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 && !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) { 851 if (jailed) { 852 if (jailed_ip(cred->cr_prison, 853 ro->ro_rt->rt_ifa->ifa_addr)) { 854 ia = ifatoia(ro->ro_rt->rt_ifa); 855 } 856 } else { 857 ia = ifatoia(ro->ro_rt->rt_ifa); 858 } 859 } 860 if (ia == NULL) { 861 u_short fport = sin->sin_port; 862 863 sin->sin_port = 0; 864 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 865 if (ia && jailed && !jailed_ip(cred->cr_prison, 866 sintosa(&ia->ia_addr))) 867 ia = NULL; 868 if (ia == NULL) 869 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 870 if (ia && jailed && !jailed_ip(cred->cr_prison, 871 sintosa(&ia->ia_addr))) 872 ia = NULL; 873 sin->sin_port = fport; 874 if (ia == NULL && 875 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 876 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 877 if (ia && jailed && !jailed_ip(cred->cr_prison, 878 sintosa(&ia->ia_addr))) 879 ia = NULL; 880 881 if (!jailed && ia == NULL) 882 goto fail; 883 } 884 /* 885 * If the destination address is multicast and an outgoing 886 * interface has been set as a multicast option, use the 887 * address of that interface as our source address. 888 */ 889 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 890 inp->inp_moptions != NULL) { 891 struct ip_moptions *imo; 892 struct ifnet *ifp; 893 894 imo = inp->inp_moptions; 895 if (imo->imo_multicast_ifp != NULL) { 896 struct in_ifaddr_container *iac; 897 898 ifp = imo->imo_multicast_ifp; 899 ia = NULL; 900 TAILQ_FOREACH(iac, 901 &in_ifaddrheads[mycpuid], ia_link) { 902 if (iac->ia->ia_ifp == ifp) { 903 ia = iac->ia; 904 break; 905 } 906 } 907 if (ia == NULL) 908 goto fail; 909 } 910 } 911 /* 912 * Don't do pcblookup call here; return interface in plocal_sin 913 * and exit to caller, that will do the lookup. 914 */ 915 if (ia == NULL && jailed) { 916 if ((jsin = prison_get_nonlocal(cred->cr_prison, AF_INET, NULL)) != NULL || 917 (jsin = prison_get_local(cred->cr_prison, AF_INET, NULL)) != NULL) { 918 *plocal_sin = satosin(jsin); 919 } else { 920 /* IPv6 only Jail */ 921 goto fail; 922 } 923 } else { 924 *plocal_sin = &ia->ia_addr; 925 } 926 } 927 return (0); 928 fail: 929 if (alloc_route) { 930 struct route *ro = &inp->inp_route; 931 932 if (ro->ro_rt != NULL) 933 RTFREE(ro->ro_rt); 934 bzero(ro, sizeof(*ro)); 935 } 936 return (EADDRNOTAVAIL); 937 } 938 939 int 940 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam, 941 struct sockaddr_in **plocal_sin, struct thread *td) 942 { 943 return in_pcbladdr_find(inp, nam, plocal_sin, td, 944 (inp->inp_laddr.s_addr == INADDR_ANY)); 945 } 946 947 /* 948 * Outer subroutine: 949 * Connect from a socket to a specified address. 950 * Both address and port must be specified in argument sin. 951 * If don't have a local address for this socket yet, 952 * then pick one. 953 */ 954 int 955 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 956 { 957 struct sockaddr_in *if_sin; 958 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 959 int error; 960 961 /* Call inner routine to assign local interface address. */ 962 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0) 963 return (error); 964 965 if (in_pcblookup_hash(inp->inp_cpcbinfo, sin->sin_addr, sin->sin_port, 966 inp->inp_laddr.s_addr ? 967 inp->inp_laddr : if_sin->sin_addr, 968 inp->inp_lport, FALSE, NULL) != NULL) { 969 return (EADDRINUSE); 970 } 971 if (inp->inp_laddr.s_addr == INADDR_ANY) { 972 if (inp->inp_lport == 0) { 973 error = in_pcbbind(inp, NULL, td); 974 if (error) 975 return (error); 976 } 977 inp->inp_laddr = if_sin->sin_addr; 978 } 979 inp->inp_faddr = sin->sin_addr; 980 inp->inp_fport = sin->sin_port; 981 in_pcbinsconnhash(inp); 982 return (0); 983 } 984 985 void 986 in_pcbdisconnect(struct inpcb *inp) 987 { 988 989 inp->inp_faddr.s_addr = INADDR_ANY; 990 inp->inp_fport = 0; 991 in_pcbremconnhash(inp); 992 if (inp->inp_socket->so_state & SS_NOFDREF) 993 in_pcbdetach(inp); 994 } 995 996 void 997 in_pcbdetach(struct inpcb *inp) 998 { 999 struct socket *so = inp->inp_socket; 1000 struct inpcbinfo *ipi = inp->inp_pcbinfo; 1001 1002 #ifdef IPSEC 1003 ipsec4_delete_pcbpolicy(inp); 1004 #endif /*IPSEC*/ 1005 inp->inp_gencnt = ++ipi->ipi_gencnt; 1006 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 1007 in_pcbremlists(inp); 1008 so->so_pcb = NULL; 1009 sofree(so); /* remove pcb ref */ 1010 if (inp->inp_options) 1011 m_free(inp->inp_options); 1012 if (inp->inp_route.ro_rt) 1013 rtfree(inp->inp_route.ro_rt); 1014 ip_freemoptions(inp->inp_moptions); 1015 inp->inp_vflag = 0; 1016 kfree(inp, M_PCB); 1017 } 1018 1019 /* 1020 * The calling convention of in_setsockaddr() and in_setpeeraddr() was 1021 * modified to match the pru_sockaddr() and pru_peeraddr() entry points 1022 * in struct pr_usrreqs, so that protocols can just reference then directly 1023 * without the need for a wrapper function. The socket must have a valid 1024 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one 1025 * except through a kernel programming error, so it is acceptable to panic 1026 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap 1027 * because there actually /is/ a programming error somewhere... XXX) 1028 */ 1029 int 1030 in_setsockaddr(struct socket *so, struct sockaddr **nam) 1031 { 1032 struct inpcb *inp; 1033 struct sockaddr_in *sin; 1034 1035 /* 1036 * Do the malloc first in case it blocks. 1037 */ 1038 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1039 sin->sin_family = AF_INET; 1040 sin->sin_len = sizeof *sin; 1041 1042 crit_enter(); 1043 inp = so->so_pcb; 1044 if (!inp) { 1045 crit_exit(); 1046 kfree(sin, M_SONAME); 1047 return (ECONNRESET); 1048 } 1049 sin->sin_port = inp->inp_lport; 1050 sin->sin_addr = inp->inp_laddr; 1051 crit_exit(); 1052 1053 *nam = (struct sockaddr *)sin; 1054 return (0); 1055 } 1056 1057 void 1058 in_setsockaddr_dispatch(netmsg_t msg) 1059 { 1060 int error; 1061 1062 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1063 lwkt_replymsg(&msg->lmsg, error); 1064 } 1065 1066 int 1067 in_setpeeraddr(struct socket *so, struct sockaddr **nam) 1068 { 1069 struct inpcb *inp; 1070 struct sockaddr_in *sin; 1071 1072 /* 1073 * Do the malloc first in case it blocks. 1074 */ 1075 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1076 sin->sin_family = AF_INET; 1077 sin->sin_len = sizeof *sin; 1078 1079 crit_enter(); 1080 inp = so->so_pcb; 1081 if (!inp) { 1082 crit_exit(); 1083 kfree(sin, M_SONAME); 1084 return (ECONNRESET); 1085 } 1086 sin->sin_port = inp->inp_fport; 1087 sin->sin_addr = inp->inp_faddr; 1088 crit_exit(); 1089 1090 *nam = (struct sockaddr *)sin; 1091 return (0); 1092 } 1093 1094 void 1095 in_setpeeraddr_dispatch(netmsg_t msg) 1096 { 1097 int error; 1098 1099 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1100 lwkt_replymsg(&msg->lmsg, error); 1101 } 1102 1103 void 1104 in_pcbnotifyall(struct inpcbhead *head, struct in_addr faddr, int err, 1105 void (*notify)(struct inpcb *, int)) 1106 { 1107 struct inpcb *inp, *ninp; 1108 1109 /* 1110 * note: if INP_PLACEMARKER is set we must ignore the rest of 1111 * the structure and skip it. 1112 */ 1113 crit_enter(); 1114 LIST_FOREACH_MUTABLE(inp, head, inp_list, ninp) { 1115 if (inp->inp_flags & INP_PLACEMARKER) 1116 continue; 1117 #ifdef INET6 1118 if (!(inp->inp_vflag & INP_IPV4)) 1119 continue; 1120 #endif 1121 if (inp->inp_faddr.s_addr != faddr.s_addr || 1122 inp->inp_socket == NULL) 1123 continue; 1124 (*notify)(inp, err); /* can remove inp from list! */ 1125 } 1126 crit_exit(); 1127 } 1128 1129 void 1130 in_pcbpurgeif0(struct inpcb *head, struct ifnet *ifp) 1131 { 1132 struct inpcb *inp; 1133 struct ip_moptions *imo; 1134 int i, gap; 1135 1136 for (inp = head; inp != NULL; inp = LIST_NEXT(inp, inp_list)) { 1137 if (inp->inp_flags & INP_PLACEMARKER) 1138 continue; 1139 imo = inp->inp_moptions; 1140 if ((inp->inp_vflag & INP_IPV4) && imo != NULL) { 1141 /* 1142 * Unselect the outgoing interface if it is being 1143 * detached. 1144 */ 1145 if (imo->imo_multicast_ifp == ifp) 1146 imo->imo_multicast_ifp = NULL; 1147 1148 /* 1149 * Drop multicast group membership if we joined 1150 * through the interface being detached. 