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 #include <net/toeplitz2.h> 94 95 #include <netinet/in.h> 96 #include <netinet/in_pcb.h> 97 #include <netinet/in_var.h> 98 #include <netinet/ip_var.h> 99 #ifdef INET6 100 #include <netinet/ip6.h> 101 #include <netinet6/ip6_var.h> 102 #endif /* INET6 */ 103 104 #ifdef IPSEC 105 #include <netinet6/ipsec.h> 106 #include <netproto/key/key.h> 107 #include <netproto/ipsec/esp_var.h> 108 #endif 109 110 #ifdef FAST_IPSEC 111 #if defined(IPSEC) || defined(IPSEC_ESP) 112 #error "Bad idea: don't compile with both IPSEC and FAST_IPSEC!" 113 #endif 114 115 #include <netproto/ipsec/ipsec.h> 116 #include <netproto/ipsec/key.h> 117 #define IPSEC 118 #endif /* FAST_IPSEC */ 119 120 #define INP_LOCALGROUP_SIZMIN 8 121 #define INP_LOCALGROUP_SIZMAX 256 122 123 static struct inpcb *in_pcblookup_local(struct inpcbporthead *porthash, 124 struct in_addr laddr, u_int lport_arg, int wild_okay, 125 struct ucred *cred); 126 127 struct in_addr zeroin_addr; 128 129 /* 130 * These configure the range of local port addresses assigned to 131 * "unspecified" outgoing connections/packets/whatever. 132 */ 133 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */ 134 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */ 135 136 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */ 137 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */ 138 139 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 140 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */ 141 142 #define RANGECHK(var, min, max) \ 143 if ((var) < (min)) { (var) = (min); } \ 144 else if ((var) > (max)) { (var) = (max); } 145 146 int udpencap_enable = 1; /* enabled by default */ 147 int udpencap_port = 4500; /* triggers decapsulation */ 148 149 /* 150 * Per-netisr inpcb markers. 151 * NOTE: they should only be used in netisrs. 152 */ 153 static struct inpcb *in_pcbmarkers; 154 static struct inpcontainer *in_pcbcontainer_markers; 155 156 static int 157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 158 { 159 int error; 160 161 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); 162 if (!error) { 163 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 164 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 165 166 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX); 167 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX); 168 169 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX); 170 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX); 171 } 172 return (error); 173 } 174 175 #undef RANGECHK 176 177 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); 178 179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW, 180 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", ""); 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW, 182 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", ""); 183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW, 184 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", ""); 185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW, 186 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", ""); 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW, 188 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", ""); 189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW, 190 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", ""); 191 192 static int ip_porthash_trycount = 15; 193 SYSCTL_INT(_net_inet_ip, OID_AUTO, porthash_trycount, CTLFLAG_RW, 194 &ip_porthash_trycount, 0, 195 "Number of tries to find local port matching hash of 4-tuple"); 196 197 /* 198 * in_pcb.c: manage the Protocol Control Blocks. 199 * 200 * NOTE: It is assumed that most of these functions will be called from 201 * a critical section. XXX - There are, unfortunately, a few exceptions 202 * to this rule that should be fixed. 