDCARD; 1800 1801 in_pcbinswildcardhash_oncpu(inp, pcbinfo); 1802 } 1803 1804 static void 1805 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1806 { 1807 struct inp_localgrphead *hdr; 1808 struct inp_localgroup *grp; 1809 1810 if (pcbinfo->localgrphashbase == NULL) 1811 return; 1812 1813 hdr = &pcbinfo->localgrphashbase[ 1814 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1815 1816 LIST_FOREACH(grp, hdr, il_list) { 1817 int i; 1818 1819 for (i = 0; i < grp->il_inpcnt; ++i) { 1820 if (grp->il_inp[i] != inp) 1821 continue; 1822 1823 if (grp->il_inpcnt == 1) { 1824 /* Free this local group */ 1825 inp_localgroup_free(grp); 1826 } else { 1827 /* Pull up inpcbs */ 1828 for (; i + 1 < grp->il_inpcnt; ++i) 1829 grp->il_inp[i] = grp->il_inp[i + 1]; 1830 grp->il_inpcnt--; 1831 inp_localgroup_factor(grp); 1832 1833 if (grp->il_inpsiz > INP_LOCALGROUP_SIZMIN && 1834 grp->il_inpcnt <= (grp->il_inpsiz / 4)) { 1835 /* Shrink this local group */ 1836 grp = inp_localgroup_resize(hdr, grp, 1837 grp->il_inpsiz / 2); 1838 } 1839 } 1840 return; 1841 } 1842 } 1843 } 1844 1845 void 1846 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1847 { 1848 struct inpcontainer *ic; 1849 struct inpcontainerhead *head; 1850 1851 in_pcbremlocalgrphash_oncpu(inp, pcbinfo); 1852 1853 /* find bucket */ 1854 head = &pcbinfo->wildcardhashbase[ 1855 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1856 1857 LIST_FOREACH(ic, head, ic_list) { 1858 if (ic->ic_inp == inp) 1859 goto found; 1860 } 1861 return; /* not found! */ 1862 1863 found: 1864 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */ 1865 kfree(ic, M_TEMP); /* deallocate container */ 1866 } 1867 1868 /* 1869 * Remove PCB from wildcard hash table. 1870 */ 1871 void 1872 in_pcbremwildcardhash(struct inpcb *inp) 1873 { 1874 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1875 1876 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard")); 1877 in_pcbremwildcardhash_oncpu(inp, pcbinfo); 1878 inp->inp_flags &= ~INP_WILDCARD; 1879 } 1880 1881 /* 1882 * Remove PCB from various lists. 1883 */ 1884 void 1885 in_pcbremlists(struct inpcb *inp) 1886 { 1887 if (inp->inp_lport) { 1888 struct inpcbportinfo *portinfo; 1889 struct inpcbport *phd; 1890 1891 /* 1892 * NOTE: 1893 * inp->inp_portinfo is _not_ necessary same as 1894 * inp->inp_pcbinfo->portinfo. 1895 */ 1896 portinfo = inp->inp_portinfo; 1897 GET_PORT_TOKEN(portinfo); 1898 1899 phd = inp->inp_phd; 1900 LIST_REMOVE(inp, inp_portlist); 1901 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1902 LIST_REMOVE(phd, phd_hash); 1903 kfree(phd, M_PCB); 1904 } 1905 1906 REL_PORT_TOKEN(portinfo); 1907 } 1908 if (inp->inp_flags & INP_WILDCARD) { 1909 in_pcbremwildcardhash(inp); 1910 } else if (inp->inp_flags & INP_CONNECTED) { 1911 in_pcbremconnhash(inp); 1912 } 1913 LIST_REMOVE(inp, inp_list); 1914 inp->inp_pcbinfo->ipi_count--; 1915 } 1916 1917 int 1918 prison_xinpcb(struct thread *td, struct inpcb *inp) 1919 { 1920 struct ucred *cr; 1921 1922 if (td->td_proc == NULL) 1923 return (0); 1924 cr = td->td_proc->p_ucred; 1925 if (cr->cr_prison == NULL) 1926 return (0); 1927 if (inp->inp_socket && inp->inp_socket->so_cred && 1928 inp->inp_socket->so_cred->cr_prison && 1929 cr->cr_prison == inp->inp_socket->so_cred->cr_prison) 1930 return (0); 1931 return (1); 1932 } 1933 1934 int 1935 in_pcblist_global(SYSCTL_HANDLER_ARGS) 1936 { 1937 struct inpcbinfo *pcbinfo = arg1; 1938 struct inpcb *inp, *marker; 1939 struct xinpcb xi; 1940 int error, i, n; 1941 1942 /* 1943 * The process of preparing the TCB list is too time-consuming and 1944 * resource-intensive to repeat twice on every request. 1945 */ 1946 if (req->oldptr == NULL) { 1947 n = pcbinfo->ipi_count; 1948 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 1949 return 0; 1950 } 1951 1952 if (req->newptr != NULL) 1953 return EPERM; 1954 1955 /* 1956 * OK, now we're committed to doing something. Re-fetch ipi_count 1957 * after obtaining the generation count. 