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 = 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_flags & (INP_WILDCARD | INP_CONNECTED)) == 0, 270 ("already linked")); 271 272 LIST_REMOVE(inp, inp_list); 273 pcbinfo->ipi_count--; 274 inp->inp_pcbinfo = NULL; 275 } 276 277 /* 278 * Relink a pcb into a new pcbinfo. 279 */ 280 void 281 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 282 { 283 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo")); 284 KASSERT((inp->inp_flags & (INP_WILDCARD | INP_CONNECTED)) == 0, 285 ("already linked")); 286 287 inp->inp_pcbinfo = pcbinfo; 288 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list); 289 pcbinfo->ipi_count++; 290 } 291 292 static int 293 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred) 294 { 295 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 296 struct inpcbportinfo *portinfo; 297 u_short first, last, lport, step; 298 u_short *lastport; 299 int count, error; 300 int portinfo_first, portinfo_idx; 301 302 inp->inp_flags |= INP_ANONPORT; 303 304 step = pcbinfo->portinfo_mask + 1; 305 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 306 portinfo_idx = portinfo_first; 307 loop: 308 portinfo = &pcbinfo->portinfo[portinfo_idx]; 309 310 if (inp->inp_flags & INP_HIGHPORT) { 311 first = ipport_hifirstauto; /* sysctl */ 312 last = ipport_hilastauto; 313 lastport = &portinfo->lasthi; 314 } else if (inp->inp_flags & INP_LOWPORT) { 315 if (cred && 316 (error = 317 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 318 inp->inp_laddr.s_addr = INADDR_ANY; 319 return error; 320 } 321 first = ipport_lowfirstauto; /* 1023 */ 322 last = ipport_lowlastauto; /* 600 */ 323 lastport = &portinfo->lastlow; 324 } else { 325 first = ipport_firstauto; /* sysctl */ 326 last = ipport_lastauto; 327 lastport = &portinfo->lastport; 328 } 329 330 /* 331 * This has to be atomic. If the porthash is shared across multiple 332 * protocol threads (aka tcp) then the token must be held. 333 */ 334 GET_PORT_TOKEN(portinfo); 335 336 /* 337 * Simple check to ensure all ports are not used up causing 338 * a deadlock here. 339 * 340 * We split the two cases (up and down) so that the direction 341 * is not being tested on each round of the loop. 342 */ 343 if (first > last) { 344 /* 345 * counting down 346 */ 347 in_pcbportrange(&first, &last, portinfo->offset, step); 348 count = (first - last) / step; 349 350 do { 351 if (count-- < 0) { /* completely used? */ 352 error = EADDRNOTAVAIL; 353 goto done; 354 } 355 *lastport -= step; 356 if (*lastport > first || *lastport < last) 357 *lastport = first; 358 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 359 portinfo->offset); 360 lport = htons(*lastport); 361 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport, 362 wild, cred)); 363 } else { 364 /* 365 * counting up 366 */ 367 in_pcbportrange(&last, &first, portinfo->offset, step); 368 count = (last - first) / step; 369 370 do { 371 if (count-- < 0) { /* completely used? */ 372 error = EADDRNOTAVAIL; 373 goto done; 374 } 375 *lastport += step; 376 if (*lastport < first || *lastport > last) 377 *lastport = first; 378 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 379 portinfo->offset); 380 lport = htons(*lastport); 381 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport, 382 wild, cred)); 383 } 384 inp->inp_lport = lport; 385 in_pcbinsporthash(portinfo, inp); 386 error = 0; 387 done: 388 REL_PORT_TOKEN(portinfo); 389 390 if (error) { 391 /* Try next portinfo */ 392 portinfo_idx++; 393 portinfo_idx &= pcbinfo->portinfo_mask; 394 if (portinfo_idx != portinfo_first) 395 goto loop; 396 inp->inp_laddr.s_addr = INADDR_ANY; 397 } 398 return error; 399 } 400 401 int 402 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 403 { 404 struct socket *so = inp->inp_socket; 405 struct sockaddr_in jsin; 406 struct ucred *cred = NULL; 407 int wild = 0; 408 409 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 410 return (EADDRNOTAVAIL); 411 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 412 return (EINVAL); /* already bound */ 413 414 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT))) 415 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */ 416 if (td->td_proc) 417 cred = td->td_proc->p_ucred; 418 419 if (nam != NULL) { 420 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 421 struct inpcbinfo *pcbinfo; 422 struct inpcbportinfo *portinfo; 423 struct inpcb *t; 424 u_short lport, lport_ho; 425 int reuseport = (so->so_options & SO_REUSEPORT); 426 int error; 427 428 if (nam->sa_len != sizeof *sin) 429 return (EINVAL); 430 #ifdef notdef 431 /* 432 * We should check the family, but old programs 433 * incorrectly fail to initialize it. 434 */ 435 if (sin->sin_family != AF_INET) 436 return (EAFNOSUPPORT); 437 #endif 438 if (!prison_replace_wildcards(td, nam)) 439 return (EINVAL); 440 441 lport = sin->sin_port; 442 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 443 /* 444 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 445 * allow complete duplication of binding if 446 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 447 * and a multicast address is bound on both 448 * new and duplicated sockets. 449 */ 450 if (so->so_options & SO_REUSEADDR) 451 reuseport = SO_REUSEADDR | SO_REUSEPORT; 452 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 453 sin->sin_port = 0; /* yech... */ 454 bzero(&sin->sin_zero, sizeof sin->sin_zero); 455 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL) 456 return (EADDRNOTAVAIL); 457 } 458 459 inp->inp_laddr = sin->sin_addr; 460 461 jsin.sin_family = AF_INET; 462 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 463 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 464 inp->inp_laddr.s_addr = INADDR_ANY; 465 return (EINVAL); 466 } 467 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 468 469 if (lport == 0) { 470 /* Auto-select local port */ 471 return in_pcbsetlport(inp, wild, cred); 472 } 473 lport_ho = ntohs(lport); 474 475 /* GROSS */ 476 if (lport_ho < IPPORT_RESERVED && cred && 477 (error = 478 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 479 inp->inp_laddr.s_addr = INADDR_ANY; 480 return (error); 481 } 482 483 /* 484 * Locate the proper portinfo based on lport 485 */ 486 pcbinfo = inp->inp_pcbinfo; 487 portinfo = 488 &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 489 