203 * 204 * NOTE: The caller should initialize the cpu field to the cpu running the 205 * protocol stack associated with this inpcbinfo. 206 */ 207 208 void 209 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared) 210 { 211 KASSERT(cpu >= 0 && cpu < ncpus, ("invalid cpu%d", cpu)); 212 pcbinfo->cpu = cpu; 213 214 LIST_INIT(&pcbinfo->pcblisthead); 215 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB, 216 M_WAITOK | M_ZERO); 217 218 if (shared) { 219 pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token), 220 M_PCB, M_WAITOK); 221 lwkt_token_init(pcbinfo->infotoken, "infotoken"); 222 } else { 223 pcbinfo->infotoken = NULL; 224 } 225 } 226 227 struct baddynamicports baddynamicports; 228 229 /* 230 * Check if the specified port is invalid for dynamic allocation. 231 */ 232 int 233 in_baddynamic(u_int16_t port, u_int16_t proto) 234 { 235 switch (proto) { 236 case IPPROTO_TCP: 237 return (DP_ISSET(baddynamicports.tcp, port)); 238 case IPPROTO_UDP: 239 #ifdef IPSEC 240 /* Cannot preset this as it is a sysctl */ 241 if (port == udpencap_port) 242 return (1); 243 #endif 244 return (DP_ISSET(baddynamicports.udp, port)); 245 default: 246 return (0); 247 } 248 } 249 250 void 251 in_pcbonlist(struct inpcb *inp) 252 { 253 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 254 255 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 256 ("not in the correct netisr")); 257 KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist")); 258 inp->inp_flags |= INP_ONLIST; 259 260 GET_PCBINFO_TOKEN(pcbinfo); 261 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list); 262 pcbinfo->ipi_count++; 263 REL_PCBINFO_TOKEN(pcbinfo); 264 } 265 266 void 267 in_pcbofflist(struct inpcb *inp) 268 { 269 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 270 271 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 272 ("not in the correct netisr")); 273 KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist")); 274 inp->inp_flags &= ~INP_ONLIST; 275 276 GET_PCBINFO_TOKEN(pcbinfo); 277 LIST_REMOVE(inp, inp_list); 278 KASSERT(pcbinfo->ipi_count > 0, 279 ("invalid inpcb count %d", pcbinfo->ipi_count)); 280 pcbinfo->ipi_count--; 281 REL_PCBINFO_TOKEN(pcbinfo); 282 } 283 284 /* 285 * Allocate a PCB and associate it with the socket. 286 */ 287 int 288 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 289 { 290 struct inpcb *inp; 291 #ifdef IPSEC 292 int error; 293 #endif 294 295 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK); 296 if (inp == NULL) 297 return (ENOMEM); 298 inp->inp_lgrpindex = -1; 299 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 300 inp->inp_pcbinfo = pcbinfo; 301 inp->inp_socket = so; 302 #ifdef IPSEC 303 error = ipsec_init_policy(so, &inp->inp_sp); 304 if (error != 0) { 305 kfree(inp, M_PCB); 306 return (error); 307 } 308 #endif 309 #ifdef INET6 310 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 311 if (ip6_auto_flowlabel) 312 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 313 inp->inp_af = AF_INET6; 314 } else 315 #endif 316 inp->inp_af = AF_INET; 317 soreference(so); 318 so->so_pcb = inp; 319 320 in_pcbonlist(inp); 321 return (0); 322 } 323 324 /* 325 * Unlink a pcb with the intention of moving it to another cpu with a 326 * different pcbinfo. While unlinked nothing should attempt to dereference 327 * inp_pcbinfo, NULL it out so we assert if it does. 328 */ 329 void 330 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags) 331 { 332 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch")); 333 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0, 334 ("already linked")); 335 336 in_pcbofflist(inp); 337 inp->inp_pcbinfo = NULL; 338 } 339 340 void 341 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 342 { 343 in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD); 344 } 345 346 /* 347 * Relink a pcb into a new pcbinfo. 