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 #include <net/netisr2.h> 93 94 #include <netinet/in.h> 95 #include <netinet/in_pcb.h> 96 #include <netinet/in_var.h> 97 #include <netinet/ip_var.h> 98 #ifdef INET6 99 #include <netinet/ip6.h> 100 #include <netinet6/ip6_var.h> 101 #endif /* INET6 */ 102 103 #ifdef IPSEC 104 #include <netinet6/ipsec.h> 105 #include <netproto/key/key.h> 106 #include <netproto/ipsec/esp_var.h> 107 #endif 108 109 #ifdef FAST_IPSEC 110 #if defined(IPSEC) || defined(IPSEC_ESP) 111 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!" 112 #endif 113 114 #include <netproto/ipsec/ipsec.h> 115 #include <netproto/ipsec/key.h> 116 #define IPSEC 117 #endif /* FAST_IPSEC */ 118 119 #define INP_LOCALGROUP_SIZMIN 8 120 #define INP_LOCALGROUP_SIZMAX 256 121 122 struct in_addr zeroin_addr; 123 124 /* 125 * These configure the range of local port addresses assigned to 126 * "unspecified" outgoing connections/packets/whatever. 127 */ 128 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */ 129 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */ 130 131 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */ 132 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */ 133 134 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 135 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */ 136 137 #define RANGECHK(var, min, max) \ 138 if ((var) < (min)) { (var) = (min); } \ 139 else if ((var) > (max)) { (var) = (max); } 140 141 int udpencap_enable = 1; /* enabled by default */ 142 int udpencap_port = 4500; /* triggers decapsulation */ 143 144 /* 145 * Per-netisr inpcb markers. 146 * NOTE: they should only be used in netisrs. 147 */ 148 static struct inpcb *in_pcbmarkers; 149 static struct inpcontainer *in_pcbcontainer_markers; 150 151 static int 152 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 153 { 154 int error; 155 156 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); 157 if (!error) { 158 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 159 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 160 161 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX); 162 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX); 163 164 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX); 165 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX); 166 } 167 return (error); 168 } 169 170 #undef RANGECHK 171 172 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); 173 174 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW, 175 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", ""); 176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW, 177 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", ""); 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW, 179 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", ""); 180 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW, 181 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", ""); 182 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW, 183 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", ""); 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW, 185 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", ""); 186 187 /* 188 * in_pcb.c: manage the Protocol Control Blocks. 189 * 190 * NOTE: It is assumed that most of these functions will be called from 191 * a critical section. XXX - There are, unfortunately, a few exceptions 192 * to this rule that should be fixed. 193 * 194 * NOTE: The caller should initialize the cpu field to the cpu running the 195 * protocol stack associated with this inpcbinfo. 196 */ 197 198 void 199 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared) 200 { 201 KASSERT(cpu >= 0 && cpu < ncpus, ("invalid cpu%d", cpu)); 202 pcbinfo->cpu = cpu; 203 204 LIST_INIT(&pcbinfo->pcblisthead); 205 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB, 206 M_WAITOK | M_ZERO); 207 208 if (shared) { 209 pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token), 210 M_PCB, M_WAITOK); 211 lwkt_token_init(pcbinfo->infotoken, "infotoken"); 212 } else { 213 pcbinfo->infotoken = NULL; 214 } 215 } 216 217 struct baddynamicports baddynamicports; 218 219 /* 220 * Check if the specified port is invalid for dynamic allocation. 221 */ 222 int 223 in_baddynamic(u_int16_t port, u_int16_t proto) 224 { 225 switch (proto) { 226 case IPPROTO_TCP: 227 return (DP_ISSET(baddynamicports.tcp, port)); 228 case IPPROTO_UDP: 229 #ifdef IPSEC 230 /* Cannot preset this as it is a sysctl */ 231 if (port == udpencap_port) 232 return (1); 233 #endif 234 return (DP_ISSET(baddynamicports.udp, port)); 235 default: 236 return (0); 237 } 238 } 239 240 void 241 in_pcbonlist(struct inpcb *inp) 242 { 243 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 244 245 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 246 ("not in the correct netisr")); 247 KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist")); 248 inp->inp_flags |= INP_ONLIST; 249 250 GET_PCBINFO_TOKEN(pcbinfo); 251 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list); 252 pcbinfo->ipi_count++; 253 REL_PCBINFO_TOKEN(pcbinfo); 254 } 255 256 void 257 in_pcbofflist(struct inpcb *inp) 258 { 259 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 260 261 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 262 ("not in the correct netisr")); 263 KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist")); 264 inp->inp_flags &= ~INP_ONLIST; 265 266 GET_PCBINFO_TOKEN(pcbinfo); 267 LIST_REMOVE(inp, inp_list); 268 KASSERT(pcbinfo->ipi_count > 0, 269 ("invalid inpcb count %d", pcbinfo->ipi_count)); 270 pcbinfo->ipi_count--; 271 REL_PCBINFO_TOKEN(pcbinfo); 272 } 273 274 /* 275 * Allocate a PCB and associate it with the socket. 276 */ 277 int 278 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 279 { 280 struct inpcb *inp; 281 #ifdef IPSEC 282 int error; 283 #endif 284 285 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK); 286 if (inp == NULL) 287 return (ENOMEM); 288 inp->inp_lgrpindex = -1; 289 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 290 inp->inp_pcbinfo = pcbinfo; 291 inp->inp_socket = so; 292 #ifdef IPSEC 293 error = ipsec_init_policy(so, &inp->inp_sp); 294 if (error != 0) { 295 kfree(inp, M_PCB); 296 return (error); 297 } 298 #endif 299 #ifdef INET6 300 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 301 if (ip6_auto_flowlabel) 302 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 303 inp->inp_af = AF_INET6; 304 } else 305 #endif 306 inp->inp_af = AF_INET; 307 soreference(so); 308 so->so_pcb = inp; 309 310 in_pcbonlist(inp); 311 return (0); 312 } 313 314 /* 315 * Unlink a pcb with the intention of moving it to another cpu with a 316 * different pcbinfo. While unlinked nothing should attempt to dereference 317 * inp_pcbinfo, NULL it out so we assert if it does. 318 */ 319 void 320 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags) 321 { 322 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch")); 323 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0, 324 ("already linked")); 325 326 in_pcbofflist(inp); 327 inp->inp_pcbinfo = NULL; 328 } 329 330 void 331 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 332 { 333 in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD); 334 } 335 336 /* 337 * Relink a pcb into a new pcbinfo. 