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == 490 portinfo->offset); 491 492 /* 493 * This has to be atomic. If the porthash is shared across 494 * multiple protocol threads (aka tcp) then the token must 495 * be held. 496 */ 497 GET_PORT_TOKEN(portinfo); 498 499 if (so->so_cred->cr_uid != 0 && 500 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 501 t = in_pcblookup_local(portinfo, sin->sin_addr, lport, 502 INPLOOKUP_WILDCARD, cred); 503 if (t && 504 (!in_nullhost(sin->sin_addr) || 505 !in_nullhost(t->inp_laddr) || 506 (t->inp_socket->so_options & SO_REUSEPORT) == 0) && 507 (so->so_cred->cr_uid != 508 t->inp_socket->so_cred->cr_uid)) { 509 #ifdef INET6 510 if (!in_nullhost(sin->sin_addr) || 511 !in_nullhost(t->inp_laddr) || 512 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket)) 513 #endif 514 { 515 inp->inp_laddr.s_addr = INADDR_ANY; 516 error = EADDRINUSE; 517 goto done; 518 } 519 } 520 } 521 if (cred && !prison_replace_wildcards(td, nam)) { 522 inp->inp_laddr.s_addr = INADDR_ANY; 523 error = EADDRNOTAVAIL; 524 goto done; 525 } 526 t = in_pcblookup_local(portinfo, sin->sin_addr, lport, 527 wild, cred); 528 if (t && !(reuseport & t->inp_socket->so_options)) { 529 #ifdef INET6 530 if (!in_nullhost(sin->sin_addr) || 531 !in_nullhost(t->inp_laddr) || 532 INP_SOCKAF(so) == INP_SOCKAF(t->inp_socket)) 533 #endif 534 { 535 inp->inp_laddr.s_addr = INADDR_ANY; 536 error = EADDRINUSE; 537 goto done; 538 } 539 } 540 inp->inp_lport = lport; 541 in_pcbinsporthash(portinfo, inp); 542 error = 0; 543 done: 544 REL_PORT_TOKEN(portinfo); 545 return (error); 546 } else { 547 jsin.sin_family = AF_INET; 548 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 549 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 550 inp->inp_laddr.s_addr = INADDR_ANY; 551 return (EINVAL); 552 } 553 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 554 555 return in_pcbsetlport(inp, wild, cred); 556 } 557 } 558 559 static struct inpcb * 560 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr, 561 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred) 562 { 563 struct inpcb *inp; 564 struct inpcbporthead *porthash; 565 struct inpcbport *phd; 566 struct inpcb *match = NULL; 567 568 /* 569 * If the porthashbase is shared across several cpus, it must 570 * have been locked. 571 */ 572 ASSERT_PORT_TOKEN_HELD(portinfo); 573 574 /* 575 * Best fit PCB lookup. 576 * 577 * First see if this local port is in use by looking on the 578 * port hash list. 579 */ 580 porthash = &portinfo->porthashbase[ 581 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 582 LIST_FOREACH(phd, porthash, phd_hash) { 583 if (phd->phd_port == lport) 584 break; 585 } 586 if (phd != NULL) { 587 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 588 #ifdef INET6 589 if ((inp->inp_vflag & INP_IPV4) == 0) 590 continue; 591 #endif 592 if (inp->inp_laddr.s_addr != INADDR_ANY && 593 inp->inp_laddr.s_addr != laddr.s_addr) 594 continue; 595 596 if (inp->inp_faddr.s_addr != INADDR_ANY && 597 inp->inp_faddr.s_addr != faddr.s_addr) 598 continue; 599 600 if (inp->inp_fport != 0 && inp->inp_fport != fport) 601 continue; 602 603 if (cred == NULL || 604 cred->cr_prison == 605 inp->inp_socket->so_cred->cr_prison) { 606 match = inp; 607 break; 608 } 609 } 610 } 611 return (match); 612 } 613 614 int 615 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote, 616 struct thread *td) 617 { 618 struct proc *p = td->td_proc; 619 unsigned short *lastport; 620 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote; 621 struct sockaddr_in jsin; 622 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 623 struct inpcbportinfo *portinfo; 624 struct ucred *cred = NULL; 625 u_short first, last, lport, step; 626 int count, error, dup; 627 int portinfo_first, portinfo_idx; 628 629 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 630 return (EADDRNOTAVAIL); 631 632 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY); 633 if (inp->inp_lport != 0) 634 return (EINVAL); /* already bound */ 635 636 KKASSERT(p); 637 cred = p->p_ucred; 638 639 jsin.sin_family = AF_INET; 640 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 641 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 642 inp->inp_laddr.s_addr = INADDR_ANY; 643 return (EINVAL); 644 } 645 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 646 647 inp->inp_flags |= INP_ANONPORT; 648 649 step = pcbinfo->portinfo_mask + 1; 650 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 651 portinfo_idx = portinfo_first; 652 loop: 653 portinfo = &pcbinfo->portinfo[portinfo_idx]; 654 dup = 0; 655 656 if (inp->inp_flags & INP_HIGHPORT) { 657 first = ipport_hifirstauto; /* sysctl */ 658 last = ipport_hilastauto; 659 lastport = &portinfo->lasthi; 660 } else if (inp->inp_flags & INP_LOWPORT) { 661 if (cred && 662 (error = 663 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 664 inp->inp_laddr.s_addr = INADDR_ANY; 665 return (error); 666 } 667 first = ipport_lowfirstauto; /* 1023 */ 668 last = ipport_lowlastauto; /* 600 */ 669 lastport = &portinfo->lastlow; 670 } else { 671 first = ipport_firstauto; /* sysctl */ 672 last = ipport_lastauto; 673 lastport = &portinfo->lastport; 674 } 675 676 /* 677 * This has to be atomic. If the porthash is shared across multiple 678 * protocol threads (aka tcp) then the token must be held. 679 */ 680 GET_PORT_TOKEN(portinfo); 681 682 again: 683 /* 684 * Simple check to ensure all ports are not used up causing 685 * a deadlock here. 686 * 687 * We split the two cases (up and down) so that the direction 688 * is not being tested on each round of the loop. 689 */ 690 if (first > last) { 691 /* 692 * counting down 693 */ 694 in_pcbportrange(&first, &last, portinfo->offset, step); 695 count = (first - last) / step; 696 697 do { 698 if (count-- < 0) { /* completely used? */ 699 error = EADDRNOTAVAIL; 700 goto done; 701 } 702 *lastport -= step; 703 if (*lastport > first || *lastport < last) 704 *lastport = first; 705 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 706 portinfo->offset); 707 lport = htons(*lastport); 708 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr, 709 lport, sin->sin_addr, sin->sin_port, cred)); 710 } else { 711 /* 712 * counting up 713 */ 714 in_pcbportrange(&last, &first, portinfo->offset, step); 715 count = (last - first) / step; 716 717 do { 718 if (count-- < 0) { /* completely used? */ 719 error = EADDRNOTAVAIL; 