348 */ 349 void 350 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags) 351 { 352 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo")); 353 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0, 354 ("already linked")); 355 356 inp->inp_pcbinfo = pcbinfo; 357 in_pcbonlist(inp); 358 } 359 360 void 361 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 362 { 363 return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD); 364 } 365 366 static boolean_t 367 in_pcbporthash_update(struct inpcbportinfo *portinfo, 368 struct inpcb *inp, u_short lport, struct ucred *cred, int wild) 369 { 370 struct inpcbporthead *porthash; 371 372 /* 373 * This has to be atomic. If the porthash is shared across multiple 374 * protocol threads, e.g. tcp and udp, then the token must be held. 375 */ 376 porthash = in_pcbporthash_head(portinfo, lport); 377 GET_PORTHASH_TOKEN(porthash); 378 379 if (in_pcblookup_local(porthash, inp->inp_laddr, lport, 380 wild, cred) != NULL) { 381 REL_PORTHASH_TOKEN(porthash); 382 return FALSE; 383 } 384 inp->inp_lport = lport; 385 in_pcbinsporthash(porthash, inp); 386 387 REL_PORTHASH_TOKEN(porthash); 388 return TRUE; 389 } 390 391 static int 392 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred) 393 { 394 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 395 struct inpcbportinfo *portinfo; 396 u_short first, last, lport, step; 397 u_short *lastport; 398 int count, error; 399 int portinfo_first, portinfo_idx; 400 401 inp->inp_flags |= INP_ANONPORT; 402 403 step = pcbinfo->portinfo_mask + 1; 404 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 405 portinfo_idx = portinfo_first; 406 loop: 407 portinfo = &pcbinfo->portinfo[portinfo_idx]; 408 409 if (inp->inp_flags & INP_HIGHPORT) { 410 first = ipport_hifirstauto; /* sysctl */ 411 last = ipport_hilastauto; 412 lastport = &portinfo->lasthi; 413 } else if (inp->inp_flags & INP_LOWPORT) { 414 if (cred && 415 (error = 416 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 417 inp->inp_laddr.s_addr = INADDR_ANY; 418 return error; 419 } 420 first = ipport_lowfirstauto; /* 1023 */ 421 last = ipport_lowlastauto; /* 600 */ 422 lastport = &portinfo->lastlow; 423 } else { 424 first = ipport_firstauto; /* sysctl */ 425 last = ipport_lastauto; 426 lastport = &portinfo->lastport; 427 } 428 429 /* 430 * Simple check to ensure all ports are not used up causing 431 * a deadlock here. 432 * 433 * We split the two cases (up and down) so that the direction 434 * is not being tested on each round of the loop. 435 */ 436 if (first > last) { 437 /* 438 * counting down 439 */ 440 in_pcbportrange(&first, &last, portinfo->offset, step); 441 count = (first - last) / step; 442 443 for (;;) { 444 if (count-- < 0) { /* completely used? */ 445 error = EADDRNOTAVAIL; 446 break; 447 } 448 lport = in_pcblastport_down(lastport, 449 first, last, step); 450 KKASSERT((lport & pcbinfo->portinfo_mask) == 451 portinfo->offset); 452 lport = htons(lport); 453 454 if (in_pcbporthash_update(portinfo, inp, lport, 455 cred, wild)) { 456 error = 0; 457 break; 458 } 459 } 460 } else { 461 /* 462 * counting up 463 */ 464 in_pcbportrange(&last, &first, portinfo->offset, step); 465 count = (last - first) / step; 466 467 for (;;) { 468 if (count-- < 0) { /* completely used? */ 469 error = EADDRNOTAVAIL; 470 break; 471 } 472 lport = in_pcblastport_up(lastport, first, last, step); 473 KKASSERT((lport & pcbinfo->portinfo_mask) == 474 portinfo->offset); 475 lport = htons(lport); 476 477 if (in_pcbporthash_update(portinfo, inp, lport, 478 cred, wild)) { 479 error = 0; 480 break; 481 } 482 } 483 } 484 485 if (error) { 486 /* Try next portinfo */ 487 portinfo_idx++; 488 portinfo_idx &= pcbinfo->portinfo_mask; 489 if (portinfo_idx != portinfo_first) 490 goto loop; 491 inp->inp_laddr.s_addr = INADDR_ANY; 492 } 493 return error; 494 } 495 496 int 497 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 498 { 499 struct socket *so = inp->inp_socket; 500 struct sockaddr_in jsin; 501 struct ucred *cred = NULL; 