338 */ 339 void 340 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags) 341 { 342 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo")); 343 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0, 344 ("already linked")); 345 346 inp->inp_pcbinfo = pcbinfo; 347 in_pcbonlist(inp); 348 } 349 350 void 351 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 352 { 353 return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD); 354 } 355 356 static int 357 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred) 358 { 359 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 360 struct inpcbportinfo *portinfo; 361 u_short first, last, lport, step; 362 u_short *lastport; 363 int count, error; 364 int portinfo_first, portinfo_idx; 365 366 inp->inp_flags |= INP_ANONPORT; 367 368 step = pcbinfo->portinfo_mask + 1; 369 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 370 portinfo_idx = portinfo_first; 371 loop: 372 portinfo = &pcbinfo->portinfo[portinfo_idx]; 373 374 if (inp->inp_flags & INP_HIGHPORT) { 375 first = ipport_hifirstauto; /* sysctl */ 376 last = ipport_hilastauto; 377 lastport = &portinfo->lasthi; 378 } else if (inp->inp_flags & INP_LOWPORT) { 379 if (cred && 380 (error = 381 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 382 inp->inp_laddr.s_addr = INADDR_ANY; 383 return error; 384 } 385 first = ipport_lowfirstauto; /* 1023 */ 386 last = ipport_lowlastauto; /* 600 */ 387 lastport = &portinfo->lastlow; 388 } else { 389 first = ipport_firstauto; /* sysctl */ 390 last = ipport_lastauto; 391 lastport = &portinfo->lastport; 392 } 393 394 /* 395 * This has to be atomic. If the porthash is shared across multiple 396 * protocol threads (aka tcp) then the token must be held. 397 */ 398 GET_PORT_TOKEN(portinfo); 399 400 /* 401 * Simple check to ensure all ports are not used up causing 402 * a deadlock here. 403 * 404 * We split the two cases (up and down) so that the direction 405 * is not being tested on each round of the loop. 406 */ 407 if (first > last) { 408 /* 409 * counting down 410 */ 411 in_pcbportrange(&first, &last, portinfo->offset, step); 412 count = (first - last) / step; 413 414 do { 415 if (count-- < 0) { /* completely used? */ 416 error = EADDRNOTAVAIL; 417 goto done; 418 } 419 *lastport -= step; 420 if (*lastport > first || *lastport < last) 421 *lastport = first; 422 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 423 portinfo->offset); 424 lport = htons(*lastport); 425 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport, 426 wild, cred)); 427 } else { 428 /* 429 * counting up 430 */ 431 in_pcbportrange(&last, &first, portinfo->offset, step); 432 count = (last - first) / step; 433 434 do { 435 if (count-- < 0) { /* completely used? */ 436 error = EADDRNOTAVAIL; 437 goto done; 438 } 439 *lastport += step; 440 if (*lastport < first || *lastport > last) 441 *lastport = first; 442 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 443 portinfo->offset); 444 lport = htons(*lastport); 445 } while (in_pcblookup_local(portinfo, inp->inp_laddr, lport, 446 wild, cred)); 447 } 448 inp->inp_lport = lport; 449 in_pcbinsporthash(portinfo, inp); 450 error = 0; 451 done: 452 REL_PORT_TOKEN(portinfo); 453 454 if (error) { 455 /* Try next portinfo */ 456 portinfo_idx++; 457 portinfo_idx &= pcbinfo->portinfo_mask; 458 if (portinfo_idx != portinfo_first) 459 goto loop; 460 inp->inp_laddr.s_addr = INADDR_ANY; 461 } 462 return error; 463 } 464 465 int 466 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 467 { 468 struct socket *so = inp->inp_socket; 469 struct sockaddr_in jsin; 470 struct ucred *cred = NULL; 471 int wild = 0; 472 473 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 474 return (EADDRNOTAVAIL); 475 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 476 return (EINVAL); /* already bound */ 477 478 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT))) 479 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */ 480 if (td->td_proc) 481 cred = td->td_proc->p_ucred; 482 483 if (nam != NULL) { 484 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 485 struct inpcbinfo *pcbinfo; 486 struct inpcbportinfo *portinfo; 487 struct inpcb *t; 488 u_short lport, lport_ho; 489 int reuseport = (so->so_options & SO_REUSEPORT); 490 int error; 491 492 if (nam->sa_len != sizeof *sin) 493 return (EINVAL); 494 #ifdef notdef 495 /* 496 * We should check the family, but old programs 497 * incorrectly fail to initialize it. 498 */ 499 if (sin->sin_family != AF_INET) 500 return (EAFNOSUPPORT); 501 #endif 502 if (!prison_replace_wildcards(td, nam)) 503 return (EINVAL); 504 505 lport = sin->sin_port; 506 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 507 /* 508 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 509 * allow complete duplication of binding if 510 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 511 * and a multicast address is bound on both 512 * new and duplicated sockets. 513 */ 514 if (so->so_options & SO_REUSEADDR) 515 reuseport = SO_REUSEADDR | SO_REUSEPORT; 516 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 517 sin->sin_port = 0; /* yech... */ 518 bzero(&sin->sin_zero, sizeof sin->sin_zero); 519 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL) 520 return (EADDRNOTAVAIL); 521 } 522 523 inp->inp_laddr = sin->sin_addr; 524 525 jsin.sin_family = AF_INET; 526 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 527 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 528 inp->inp_laddr.s_addr = INADDR_ANY; 529 return (EINVAL); 530 } 531 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 532 533 if (lport == 0) { 534 /* Auto-select local port */ 535 return in_pcbsetlport(inp, wild, cred); 536 } 537 lport_ho = ntohs(lport); 538 539 /* GROSS */ 540 if (lport_ho < IPPORT_RESERVED && cred && 541 (error = 542 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 543 inp->inp_laddr.s_addr = INADDR_ANY; 544 return (error); 545 } 546 547 /* 548 * Locate the proper portinfo based on lport 549 */ 550 pcbinfo = inp->inp_pcbinfo; 551 portinfo = 552 &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 553 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == 554 portinfo->offset); 555 556 /* 557 * This has to be atomic. If the porthash is shared across 558 * multiple protocol threads (aka tcp) then the token must 559 * be held. 560 */ 561 GET_PORT_TOKEN(portinfo); 562 563 if (so->so_cred->cr_uid != 0 && 564 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 565 t = in_pcblookup_local(portinfo, sin->sin_addr, lport, 566 INPLOOKUP_WILDCARD, cred); 567 if (t && 568 (so->so_cred->cr_uid != 569 t->inp_socket->so_cred->cr_uid)) { 570 inp->inp_laddr.s_addr = INADDR_ANY; 571 error = EADDRINUSE; 572 goto done; 573 } 574 } 575 if (cred && !prison_replace_wildcards(td, nam)) { 576 inp->inp_laddr.s_addr = INADDR_ANY; 577 error = EADDRNOTAVAIL; 578 goto done; 579 } 580 t = in_pcblookup_local(portinfo, sin->sin_addr, lport, 581 wild, cred); 582 if (t && !(reuseport & t->inp_socket->so_options)) { 583 inp->inp_laddr.s_addr = INADDR_ANY; 584 error = EADDRINUSE; 585 goto done; 586 } 587 inp->inp_lport = lport; 588 in_pcbinsporthash(portinfo, inp); 589 error = 0; 590 done: 591 REL_PORT_TOKEN(portinfo); 592 return (error); 593 } else { 594 jsin.sin_family = AF_INET; 595 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 596 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 597 inp->inp_laddr.s_addr = INADDR_ANY; 598 return (EINVAL); 599 } 600 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 601 602 return in_pcbsetlport(inp, wild, cred); 603 } 604 } 605 606 static struct inpcb * 607 in_pcblookup_localremote(struct inpcbportinfo *portinfo, struct in_addr laddr, 608 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred) 609 { 610 struct inpcb *inp; 611 struct inpcbporthead *porthash; 612 struct inpcbport *phd; 613 struct inpcb *match = NULL; 614 615 /* 616 * If the porthashbase is shared across several cpus, it must 617 * have been locked. 618 */ 619 ASSERT_PORT_TOKEN_HELD(portinfo); 620 621 /* 622 * Best fit PCB lookup. 