720 goto done; 721 } 722 *lastport += step; 723 if (*lastport < first || *lastport > last) 724 *lastport = first; 725 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 726 portinfo->offset); 727 lport = htons(*lastport); 728 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr, 729 lport, sin->sin_addr, sin->sin_port, cred)); 730 } 731 732 /* This could happen on loopback interface */ 733 if (sin->sin_port == lport && 734 sin->sin_addr.s_addr == inp->inp_laddr.s_addr) { 735 if (dup) { 736 /* 737 * Duplicate again; give up 738 */ 739 error = EADDRNOTAVAIL; 740 goto done; 741 } 742 dup = 1; 743 goto again; 744 } 745 inp->inp_lport = lport; 746 in_pcbinsporthash(portinfo, inp); 747 error = 0; 748 done: 749 REL_PORT_TOKEN(portinfo); 750 751 if (error) { 752 /* Try next portinfo */ 753 portinfo_idx++; 754 portinfo_idx &= pcbinfo->portinfo_mask; 755 if (portinfo_idx != portinfo_first) 756 goto loop; 757 inp->inp_laddr.s_addr = INADDR_ANY; 758 } 759 return error; 760 } 761 762 /* 763 * Transform old in_pcbconnect() into an inner subroutine for new 764 * in_pcbconnect(): Do some validity-checking on the remote 765 * address (in mbuf 'nam') and then determine local host address 766 * (i.e., which interface) to use to access that remote host. 767 * 768 * This preserves definition of in_pcbconnect(), while supporting a 769 * slightly different version for T/TCP. (This is more than 770 * a bit of a kludge, but cleaning up the internal interfaces would 771 * have forced minor changes in every protocol). 772 */ 773 int 774 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam, 775 struct sockaddr_in **plocal_sin, struct thread *td, int find) 776 { 777 struct in_ifaddr *ia; 778 struct ucred *cred = NULL; 779 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 780 struct sockaddr *jsin; 781 int jailed = 0, alloc_route = 0; 782 783 if (nam->sa_len != sizeof *sin) 784 return (EINVAL); 785 if (sin->sin_family != AF_INET) 786 return (EAFNOSUPPORT); 787 if (sin->sin_port == 0) 788 return (EADDRNOTAVAIL); 789 if (td && td->td_proc && td->td_proc->p_ucred) 790 cred = td->td_proc->p_ucred; 791 if (cred && cred->cr_prison) 792 jailed = 1; 793 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) { 794 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 795 /* 796 * If the destination address is INADDR_ANY, 797 * use the primary local address. 798 * If the supplied address is INADDR_BROADCAST, 799 * and the primary interface supports broadcast, 800 * choose the broadcast address for that interface. 801 */ 802 if (sin->sin_addr.s_addr == INADDR_ANY) 803 sin->sin_addr = IA_SIN(ia)->sin_addr; 804 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST && 805 (ia->ia_ifp->if_flags & IFF_BROADCAST)) 806 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr; 807 } 808 if (find) { 809 struct route *ro; 810 811 ia = NULL; 812 /* 813 * If route is known or can be allocated now, 814 * our src addr is taken from the i/f, else punt. 815 * Note that we should check the address family of the cached 816 * destination, in case of sharing the cache with IPv6. 817 */ 818 ro = &inp->inp_route; 819 if (ro->ro_rt && 820 (!(ro->ro_rt->rt_flags & RTF_UP) || 821 ro->ro_dst.sa_family != AF_INET || 822 satosin(&ro->ro_dst)->sin_addr.s_addr != 823 sin->sin_addr.s_addr || 824 inp->inp_socket->so_options & SO_DONTROUTE)) { 825 RTFREE(ro->ro_rt); 826 ro->ro_rt = NULL; 827 } 828 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 829 (ro->ro_rt == NULL || 830 ro->ro_rt->rt_ifp == NULL)) { 831 /* No route yet, so try to acquire one */ 832 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 833 ro->ro_dst.sa_family = AF_INET; 834 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 835 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 836 sin->sin_addr; 837 rtalloc(ro); 838 alloc_route = 1; 839 } 840 /* 841 * If we found a route, use the address 842 * corresponding to the outgoing interface 843 * unless it is the loopback (in case a route 844 * to our address on another net goes to loopback). 845 */ 846 if (ro->ro_rt && 847 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) { 848 if (jailed) { 849 if (jailed_ip(cred->cr_prison, 850 ro->ro_rt->rt_ifa->ifa_addr)) { 851 ia = ifatoia(ro->ro_rt->rt_ifa); 852 } 853 } else { 854 ia = ifatoia(ro->ro_rt->rt_ifa); 855 } 856 } 857 if (ia == NULL) { 858 u_short fport = sin->sin_port; 859 860 sin->sin_port = 0; 861 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 862 if (ia && jailed && !jailed_ip(cred->cr_prison, 863 sintosa(&ia->ia_addr))) 864 ia = NULL; 865 if (ia == NULL) 866 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 867 if (ia && jailed && !jailed_ip(cred->cr_prison, 868 sintosa(&ia->ia_addr))) 869 ia = NULL; 870 sin->sin_port = fport; 871 if (ia == NULL && 872 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 873 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 874 if (ia && jailed && !jailed_ip(cred->cr_prison, 875 sintosa(&ia->ia_addr))) 876 ia = NULL; 877 878 if (!jailed && ia == NULL) 879 goto fail; 880 } 881 /* 882 * If the destination address is multicast and an outgoing 883 * interface has been set as a multicast option, use the 884 * address of that interface as our source address. 885 */ 886 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 887 inp->inp_moptions != NULL) { 888 struct ip_moptions *imo; 889 struct ifnet *ifp; 890 891 imo = inp->inp_moptions; 892 if (imo->imo_multicast_ifp != NULL) { 893 struct in_ifaddr_container *iac; 894 895 ifp = imo->imo_multicast_ifp; 896 ia = NULL; 897 TAILQ_FOREACH(iac, 898 &in_ifaddrheads[mycpuid], ia_link) { 899 if (iac->ia->ia_ifp == ifp) { 900 ia = iac->ia; 901 break; 902 } 903 } 904 if (ia == NULL) 905 goto fail; 906 } 907 } 908 /* 909 * Don't do pcblookup call here; return interface in plocal_sin 910 * and exit to caller, that will do the lookup. 911 */ 912 if (ia == NULL && jailed) { 913 if ((jsin = prison_get_nonlocal( 914 cred->cr_prison, AF_INET, NULL)) != NULL || 915 (jsin = prison_get_local( 916 cred->cr_prison, AF_INET, NULL)) != NULL) { 917 *plocal_sin = satosin(jsin); 918 } else { 919 /* IPv6 only Jail */ 920 goto fail; 921 } 922 } else { 923 *plocal_sin = &ia->ia_addr; 924 } 925 } 926 return (0); 927 fail: 928 if (alloc_route) { 929 struct route *ro = &inp->inp_route; 930 931 if (ro->ro_rt != NULL) 932 RTFREE(ro->ro_rt); 933 bzero(ro, sizeof(*ro)); 934 } 935 return (EADDRNOTAVAIL); 936 } 937 938 int 939 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam, 940 struct sockaddr_in **plocal_sin, struct thread *td) 941 { 942 return in_pcbladdr_find(inp, nam, plocal_sin, td, 943 (inp->inp_laddr.s_addr == INADDR_ANY)); 944 } 945 946 /* 947 * Outer subroutine: 948 * Connect from a socket to a specified address. 