1958 */ 1959 n = pcbinfo->ipi_count; 1960 1961 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1962 marker->inp_flags |= INP_PLACEMARKER; 1963 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1964 1965 i = 0; 1966 error = 0; 1967 1968 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 1969 LIST_REMOVE(marker, inp_list); 1970 LIST_INSERT_AFTER(inp, marker, inp_list); 1971 1972 if (inp->inp_flags & INP_PLACEMARKER) 1973 continue; 1974 if (prison_xinpcb(req->td, inp)) 1975 continue; 1976 bzero(&xi, sizeof xi); 1977 xi.xi_len = sizeof xi; 1978 bcopy(inp, &xi.xi_inp, sizeof *inp); 1979 if (inp->inp_socket) 1980 sotoxsocket(inp->inp_socket, &xi.xi_socket); 1981 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0) 1982 break; 1983 ++i; 1984 } 1985 LIST_REMOVE(marker, inp_list); 1986 if (error == 0 && i < n) { 1987 bzero(&xi, sizeof xi); 1988 xi.xi_len = sizeof xi; 1989 while (i < n) { 1990 error = SYSCTL_OUT(req, &xi, sizeof xi); 1991 ++i; 1992 } 1993 } 1994 kfree(marker, M_TEMP); 1995 return(error); 1996 } 1997 1998 int 1999 in_pcblist_global_nomarker(SYSCTL_HANDLER_ARGS, struct xinpcb **xi0, int *nxi0) 2000 { 2001 struct inpcbinfo *pcbinfo = arg1; 2002 struct inpcb *inp; 2003 struct xinpcb *xi; 2004 int nxi; 2005 2006 *nxi0 = 0; 2007 *xi0 = NULL; 2008 2009 /* 2010 * The process of preparing the PCB list is too time-consuming and 2011 * resource-intensive to repeat twice on every request. 2012 */ 2013 if (req->oldptr == NULL) { 2014 int n = pcbinfo->ipi_count; 2015 2016 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 2017 return 0; 2018 } 2019 2020 if (req->newptr != NULL) 2021 return EPERM; 2022 2023 if (pcbinfo->ipi_count == 0) 2024 return 0; 2025 2026 nxi = 0; 2027 xi = kmalloc(pcbinfo->ipi_count * sizeof(*xi), M_TEMP, 2028 M_WAITOK | M_ZERO | M_NULLOK); 2029 if (xi == NULL) 2030 return ENOMEM; 2031 2032 LIST_FOREACH(inp, &pcbinfo->pcblisthead, inp_list) { 2033 struct xinpcb *xi_ptr = &xi[nxi]; 2034 2035 if (prison_xinpcb(req->td, inp)) 2036 continue; 2037 2038 xi_ptr->xi_len = sizeof(*xi_ptr); 2039 bcopy(inp, &xi_ptr->xi_inp, sizeof(*inp)); 2040 if (inp->inp_socket) 2041 sotoxsocket(inp->inp_socket, &xi_ptr->xi_socket); 2042 ++nxi; 2043 } 2044 2045 if (nxi == 0) { 2046 kfree(xi, M_TEMP); 2047 return 0; 2048 } 2049 2050 *nxi0 = nxi; 2051 *xi0 = xi; 2052 2053 return 0; 2054 } 2055 2056 void 2057 in_savefaddr(struct socket *so, const struct sockaddr *faddr) 2058 { 2059 struct sockaddr_in *sin; 2060 2061 KASSERT(faddr->sa_family == AF_INET, 2062 ("not AF_INET faddr %d", faddr->sa_family)); 2063 2064 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO); 2065 sin->sin_family = AF_INET; 2066 sin->sin_len = sizeof(*sin); 2067 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port; 2068 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr; 2069 2070 so->so_faddr = (struct sockaddr *)sin; 2071 } 2072 2073 void 2074 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize, 2075 boolean_t shared, u_short offset) 2076 { 2077 memset(portinfo, 0, sizeof(*portinfo)); 2078 2079 portinfo->offset = offset; 2080 portinfo->lastport = offset; 2081 portinfo->lastlow = offset; 2082 portinfo->lasthi = offset; 2083 2084 portinfo->porthashbase = hashinit(hashsize, M_PCB, 2085 &portinfo->porthashmask); 2086 2087 if (shared) { 2088 portinfo->porttoken = kmalloc(sizeof(struct lwkt_token), 2089 M_PCB, M_WAITOK); 2090 lwkt_token_init(portinfo->porttoken, "porttoken"); 2091 } 2092 } 2093 2094 void 2095 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step) 2096 { 2097 int hi, lo; 2098 2099 if (step == 1) 2100 return; 2101 2102 hi = *hi0; 2103 lo = *lo0; 2104 2105 hi = rounddown2(hi, step); 2106 hi += ofs; 2107 if (hi > (int)*hi0) 2108 hi -= step; 2109 2110 lo = roundup2(lo, step); 2111 lo -= (step - ofs); 2112 if (lo < (int)*lo0) 2113 lo += step; 2114 2115 *hi0 = hi; 2116 *lo0 = lo; 2117 } 2118