623 * 624 * First see if this local port is in use by looking on the 625 * port hash list. 626 */ 627 porthash = &portinfo->porthashbase[ 628 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 629 LIST_FOREACH(phd, porthash, phd_hash) { 630 if (phd->phd_port == lport) 631 break; 632 } 633 if (phd != NULL) { 634 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 635 #ifdef INET6 636 if (!INP_ISIPV4(inp)) 637 continue; 638 #endif 639 if (inp->inp_laddr.s_addr != INADDR_ANY && 640 inp->inp_laddr.s_addr != laddr.s_addr) 641 continue; 642 643 if (inp->inp_faddr.s_addr != INADDR_ANY && 644 inp->inp_faddr.s_addr != faddr.s_addr) 645 continue; 646 647 if (inp->inp_fport != 0 && inp->inp_fport != fport) 648 continue; 649 650 if (cred == NULL || 651 cred->cr_prison == 652 inp->inp_socket->so_cred->cr_prison) { 653 match = inp; 654 break; 655 } 656 } 657 } 658 return (match); 659 } 660 661 int 662 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote, 663 struct thread *td) 664 { 665 struct proc *p = td->td_proc; 666 unsigned short *lastport; 667 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote; 668 struct sockaddr_in jsin; 669 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 670 struct inpcbportinfo *portinfo; 671 struct ucred *cred = NULL; 672 u_short first, last, lport, step; 673 int count, error, dup; 674 int portinfo_first, portinfo_idx; 675 676 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 677 return (EADDRNOTAVAIL); 678 679 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY); 680 if (inp->inp_lport != 0) 681 return (EINVAL); /* already bound */ 682 683 KKASSERT(p); 684 cred = p->p_ucred; 685 686 jsin.sin_family = AF_INET; 687 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 688 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 689 inp->inp_laddr.s_addr = INADDR_ANY; 690 return (EINVAL); 691 } 692 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 693 694 inp->inp_flags |= INP_ANONPORT; 695 696 step = pcbinfo->portinfo_mask + 1; 697 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 698 portinfo_idx = portinfo_first; 699 loop: 700 portinfo = &pcbinfo->portinfo[portinfo_idx]; 701 dup = 0; 702 703 if (inp->inp_flags & INP_HIGHPORT) { 704 first = ipport_hifirstauto; /* sysctl */ 705 last = ipport_hilastauto; 706 lastport = &portinfo->lasthi; 707 } else if (inp->inp_flags & INP_LOWPORT) { 708 if (cred && 709 (error = 710 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 711 inp->inp_laddr.s_addr = INADDR_ANY; 712 return (error); 713 } 714 first = ipport_lowfirstauto; /* 1023 */ 715 last = ipport_lowlastauto; /* 600 */ 716 lastport = &portinfo->lastlow; 717 } else { 718 first = ipport_firstauto; /* sysctl */ 719 last = ipport_lastauto; 720 lastport = &portinfo->lastport; 721 } 722 723 /* 724 * This has to be atomic. If the porthash is shared across multiple 725 * protocol threads (aka tcp) then the token must be held. 726 */ 727 GET_PORT_TOKEN(portinfo); 728 729 again: 730 /* 731 * Simple check to ensure all ports are not used up causing 732 * a deadlock here. 733 * 734 * We split the two cases (up and down) so that the direction 735 * is not being tested on each round of the loop. 736 */ 737 if (first > last) { 738 /* 739 * counting down 740 */ 741 in_pcbportrange(&first, &last, portinfo->offset, step); 742 count = (first - last) / step; 743 744 do { 745 if (count-- < 0) { /* completely used? */ 746 error = EADDRNOTAVAIL; 747 goto done; 748 } 749 *lastport -= step; 750 if (*lastport > first || *lastport < last) 751 *lastport = first; 752 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 753 portinfo->offset); 754 lport = htons(*lastport); 755 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr, 756 lport, sin->sin_addr, sin->sin_port, cred)); 757 } else { 758 /* 759 * counting up 760 */ 761 in_pcbportrange(&last, &first, portinfo->offset, step); 762 count = (last - first) / step; 763 764 do { 765 if (count-- < 0) { /* completely used? */ 766 error = EADDRNOTAVAIL; 767 goto done; 768 } 769 *lastport += step; 770 if (*lastport < first || *lastport > last) 771 *lastport = first; 772 KKASSERT((*lastport & pcbinfo->portinfo_mask) == 773 portinfo->offset); 774 lport = htons(*lastport); 775 } while (in_pcblookup_localremote(portinfo, inp->inp_laddr, 776 lport, sin->sin_addr, sin->sin_port, cred)); 777 } 778 779 /* This could happen on loopback interface */ 780 if (sin->sin_port == lport && 781 sin->sin_addr.s_addr == inp->inp_laddr.s_addr) { 782 if (dup) { 783 /* 784 * Duplicate again; give up 785 */ 786 error = EADDRNOTAVAIL; 787 goto done; 788 } 789 dup = 1; 790 goto again; 791 } 792 inp->inp_lport = lport; 793 in_pcbinsporthash(portinfo, inp); 794 error = 0; 795 done: 796 REL_PORT_TOKEN(portinfo); 797 798 if (error) { 799 /* Try next portinfo */ 800 portinfo_idx++; 801 portinfo_idx &= pcbinfo->portinfo_mask; 802 if (portinfo_idx != portinfo_first) 803 goto loop; 804 inp->inp_laddr.s_addr = INADDR_ANY; 805 } 806 return error; 807 } 808 809 /* 810 * Transform old in_pcbconnect() into an inner subroutine for new 811 * in_pcbconnect(): Do some validity-checking on the remote 812 * address (in mbuf 'nam') and then determine local host address 813 * (i.e., which interface) to use to access that remote host. 814 * 815 * This preserves definition of in_pcbconnect(), while supporting a 816 * slightly different version for T/TCP. (This is more than 817 * a bit of a kludge, but cleaning up the internal interfaces would 818 * have forced minor changes in every protocol). 819 */ 820 int 821 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam, 822 struct sockaddr_in **plocal_sin, struct thread *td, int find) 823 { 824 struct in_ifaddr *ia; 825 struct ucred *cred = NULL; 826 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 827 struct sockaddr *jsin; 828 int jailed = 0, alloc_route = 0; 829 830 if (nam->sa_len != sizeof *sin) 831 return (EINVAL); 832 if (sin->sin_family != AF_INET) 833 return (EAFNOSUPPORT); 834 if (sin->sin_port == 0) 835 return (EADDRNOTAVAIL); 836 if (td && td->td_proc && td->td_proc->p_ucred) 837 cred = td->td_proc->p_ucred; 838 if (cred && cred->cr_prison) 839 jailed = 1; 840 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) { 841 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 842 /* 843 * If the destination address is INADDR_ANY, 844 * use the primary local address. 845 * If the supplied address is INADDR_BROADCAST, 846 * and the primary interface supports broadcast, 847 * choose the broadcast address for that interface. 848 */ 849 if (sin->sin_addr.s_addr == INADDR_ANY) 850 sin->sin_addr = IA_SIN(ia)->sin_addr; 851 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST && 852 (ia->ia_ifp->if_flags & IFF_BROADCAST)) 853 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr; 854 } 855 if (find) { 856 struct route *ro; 857 858 ia = NULL; 859 /* 860 * If route is known or can be allocated now, 861 * our src addr is taken from the i/f, else punt. 862 * Note that we should check the address family of the cached 863 * destination, in case of sharing the cache with IPv6. 864 */ 865 ro = &inp->inp_route; 866 if (ro->ro_rt && 867 (!(ro->ro_rt->rt_flags & RTF_UP) || 868 ro->ro_dst.sa_family != AF_INET || 869 satosin(&ro->ro_dst)->sin_addr.s_addr != 870 sin->sin_addr.s_addr || 871 inp->inp_socket->so_options & SO_DONTROUTE)) { 872 RTFREE(ro->ro_rt); 873 ro->ro_rt = NULL; 874 } 875 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 876 (ro->ro_rt == NULL || 877 ro->ro_rt->rt_ifp == NULL)) { 878 /* No route yet, so try to acquire one */ 879 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 880 ro->ro_dst.sa_family = AF_INET; 881 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 882 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 883 sin->sin_addr; 884 rtalloc(ro); 885 alloc_route = 1; 886 } 887 /* 888 * If we found a route, use the address 889 * corresponding to the outgoing interface 890 * unless it is the loopback (in case a route 891 * to our address on another net goes to loopback). 