949 * Both address and port must be specified in argument sin. 950 * If don't have a local address for this socket yet, 951 * then pick one. 952 */ 953 int 954 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 955 { 956 struct sockaddr_in *if_sin; 957 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 958 int error; 959 960 /* Call inner routine to assign local interface address. */ 961 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0) 962 return (error); 963 964 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port, 965 inp->inp_laddr.s_addr ? 966 inp->inp_laddr : if_sin->sin_addr, 967 inp->inp_lport, FALSE, NULL) != NULL) { 968 return (EADDRINUSE); 969 } 970 if (inp->inp_laddr.s_addr == INADDR_ANY) { 971 if (inp->inp_lport == 0) { 972 error = in_pcbbind(inp, NULL, td); 973 if (error) 974 return (error); 975 } 976 inp->inp_laddr = if_sin->sin_addr; 977 } 978 inp->inp_faddr = sin->sin_addr; 979 inp->inp_fport = sin->sin_port; 980 in_pcbinsconnhash(inp); 981 return (0); 982 } 983 984 void 985 in_pcbdisconnect(struct inpcb *inp) 986 { 987 988 inp->inp_faddr.s_addr = INADDR_ANY; 989 inp->inp_fport = 0; 990 in_pcbremconnhash(inp); 991 } 992 993 void 994 in_pcbdetach(struct inpcb *inp) 995 { 996 struct socket *so = inp->inp_socket; 997 struct inpcbinfo *ipi = inp->inp_pcbinfo; 998 999 #ifdef IPSEC 1000 ipsec4_delete_pcbpolicy(inp); 1001 #endif /*IPSEC*/ 1002 inp->inp_gencnt = ++ipi->ipi_gencnt; 1003 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 1004 in_pcbremlists(inp); 1005 so->so_pcb = NULL; 1006 sofree(so); /* remove pcb ref */ 1007 if (inp->inp_options) 1008 m_free(inp->inp_options); 1009 if (inp->inp_route.ro_rt) 1010 rtfree(inp->inp_route.ro_rt); 1011 ip_freemoptions(inp->inp_moptions); 1012 inp->inp_vflag = 0; 1013 kfree(inp, M_PCB); 1014 } 1015 1016 /* 1017 * The calling convention of in_setsockaddr() and in_setpeeraddr() was 1018 * modified to match the pru_sockaddr() and pru_peeraddr() entry points 1019 * in struct pr_usrreqs, so that protocols can just reference then directly 1020 * without the need for a wrapper function. The socket must have a valid 1021 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one 1022 * except through a kernel programming error, so it is acceptable to panic 1023 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap 1024 * because there actually /is/ a programming error somewhere... XXX) 1025 */ 1026 int 1027 in_setsockaddr(struct socket *so, struct sockaddr **nam) 1028 { 1029 struct inpcb *inp; 1030 struct sockaddr_in *sin; 1031 1032 /* 1033 * Do the malloc first in case it blocks. 1034 */ 1035 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1036 sin->sin_family = AF_INET; 1037 sin->sin_len = sizeof *sin; 1038 1039 crit_enter(); 1040 inp = so->so_pcb; 1041 if (!inp) { 1042 crit_exit(); 1043 kfree(sin, M_SONAME); 1044 return (ECONNRESET); 1045 } 1046 sin->sin_port = inp->inp_lport; 1047 sin->sin_addr = inp->inp_laddr; 1048 crit_exit(); 1049 1050 *nam = (struct sockaddr *)sin; 1051 return (0); 1052 } 1053 1054 void 1055 in_setsockaddr_dispatch(netmsg_t msg) 1056 { 1057 int error; 1058 1059 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1060 lwkt_replymsg(&msg->lmsg, error); 1061 } 1062 1063 int 1064 in_setpeeraddr(struct socket *so, struct sockaddr **nam) 1065 { 1066 struct inpcb *inp; 1067 struct sockaddr_in *sin; 1068 1069 /* 1070 * Do the malloc first in case it blocks. 1071 */ 1072 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1073 sin->sin_family = AF_INET; 1074 sin->sin_len = sizeof *sin; 1075 1076 crit_enter(); 1077 inp = so->so_pcb; 1078 if (!inp) { 1079 crit_exit(); 1080 kfree(sin, M_SONAME); 1081 return (ECONNRESET); 1082 } 1083 sin->sin_port = inp->inp_fport; 1084 sin->sin_addr = inp->inp_faddr; 1085 crit_exit(); 1086 1087 *nam = (struct sockaddr *)sin; 1088 return (0); 1089 } 1090 1091 void 1092 in_setpeeraddr_dispatch(netmsg_t msg) 1093 { 1094 int error; 1095 1096 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1097 lwkt_replymsg(&msg->lmsg, error); 1098 } 1099 1100 void 1101 in_pcbnotifyall(struct inpcbhead *head, struct in_addr faddr, int err, 1102 void (*notify)(struct inpcb *, int)) 1103 { 1104 struct inpcb *inp, *ninp; 1105 1106 /* 1107 * note: if INP_PLACEMARKER is set we must ignore the rest of 1108 * the structure and skip it. 1109 */ 1110 crit_enter(); 1111 LIST_FOREACH_MUTABLE(inp, head, inp_list, ninp) { 1112 if (inp->inp_flags & INP_PLACEMARKER) 1113 continue; 1114 #ifdef INET6 1115 if (!(inp->inp_vflag & INP_IPV4)) 1116 continue; 1117 #endif 1118 if (inp->inp_faddr.s_addr != faddr.s_addr || 1119 inp->inp_socket == NULL) 1120 continue; 1121 (*notify)(inp, err); /* can remove inp from list! */ 1122 } 1123 crit_exit(); 1124 } 1125 1126 void 1127 in_pcbpurgeif0(struct inpcb *head, struct ifnet *ifp) 1128 { 1129 struct inpcb *inp; 1130 struct ip_moptions *imo; 1131 int i, gap; 1132 1133 for (inp = head; inp != NULL; inp = LIST_NEXT(inp, inp_list)) { 1134 if (inp->inp_flags & INP_PLACEMARKER) 1135 continue; 1136 imo = inp->inp_moptions; 1137 if ((inp->inp_vflag & INP_IPV4) && imo != NULL) { 1138 /* 1139 * Unselect the outgoing interface if it is being 1140 * detached. 1141 */ 1142 if (imo->imo_multicast_ifp == ifp) 1143 imo->imo_multicast_ifp = NULL; 1144 1145 /* 1146 * Drop multicast group membership if we joined 1147 * through the interface being detached. 1148 */ 1149 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1150 i++) { 1151 if (imo->imo_membership[i]->inm_ifp == ifp) { 1152 in_delmulti(imo->imo_membership[i]); 1153 gap++; 1154 } else if (gap != 0) 1155 imo->imo_membership[i - gap] = 1156 imo->imo_membership[i]; 1157 } 1158 imo->imo_num_memberships -= gap; 1159 } 1160 } 1161 } 1162 1163 /* 1164 * Check for alternatives when higher level complains 1165 * about service problems. For now, invalidate cached 1166 * routing information. If the route was created dynamically 1167 * (by a redirect), time to try a default gateway again. 