1151 */ 1152 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1153 i++) { 1154 if (imo->imo_membership[i]->inm_ifp == ifp) { 1155 in_delmulti(imo->imo_membership[i]); 1156 gap++; 1157 } else if (gap != 0) 1158 imo->imo_membership[i - gap] = 1159 imo->imo_membership[i]; 1160 } 1161 imo->imo_num_memberships -= gap; 1162 } 1163 } 1164 } 1165 1166 /* 1167 * Check for alternatives when higher level complains 1168 * about service problems. For now, invalidate cached 1169 * routing information. If the route was created dynamically 1170 * (by a redirect), time to try a default gateway again. 1171 */ 1172 void 1173 in_losing(struct inpcb *inp) 1174 { 1175 struct rtentry *rt; 1176 struct rt_addrinfo rtinfo; 1177 1178 if ((rt = inp->inp_route.ro_rt)) { 1179 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1180 rtinfo.rti_info[RTAX_DST] = rt_key(rt); 1181 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1182 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt); 1183 rtinfo.rti_flags = rt->rt_flags; 1184 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0); 1185 if (rt->rt_flags & RTF_DYNAMIC) { 1186 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1187 rt_mask(rt), rt->rt_flags, NULL); 1188 } 1189 inp->inp_route.ro_rt = NULL; 1190 rtfree(rt); 1191 /* 1192 * A new route can be allocated 1193 * the next time output is attempted. 1194 */ 1195 } 1196 } 1197 1198 /* 1199 * After a routing change, flush old routing 1200 * and allocate a (hopefully) better one. 1201 */ 1202 void 1203 in_rtchange(struct inpcb *inp, int err) 1204 { 1205 if (inp->inp_route.ro_rt) { 1206 rtfree(inp->inp_route.ro_rt); 1207 inp->inp_route.ro_rt = NULL; 1208 /* 1209 * A new route can be allocated the next time 1210 * output is attempted. 1211 */ 1212 } 1213 } 1214 1215 /* 1216 * Lookup a PCB based on the local address and port. 1217 */ 1218 struct inpcb * 1219 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr, 1220 u_int lport_arg, int wild_okay, struct ucred *cred) 1221 { 1222 struct inpcb *inp; 1223 int matchwild = 3, wildcard; 1224 u_short lport = lport_arg; 1225 struct inpcbporthead *porthash; 1226 struct inpcbport *phd; 1227 struct inpcb *match = NULL; 1228 1229 /* 1230 * If the porthashbase is shared across several cpus, it must 1231 * have been locked. 1232 */ 1233 ASSERT_PORT_TOKEN_HELD(portinfo); 1234 1235 /* 1236 * Best fit PCB lookup. 1237 * 1238 * First see if this local port is in use by looking on the 1239 * port hash list. 1240 */ 1241 porthash = &portinfo->porthashbase[ 1242 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 1243 LIST_FOREACH(phd, porthash, phd_hash) { 1244 if (phd->phd_port == lport) 1245 break; 1246 } 1247 if (phd != NULL) { 1248 /* 1249 * Port is in use by one or more PCBs. Look for best 1250 * fit. 1251 */ 1252 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1253 wildcard = 0; 1254 #ifdef INET6 1255 if ((inp->inp_vflag & INP_IPV4) == 0) 1256 continue; 1257 #endif 1258 if (inp->inp_faddr.s_addr != INADDR_ANY) 1259 wildcard++; 1260 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1261 if (laddr.s_addr == INADDR_ANY) 1262 wildcard++; 1263 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1264 continue; 1265 } else { 1266 if (laddr.s_addr != INADDR_ANY) 1267 wildcard++; 1268 } 1269 if (wildcard && !wild_okay) 1270 continue; 1271 if (wildcard < matchwild && 1272 (cred == NULL || 1273 cred->cr_prison == 1274 inp->inp_socket->so_cred->cr_prison)) { 1275 match = inp; 1276 matchwild = wildcard; 1277 if (matchwild == 0) { 1278 break; 1279 } 1280 } 1281 } 1282 } 1283 return (match); 1284 } 1285 1286 struct inpcb * 1287 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo, 1288 const struct inpcb *inp) 1289 { 1290 const struct inp_localgrphead *hdr; 1291 const struct inp_localgroup *grp; 1292 int i; 1293 1294 if (pcbinfo->localgrphashbase == NULL) 1295 return NULL; 1296 1297 hdr = &pcbinfo->localgrphashbase[ 1298 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1299 1300 LIST_FOREACH(grp, hdr, il_list) { 1301 if (grp->il_vflag == inp->inp_vflag && 1302 grp->il_lport == inp->inp_lport && 1303 memcmp(&grp->il_dependladdr, 1304 &inp->inp_inc.inc_ie.ie_dependladdr, 1305 sizeof(grp->il_dependladdr)) == 0) { 1306 break; 1307 } 1308 } 1309 if (grp == NULL || grp->il_inpcnt == 1) 1310 return NULL; 1311 1312 KASSERT(grp->il_inpcnt >= 2, 1313 ("invalid localgroup inp count %d", grp->il_inpcnt)); 1314 for (i = 0; i < grp->il_inpcnt; ++i) { 1315 if (grp->il_inp[i] == inp) { 1316 int