892 */ 893 if (ro->ro_rt && 894 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) { 895 if (jailed) { 896 if (jailed_ip(cred->cr_prison, 897 ro->ro_rt->rt_ifa->ifa_addr)) { 898 ia = ifatoia(ro->ro_rt->rt_ifa); 899 } 900 } else { 901 ia = ifatoia(ro->ro_rt->rt_ifa); 902 } 903 } 904 if (ia == NULL) { 905 u_short fport = sin->sin_port; 906 907 sin->sin_port = 0; 908 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 909 if (ia && jailed && !jailed_ip(cred->cr_prison, 910 sintosa(&ia->ia_addr))) 911 ia = NULL; 912 if (ia == NULL) 913 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 914 if (ia && jailed && !jailed_ip(cred->cr_prison, 915 sintosa(&ia->ia_addr))) 916 ia = NULL; 917 sin->sin_port = fport; 918 if (ia == NULL && 919 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 920 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 921 if (ia && jailed && !jailed_ip(cred->cr_prison, 922 sintosa(&ia->ia_addr))) 923 ia = NULL; 924 925 if (!jailed && ia == NULL) 926 goto fail; 927 } 928 /* 929 * If the destination address is multicast and an outgoing 930 * interface has been set as a multicast option, use the 931 * address of that interface as our source address. 932 */ 933 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 934 inp->inp_moptions != NULL) { 935 struct ip_moptions *imo; 936 struct ifnet *ifp; 937 938 imo = inp->inp_moptions; 939 if (imo->imo_multicast_ifp != NULL) { 940 struct in_ifaddr_container *iac; 941 942 ifp = imo->imo_multicast_ifp; 943 ia = NULL; 944 TAILQ_FOREACH(iac, 945 &in_ifaddrheads[mycpuid], ia_link) { 946 if (iac->ia->ia_ifp == ifp) { 947 ia = iac->ia; 948 break; 949 } 950 } 951 if (ia == NULL) 952 goto fail; 953 } 954 } 955 /* 956 * Don't do pcblookup call here; return interface in plocal_sin 957 * and exit to caller, that will do the lookup. 958 */ 959 if (ia == NULL && jailed) { 960 if ((jsin = prison_get_nonlocal( 961 cred->cr_prison, AF_INET, NULL)) != NULL || 962 (jsin = prison_get_local( 963 cred->cr_prison, AF_INET, NULL)) != NULL) { 964 *plocal_sin = satosin(jsin); 965 } else { 966 /* IPv6 only Jail */ 967 goto fail; 968 } 969 } else { 970 *plocal_sin = &ia->ia_addr; 971 } 972 } 973 return (0); 974 fail: 975 if (alloc_route) 976 in_pcbresetroute(inp); 977 return (EADDRNOTAVAIL); 978 } 979 980 int 981 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam, 982 struct sockaddr_in **plocal_sin, struct thread *td) 983 { 984 return in_pcbladdr_find(inp, nam, plocal_sin, td, 985 (inp->inp_laddr.s_addr == INADDR_ANY)); 986 } 987 988 /* 989 * Outer subroutine: 990 * Connect from a socket to a specified address. 991 * Both address and port must be specified in argument sin. 992 * If don't have a local address for this socket yet, 993 * then pick one. 994 */ 995 int 996 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 997 { 998 struct sockaddr_in *if_sin; 999 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1000 int error; 1001 1002 /* Call inner routine to assign local interface address. */ 1003 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0) 1004 return (error); 1005 1006 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port, 1007 inp->inp_laddr.s_addr ? 1008 inp->inp_laddr : if_sin->sin_addr, 1009 inp->inp_lport, FALSE, NULL) != NULL) { 1010 return (EADDRINUSE); 1011 } 1012 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1013 if (inp->inp_lport == 0) { 1014 error = in_pcbbind(inp, NULL, td); 1015 if (error) 1016 return (error); 1017 } 1018 inp->inp_laddr = if_sin->sin_addr; 1019 } 1020 inp->inp_faddr = sin->sin_addr; 1021 inp->inp_fport = sin->sin_port; 1022 in_pcbinsconnhash(inp); 1023 return (0); 1024 } 1025 1026 void 1027 in_pcbdisconnect(struct inpcb *inp) 1028 { 1029 1030 in_pcbremconnhash(inp); 1031 inp->inp_faddr.s_addr = INADDR_ANY; 1032 inp->inp_fport = 0; 1033 } 1034 1035 void 1036 in_pcbdetach(struct inpcb *inp) 1037 { 1038 struct socket *so = inp->inp_socket; 1039 struct inpcbinfo *ipi = inp->inp_pcbinfo; 1040 1041 #ifdef IPSEC 1042 ipsec4_delete_pcbpolicy(inp); 1043 #endif /*IPSEC*/ 1044 inp->inp_gencnt = ++ipi->ipi_gencnt; 1045 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 1046 in_pcbremlists(inp); 1047 so->so_pcb = NULL; 1048 sofree(so); /* remove pcb ref */ 1049 if (inp->inp_options) 1050 m_free(inp->inp_options); 1051 if (inp->inp_route.ro_rt) 1052 rtfree(inp->inp_route.ro_rt); 1053 ip_freemoptions(inp->inp_moptions); 1054 kfree(inp, M_PCB); 1055 } 1056 1057 /* 1058 * The calling convention of in_setsockaddr() and in_setpeeraddr() was 1059 * modified to match the pru_sockaddr() and pru_peeraddr() entry points 1060 * in struct pr_usrreqs, so that protocols can just reference then directly 1061 * without the need for a wrapper function. The socket must have a valid 1062 * (i.e., non-nil) PCB, but it should be impossible to get an invalid one 1063 * except through a kernel programming error, so it is acceptable to panic 1064 * (or in this case trap) if the PCB is invalid. (Actually, we don't trap 1065 * because there actually /is/ a programming error somewhere... XXX) 1066 */ 1067 int 1068 in_setsockaddr(struct socket *so, struct sockaddr **nam) 1069 { 1070 struct inpcb *inp; 1071 struct sockaddr_in *sin; 1072 1073 /* 1074 * Do the malloc first in case it blocks. 1075 */ 1076 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1077 sin->sin_family = AF_INET; 1078 sin->sin_len = sizeof *sin; 1079 1080 crit_enter(); 1081 inp = so->so_pcb; 1082 if (!inp) { 1083 crit_exit(); 1084 kfree(sin, M_SONAME); 1085 return (ECONNRESET); 1086 } 1087 sin->sin_port = inp->inp_lport; 1088 sin->sin_addr = inp->inp_laddr; 1089 crit_exit(); 1090 1091 *nam = (struct sockaddr *)sin; 1092 return (0); 1093 } 1094 1095 void 1096 in_setsockaddr_dispatch(netmsg_t msg) 1097 { 1098 int error; 1099 1100 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1101 lwkt_replymsg(&msg->lmsg, error); 1102 } 1103 1104 int 1105 in_setpeeraddr(struct socket *so, struct sockaddr **nam) 1106 { 1107 struct inpcb *inp; 1108 struct sockaddr_in *sin; 1109 1110 /* 1111 * Do the malloc first in case it blocks. 1112 */ 1113 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1114 sin->sin_family = AF_INET; 1115 sin->sin_len = sizeof *sin; 1116 1117 crit_enter(); 1118 inp = so->so_pcb; 1119 if (!inp) { 1120 crit_exit(); 1121 kfree(sin, M_SONAME); 1122 return (ECONNRESET); 1123 } 1124 sin->sin_port = inp->inp_fport; 1125 sin->sin_addr = inp->inp_faddr; 1126 crit_exit(); 1127 1128 *nam = (struct sockaddr *)sin; 1129 return (0); 1130 } 1131 1132 void 1133 in_setpeeraddr_dispatch(netmsg_t msg) 1134 { 1135 int error; 1136 1137 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1138 lwkt_replymsg(&msg->lmsg, error); 1139 } 1140 1141 void 1142 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err, 1143 inp_notify_t notify) 1144 { 1145 struct inpcb *inp, *marker; 1146 1147 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1148 ("not in the correct netisr")); 1149 marker = &in_pcbmarkers[mycpuid]; 1150 1151 /* 1152 * NOTE: 1153 * - If INP_PLACEMARKER is set we must ignore the rest of the 1154 * structure and skip it. 1155 * - It is safe to nuke inpcbs here, since we are in their own 1156 * netisr. 