1168 */ 1169 void 1170 in_losing(struct inpcb *inp) 1171 { 1172 struct rtentry *rt; 1173 struct rt_addrinfo rtinfo; 1174 1175 if ((rt = inp->inp_route.ro_rt)) { 1176 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1177 rtinfo.rti_info[RTAX_DST] = rt_key(rt); 1178 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1179 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt); 1180 rtinfo.rti_flags = rt->rt_flags; 1181 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0); 1182 if (rt->rt_flags & RTF_DYNAMIC) { 1183 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1184 rt_mask(rt), rt->rt_flags, NULL); 1185 } 1186 inp->inp_route.ro_rt = NULL; 1187 rtfree(rt); 1188 /* 1189 * A new route can be allocated 1190 * the next time output is attempted. 1191 */ 1192 } 1193 } 1194 1195 /* 1196 * After a routing change, flush old routing 1197 * and allocate a (hopefully) better one. 1198 */ 1199 void 1200 in_rtchange(struct inpcb *inp, int err) 1201 { 1202 if (inp->inp_route.ro_rt) { 1203 rtfree(inp->inp_route.ro_rt); 1204 inp->inp_route.ro_rt = NULL; 1205 /* 1206 * A new route can be allocated the next time 1207 * output is attempted. 1208 */ 1209 } 1210 } 1211 1212 /* 1213 * Lookup a PCB based on the local address and port. 1214 */ 1215 struct inpcb * 1216 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr, 1217 u_int lport_arg, int wild_okay, struct ucred *cred) 1218 { 1219 struct inpcb *inp; 1220 int matchwild = 3, wildcard; 1221 u_short lport = lport_arg; 1222 struct inpcbporthead *porthash; 1223 struct inpcbport *phd; 1224 struct inpcb *match = NULL; 1225 1226 /* 1227 * If the porthashbase is shared across several cpus, it must 1228 * have been locked. 1229 */ 1230 ASSERT_PORT_TOKEN_HELD(portinfo); 1231 1232 /* 1233 * Best fit PCB lookup. 1234 * 1235 * First see if this local port is in use by looking on the 1236 * port hash list. 1237 */ 1238 porthash = &portinfo->porthashbase[ 1239 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 1240 LIST_FOREACH(phd, porthash, phd_hash) { 1241 if (phd->phd_port == lport) 1242 break; 1243 } 1244 if (phd != NULL) { 1245 /* 1246 * Port is in use by one or more PCBs. Look for best 1247 * fit. 1248 */ 1249 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1250 wildcard = 0; 1251 #ifdef INET6 1252 if ((inp->inp_vflag & INP_IPV4) == 0) 1253 continue; 1254 #endif 1255 if (inp->inp_faddr.s_addr != INADDR_ANY) 1256 wildcard++; 1257 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1258 if (laddr.s_addr == INADDR_ANY) 1259 wildcard++; 1260 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1261 continue; 1262 } else { 1263 if (laddr.s_addr != INADDR_ANY) 1264 wildcard++; 1265 } 1266 if (wildcard && !wild_okay) 1267 continue; 1268 if (wildcard < matchwild && 1269 (cred == NULL || 1270 cred->cr_prison == 1271 inp->inp_socket->so_cred->cr_prison)) { 1272 match = inp; 1273 matchwild = wildcard; 1274 if (matchwild == 0) { 1275 break; 1276 } 1277 } 1278 } 1279 } 1280 return (match); 1281 } 1282 1283 struct inpcb * 1284 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo, 1285 const struct inpcb *inp) 1286 { 1287 const struct inp_localgrphead *hdr; 1288 const struct inp_localgroup *grp; 1289 int i; 1290 1291 if (pcbinfo->localgrphashbase == NULL) 1292 return NULL; 1293 1294 hdr = &pcbinfo->localgrphashbase[ 1295 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1296 1297 LIST_FOREACH(grp, hdr, il_list) { 1298 if (grp->il_vflag == inp->inp_vflag && 1299 grp->il_lport == inp->inp_lport && 1300 memcmp(&grp->il_dependladdr, 1301 &inp->inp_inc.inc_ie.ie_dependladdr, 1302 sizeof(grp->il_dependladdr)) == 0) { 1303 break; 1304 } 1305 } 1306 if (grp == NULL || grp->il_inpcnt == 1) 1307 return NULL; 1308 1309 KASSERT(grp->il_inpcnt >= 2, 1310 ("invalid localgroup inp count %d", grp->il_inpcnt)); 1311 for (i = 0; i < grp->il_inpcnt; ++i) { 1312 if (grp->il_inp[i] == inp) { 1313 int last = grp->il_inpcnt - 1; 1314 1315 if (i == last) 1316 last = grp->il_inpcnt - 2; 1317 return grp->il_inp[last]; 1318 } 1319 } 1320 return NULL; 1321 } 1322 1323 static struct inpcb * 1324 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo, 1325 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash) 1326 { 1327 struct inpcb *local_wild = NULL; 1328 const struct inp_localgrphead *hdr; 1329 const struct inp_localgroup *grp; 1330 1331 hdr = &pcbinfo->localgrphashbase[ 1332 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)]; 1333 #ifdef INP_LOCALGROUP_HASHTHR 1334 pkt_hash >>= ncpus2_shift; 1335 #endif 1336 1337 /* 1338 * Order of socket selection: 1339 * 1. non-wild. 1340 * 2. wild. 1341 * 1342 * NOTE: 1343 * - Local group does not contain jailed sockets 1344 * - Local group does not contain IPv4 mapped INET6 wild sockets 1345 */ 1346 LIST_FOREACH(grp, hdr, il_list) { 1347 #ifdef INET6 1348 if (!(grp->il_vflag & INP_IPV4)) 1349 continue; 1350 #endif 1351 if (grp->il_lport == lport) { 1352 int idx; 1353 1354 #ifdef INP_LOCALGROUP_HASHTHR 1355 idx = pkt_hash / grp->il_factor; 1356 KASSERT(idx < grp->il_inpcnt && idx >= 0, 1357 ("invalid hash %04x, cnt %d or fact %d", 1358 pkt_hash, grp->il_inpcnt, grp->il_factor)); 1359 #else 1360 /* 1361 * Modulo-N is used here, which greatly reduces 1362 * completion queue token contention, thus more 1363 * cpu time is saved. 1364 */ 1365 idx = pkt_hash % grp->il_inpcnt; 1366 #endif 1367 1368 if (grp->il_laddr.s_addr == laddr.s_addr) 1369 return grp->il_inp[idx]; 1370 else if (grp->il_laddr.s_addr == INADDR_ANY) 1371 local_wild = grp->il_inp[idx]; 1372 } 1373 } 1374 if (local_wild != NULL) 1375 return local_wild; 1376 return NULL; 1377 } 1378 1379 /* 1380 * Lookup PCB in hash list. 1381 */ 1382 struct inpcb * 1383 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1384 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1385 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m) 1386 { 1387 struct inpcbhead *head; 1388 struct inpcb *inp, *jinp=NULL; 1389 u_short fport = fport_arg, lport = lport_arg; 1390 1391 /* 1392 * First look for an exact match. 1393 */ 1394 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport, 1395 laddr.s_addr, lport, pcbinfo->hashmask)]; 1396 LIST_FOREACH(inp, head, inp_hash) { 1397 #ifdef INET6 1398 if (!