last = grp->il_inpcnt - 1; 1317 1318 if (i == last) 1319 last = grp->il_inpcnt - 2; 1320 return grp->il_inp[last]; 1321 } 1322 } 1323 return NULL; 1324 } 1325 1326 static struct inpcb * 1327 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo, 1328 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash) 1329 { 1330 struct inpcb *local_wild = NULL; 1331 const struct inp_localgrphead *hdr; 1332 const struct inp_localgroup *grp; 1333 1334 hdr = &pcbinfo->localgrphashbase[ 1335 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)]; 1336 #ifdef INP_LOCALGROUP_HASHTHR 1337 pkt_hash >>= ncpus2_shift; 1338 #endif 1339 1340 /* 1341 * Order of socket selection: 1342 * 1. non-wild. 1343 * 2. wild. 1344 * 1345 * NOTE: 1346 * - Local group does not contain jailed sockets 1347 * - Local group does not contain IPv4 mapped INET6 wild sockets 1348 */ 1349 LIST_FOREACH(grp, hdr, il_list) { 1350 #ifdef INET6 1351 if (!(grp->il_vflag & INP_IPV4)) 1352 continue; 1353 #endif 1354 if (grp->il_lport == lport) { 1355 int idx; 1356 1357 #ifdef INP_LOCALGROUP_HASHTHR 1358 idx = pkt_hash / grp->il_factor; 1359 KASSERT(idx < grp->il_inpcnt && idx >= 0, 1360 ("invalid hash %04x, cnt %d or fact %d", 1361 pkt_hash, grp->il_inpcnt, grp->il_factor)); 1362 #else 1363 /* 1364 * Modulo-N is used here, which greatly reduces 1365 * completion queue token contention, thus more 1366 * cpu time is saved. 1367 */ 1368 idx = pkt_hash % grp->il_inpcnt; 1369 #endif 1370 1371 if (grp->il_laddr.s_addr == laddr.s_addr) 1372 return grp->il_inp[idx]; 1373 else if (grp->il_laddr.s_addr == INADDR_ANY) 1374 local_wild = grp->il_inp[idx]; 1375 } 1376 } 1377 if (local_wild != NULL) 1378 return local_wild; 1379 return NULL; 1380 } 1381 1382 /* 1383 * Lookup PCB in hash list. 1384 */ 1385 struct inpcb * 1386 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1387 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1388 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m) 1389 { 1390 struct inpcbhead *head; 1391 struct inpcb *inp, *jinp=NULL; 1392 u_short fport = fport_arg, lport = lport_arg; 1393 1394 /* 1395 * First look for an exact match. 1396 */ 1397 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport, 1398 laddr.s_addr, lport, pcbinfo->hashmask)]; 1399 LIST_FOREACH(inp, head, inp_hash) { 1400 #ifdef INET6 1401 if (!(inp->inp_vflag & INP_IPV4)) 1402 continue; 1403 #endif 1404 if (in_hosteq(inp->inp_faddr, faddr) && 1405 in_hosteq(inp->inp_laddr, laddr) && 1406 inp->inp_fport == fport && inp->inp_lport == lport) { 1407 /* found */ 1408 if (inp->inp_socket == NULL || 1409 inp->inp_socket->so_cred->cr_prison == NULL) { 1410 return (inp); 1411 } else { 1412 if (jinp == NULL) 1413 jinp = inp; 1414 } 1415 } 1416 } 1417 if (jinp != NULL) 1418 return (jinp); 1419 if (wildcard) { 1420 struct inpcb *local_wild = NULL; 1421 struct inpcb *jinp_wild = NULL; 1422 #ifdef INET6 1423 struct inpcb *local_wild_mapped = NULL; 1424 #endif 1425 struct inpcontainer *ic; 1426 struct inpcontainerhead *chead; 1427 struct sockaddr_in jsin; 1428 struct ucred *cred; 1429 1430 /* 1431 * Check local group first 1432 */ 1433 if (pcbinfo->localgrphashbase != NULL && 1434 m != NULL && (m->m_flags & M_HASH) && 1435 !(ifp && ifp->if_type == IFT_FAITH)) { 1436 inp = inp_localgroup_lookup(pcbinfo, 1437 laddr, lport, m->m_pkthdr.hash); 1438 if (inp != NULL) 1439 return inp; 1440 } 1441 1442 /* 1443 * Order of socket selection: 1444 * 1. non-jailed, non-wild. 1445 * 2. non-jailed, wild. 1446 * 3. jailed, non-wild. 1447 * 4. jailed, wild. 1448 */ 1449 jsin.sin_family = AF_INET; 1450 chead = &pcbinfo->wildcardhashbase[ 1451 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)]; 1452 LIST_FOREACH(ic, chead, ic_list) { 1453 inp = ic->ic_inp; 1454 jsin.sin_addr.s_addr = laddr.s_addr; 1455 #ifdef INET6 1456 if (!(inp->inp_vflag & INP_IPV4)) 1457 continue; 1458 #endif 1459 if (inp->inp_socket != NULL) 1460 cred = inp->inp_socket->so_cred; 1461 else 1462 cred = NULL; 1463 if (cred != NULL && jailed(cred)) { 1464 if (jinp != NULL) 1465 continue; 1466 else 1467 if (!jailed_ip(cred->cr_prison, 1468 (struct sockaddr *)&jsin)) 1469 continue; 1470 } 1471 if (inp->inp_lport == lport) { 1472 if (ifp && ifp->if_type == IFT_FAITH && 1473 !