1157 */ 1158 GET_PCBINFO_TOKEN(pcbinfo); 1159 1160 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1161 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) { 1162 LIST_REMOVE(marker, inp_list); 1163 LIST_INSERT_AFTER(inp, marker, inp_list); 1164 1165 if (inp->inp_flags & INP_PLACEMARKER) 1166 continue; 1167 #ifdef INET6 1168 if (!INP_ISIPV4(inp)) 1169 continue; 1170 #endif 1171 if (inp->inp_faddr.s_addr != faddr.s_addr || 1172 inp->inp_socket == NULL) 1173 continue; 1174 (*notify)(inp, err); /* can remove inp from list! */ 1175 } 1176 LIST_REMOVE(marker, inp_list); 1177 1178 REL_PCBINFO_TOKEN(pcbinfo); 1179 } 1180 1181 void 1182 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1183 { 1184 struct inpcb *inp, *marker; 1185 1186 /* 1187 * We only need to make sure that we are in netisr0, where all 1188 * multicast operation happen. We could check inpcbinfo which 1189 * does not belong to netisr0 by holding the inpcbinfo's token. 1190 * In this case, the pcbinfo must be able to be shared, i.e. 1191 * pcbinfo->infotoken is not NULL. 1192 */ 1193 ASSERT_IN_NETISR(0); 1194 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL, 1195 ("pcbinfo could not be shared")); 1196 1197 /* 1198 * Get a marker for the current netisr (netisr0). 1199 * 1200 * It is possible that the multicast address deletion blocks, 1201 * which could cause temporary token releasing. So we use 1202 * inpcb marker here to get a coherent view of the inpcb list. 1203 * 1204 * While, on the other hand, moptions are only added and deleted 1205 * in netisr0, so we would not see staled moption or miss moption 1206 * even if the token was released due to the blocking multicast 1207 * address deletion. 1208 */ 1209 marker = &in_pcbmarkers[mycpuid]; 1210 1211 GET_PCBINFO_TOKEN(pcbinfo); 1212 1213 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1214 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) { 1215 struct ip_moptions *imo; 1216 1217 LIST_REMOVE(marker, inp_list); 1218 LIST_INSERT_AFTER(inp, marker, inp_list); 1219 1220 if (inp->inp_flags & INP_PLACEMARKER) 1221 continue; 1222 imo = inp->inp_moptions; 1223 if (INP_ISIPV4(inp) && imo != NULL) { 1224 int i, gap; 1225 1226 /* 1227 * Unselect the outgoing interface if it is being 1228 * detached. 1229 */ 1230 if (imo->imo_multicast_ifp == ifp) 1231 imo->imo_multicast_ifp = NULL; 1232 1233 /* 1234 * Drop multicast group membership if we joined 1235 * through the interface being detached. 1236 */ 1237 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1238 i++) { 1239 if (imo->imo_membership[i]->inm_ifp == ifp) { 1240 /* 1241 * NOTE: 1242 * This could block and the pcbinfo 1243 * token could be passively released. 1244 */ 1245 in_delmulti(imo->imo_membership[i]); 1246 gap++; 1247 } else if (gap != 0) 1248 imo->imo_membership[i - gap] = 1249 imo->imo_membership[i]; 1250 } 1251 imo->imo_num_memberships -= gap; 1252 } 1253 } 1254 LIST_REMOVE(marker, inp_list); 1255 1256 REL_PCBINFO_TOKEN(pcbinfo); 1257 } 1258 1259 /* 1260 * Check for alternatives when higher level complains 1261 * about service problems. For now, invalidate cached 1262 * routing information. If the route was created dynamically 1263 * (by a redirect), time to try a default gateway again. 1264 */ 1265 void 1266 in_losing(struct inpcb *inp) 1267 { 1268 struct rtentry *rt; 1269 struct rt_addrinfo rtinfo; 1270 1271 if ((rt = inp->inp_route.ro_rt)) { 1272 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1273 rtinfo.rti_info[RTAX_DST] = rt_key(rt); 1274 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1275 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt); 1276 rtinfo.rti_flags = rt->rt_flags; 1277 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0); 1278 if (rt->rt_flags & RTF_DYNAMIC) { 1279 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1280 rt_mask(rt), rt->rt_flags, NULL); 1281 } 1282 inp->inp_route.ro_rt = NULL; 1283 rtfree(rt); 1284 /* 1285 * A new route can be allocated 1286 * the next time output is attempted. 1287 */ 1288 } 1289 } 1290 1291 /* 1292 * After a routing change, flush old routing 1293 * and allocate a (hopefully) better one. 1294 */ 1295 void 1296 in_rtchange(struct inpcb *inp, int err) 1297 { 1298 if (inp->inp_route.ro_rt) { 1299 rtfree(inp->inp_route.ro_rt); 1300 inp->inp_route.ro_rt = NULL; 1301 /* 1302 * A new route can be allocated the next time 1303 * output is attempted. 1304 */ 1305 } 1306 } 1307 1308 /* 1309 * Lookup a PCB based on the local address and port. 1310 */ 1311 struct inpcb * 1312 in_pcblookup_local(struct inpcbportinfo *portinfo, struct in_addr laddr, 1313 u_int lport_arg, int wild_okay, struct ucred *cred) 1314 { 1315 struct inpcb *inp; 1316 int matchwild = 3, wildcard; 1317 u_short lport = lport_arg; 1318 struct inpcbporthead *porthash; 1319 struct inpcbport *phd; 1320 struct inpcb *match = NULL; 1321 1322 /* 1323 * If the porthashbase is shared across several cpus, it must 1324 * have been locked. 1325 */ 1326 ASSERT_PORT_TOKEN_HELD(portinfo); 1327 1328 /* 1329 * Best fit PCB lookup. 1330 * 1331 * First see if this local port is in use by looking on the 1332 * port hash list. 1333 */ 1334 porthash = &portinfo->porthashbase[ 1335 INP_PCBPORTHASH(lport, portinfo->porthashmask)]; 1336 LIST_FOREACH(phd, porthash, phd_hash) { 1337 if (phd->phd_port == lport) 1338 break; 1339 } 1340 if (phd != NULL) { 1341 /* 1342 * Port is in use by one or more PCBs. Look for best 1343 * fit. 1344 */ 1345 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1346 wildcard = 0; 1347 #ifdef INET6 1348 if (!INP_ISIPV4(inp)) 1349 continue; 1350 #endif 1351 if (inp->inp_faddr.s_addr != INADDR_ANY) 1352 wildcard++; 1353 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1354 if (laddr.s_addr == INADDR_ANY) 1355 wildcard++; 1356 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1357 continue; 1358 } else { 1359 if (laddr.s_addr != INADDR_ANY) 1360 wildcard++; 1361 } 1362 if (wildcard && !wild_okay) 1363 continue; 1364 if (wildcard < matchwild && 1365 (cred == NULL || 1366 cred->cr_prison == 1367 inp->inp_socket->so_cred->cr_prison)) { 1368 match = inp; 1369 matchwild = wildcard; 1370 if (matchwild == 0) { 1371 break; 1372 } 1373 } 1374 } 1375 } 1376 return (match); 1377 } 1378 1379 struct inpcb * 1380 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo, 1381 const struct inpcb *inp) 1382 { 1383 const struct inp_localgrphead *hdr; 1384 const struct inp_localgroup *grp; 1385 int i; 1386 1387 if (pcbinfo->localgrphashbase == NULL) 1388 return NULL; 1389 1390 GET_PCBINFO_TOKEN(pcbinfo); 1391 1392 hdr = &pcbinfo->localgrphashbase[ 1393 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1394 1395 LIST_FOREACH(grp, hdr, il_list) { 1396 if (grp->il_af == inp->inp_af && 1397 grp->il_lport == inp->inp_lport && 1398 memcmp(&grp->il_dependladdr, 1399 &inp->inp_inc.inc_ie.ie_dependladdr, 1400 sizeof(grp->il_dependladdr)) == 0) { 1401 break; 1402 } 1403 } 1404 if (grp == NULL || grp->il_inpcnt == 1) { 1405 REL_PCBINFO_TOKEN(pcbinfo); 1406 return NULL; 1407 } 1408 1409 KASSERT(grp->il_inpcnt >= 2, 1410 ("invalid localgroup inp count %d", grp->il_inpcnt)); 1411 for (i = 0; i < grp->il_inpcnt; ++i) { 1412 if (grp->il_inp[i] == inp) { 1413 int last = grp->il_inpcnt - 1; 1414 1415 if (i == last) 1416 last = grp->il_inpcnt - 2; 1417 REL_PCBINFO_TOKEN(pcbinfo); 1418 return grp->il_inp[last]; 1419 } 1420 } 1421 REL_PCBINFO_TOKEN(pcbinfo); 1422 return NULL; 1423 } 1424 1425 static struct inpcb * 1426 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo, 1427 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash) 1428 { 1429 struct inpcb *local_wild = NULL; 1430 const struct inp_localgrphead *hdr; 1431 const struct inp_localgroup *grp; 1432 1433 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 1434 1435 hdr = &pcbinfo->localgrphashbase[ 1436 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)]; 1437 1438 /* 1439 * Order of socket selection: 1440 * 1. non-wild. 1441 * 2. wild. 1442 * 1443 * NOTE: Local group does not contain jailed sockets 1444 */ 1445 LIST_FOREACH(grp, hdr, il_list) { 1446 #ifdef INET6 1447 if (grp->il_af != AF_INET) 1448 continue; 1449 #endif 1450 if (grp->il_lport == lport) { 1451 int idx; 1452 1453 /* 1454 * Modulo-N is used here, which greatly reduces 1455 * completion queue token contention, thus more 1456 * cpu time is saved. 