(inp->inp_vflag & INP_IPV4)) 1399 continue; 1400 #endif 1401 if (in_hosteq(inp->inp_faddr, faddr) && 1402 in_hosteq(inp->inp_laddr, laddr) && 1403 inp->inp_fport == fport && inp->inp_lport == lport) { 1404 /* found */ 1405 if (inp->inp_socket == NULL || 1406 inp->inp_socket->so_cred->cr_prison == NULL) { 1407 return (inp); 1408 } else { 1409 if (jinp == NULL) 1410 jinp = inp; 1411 } 1412 } 1413 } 1414 if (jinp != NULL) 1415 return (jinp); 1416 if (wildcard) { 1417 struct inpcb *local_wild = NULL; 1418 struct inpcb *jinp_wild = NULL; 1419 #ifdef INET6 1420 struct inpcb *local_wild_mapped = NULL; 1421 #endif 1422 struct inpcontainer *ic; 1423 struct inpcontainerhead *chead; 1424 struct sockaddr_in jsin; 1425 struct ucred *cred; 1426 1427 /* 1428 * Check local group first 1429 */ 1430 if (pcbinfo->localgrphashbase != NULL && 1431 m != NULL && (m->m_flags & M_HASH) && 1432 !(ifp && ifp->if_type == IFT_FAITH)) { 1433 inp = inp_localgroup_lookup(pcbinfo, 1434 laddr, lport, m->m_pkthdr.hash); 1435 if (inp != NULL) 1436 return inp; 1437 } 1438 1439 /* 1440 * Order of socket selection: 1441 * 1. non-jailed, non-wild. 1442 * 2. non-jailed, wild. 1443 * 3. jailed, non-wild. 1444 * 4. jailed, wild. 1445 */ 1446 jsin.sin_family = AF_INET; 1447 chead = &pcbinfo->wildcardhashbase[ 1448 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)]; 1449 LIST_FOREACH(ic, chead, ic_list) { 1450 inp = ic->ic_inp; 1451 jsin.sin_addr.s_addr = laddr.s_addr; 1452 #ifdef INET6 1453 if (!(inp->inp_vflag & INP_IPV4)) 1454 continue; 1455 #endif 1456 if (inp->inp_socket != NULL) 1457 cred = inp->inp_socket->so_cred; 1458 else 1459 cred = NULL; 1460 if (cred != NULL && jailed(cred)) { 1461 if (jinp != NULL) 1462 continue; 1463 else 1464 if (!jailed_ip(cred->cr_prison, 1465 (struct sockaddr *)&jsin)) 1466 continue; 1467 } 1468 if (inp->inp_lport == lport) { 1469 if (ifp && ifp->if_type == IFT_FAITH && 1470 !(inp->inp_flags & INP_FAITH)) 1471 continue; 1472 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1473 if (cred != NULL && jailed(cred)) 1474 jinp = inp; 1475 else 1476 return (inp); 1477 } 1478 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1479 #ifdef INET6 1480 if (INP_CHECK_SOCKAF(inp->inp_socket, 1481 AF_INET6)) 1482 local_wild_mapped = inp; 1483 else 1484 #endif 1485 if (cred != NULL && 1486 jailed(cred)) 1487 jinp_wild = inp; 1488 else 1489 local_wild = inp; 1490 } 1491 } 1492 } 1493 if (local_wild != NULL) 1494 return (local_wild); 1495 #ifdef INET6 1496 if (local_wild_mapped != NULL) 1497 return (local_wild_mapped); 1498 #endif 1499 if (jinp != NULL) 1500 return (jinp); 1501 return (jinp_wild); 1502 } 1503 1504 /* 1505 * Not found. 1506 */ 1507 return (NULL); 1508 } 1509 1510 struct inpcb * 1511 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1512 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1513 boolean_t wildcard, struct ifnet *ifp) 1514 { 1515 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg, 1516 laddr, lport_arg, wildcard, ifp, NULL); 1517 } 1518 1519 /* 1520 * Insert PCB into connection hash table. 1521 */ 1522 void 1523 in_pcbinsconnhash(struct inpcb *inp) 1524 { 1525 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1526 struct inpcbhead *bucket; 1527 u_int32_t hashkey_faddr, hashkey_laddr; 1528 1529 #ifdef INET6 1530 if (inp->inp_vflag & INP_IPV6) { 1531 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */; 1532 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */; 1533 } else { 1534 #endif 1535 hashkey_faddr = inp->inp_faddr.s_addr; 1536 hashkey_laddr = inp->inp_laddr.s_addr; 1537 #ifdef INET6 1538 } 1539 #endif 1540 1541 KASSERT(!(inp->inp_flags & INP_WILDCARD), 1542 ("already on wildcardhash")); 1543 KASSERT(!(inp->inp_flags & INP_CONNECTED), 1544 ("already on connhash")); 1545 inp->inp_flags |= INP_CONNECTED; 1546 1547 /* 1548 * Insert into the connection hash table. 1549 */ 1550 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr, 1551 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)]; 1552 LIST_INSERT_HEAD(bucket, inp, inp_hash); 1553 } 1554 1555 /* 1556 * Remove PCB from connection hash table. 1557 */ 1558 void 1559 in_pcbremconnhash(struct inpcb *inp) 1560 { 1561 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected")); 1562 LIST_REMOVE(inp, inp_hash); 1563 inp->inp_flags &= ~INP_CONNECTED; 1564 } 1565 1566 /* 1567 * Insert PCB into port hash table. 1568 */ 1569 void 1570 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp) 1571 { 1572 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1573 struct inpcbporthead *pcbporthash; 1574 struct inpcbport *phd; 1575 1576 /* 1577 * If the porthashbase is shared across several cpus, it must 1578 * have been locked. 1579 */ 1580 ASSERT_PORT_TOKEN_HELD(portinfo); 1581 1582 /* 1583 * Insert into the port hash table. 1584 */ 1585 pcbporthash = &portinfo->porthashbase[ 1586 INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)]; 1587 1588 /* Go through port list and look for a head for this lport. */ 1589 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1590 if (phd->phd_port == inp->inp_lport) 1591 break; 1592 } 1593 1594 /* If none exists, use saved one and tack it on. */ 1595 if (phd == NULL) { 1596 KKASSERT(pcbinfo->portsave != NULL); 1597 phd = pcbinfo->portsave; 1598 pcbinfo->portsave = NULL; 1599 phd->phd_port = inp->inp_lport; 1600 LIST_INIT(&phd->phd_pcblist); 1601 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1602 } 1603 1604 inp->inp_portinfo = portinfo; 1605 inp->inp_phd = phd; 1606 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1607 1608 /* 1609 * Malloc one inpcbport for later use. It is safe to use 1610 * "wait" malloc here (port token would be released, if 1611 * malloc ever blocked), since all changes to the porthash 1612 * are done. 1613 */ 1614 if (pcbinfo->portsave == NULL) { 1615 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), 1616 M_PCB, M_INTWAIT | M_ZERO); 1617 } 1618 } 1619 1620 void 1621 in_pcbinsporthash_lport(struct inpcb *inp) 1622 { 1623 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1624 struct inpcbportinfo *portinfo; 1625 u_short lport_ho; 1626 1627 /* Locate the proper portinfo based on lport */ 1628 lport_ho = ntohs(inp->inp_lport); 1629 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 1630 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset); 1631 1632 GET_PORT_TOKEN(portinfo); 1633 in_pcbinsporthash(portinfo, inp); 1634 REL_PORT_TOKEN(portinfo); 1635 } 1636 1637 static struct inp_localgroup * 1638 inp_localgroup_alloc(u_char vflag, 1639 uint16_t port, const union