(inp->inp_flags & INP_FAITH)) 1474 continue; 1475 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1476 if (cred != NULL && jailed(cred)) 1477 jinp = inp; 1478 else 1479 return (inp); 1480 } 1481 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1482 #ifdef INET6 1483 if (INP_CHECK_SOCKAF(inp->inp_socket, 1484 AF_INET6)) 1485 local_wild_mapped = inp; 1486 else 1487 #endif 1488 if (cred != NULL && 1489 jailed(cred)) 1490 jinp_wild = inp; 1491 else 1492 local_wild = inp; 1493 } 1494 } 1495 } 1496 if (local_wild != NULL) 1497 return (local_wild); 1498 #ifdef INET6 1499 if (local_wild_mapped != NULL) 1500 return (local_wild_mapped); 1501 #endif 1502 if (jinp != NULL) 1503 return (jinp); 1504 return (jinp_wild); 1505 } 1506 1507 /* 1508 * Not found. 1509 */ 1510 return (NULL); 1511 } 1512 1513 struct inpcb * 1514 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1515 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1516 boolean_t wildcard, struct ifnet *ifp) 1517 { 1518 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg, 1519 laddr, lport_arg, wildcard, ifp, NULL); 1520 } 1521 1522 /* 1523 * Insert PCB into connection hash table. 1524 */ 1525 void 1526 in_pcbinsconnhash(struct inpcb *inp) 1527 { 1528 struct inpcbinfo *pcbinfo = inp->inp_cpcbinfo; 1529 struct inpcbhead *bucket; 1530 u_int32_t hashkey_faddr, hashkey_laddr; 1531 1532 #ifdef INET6 1533 if (inp->inp_vflag & INP_IPV6) { 1534 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */; 1535 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */; 1536 } else { 1537 #endif 1538 hashkey_faddr = inp->inp_faddr.s_addr; 1539 hashkey_laddr = inp->inp_laddr.s_addr; 1540 #ifdef INET6 1541 } 1542 #endif 1543 1544 KASSERT(!(inp->inp_flags & INP_WILDCARD), 1545 ("already on wildcardhash")); 1546 KASSERT(!(inp->inp_flags & INP_CONNECTED), 1547 ("already on connhash")); 1548 inp->inp_flags |= INP_CONNECTED; 1549 1550 /* 1551 * Insert into the connection hash table. 1552 */ 1553 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr, 1554 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)]; 1555 LIST_INSERT_HEAD(bucket, inp, inp_hash); 1556 } 1557 1558 /* 1559 * Remove PCB from connection hash table. 1560 */ 1561 void 1562 in_pcbremconnhash(struct inpcb *inp) 1563 { 1564 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected")); 1565 LIST_REMOVE(inp, inp_hash); 1566 inp->inp_flags &= ~INP_CONNECTED; 1567 } 1568 1569 /* 1570 * Insert PCB into port hash table. 1571 */ 1572 void 1573 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp) 1574 { 1575 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1576 struct inpcbporthead *pcbporthash; 1577 struct inpcbport *phd; 1578 1579 /* 1580 * If the porthashbase is shared across several cpus, it must 1581 * have been locked. 1582 */ 1583 ASSERT_PORT_TOKEN_HELD(portinfo); 1584 1585 /* 1586 * Insert into the port hash table. 1587 */ 1588 pcbporthash = &portinfo->porthashbase[ 1589 INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)]; 1590 1591 /* Go through port list and look for a head for this lport. */ 1592 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1593 if (phd->phd_port == inp->inp_lport) 1594 break; 1595 } 1596 1597 /* If none exists, use saved one and tack it on. */ 1598 if (phd == NULL) { 1599 KKASSERT(pcbinfo->portsave != NULL); 1600 phd = pcbinfo->portsave; 1601 pcbinfo->portsave = NULL; 1602 phd->phd_port = inp->inp_lport; 1603 LIST_INIT(&phd->phd_pcblist); 1604 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1605 } 1606 1607 inp->inp_portinfo = portinfo; 1608 inp->inp_phd = phd; 1609 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1610 1611 /* 1612 * Malloc one inpcbport for later use. It is safe to use 1613 * "wait" malloc here (port token would be released, if 1614 * malloc ever blocked), since all changes to the porthash 1615 * are done. 1616 */ 1617 if (pcbinfo->portsave == NULL) { 1618 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), 1619 M_PCB, M_INTWAIT | M_ZERO); 1620 } 1621 } 1622 1623 void 1624 in_pcbinsporthash_lport(struct inpcb *inp) 1625 { 1626 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1627 struct inpcbportinfo *portinfo; 1628 u_short lport_ho; 1629 1630 /* Locate