1457 */ 1458 idx = pkt_hash % grp->il_inpcnt; 1459 if (grp->il_laddr.s_addr == laddr.s_addr) 1460 return grp->il_inp[idx]; 1461 else if (grp->il_laddr.s_addr == INADDR_ANY) 1462 local_wild = grp->il_inp[idx]; 1463 } 1464 } 1465 if (local_wild != NULL) 1466 return local_wild; 1467 return NULL; 1468 } 1469 1470 /* 1471 * Lookup PCB in hash list. 1472 */ 1473 struct inpcb * 1474 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1475 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1476 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m) 1477 { 1478 struct inpcbhead *head; 1479 struct inpcb *inp, *jinp=NULL; 1480 u_short fport = fport_arg, lport = lport_arg; 1481 1482 /* 1483 * First look for an exact match. 1484 */ 1485 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport, 1486 laddr.s_addr, lport, pcbinfo->hashmask)]; 1487 LIST_FOREACH(inp, head, inp_hash) { 1488 #ifdef INET6 1489 if (!INP_ISIPV4(inp)) 1490 continue; 1491 #endif 1492 if (in_hosteq(inp->inp_faddr, faddr) && 1493 in_hosteq(inp->inp_laddr, laddr) && 1494 inp->inp_fport == fport && inp->inp_lport == lport) { 1495 /* found */ 1496 if (inp->inp_socket == NULL || 1497 inp->inp_socket->so_cred->cr_prison == NULL) { 1498 return (inp); 1499 } else { 1500 if (jinp == NULL) 1501 jinp = inp; 1502 } 1503 } 1504 } 1505 if (jinp != NULL) 1506 return (jinp); 1507 1508 if (wildcard) { 1509 struct inpcb *local_wild = NULL; 1510 struct inpcb *jinp_wild = NULL; 1511 struct inpcontainer *ic; 1512 struct inpcontainerhead *chead; 1513 struct sockaddr_in jsin; 1514 struct ucred *cred; 1515 1516 GET_PCBINFO_TOKEN(pcbinfo); 1517 1518 /* 1519 * Check local group first 1520 */ 1521 if (pcbinfo->localgrphashbase != NULL && 1522 m != NULL && (m->m_flags & M_HASH) && 1523 !(ifp && ifp->if_type == IFT_FAITH)) { 1524 inp = inp_localgroup_lookup(pcbinfo, 1525 laddr, lport, m->m_pkthdr.hash); 1526 if (inp != NULL) { 1527 REL_PCBINFO_TOKEN(pcbinfo); 1528 return inp; 1529 } 1530 } 1531 1532 /* 1533 * Order of socket selection: 1534 * 1. non-jailed, non-wild. 1535 * 2. non-jailed, wild. 1536 * 3. jailed, non-wild. 1537 * 4. jailed, wild. 1538 */ 1539 jsin.sin_family = AF_INET; 1540 chead = &pcbinfo->wildcardhashbase[ 1541 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)]; 1542 LIST_FOREACH(ic, chead, ic_list) { 1543 inp = ic->ic_inp; 1544 if (inp->inp_flags & INP_PLACEMARKER) 1545 continue; 1546 1547 jsin.sin_addr.s_addr = laddr.s_addr; 1548 #ifdef INET6 1549 if (!INP_ISIPV4(inp)) 1550 continue; 1551 #endif 1552 if (inp->inp_socket != NULL) 1553 cred = inp->inp_socket->so_cred; 1554 else 1555 cred = NULL; 1556 if (cred != NULL && jailed(cred)) { 1557 if (jinp != NULL) 1558 continue; 1559 else 1560 if (!jailed_ip(cred->cr_prison, 1561 (struct sockaddr *)&jsin)) 1562 continue; 1563 } 1564 if (inp->inp_lport == lport) { 1565 if (ifp && ifp->if_type == IFT_FAITH && 1566 !(inp->inp_flags & INP_FAITH)) 1567 continue; 1568 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1569 if (cred != NULL && jailed(cred)) { 1570 jinp = inp; 1571 } else { 1572 REL_PCBINFO_TOKEN(pcbinfo); 1573 return (inp); 1574 } 1575 } 1576 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1577 if (cred != NULL && jailed(cred)) 1578 jinp_wild = inp; 1579 else 1580 local_wild = inp; 1581 } 1582 } 1583 } 1584 1585 REL_PCBINFO_TOKEN(pcbinfo); 1586 1587 if (local_wild != NULL) 1588 return (local_wild); 1589 if (jinp != NULL) 1590 return (jinp); 1591 return (jinp_wild); 1592 } 1593 1594 /* 1595 * Not found. 1596 */ 1597 return (NULL); 1598 } 1599 1600 struct inpcb * 1601 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1602 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1603 boolean_t wildcard, struct ifnet *ifp) 1604 { 1605 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg, 1606 laddr, lport_arg, wildcard, ifp, NULL); 1607 } 1608 1609 /* 1610 * Insert PCB into connection hash table. 1611 */ 1612 void 1613 in_pcbinsconnhash(struct inpcb *inp) 1614 { 1615 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1616 struct inpcbhead *bucket; 1617 u_int32_t hashkey_faddr, hashkey_laddr; 1618 1619 #ifdef INET6 1620 if (INP_ISIPV6(inp)) { 1621 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */; 1622 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */; 1623 } else { 1624 #endif 1625 hashkey_faddr = inp->inp_faddr.s_addr; 1626 hashkey_laddr = inp->inp_laddr.s_addr; 1627 #ifdef INET6 1628 } 1629 #endif 1630 1631 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1632 ("not in the correct netisr")); 1633 ASSERT_INP_NOTINHASH(inp); 1634 inp->inp_flags |= INP_CONNECTED; 1635 1636 /* 1637 * Insert into the connection hash table. 1638 */ 1639 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr, 1640 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)]; 1641 LIST_INSERT_HEAD(bucket, inp, inp_hash); 1642 } 1643 1644 /* 1645 * Remove PCB from connection hash table. 1646 */ 1647 void 1648 in_pcbremconnhash(struct inpcb *inp) 1649 { 1650 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo; 1651 1652 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1653 ("not in the correct netisr")); 1654 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected")); 1655 1656 LIST_REMOVE(inp, inp_hash); 1657 inp->inp_flags &= ~INP_CONNECTED; 1658 } 1659 1660 /* 1661 * Insert PCB into port hash table. 1662 */ 1663 void 1664 in_pcbinsporthash(struct inpcbportinfo *portinfo, struct inpcb *inp) 1665 { 1666 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1667 struct inpcbporthead *pcbporthash; 1668 struct inpcbport *phd; 1669 1670 /* 1671 * If the porthashbase is shared across several cpus, it must 1672 * have been locked. 1673 */ 1674 ASSERT_PORT_TOKEN_HELD(portinfo); 1675 1676 /* 1677 * Insert into the port hash table. 1678 */ 1679 pcbporthash = &portinfo->porthashbase[ 1680 INP_PCBPORTHASH(inp->inp_lport, portinfo->porthashmask)]; 1681 1682 /* Go through port list and look for a head for this lport. */ 1683 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1684 if (phd->phd_port == inp->inp_lport) 1685 break; 1686 } 1687 1688 /* If none exists, use saved one and tack it on. */ 1689 if (phd == NULL) { 1690 KKASSERT(pcbinfo->portsave != NULL); 1691 phd = pcbinfo->portsave; 1692 pcbinfo->portsave = NULL; 1693 phd->phd_port = inp->inp_lport; 1694 LIST_INIT(&phd->phd_pcblist); 1695 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1696 } 1697 1698 inp->inp_portinfo = portinfo; 1699 inp->inp_phd = phd; 1700 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1701 1702 /* 1703 * Malloc one inpcbport for later use. It is safe to use 1704 * "wait" malloc here (port token would be released, if 1705 * malloc ever blocked), since all changes to the porthash 1706 * are done. 1707 */ 1708 if (pcbinfo->portsave == NULL) { 1709 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), 1710 M_PCB, M_INTWAIT | M_ZERO); 1711 } 1712 } 1713 1714 void 1715 in_pcbinsporthash_lport(struct inpcb *inp) 1716 { 1717 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1718 struct inpcbportinfo *portinfo; 1719 u_short lport_ho; 1720 1721 /* Locate the proper portinfo based on lport */ 1722 lport_ho = ntohs(inp->inp_lport); 1723 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 1724 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset); 1725 1726 GET_PORT_TOKEN(portinfo); 1727 in_pcbinsporthash(portinfo, inp); 1728 REL_PORT_TOKEN(portinfo); 1729 } 1730 1731 static struct inp_localgroup * 1732 inp_localgroup_alloc(u_char af, uint16_t