in_dependaddr *addr, int size) 1640 { 1641 struct inp_localgroup *grp; 1642 1643 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]), 1644 M_TEMP, M_INTWAIT | M_ZERO); 1645 grp->il_vflag = vflag; 1646 grp->il_lport = port; 1647 grp->il_dependladdr = *addr; 1648 grp->il_inpsiz = size; 1649 1650 return grp; 1651 } 1652 1653 static struct inp_localgroup * 1654 inp_localgroup_create(struct inp_localgrphead *hdr, u_char vflag, 1655 uint16_t port, const union in_dependaddr *addr, int size) 1656 { 1657 struct inp_localgroup *grp; 1658 1659 grp = inp_localgroup_alloc(vflag, port, addr, size); 1660 LIST_INSERT_HEAD(hdr, grp, il_list); 1661 1662 return grp; 1663 } 1664 1665 static void 1666 inp_localgroup_destroy(struct inp_localgroup *grp) 1667 { 1668 LIST_REMOVE(grp, il_list); 1669 kfree(grp, M_TEMP); 1670 } 1671 1672 static void 1673 inp_localgroup_copy(struct inp_localgroup *grp, 1674 const struct inp_localgroup *old_grp) 1675 { 1676 int i; 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 1687 static struct inp_localgroup * 1688 inp_localgroup_resize(struct inp_localgrphead *hdr, 1689 struct inp_localgroup *old_grp, int size) 1690 { 1691 struct inp_localgroup *grp; 1692 1693 grp = inp_localgroup_create(hdr, old_grp->il_vflag, 1694 old_grp->il_lport, &old_grp->il_dependladdr, size); 1695 inp_localgroup_copy(grp, old_grp); 1696 inp_localgroup_destroy(old_grp); 1697 1698 return grp; 1699 } 1700 1701 static void 1702 inp_localgroup_factor(struct inp_localgroup *grp) 1703 { 1704 grp->il_factor = 1705 ((uint32_t)(0xffff >> ncpus2_shift) / grp->il_inpcnt) + 1; 1706 KASSERT(grp->il_factor != 0, ("invalid local group factor, " 1707 "ncpus2_shift %d, inpcnt %d", ncpus2_shift, grp->il_inpcnt)); 1708 } 1709 1710 static void 1711 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1712 { 1713 struct inp_localgrphead *hdr; 1714 struct inp_localgroup *grp; 1715 struct ucred *cred; 1716 1717 if (pcbinfo->localgrphashbase == NULL) 1718 return; 1719 1720 /* 1721 * XXX don't allow jailed socket to join local group 1722 */ 1723 if (inp->inp_socket != NULL) 1724 cred = inp->inp_socket->so_cred; 1725 else 1726 cred = NULL; 1727 if (cred != NULL && jailed(cred)) 1728 return; 1729 1730 #ifdef INET6 1731 /* 1732 * XXX don't allow IPv4 mapped INET6 wild socket 1733 */ 1734 if ((inp->inp_vflag & INP_IPV4) && 1735 inp->inp_laddr.s_addr == INADDR_ANY && 1736 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) 1737 return; 1738 #endif 1739 1740 hdr = &pcbinfo->localgrphashbase[ 1741 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1742 1743 LIST_FOREACH(grp, hdr, il_list) { 1744 if (grp->il_vflag == inp->inp_vflag && 1745 grp->il_lport == inp->inp_lport && 1746 memcmp(&grp->il_dependladdr, 1747 &inp->inp_inc.inc_ie.ie_dependladdr, 1748 sizeof(grp->il_dependladdr)) == 0) { 1749 break; 1750 } 1751 } 1752 if (grp == NULL) { 1753 /* Create new local group */ 1754 grp = inp_localgroup_create(hdr, inp->inp_vflag, 1755 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 1756 INP_LOCALGROUP_SIZMIN); 1757 } else if (grp->il_inpcnt == grp->il_inpsiz) { 1758 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) { 1759 static int limit_logged = 0; 1760 1761 if (!limit_logged) { 1762 limit_logged = 1; 1763 kprintf("local group port %d, " 1764 "limit reached\n", ntohs(grp->il_lport)); 1765 } 1766 return; 1767 } 1768 1769 /* Expand this local group */ 1770 grp = inp_localgroup_resize(hdr, grp, grp->il_inpsiz * 2); 1771 } 1772 1773 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 1774 ("invalid local group size %d and count %d", 1775 grp->il_inpsiz, grp->il_inpcnt)); 1776 grp->il_inp[grp->il_inpcnt] = inp; 1777 grp->il_inpcnt++; 1778 inp_localgroup_factor(grp); 1779 } 1780 1781 void 1782 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1783 { 1784 struct inpcontainer *ic; 1785 struct inpcontainerhead *bucket; 1786 1787 in_pcbinslocalgrphash_oncpu(inp, pcbinfo); 1788 1789 bucket = &pcbinfo->wildcardhashbase[ 1790 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1791 1792 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT); 1793 ic->ic_inp = inp; 1794 LIST_INSERT_HEAD(bucket, ic, ic_list); 1795 } 1796 1797 /* 1798 * Insert PCB into wildcard hash table. 1799 */ 1800 void 1801 in_pcbinswildcardhash(struct inpcb *inp) 1802 { 1803 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1804 1805 KASSERT(!(inp->inp_flags & INP_CONNECTED), 1806 ("already on connhash")); 1807 KASSERT(!(inp->inp_flags & INP_WILDCARD), 1808 ("already on wildcardhash")); 1809 inp->inp_flags |= INP_WILDCARD; 1810 1811 in_pcbinswildcardhash_oncpu(inp, pcbinfo); 1812 } 1813 1814 static void 1815 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1816 { 1817 struct inp_localgrphead *hdr; 1818 struct inp_localgroup *grp; 1819 1820 if (pcbinfo->localgrphashbase == NULL) 1821 return; 1822 1823 hdr = &pcbinfo->localgrphashbase[ 1824 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1825 1826 LIST_FOREACH(grp, hdr, il_list) { 1827 int i; 1828 1829 for (i = 0; i < grp->il_inpcnt; ++i) { 1830 if (grp->il_inp[i] != inp) 1831 continue; 1832 1833 if (grp->il_inpcnt == 1) { 1834 /* Destroy this local group */ 1835 inp_localgroup_destroy(grp); 1836 } else { 1837 /* Pull up inpcbs */ 1838 for (; i + 1 < grp->il_inpcnt; ++i) 1839 grp->il_inp[i] = grp->il_inp[i + 1]; 1840 grp->il_inpcnt--; 1841 inp_localgroup_factor(grp); 1842 1843 if (grp->il_inpsiz > INP_LOCALGROUP_SIZMIN && 1844 grp->il_inpcnt <= (grp->il_inpsiz / 4)) { 1845 /* Shrink this local group */ 1846 grp = inp_localgroup_resize(hdr, grp, 1847 grp->il_inpsiz / 2); 1848 } 1849 } 1850 return; 1851 } 1852 } 1853 } 1854 1855 void 1856 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1857 { 1858 struct inpcontainer *ic; 1859 struct inpcontainerhead *head; 1860 1861 in_pcbremlocalgrphash_oncpu(inp, pcbinfo); 1862 1863 /* find bucket */ 1864 head = &pcbinfo->wildcardhashbase[ 1865 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1866 1867 LIST_FOREACH(ic, head, ic_list) { 1868 if (ic->ic_inp == inp) 1869 goto found; 1870 } 1871 return; /* not found! */ 1872 1873 found: 1874 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */ 1875 kfree(ic, M_TEMP); /* deallocate container */ 1876 } 1877 1878 /* 1879 * Remove PCB from wildcard hash table. 