502 int wild = 0; 503 504 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 505 return (EADDRNOTAVAIL); 506 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 507 return (EINVAL); /* already bound */ 508 509 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT))) 510 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */ 511 if (td->td_proc) 512 cred = td->td_proc->p_ucred; 513 514 if (nam != NULL) { 515 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 516 struct inpcbinfo *pcbinfo; 517 struct inpcbportinfo *portinfo; 518 struct inpcbporthead *porthash; 519 struct inpcb *t; 520 u_short lport, lport_ho; 521 int reuseport = (so->so_options & SO_REUSEPORT); 522 int error; 523 524 if (nam->sa_len != sizeof *sin) 525 return (EINVAL); 526 #ifdef notdef 527 /* 528 * We should check the family, but old programs 529 * incorrectly fail to initialize it. 530 */ 531 if (sin->sin_family != AF_INET) 532 return (EAFNOSUPPORT); 533 #endif 534 if (!prison_replace_wildcards(td, nam)) 535 return (EINVAL); 536 537 lport = sin->sin_port; 538 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 539 /* 540 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 541 * allow complete duplication of binding if 542 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 543 * and a multicast address is bound on both 544 * new and duplicated sockets. 545 */ 546 if (so->so_options & SO_REUSEADDR) 547 reuseport = SO_REUSEADDR | SO_REUSEPORT; 548 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 549 sin->sin_port = 0; /* yech... */ 550 bzero(&sin->sin_zero, sizeof sin->sin_zero); 551 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL) 552 return (EADDRNOTAVAIL); 553 } 554 555 inp->inp_laddr = sin->sin_addr; 556 557 jsin.sin_family = AF_INET; 558 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 559 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 560 inp->inp_laddr.s_addr = INADDR_ANY; 561 return (EINVAL); 562 } 563 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 564 565 if (lport == 0) { 566 /* Auto-select local port */ 567 return in_pcbsetlport(inp, wild, cred); 568 } 569 lport_ho = ntohs(lport); 570 571 /* GROSS */ 572 if (lport_ho < IPPORT_RESERVED && cred && 573 (error = 574 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 575 inp->inp_laddr.s_addr = INADDR_ANY; 576 return (error); 577 } 578 579 /* 580 * Locate the proper portinfo based on lport 581 */ 582 pcbinfo = inp->inp_pcbinfo; 583 portinfo = 584 &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 585 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == 586 portinfo->offset); 587 588 /* 589 * This has to be atomic. If the porthash is shared across 590 * multiple protocol threads, e.g. tcp and udp then the token 591 * must be held. 592 */ 593 porthash = in_pcbporthash_head(portinfo, lport); 594 GET_PORTHASH_TOKEN(porthash); 595 596 if (so->so_cred->cr_uid != 0 && 597 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 598 t = in_pcblookup_local(porthash, sin->sin_addr, lport, 599 INPLOOKUP_WILDCARD, cred); 600 if (t && 601 (so->so_cred->cr_uid != 602 t->inp_socket->so_cred->cr_uid)) { 603 inp->inp_laddr.s_addr = INADDR_ANY; 604 error = EADDRINUSE; 605 goto done; 606 } 607 } 608 if (cred && !prison_replace_wildcards(td, nam)) { 609 inp->inp_laddr.s_addr = INADDR_ANY; 610 error = EADDRNOTAVAIL; 611 goto done; 612 } 613 t = in_pcblookup_local(porthash, sin->sin_addr, lport, 614 wild, cred); 615 if (t && !