port, 1733 const union in_dependaddr *addr, int size) 1734 { 1735 struct inp_localgroup *grp; 1736 1737 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]), 1738 M_TEMP, M_INTWAIT | M_ZERO); 1739 grp->il_af = af; 1740 grp->il_lport = port; 1741 grp->il_dependladdr = *addr; 1742 grp->il_inpsiz = size; 1743 1744 return grp; 1745 } 1746 1747 static void 1748 inp_localgroup_free(struct inp_localgroup *grp) 1749 { 1750 kfree(grp, M_TEMP); 1751 } 1752 1753 static void 1754 inp_localgroup_destroy(struct inp_localgroup *grp) 1755 { 1756 LIST_REMOVE(grp, il_list); 1757 inp_localgroup_free(grp); 1758 } 1759 1760 static void 1761 inp_localgroup_copy(struct inp_localgroup *grp, 1762 const struct inp_localgroup *old_grp) 1763 { 1764 int i; 1765 1766 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 1767 ("invalid new local group size %d and old local group count %d", 1768 grp->il_inpsiz, old_grp->il_inpcnt)); 1769 for (i = 0; i < old_grp->il_inpcnt; ++i) 1770 grp->il_inp[i] = old_grp->il_inp[i]; 1771 grp->il_inpcnt = old_grp->il_inpcnt; 1772 } 1773 1774 static void 1775 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1776 { 1777 struct inp_localgrphead *hdr; 1778 struct inp_localgroup *grp, *grp_alloc = NULL; 1779 struct ucred *cred; 1780 int i, idx; 1781 1782 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 1783 1784 if (pcbinfo->localgrphashbase == NULL) 1785 return; 1786 1787 /* 1788 * XXX don't allow jailed socket to join local group 1789 */ 1790 if (inp->inp_socket != NULL) 1791 cred = inp->inp_socket->so_cred; 1792 else 1793 cred = NULL; 1794 if (cred != NULL && jailed(cred)) 1795 return; 1796 1797 hdr = &pcbinfo->localgrphashbase[ 1798 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1799 1800 again: 1801 LIST_FOREACH(grp, hdr, il_list) { 1802 if (grp->il_af == inp->inp_af && 1803 grp->il_lport == inp->inp_lport && 1804 memcmp(&grp->il_dependladdr, 1805 &inp->inp_inc.inc_ie.ie_dependladdr, 1806 sizeof(grp->il_dependladdr)) == 0) { 1807 break; 1808 } 1809 } 1810 if (grp == NULL) { 1811 /* 1812 * Create a new local group 1813 */ 1814 if (grp_alloc == NULL) { 1815 grp_alloc = inp_localgroup_alloc(inp->inp_af, 1816 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 1817 INP_LOCALGROUP_SIZMIN); 1818 /* 1819 * Local group allocation could block and the 1820 * local group w/ the same property might have 1821 * been added by others when we were blocked; 1822 * check again. 1823 */ 1824 goto again; 1825 } else { 1826 /* Local group has been allocated; link it */ 1827 grp = grp_alloc; 1828 grp_alloc = NULL; 1829 LIST_INSERT_HEAD(hdr, grp, il_list); 1830 } 1831 } else if (grp->il_inpcnt == grp->il_inpsiz) { 1832 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) { 1833 static int limit_logged = 0; 1834 1835 if (!limit_logged) { 1836 limit_logged = 1; 1837 kprintf("local group port %d, " 1838 "limit reached\n", ntohs(grp->il_lport)); 1839 } 1840 if (grp_alloc != NULL) { 1841 /* 1842 * This would happen if the local group 1843 * w/ the same property was expanded when 1844 * our local group allocation blocked. 1845 */ 1846 inp_localgroup_free(grp_alloc); 1847 } 1848 return; 1849 } 1850 1851 /* 1852 * Expand this local group 1853 */ 1854 if (grp_alloc == NULL || 1855 grp->il_inpcnt >= grp_alloc->il_inpsiz) { 1856 if (grp_alloc != NULL) 1857 inp_localgroup_free(grp_alloc); 1858 grp_alloc = inp_localgroup_alloc(grp->il_af, 1859 grp->il_lport, &grp->il_dependladdr, 1860 grp->il_inpsiz * 2); 1861 /* 1862 * Local group allocation could block and the 1863 * local group w/ the same property might have 1864 * been expanded by others when we were blocked; 1865 * check again. 1866 */ 1867 goto again; 1868 } 1869 1870 /* 1871 * Save the old local group, link the new one, and then 1872 * destroy the old local group 1873 */ 1874 inp_localgroup_copy(grp_alloc, grp); 1875 LIST_INSERT_HEAD(hdr, grp_alloc, il_list); 1876 inp_localgroup_destroy(grp); 1877 1878 grp = grp_alloc; 1879 grp_alloc = NULL; 1880 } else { 1881 /* 1882 * Found the local group 1883 */ 1884 if (grp_alloc != NULL) { 1885 /* 1886 * This would happen if the local group w/ the 1887 * same property was added or expanded when our 1888 * local group allocation blocked. 1889 */ 1890 inp_localgroup_free(grp_alloc); 1891 grp_alloc = NULL; 1892 } 1893 } 1894 1895 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 1896 ("invalid local group size %d and count %d", 1897 grp->il_inpsiz, grp->il_inpcnt)); 1898 1899 /* 1900 * Keep the local group sorted by the inpcb local group index 1901 * in ascending order. 1902 * 1903 * This eases the multi-process userland application which uses 1904 * SO_REUSEPORT sockets and binds process to the owner cpu of 1905 * the SO_REUSEPORT socket: 1906 * If we didn't sort the local group by the inpcb local group 1907 * index and one of the process owning an inpcb in this local 1908 * group restarted, e.g. crashed and restarted by watchdog, 1909 * other processes owning a inpcb in this local group would have 1910 * to detect that event, refetch its socket's owner cpu, and 1911 * re-bind. 1912 */ 1913 idx = grp->il_inpcnt; 1914 for (i = 0; i < idx; ++i) { 1915 struct inpcb *oinp = grp->il_inp[i]; 1916 1917 if (oinp->inp_lgrpindex > i) { 1918 if (inp->inp_lgrpindex < 0) { 1919 inp->inp_lgrpindex = i; 1920 } else if (inp->inp_lgrpindex != i) { 1921 if (bootverbose) { 1922 kprintf("inp %p: grpidx %d, " 1923 "assigned to %d, cpu%d\n", 1924 inp, inp->inp_lgrpindex, i, 1925 mycpuid); 1926 } 1927 } 1928 grp->il_inp[i] = inp; 1929 1930 /* Pull down inpcbs */ 1931 for (; i < grp->il_inpcnt; ++i) { 1932 struct inpcb *oinp1 = grp->il_inp[i + 1]; 1933 1934 grp->il_inp[i + 1] = oinp; 1935 oinp = oinp1; 1936 } 1937 grp->il_inpcnt++; 1938 return; 1939 } 1940 } 1941 1942 if (inp->inp_lgrpindex < 0) { 1943 inp->inp_lgrpindex = idx; 1944 } else if (inp->inp_lgrpindex != idx) { 1945 if (bootverbose) { 1946 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n", 1947 inp, inp->inp_lgrpindex, idx, mycpuid); 1948 } 1949 } 1950 grp->il_inp[idx] = inp; 1951 grp->il_inpcnt++; 1952 } 1953 1954 void 1955 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1956 { 1957 struct inpcontainer *ic; 1958 struct inpcontainerhead *bucket; 1959 1960 GET_PCBINFO_TOKEN(pcbinfo); 1961 1962 in_pcbinslocalgrphash_oncpu(inp, pcbinfo); 1963 1964 bucket = &pcbinfo->wildcardhashbase[ 1965 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 1966 1967 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT); 1968 ic->ic_inp = inp; 1969 LIST_INSERT_HEAD(bucket, ic, ic_list); 1970 1971 REL_PCBINFO_TOKEN(pcbinfo); 1972 } 1973 1974 /* 1975 * Insert PCB into wildcard hash table. 1976 */ 1977 void 1978 in_pcbinswildcardhash(struct inpcb *inp) 1979 { 1980 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1981 1982 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1983 ("not in correct netisr")); 1984 ASSERT_INP_NOTINHASH(inp); 1985 inp->inp_flags |= INP_WILDCARD; 1986 1987 in_pcbinswildcardhash_oncpu(inp, pcbinfo); 1988 } 1989 1990 static void 1991 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1992 { 1993 struct inp_localgrphead *hdr; 1994 struct inp_localgroup *grp; 1995 1996 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 1997 1998 if (pcbinfo->localgrphashbase == NULL) 1999 return; 2000 2001 hdr = &pcbinfo->localgrphashbase[ 2002 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 2003 2004 LIST_FOREACH(grp, hdr, il_list) { 2005 int i; 2006 2007 for (i = 0; i < grp->il_inpcnt; ++i) { 2008 if (grp->il_inp[i] != inp) 2009 continue; 2010 2011 if (grp->il_inpcnt == 1) { 2012 /* Destroy this local group */ 2013 inp_localgroup_destroy(grp); 2014 } else { 2015 /* Pull up inpcbs */ 2016 for (; i + 1 < grp->il_inpcnt; ++i) 2017 grp->il_inp[i] = grp->il_inp[i + 1]; 2018 grp->il_inpcnt--; 2019 } 2020 return; 2021 } 2022 } 2023 } 2024 2025 void 2026 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2027 { 2028 struct inpcontainer *ic; 2029 struct inpcontainerhead *head; 2030 2031 GET_PCBINFO_TOKEN(pcbinfo); 2032 2033 in_pcbremlocalgrphash_oncpu(inp, pcbinfo); 2034 2035 /* find bucket */ 2036 head = &pcbinfo->wildcardhashbase[ 2037 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 2038 2039 LIST_FOREACH(ic, head, ic_list) { 2040 if (ic->ic_inp == inp) 2041 goto found; 2042 } 2043 REL_PCBINFO_TOKEN(pcbinfo); 2044 return; /* not found! */ 2045 2046 found: 2047 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */ 2048 REL_PCBINFO_TOKEN(pcbinfo); 2049 kfree(ic, M_TEMP); /* deallocate container */ 2050 } 2051 2052 /* 2053 * Remove PCB from wildcard hash table. 