1880 */ 1881 void 1882 in_pcbremwildcardhash(struct inpcb *inp) 1883 { 1884 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1885 1886 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard")); 1887 in_pcbremwildcardhash_oncpu(inp, pcbinfo); 1888 inp->inp_flags &= ~INP_WILDCARD; 1889 } 1890 1891 /* 1892 * Remove PCB from various lists. 1893 */ 1894 void 1895 in_pcbremlists(struct inpcb *inp) 1896 { 1897 if (inp->inp_lport) { 1898 struct inpcbportinfo *portinfo; 1899 struct inpcbport *phd; 1900 1901 /* 1902 * NOTE: 1903 * inp->inp_portinfo is _not_ necessary same as 1904 * inp->inp_pcbinfo->portinfo. 1905 */ 1906 portinfo = inp->inp_portinfo; 1907 GET_PORT_TOKEN(portinfo); 1908 1909 phd = inp->inp_phd; 1910 LIST_REMOVE(inp, inp_portlist); 1911 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1912 LIST_REMOVE(phd, phd_hash); 1913 kfree(phd, M_PCB); 1914 } 1915 1916 REL_PORT_TOKEN(portinfo); 1917 } 1918 if (inp->inp_flags & INP_WILDCARD) { 1919 in_pcbremwildcardhash(inp); 1920 } else if (inp->inp_flags & INP_CONNECTED) { 1921 in_pcbremconnhash(inp); 1922 } 1923 LIST_REMOVE(inp, inp_list); 1924 inp->inp_pcbinfo->ipi_count--; 1925 } 1926 1927 int 1928 prison_xinpcb(struct thread *td, struct inpcb *inp) 1929 { 1930 struct ucred *cr; 1931 1932 if (td->td_proc == NULL) 1933 return (0); 1934 cr = td->td_proc->p_ucred; 1935 if (cr->cr_prison == NULL) 1936 return (0); 1937 if (inp->inp_socket && inp->inp_socket->so_cred && 1938 inp->inp_socket->so_cred->cr_prison && 1939 cr->cr_prison == inp->inp_socket->so_cred->cr_prison) 1940 return (0); 1941 return (1); 1942 } 1943 1944 int 1945 in_pcblist_global(SYSCTL_HANDLER_ARGS) 1946 { 1947 struct inpcbinfo *pcbinfo = arg1; 1948 struct inpcb *inp, *marker; 1949 struct xinpcb xi; 1950 int error, i, n; 1951 1952 /* 1953 * The process of preparing the TCB list is too time-consuming and 1954 * resource-intensive to repeat twice on every request. 1955 */ 1956 if (req->oldptr == NULL) { 1957 n = pcbinfo->ipi_count; 1958 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 1959 return 0; 1960 } 1961 1962 if (req->newptr != NULL) 1963 return EPERM; 1964 1965 /* 1966 * OK, now we're committed to doing something. Re-fetch ipi_count 1967 * after obtaining the generation count. 1968 */ 1969 n = pcbinfo->ipi_count; 1970 1971 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 1972 marker->inp_flags |= INP_PLACEMARKER; 1973 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1974 1975 i = 0; 1976 error = 0; 1977 1978 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 1979 LIST_REMOVE(marker, inp_list); 1980 LIST_INSERT_AFTER(inp, marker, inp_list); 1981 1982 if (inp->inp_flags & INP_PLACEMARKER) 1983 continue; 1984 if (prison_xinpcb(req->td, inp)) 1985 continue; 1986 bzero(&xi, sizeof xi); 1987 xi.xi_len = sizeof xi; 1988 bcopy(inp, &xi.xi_inp, sizeof *inp); 1989 if (inp->inp_socket) 1990 sotoxsocket(inp->inp_socket, &xi.xi_socket); 1991 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0) 1992 break; 1993 ++i; 1994 } 1995 LIST_REMOVE(marker, inp_list); 1996 if (error == 0 && i < n) { 1997 bzero(&xi, sizeof xi); 1998 xi.xi_len = sizeof xi; 1999 while (i < n) { 2000 error = SYSCTL_OUT(req, &xi, sizeof xi); 2001 ++i; 2002 } 2003 } 2004 kfree(marker, M_TEMP); 2005 return(error); 2006 } 2007 2008 int 2009 in_pcblist_global_cpu0(SYSCTL_HANDLER_ARGS) 2010 { 2011 boolean_t migrate = FALSE; 2012 int origcpu = mycpuid; 2013 int error; 2014 2015 if (origcpu != 0) { 2016 migrate = TRUE; 2017 lwkt_migratecpu(0); 2018 } 2019 2020 error = in_pcblist_global(oidp, arg1, arg2, req); 2021 2022 if (migrate) 2023 lwkt_migratecpu(origcpu); 2024 return error; 2025 } 2026 2027 int 2028 in_pcblist_global_nomarker(SYSCTL_HANDLER_ARGS, struct xinpcb **xi0, int *nxi0) 2029 { 2030 struct inpcbinfo *pcbinfo = arg1; 2031 struct inpcb *inp; 2032 struct xinpcb *xi; 2033 int nxi; 2034 2035 *nxi0 = 0; 2036 *xi0 = NULL; 2037 2038 /* 2039 * The process of preparing the PCB list is too time-consuming and 2040 * resource-intensive to repeat twice on every request. 2041 */ 2042 if (req->oldptr == NULL) { 2043 int n = pcbinfo->ipi_count; 2044 2045 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 2046 return 0; 2047 } 2048 2049 if (req->newptr != NULL) 2050 return EPERM; 2051 2052 if (pcbinfo->ipi_count == 0) 2053 return 0; 2054 2055 nxi = 0; 2056 xi = kmalloc(pcbinfo->ipi_count * sizeof(*xi), M_TEMP, 2057 M_WAITOK | M_ZERO | M_NULLOK); 2058 if (xi == NULL) 2059 return ENOMEM; 2060 2061 LIST_FOREACH(inp, &pcbinfo->pcblisthead, inp_list) { 2062 struct xinpcb *xi_ptr = &xi[nxi]; 2063 2064 if (prison_xinpcb(req->td, inp)) 2065 continue; 2066 2067 xi_ptr->xi_len = sizeof(*xi_ptr); 2068 bcopy(inp, &xi_ptr->xi_inp, sizeof(*inp)); 2069 if (inp->inp_socket) 2070 sotoxsocket(inp->inp_socket, &xi_ptr->xi_socket); 2071 ++nxi; 2072 } 2073 2074 if (nxi == 0) { 2075 kfree(xi, M_TEMP); 2076 return 0; 2077 } 2078 2079 *nxi0 = nxi; 2080 *xi0 = xi; 2081 2082 return 0; 2083 } 2084 2085 void 2086 in_savefaddr(struct socket *so, const struct sockaddr *faddr) 2087 { 2088 struct sockaddr_in *sin; 2089 2090 KASSERT(faddr->sa_family == AF_INET, 2091 ("not AF_INET faddr %d", faddr->sa_family)); 2092 2093 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO); 2094 sin->sin_family = AF_INET; 2095 sin->sin_len = sizeof(*sin); 2096 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port; 2097 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr; 2098 2099 so->so_faddr = (struct sockaddr *)sin; 2100 } 2101 2102 void 2103 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize, 2104 boolean_t shared, u_short offset) 2105 { 2106 memset(portinfo, 0, sizeof(*portinfo)); 2107 2108 portinfo->offset = offset; 2109 portinfo->lastport = offset; 2110 portinfo->lastlow = offset; 2111 portinfo->lasthi = offset; 2112 2113 portinfo->porthashbase = hashinit(hashsize, M_PCB, 2114 &portinfo->porthashmask); 2115 2116 if (shared) { 2117 portinfo->porttoken = kmalloc(sizeof(struct lwkt_token), 2118 M_PCB, M_WAITOK); 2119 lwkt_token_init(portinfo->porttoken, "porttoken"); 2120 } 2121 } 2122 2123 void 2124 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step) 2125 { 2126 int hi, lo; 2127 2128 if (step == 1) 2129 return; 2130 2131 hi = *hi0; 2132 lo = *lo0; 2133 2134 hi = rounddown2(hi, step); 2135 hi += ofs; 2136 if (hi > (int)*hi0) 2137 hi -= step; 2138 2139 lo = roundup2(lo, step); 2140 lo -= (step - ofs); 2141 if (lo < (int)*lo0) 2142 lo += step; 2143 2144 *hi0 = hi; 2145 *lo0 = lo; 2146 } 2147