the proper portinfo based on lport */ 1631 lport_ho = ntohs(inp->inp_lport); 1632 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 1633 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset); 1634 1635 GET_PORT_TOKEN(portinfo); 1636 in_pcbinsporthash(portinfo, inp); 1637 REL_PORT_TOKEN(portinfo); 1638 } 1639 1640 static struct inp_localgroup * 1641 inp_localgroup_alloc(struct inp_localgrphead *hdr, u_char vflag, 1642 uint16_t port, const union in_dependaddr *addr, int size) 1643 { 1644 struct inp_localgroup *grp; 1645 1646 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]), 1647 M_TEMP, M_INTWAIT | M_ZERO); 1648 grp->il_vflag = vflag; 1649 grp->il_lport = port; 1650 grp->il_dependladdr = *addr; 1651 grp->il_inpsiz = size; 1652 1653 LIST_INSERT_HEAD(hdr, grp, il_list); 1654 1655 return grp; 1656 } 1657 1658 static void 1659 inp_localgroup_free(struct inp_localgroup *grp) 1660 { 1661 LIST_REMOVE(grp, il_list); 1662 kfree(grp, M_TEMP); 1663 } 1664 1665 static struct inp_localgroup * 1666 inp_localgroup_resize(struct inp_localgrphead *hdr, 1667 struct inp_localgroup *old_grp, int size) 1668 { 1669 struct inp_localgroup *grp; 1670 int i; 1671 1672 grp = inp_localgroup_alloc(hdr, old_grp->il_vflag, 1673 old_grp->il_lport, &old_grp->il_dependladdr, size); 1674 1675 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 1676 ("invalid new local group size %d and old local group count %d", 1677 grp->il_inpsiz, old_grp->il_inpcnt)); 1678 for (i = 0; i < old_grp->il_inpcnt; ++i) 1679 grp->il_inp[i] = old_grp->il_inp[i]; 1680 grp->il_inpcnt = old_grp->il_inpcnt; 1681 grp->il_factor = old_grp->il_factor; 1682 1683 inp_localgroup_free(old_grp); 1684 1685 return grp; 1686 } 1687 1688 static void 1689 inp_localgroup_factor(struct inp_localgroup *grp) 1690 { 1691 grp->il_factor = 1692 ((uint32_t)(0xffff >> ncpus2_shift) / grp->il_inpcnt) + 1; 1693 KASSERT(grp->il_factor != 0, ("invalid local group factor, " 1694 "ncpus2_shift %d, inpcnt %d", ncpus2_shift, grp->il_inpcnt)); 1695 } 1696 1697 static void 1698 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1699 { 1700 struct inp_localgrphead *hdr; 1701 struct inp_localgroup *grp; 1702 struct ucred *cred; 1703 1704 if (pcbinfo->localgrphashbase == NULL) 1705 return; 1706 1707 /* 1708 * XXX don't allow jailed socket to join local group 1709 */ 1710 if (inp->inp_socket != NULL) 1711 cred = inp->inp_socket->so_cred; 1712 else 1713 cred = NULL; 1714 if (cred != NULL && jailed(cred)) 1715 return; 1716 1717 #ifdef INET6 1718 /* 1719 * XXX don't allow IPv4 mapped INET6 wild socket 1720 */ 1721 if ((inp->inp_vflag & INP_IPV4) && 1722 inp->inp_laddr.s_addr == INADDR_ANY && 1723 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) 1724 return; 1725 #endif 1726 1727 hdr = &pcbinfo->localgrphashbase[ 1728 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1729 1730 LIST_FOREACH(grp, hdr, il_list) { 1731 if (grp->il_vflag == inp->inp_vflag && 1732 grp->il_lport == inp->inp_lport && 1733 memcmp(&grp->il_dependladdr, 1734 &inp->inp_inc.inc_ie.ie_dependladdr, 1735 sizeof(grp->il_dependladdr)) == 0) { 1736 break; 1737 } 1738 } 1739 if (grp == NULL) { 1740 /* Create new local group */ 1741 grp = inp_localgroup_alloc(hdr, inp->inp_vflag, 1742 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 1743 INP_LOCALGROUP_SIZMIN); 1744 } else if (grp->il_inpcnt == grp->il_inpsiz) { 1745 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) { 1746 static int limit_logged = 0; 1747 1748 if (!limit_logged) { 1749 limit_logged = 1; 1750 kprintf("local group port %d, " 1751 "limit reached\n", ntohs(grp->il_lport)); 1752 } 1753 return; 1754 } 1755 1756 /* Expand this local group */ 1757 grp = inp_localgroup_resize(hdr, grp, grp->il_inpsiz * 2); 1758 } 1759 1760 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 1761 ("invalid local group size %d and count %d", 1762 grp->il_inpsiz, grp->il_inpcnt)); 1763 grp->il_inp[grp->il_inpcnt] = inp; 1764 grp->il_inpcnt++; 1765 inp_localgroup_factor(grp); 1766 } 1767 1768 void 1769 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1770 { 1771 struct inpcontainer *ic; 1772 struct inpcontainerhead *bucket; 1773 1774 in_pcbinslocalgrphash_oncpu(inp, pcbinfo); 1775 1776 bucket = &pcbinfo->wildcardhashbase[ 1777 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1778 1779 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT); 1780 ic->ic_inp = inp; 1781 LIST_INSERT_HEAD(bucket, ic, ic_list); 1782 } 1783 1784 /* 1785 * Insert PCB into wildcard hash table. 