(reuseport & t->inp_socket->so_options)) { 616 inp->inp_laddr.s_addr = INADDR_ANY; 617 error = EADDRINUSE; 618 goto done; 619 } 620 inp->inp_lport = lport; 621 in_pcbinsporthash(porthash, inp); 622 error = 0; 623 done: 624 REL_PORTHASH_TOKEN(porthash); 625 return (error); 626 } else { 627 jsin.sin_family = AF_INET; 628 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 629 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 630 inp->inp_laddr.s_addr = INADDR_ANY; 631 return (EINVAL); 632 } 633 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 634 635 return in_pcbsetlport(inp, wild, cred); 636 } 637 } 638 639 static struct inpcb * 640 in_pcblookup_localremote(struct inpcbporthead *porthash, struct in_addr laddr, 641 u_short lport, struct in_addr faddr, u_short fport, struct ucred *cred) 642 { 643 struct inpcb *inp; 644 struct inpcbport *phd; 645 struct inpcb *match = NULL; 646 647 /* 648 * If the porthashbase is shared across several cpus, it must 649 * have been locked. 650 */ 651 ASSERT_PORTHASH_TOKEN_HELD(porthash); 652 653 /* 654 * Best fit PCB lookup. 655 * 656 * First see if this local port is in use by looking on the 657 * port hash list. 658 */ 659 LIST_FOREACH(phd, porthash, phd_hash) { 660 if (phd->phd_port == lport) 661 break; 662 } 663 if (phd != NULL) { 664 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 665 #ifdef INET6 666 if (!INP_ISIPV4(inp)) 667 continue; 668 #endif 669 if (inp->inp_laddr.s_addr != INADDR_ANY && 670 inp->inp_laddr.s_addr != laddr.s_addr) 671 continue; 672 673 if (inp->inp_faddr.s_addr != INADDR_ANY && 674 inp->inp_faddr.s_addr != faddr.s_addr) 675 continue; 676 677 if (inp->inp_fport != 0 && inp->inp_fport != fport) 678 continue; 679 680 if (cred == NULL || 681 cred->cr_prison == 682 inp->inp_socket->so_cred->cr_prison) { 683 match = inp; 684 break; 685 } 686 } 687 } 688 return (match); 689 } 690 691 static boolean_t 692 in_pcbporthash_update4(struct inpcbportinfo *portinfo, 693 struct inpcb *inp, u_short lport, const struct sockaddr_in *sin, 694 struct ucred *cred) 695 { 696 struct inpcbporthead *porthash; 697 698 /* 699 * This has to be atomic. If the porthash is shared across multiple 700 * protocol threads, e.g. tcp and udp, then the token must be held. 701 */ 702 porthash = in_pcbporthash_head(portinfo, lport); 703 GET_PORTHASH_TOKEN(porthash); 704 705 if (in_pcblookup_localremote(porthash, inp->inp_laddr, 706 lport, sin->sin_addr, sin->sin_port, cred) != NULL) { 707 REL_PORTHASH_TOKEN(porthash); 708 return FALSE; 709 } 710 inp->inp_lport = lport; 711 in_pcbinsporthash(porthash, inp); 712 713 REL_PORTHASH_TOKEN(porthash); 714 return TRUE; 715 } 716 717 int 718 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote, 719 struct thread *td) 720 { 721 struct proc *p = td->td_proc; 722 u_short *lastport; 723 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote; 724 struct sockaddr_in jsin; 725 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 726 struct inpcbportinfo *portinfo; 727 struct ucred *cred = NULL; 728 u_short first, last, lport, step; 729 int count, error, selfconn; 730 int portinfo_first, portinfo_idx; 731 int hash_cpu, hash_count, hash_count0; 732 uint32_t hash_base = 0, hash; 733 734 if (TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) /* XXX broken! */ 735 return (EADDRNOTAVAIL); 736 737 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY); 738 if (inp->inp_lport != 0) 739 return (EINVAL); /* already bound */ 740 741 KKASSERT(p); 742 cred = p->p_ucred; 743 744 jsin.sin_family = AF_INET; 745 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 746 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 747 inp->inp_laddr.s_addr = INADDR_ANY; 748 return (EINVAL); 749 } 750 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 751 752 hash_count0 = ip_porthash_trycount; 753 if (hash_count0 > 0) { 754 hash_base = toeplitz_piecemeal_addr(sin->sin_addr.s_addr) ^ 755 toeplitz_piecemeal_addr(inp->inp_laddr.s_addr) ^ 756 toeplitz_piecemeal_port(sin->sin_port); 757 } else { 758 hash_count0 = 0; 759 } 760 761 inp->inp_flags |= INP_ANONPORT; 762 763 step = pcbinfo->portinfo_mask + 1; 764 portinfo_first = mycpuid & pcbinfo->portinfo_mask; 765 portinfo_idx = portinfo_first; 766 loop: 767 hash_cpu = portinfo_idx & ncpus2_mask; 768 portinfo = &pcbinfo->portinfo[portinfo_idx]; 769 selfconn = 0; 770 771 if (inp->inp_flags & INP_HIGHPORT) { 772 first = ipport_hifirstauto; /* sysctl */ 773 last = ipport_hilastauto; 774 lastport = &portinfo->lasthi; 775 } else if (inp->inp_flags & INP_LOWPORT) { 776 if (cred && 777 (error = 778 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0))) { 779 inp->inp_laddr.s_addr = INADDR_ANY; 780 return (error); 781 } 782 first = ipport_lowfirstauto; /* 1023 */ 783 last = ipport_lowlastauto; /* 600 */ 784 lastport = &portinfo->lastlow; 785 } else { 786 first = ipport_firstauto; /* sysctl */ 787 last = ipport_lastauto; 788 lastport = &portinfo->lastport; 789 } 790 791 /* This could happen on loopback interface */ 792 #define IS_SELFCONNECT(inp, lport, sin) \ 793 (__predict_false((sin)->sin_port == (lport) && \ 794 (sin)->sin_addr.s_addr == (inp)->inp_laddr.s_addr)) 795 796 /* 797 * Simple check to ensure all ports are not used up causing 798 * a deadlock here. 799 * 800 * We split the two cases (up and down) so that the direction 801 * is not being tested on each round of the loop. 802 */ 803 hash_count = hash_count0; 804 if (first > last) { 805 /* 806 * counting down 807 */ 808 in_pcbportrange(&first, &last, portinfo->offset, step); 809 count = ((first - last) / step) + hash_count; 810 811 for (;;) { 812 if (count-- < 0) { /* completely used? */ 813 error = EADDRNOTAVAIL; 814 break; 815 } 816 817 lport = in_pcblastport_down(lastport, first, last, 818 step); 819 KKASSERT((lport & pcbinfo->portinfo_mask) == 820 portinfo->offset); 821 lport = htons(lport); 822 if (IS_SELFCONNECT(inp, lport, sin)) { 823 if (!selfconn) { 824 ++count; /* don't count this try */ 825 selfconn = 1; 826 } 827 continue; 828 } 829 830 if (hash_count) { 831 --hash_count; 832 hash = hash_base ^ 833 toeplitz_piecemeal_port(lport); 834 if ((hash & ncpus2_mask) != hash_cpu && hash_count) 835 continue; 836 } 837 838 if (in_pcbporthash_update4(portinfo, 839 inp, lport, sin, cred)) { 840 error = 0; 841 break; 842 } 843 } 844 } else { 845 /* 846 * counting up 847 */ 848 in_pcbportrange(&last, &first, portinfo->offset, step); 849 count = ((last - first) / step) + hash_count; 850 851 for (;;) { 852 if (count-- < 0) { /* completely used? */ 853 error = EADDRNOTAVAIL; 854 break; 855 } 856 857 lport = in_pcblastport_up(lastport, first, last, step); 858 KKASSERT((lport & pcbinfo->portinfo_mask) == 859 portinfo->offset); 860 lport = htons(lport); 861 if (IS_SELFCONNECT(inp, lport, sin)) { 862 if (!selfconn) { 863 ++count; /* don't count this try */ 864 selfconn = 1; 865 } 866 continue; 867 } 868 869 if (hash_count) { 870 --hash_count; 871 hash = hash_base ^ 872 toeplitz_piecemeal_port(lport); 873 if ((hash & ncpus2_mask) != hash_cpu && hash_count) 874 continue; 875 } 876 877 if (in_pcbporthash_update4(portinfo, 878 inp, lport, sin, cred)) { 879 error = 0; 880 break; 881 } 882 } 883 } 884 885 #undef IS_SELFCONNECT 886 887 if (error) { 888 /* Try next portinfo */ 889 portinfo_idx++; 890 portinfo_idx &= pcbinfo->portinfo_mask; 891 if (portinfo_idx != portinfo_first) 892 goto loop; 893 inp->inp_laddr.s_addr = INADDR_ANY; 894 } 895 return error; 896 } 897 898 /* 899 * Transform old in_pcbconnect() into an inner subroutine for new 900 * in_pcbconnect(): Do some validity-checking on the remote 901 * address (in mbuf 'nam') and then determine local host address 902 * (i.e., which interface) to use to access that remote host. 903 * 904 * This preserves definition of in_pcbconnect(), while supporting a 905 * slightly different version for T/TCP. (This is more than 906 * a bit of a kludge, but cleaning up the internal interfaces would 907 * have forced minor changes in every protocol). 