2054 */ 2055 void 2056 in_pcbremwildcardhash(struct inpcb *inp) 2057 { 2058 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2059 2060 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 2061 ("not in correct netisr")); 2062 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard")); 2063 2064 in_pcbremwildcardhash_oncpu(inp, pcbinfo); 2065 inp->inp_lgrpindex = -1; 2066 inp->inp_flags &= ~INP_WILDCARD; 2067 } 2068 2069 /* 2070 * Remove PCB from various lists. 2071 */ 2072 void 2073 in_pcbremlists(struct inpcb *inp) 2074 { 2075 if (inp->inp_lport) { 2076 struct inpcbportinfo *portinfo; 2077 struct inpcbport *phd; 2078 2079 /* 2080 * NOTE: 2081 * inp->inp_portinfo is _not_ necessary same as 2082 * inp->inp_pcbinfo->portinfo. 2083 */ 2084 portinfo = inp->inp_portinfo; 2085 GET_PORT_TOKEN(portinfo); 2086 2087 phd = inp->inp_phd; 2088 LIST_REMOVE(inp, inp_portlist); 2089 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 2090 LIST_REMOVE(phd, phd_hash); 2091 kfree(phd, M_PCB); 2092 } 2093 2094 REL_PORT_TOKEN(portinfo); 2095 } 2096 if (inp->inp_flags & INP_WILDCARD) { 2097 in_pcbremwildcardhash(inp); 2098 } else if (inp->inp_flags & INP_CONNECTED) { 2099 in_pcbremconnhash(inp); 2100 } 2101 2102 if (inp->inp_flags & INP_ONLIST) 2103 in_pcbofflist(inp); 2104 } 2105 2106 int 2107 prison_xinpcb(struct thread *td, struct inpcb *inp) 2108 { 2109 struct ucred *cr; 2110 2111 if (td->td_proc == NULL) 2112 return (0); 2113 cr = td->td_proc->p_ucred; 2114 if (cr->cr_prison == NULL) 2115 return (0); 2116 if (inp->inp_socket && inp->inp_socket->so_cred && 2117 inp->inp_socket->so_cred->cr_prison && 2118 cr->cr_prison == inp->inp_socket->so_cred->cr_prison) 2119 return (0); 2120 return (1); 2121 } 2122 2123 int 2124 in_pcblist_global(SYSCTL_HANDLER_ARGS) 2125 { 2126 struct inpcbinfo *pcbinfo_arr = arg1; 2127 int pcbinfo_arrlen = arg2; 2128 struct inpcb *marker; 2129 int cpu, origcpu; 2130 int error, n; 2131 2132 KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1, 2133 ("invalid pcbinfo count %d", pcbinfo_arrlen)); 2134 2135 /* 2136 * The process of preparing the TCB list is too time-consuming and 2137 * resource-intensive to repeat twice on every request. 2138 */ 2139 n = 0; 2140 if (req->oldptr == NULL) { 2141 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu) 2142 n += pcbinfo_arr[cpu].ipi_count; 2143 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 2144 return 0; 2145 } 2146 2147 if (req->newptr != NULL) 2148 return EPERM; 2149 2150 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 2151 marker->inp_flags |= INP_PLACEMARKER; 2152 2153 /* 2154 * OK, now we're committed to doing something. Re-fetch ipi_count 2155 * after obtaining the generation count. 2156 */ 2157 error = 0; 2158 origcpu = mycpuid; 2159 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) { 2160 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu]; 2161 struct inpcb *inp; 2162 struct xinpcb xi; 2163 int i; 2164 2165 lwkt_migratecpu(cpu); 2166 2167 GET_PCBINFO_TOKEN(pcbinfo); 2168 2169 n = pcbinfo->ipi_count; 2170 2171 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 2172 i = 0; 2173 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 2174 LIST_REMOVE(marker, inp_list); 2175 LIST_INSERT_AFTER(inp, marker, inp_list); 2176 2177 if (inp->inp_flags & INP_PLACEMARKER) 2178 continue; 2179 if (prison_xinpcb(req->td, inp)) 2180 continue; 2181 2182 bzero(&xi, sizeof xi); 2183 xi.xi_len = sizeof xi; 2184 bcopy(inp, &xi.xi_inp, sizeof *inp); 2185 if (inp->inp_socket) 2186 sotoxsocket(inp->inp_socket, &xi.xi_socket); 2187 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0) 2188 break; 2189 ++i; 2190 } 2191 LIST_REMOVE(marker, inp_list); 2192 2193 REL_PCBINFO_TOKEN(pcbinfo); 2194 2195 if (error == 0 && i < n) { 2196 bzero(&xi, sizeof xi); 2197 xi.xi_len = sizeof xi; 2198 while (i < n) { 2199 error = SYSCTL_OUT(req, &xi, sizeof xi); 2200 if (error) 2201 break; 2202 ++i; 2203 } 2204 } 2205 } 2206 2207 lwkt_migratecpu(origcpu); 2208 kfree(marker, M_TEMP); 2209 return error; 2210 } 2211 2212 int 2213 in_pcblist_global_ncpus2(SYSCTL_HANDLER_ARGS) 2214 { 2215 return in_pcblist_global(oidp, arg1, ncpus2, req); 2216 } 2217 2218 void 2219 in_savefaddr(struct socket *so, const struct sockaddr *faddr) 2220 { 2221 struct sockaddr_in *sin; 2222 2223 KASSERT(faddr->sa_family == AF_INET, 2224 ("not AF_INET faddr %d", faddr->sa_family)); 2225 2226 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO); 2227 sin->sin_family = AF_INET; 2228 sin->sin_len = sizeof(*sin); 2229 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port; 2230 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr; 2231 2232 so->so_faddr = (struct sockaddr *)sin; 2233 } 2234 2235 void 2236 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize, 2237 boolean_t shared, u_short offset) 2238 { 2239 memset(portinfo, 0, sizeof(*portinfo)); 2240 2241 portinfo->offset = offset; 2242 portinfo->lastport = offset; 2243 portinfo->lastlow = offset; 2244 portinfo->lasthi = offset; 2245 2246 portinfo->porthashbase = hashinit(hashsize, M_PCB, 2247 &portinfo->porthashmask); 2248 2249 if (shared) { 2250 portinfo->porttoken = kmalloc(sizeof(struct lwkt_token), 2251 M_PCB, M_WAITOK); 2252 lwkt_token_init(portinfo->porttoken, "porttoken"); 2253 } 2254 } 2255 2256 void 2257 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step) 2258 { 2259 int hi, lo; 2260 2261 if (step == 1) 2262 return; 2263 2264 hi = *hi0; 2265 lo = *lo0; 2266 2267 hi = rounddown2(hi, step); 2268 hi += ofs; 2269 if (hi > (int)*hi0) 2270 hi -= step; 2271 2272 lo = roundup2(lo, step); 2273 lo -= (step - ofs); 2274 if (lo < (int)*lo0) 2275 lo += step; 2276 2277 *hi0 = hi; 2278 *lo0 = lo; 2279 } 2280 2281 void 2282 in_pcbglobalinit(void) 2283 { 2284 int cpu; 2285 2286 in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB, 2287 M_WAITOK | M_ZERO); 2288 in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer), 2289 M_PCB, M_WAITOK | M_ZERO); 2290 2291 for (cpu = 0; cpu < ncpus; ++cpu) { 2292 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu]; 2293 struct inpcb *marker = &in_pcbmarkers[cpu]; 2294 2295 marker->inp_flags |= INP_PLACEMARKER; 2296 ic->ic_inp = marker; 2297 } 2298 } 2299 2300 struct inpcb * 2301 in_pcbmarker(int cpuid) 2302 { 2303 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid)); 2304 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 2305 2306 return &in_pcbmarkers[cpuid]; 2307 } 2308 2309 struct inpcontainer * 2310 in_pcbcontainer_marker(int cpuid) 2311 { 2312 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid)); 2313 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 2314 2315 return &in_pcbcontainer_markers[cpuid]; 2316 } 2317 2318 void 2319 in_pcbresetroute(struct inpcb *inp) 2320 { 2321 struct route *ro = &inp->inp_route; 2322 2323 if (ro->ro_rt != NULL) 2324 RTFREE(ro->ro_rt); 2325 bzero(ro, sizeof(*ro)); 2326 } 2327