1786 */ 1787 void 1788 in_pcbinswildcardhash(struct inpcb *inp) 1789 { 1790 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1791 1792 KASSERT(!(inp->inp_flags & INP_CONNECTED), 1793 ("already on connhash")); 1794 KASSERT(!(inp->inp_flags & INP_WILDCARD), 1795 ("already on wildcardhash")); 1796 inp->inp_flags |= INP_WILDCARD; 1797 1798 in_pcbinswildcardhash_oncpu(inp, pcbinfo); 1799 } 1800 1801 static void 1802 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1803 { 1804 struct inp_localgrphead *hdr; 1805 struct inp_localgroup *grp; 1806 1807 if (pcbinfo->localgrphashbase == NULL) 1808 return; 1809 1810 hdr = &pcbinfo->localgrphashbase[ 1811 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1812 1813 LIST_FOREACH(grp, hdr, il_list) { 1814 int i; 1815 1816 for (i = 0; i < grp->il_inpcnt; ++i) { 1817 if (grp->il_inp[i] != inp) 1818 continue; 1819 1820 if (grp->il_inpcnt == 1) { 1821 /* Free this local group */ 1822 inp_localgroup_free(grp); 1823 } else { 1824 /* Pull up inpcbs */ 1825 for (; i + 1 < grp->il_inpcnt; ++i) 1826 grp->il_inp[i] = grp->il_inp[i + 1]; 1827 grp->il_inpcnt--; 1828 inp_localgroup_factor(grp); 1829 1830 if (grp->il_inpsiz > INP_LOCALGROUP_SIZMIN && 1831 grp->il_inpcnt <= (grp->il_inpsiz / 4)) { 1832 /* Shrink this local group */ 1833 grp = inp_localgroup_resize(hdr, grp, 1834 grp->il_inpsiz / 2); 1835 } 1836 } 1837 return; 1838 } 1839 } 1840 } 1841 1842 void 1843 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1844 { 1845 struct inpcontainer *ic; 1846 struct inpcontainerhead *head; 1847 1848 in_pcbremlocalgrphash_oncpu(inp, pcbinfo); 1849 1850 /* find bucket */ 1851 head = &pcbinfo->wildcardhashbase[ 1852 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1853 1854 LIST_FOREACH(ic, head, ic_list) { 1855 if (ic->ic_inp == inp) 1856 goto found; 1857 } 1858 return; /* not found! */ 1859 1860 found: 1861 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */ 1862 kfree(ic, M_TEMP); /* deallocate container */ 1863 } 1864 1865 /* 1866 * Remove PCB from wildcard hash table. 1867 */ 1868 void 1869 in_pcbremwildcardhash(struct inpcb *inp) 1870 { 1871 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1872 1873 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard")); 1874 in_pcbremwildcardhash_oncpu(inp, pcbinfo); 1875 inp->inp_flags &= ~INP_WILDCARD; 1876 } 1877 1878 /* 1879 * Remove PCB from various lists. 1880 */ 1881 void 1882 in_pcbremlists(struct inpcb *inp) 1883 { 1884 if (inp->inp_lport) { 1885 struct inpcbportinfo *portinfo; 1886 struct inpcbport *phd; 1887 1888 /* 1889 * NOTE: 1890 * inp->inp_portinfo is _not_ necessary same as 1891 * inp->inp_pcbinfo->portinfo. 1892 */ 1893 portinfo = inp->inp_portinfo; 1894 GET_PORT_TOKEN(portinfo); 1895 1896 phd = inp->inp_phd; 1897 LIST_REMOVE(inp, inp_portlist); 1898 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1899 LIST_REMOVE(phd, phd_hash); 1900 kfree(phd, M_PCB); 1901 } 1902 1903 REL_PORT_TOKEN(portinfo); 1904 } 1905 if (inp->inp_flags & INP_WILDCARD) { 1906 in_pcbremwildcardhash(inp); 1907 } else if (inp->inp_flags & INP_CONNECTED) { 1908 in_pcbremconnhash(inp); 1909 } 1910 LIST_REMOVE(inp, inp_list); 1911 inp->inp_pcbinfo->ipi_count--; 1912 } 1913 1914 int 1915 prison_xinpcb(struct thread *td, struct inpcb *inp) 1916 { 1917 struct ucred *cr; 1918 1919 if (td->td_proc == NULL) 1920 return (0); 1921 cr = td->td_proc->p_ucred; 1922 if (cr->cr_prison == NULL) 1923 return (0); 1924 if (inp->inp_socket && inp->inp_socket->so_cred && 1925 inp->inp_socket->so_cred->cr_prison && 1926 cr->cr_prison == inp->inp_socket->so_cred->cr_prison) 1927 return (0); 1928 return (1); 1929 } 1930 1931 int 1932 in_pcblist_global(SYSCTL_HANDLER_ARGS) 1933 { 1934 struct inpcbinfo *pcbinfo = arg1; 1935 struct inpcb *inp, *marker; 1936 struct xinpcb xi; 1937 int error, i, n; 1938 1939 /* 1940 * The process of preparing the TCB list is too time-consuming and 1941 * resource-intensive to repeat twice on every request. 1942 */ 1943 if (req->oldptr == NULL) { 1944 n = pcbinfo->ipi_count; 1945 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 1946 return 0; 1947 } 1948 1949 if (req->newptr != NULL) 1950 return EPERM; 1951 1952 /* 1953 * OK, now we're committed to doing something. Re-fetch ipi_count 1954 * after obtaining the generation count. 1955 */ 1956 n = pcbinfo->ipi_count; 1957 1958 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1959 marker->inp_flags |= INP_PLACEMARKER; 1960 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1961 1962 i = 0; 1963 error = 0; 1964 1965 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 1966 LIST_REMOVE(marker, inp_list); 1967 LIST_INSERT_AFTER(inp, marker, inp_list); 1968 1969 if (inp->inp_flags & INP_PLACEMARKER) 1970 continue; 1971 if (prison_xinpcb(req->td, inp)) 1972 continue; 1973 bzero(&xi, sizeof xi); 1974 xi.xi_len = sizeof xi; 1975 bcopy(inp, &xi.xi_inp, sizeof *inp); 1976 if (inp->inp_socket) 1977 sotoxsocket(inp->inp_socket, &xi.xi_socket); 1978 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0) 1979 break; 1980 ++i; 1981 } 1982 LIST_REMOVE(marker, inp_list); 1983 if (error == 0 && i < n) { 1984 bzero(&xi, sizeof xi); 1985 xi.xi_len = sizeof xi; 1986 while (i < n) { 1987 error = SYSCTL_OUT(req, &xi, sizeof xi); 1988 ++i; 1989 } 1990 } 1991 kfree(marker, M_TEMP); 1992 return(error); 1993 } 1994 1995 int 1996 in_pcblist_global_nomarker(SYSCTL_HANDLER_ARGS, struct xinpcb **xi0, int *nxi0) 1997 { 1998 struct inpcbinfo *pcbinfo = arg1; 1999 struct inpcb *inp; 2000 struct xinpcb *xi; 2001 int nxi; 2002 2003 *nxi0 = 0; 2004 *xi0 = NULL; 2005 2006 /* 2007 * The process of preparing the PCB list is too time-consuming and 2008 * resource-intensive to repeat twice on every request. 2009 */ 2010 if (req->oldptr == NULL) { 2011 int n = pcbinfo->ipi_count; 2012 2013 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 2014 return 0; 2015 } 2016 2017 if (req->newptr != NULL) 2018 return EPERM; 2019 2020 if (pcbinfo->ipi_count == 0) 2021 return 0; 2022 2023 nxi = 0; 2024 xi = kmalloc(pcbinfo->ipi_count * sizeof(*xi), M_TEMP, 2025 M_WAITOK | M_ZERO | M_NULLOK); 2026 if (xi == NULL) 2027 return ENOMEM; 2028 2029 LIST_FOREACH(inp, &pcbinfo->pcblisthead, inp_list) { 2030 struct xinpcb *xi_ptr = &xi[nxi]; 2031 2032 if (prison_xinpcb(req->td, inp)) 2033 continue; 2034 2035 xi_ptr->xi_len = sizeof(*xi_ptr); 2036 bcopy(inp, &xi_ptr->xi_inp, sizeof(*inp)); 2037 if (inp->inp_socket) 2038 sotoxsocket(inp->inp_socket, &xi_ptr->xi_socket); 2039 ++nxi; 2040 } 2041 2042 if (nxi == 0) { 2043 kfree(xi, M_TEMP); 2044 return 0; 2045 } 2046 2047 *nxi0 = nxi; 2048 *xi0 = xi; 2049 2050 return 0; 2051 } 2052 2053 void 2054 in_savefaddr(struct socket *so, const struct sockaddr *faddr) 2055 { 2056 struct sockaddr_in *sin; 2057 2058 KASSERT(faddr->sa_family == AF_INET, 2059 ("not AF_INET faddr %d", faddr->sa_family)); 2060 2061 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO); 2062 sin->sin_family = AF_INET; 2063 sin->sin_len = sizeof(*sin); 2064 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port; 2065 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr; 2066 2067 so->so_faddr = (struct sockaddr *)sin; 2068 } 2069 2070 void 2071 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize, 2072 boolean_t shared, u_short offset) 2073 { 2074 memset(portinfo, 0, sizeof(*portinfo)); 2075 2076 portinfo->offset = offset; 2077 portinfo->lastport = offset; 2078 portinfo->lastlow = offset; 2079 portinfo->lasthi = offset; 2080 2081 portinfo->porthashbase = hashinit(hashsize, M_PCB, 2082 &portinfo->porthashmask); 2083 2084 if (shared) { 2085 portinfo->porttoken = kmalloc(sizeof(struct lwkt_token), 2086 M_PCB, M_WAITOK); 2087 lwkt_token_init(portinfo->porttoken, "porttoken"); 2088 } 2089 } 2090 2091 void 2092 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step) 2093 { 2094 int hi, lo; 2095 2096 if (step == 1) 2097 return; 2098 2099 hi = *hi0; 2100 lo = *lo0; 2101 2102 hi = rounddown2(hi, step); 2103 hi += ofs; 2104 if (hi > (int)*hi0) 2105 hi -= step; 2106 2107 lo = roundup2(lo, step); 2108 lo -= (step - ofs); 2109 if (lo < (int)*lo0) 2110 lo += step; 2111 2112 *hi0 = hi; 2113 *lo0 = lo; 2114 } 2115