908 */ 909 int 910 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam, 911 struct sockaddr_in **plocal_sin, struct thread *td, int find) 912 { 913 struct in_ifaddr *ia; 914 struct ucred *cred = NULL; 915 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 916 struct sockaddr *jsin; 917 int jailed = 0, alloc_route = 0; 918 919 if (nam->sa_len != sizeof *sin) 920 return (EINVAL); 921 if (sin->sin_family != AF_INET) 922 return (EAFNOSUPPORT); 923 if (sin->sin_port == 0) 924 return (EADDRNOTAVAIL); 925 if (td && td->td_proc && td->td_proc->p_ucred) 926 cred = td->td_proc->p_ucred; 927 if (cred && cred->cr_prison) 928 jailed = 1; 929 if (!TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) { 930 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 931 /* 932 * If the destination address is INADDR_ANY, 933 * use the primary local address. 934 * If the supplied address is INADDR_BROADCAST, 935 * and the primary interface supports broadcast, 936 * choose the broadcast address for that interface. 937 */ 938 if (sin->sin_addr.s_addr == INADDR_ANY) 939 sin->sin_addr = IA_SIN(ia)->sin_addr; 940 else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST && 941 (ia->ia_ifp->if_flags & IFF_BROADCAST)) 942 sin->sin_addr = satosin(&ia->ia_broadaddr)->sin_addr; 943 } 944 if (find) { 945 struct route *ro; 946 947 ia = NULL; 948 /* 949 * If route is known or can be allocated now, 950 * our src addr is taken from the i/f, else punt. 951 * Note that we should check the address family of the cached 952 * destination, in case of sharing the cache with IPv6. 953 */ 954 ro = &inp->inp_route; 955 if (ro->ro_rt && 956 (!(ro->ro_rt->rt_flags & RTF_UP) || 957 ro->ro_dst.sa_family != AF_INET || 958 satosin(&ro->ro_dst)->sin_addr.s_addr != 959 sin->sin_addr.s_addr || 960 inp->inp_socket->so_options & SO_DONTROUTE)) { 961 RTFREE(ro->ro_rt); 962 ro->ro_rt = NULL; 963 } 964 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 965 (ro->ro_rt == NULL || 966 ro->ro_rt->rt_ifp == NULL)) { 967 /* No route yet, so try to acquire one */ 968 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 969 ro->ro_dst.sa_family = AF_INET; 970 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 971 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr = 972 sin->sin_addr; 973 rtalloc(ro); 974 alloc_route = 1; 975 } 976 /* 977 * If we found a route, use the address 978 * corresponding to the outgoing interface 979 * unless it is the loopback (in case a route 980 * to our address on another net goes to loopback). 981 */ 982 if (ro->ro_rt && 983 !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK)) { 984 if (jailed) { 985 if (jailed_ip(cred->cr_prison, 986 ro->ro_rt->rt_ifa->ifa_addr)) { 987 ia = ifatoia(ro->ro_rt->rt_ifa); 988 } 989 } else { 990 ia = ifatoia(ro->ro_rt->rt_ifa); 991 } 992 } 993 if (ia == NULL) { 994 u_short fport = sin->sin_port; 995 996 sin->sin_port = 0; 997 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 998 if (ia && jailed && !jailed_ip(cred->cr_prison, 999 sintosa(&ia->ia_addr))) 1000 ia = NULL; 1001 if (ia == NULL) 1002 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 1003 if (ia && jailed && !jailed_ip(cred->cr_prison, 1004 sintosa(&ia->ia_addr))) 1005 ia = NULL; 1006 sin->sin_port = fport; 1007 if (ia == NULL && 1008 !TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1009 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 1010 if (ia && jailed && !jailed_ip(cred->cr_prison, 1011 sintosa(&ia->ia_addr))) 1012 ia = NULL; 1013 1014 if (!jailed && ia == NULL) 1015 goto fail; 1016 } 1017 /* 1018 * If the destination address is multicast and an outgoing 1019 * interface has been set as a multicast option, use the 1020 * address of that interface as our source address. 1021 */ 1022 if (!jailed && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 1023 inp->inp_moptions != NULL) { 1024 struct ip_moptions *imo; 1025 struct ifnet *ifp; 1026 1027 imo = inp->inp_moptions; 1028 if ((ifp = imo->imo_multicast_ifp) != NULL) { 1029 struct in_ifaddr_container *iac; 1030 1031 ia = NULL; 1032 TAILQ_FOREACH(iac, 1033 &in_ifaddrheads[mycpuid], ia_link) { 1034 if (iac->ia->ia_ifp == ifp) { 1035 ia = iac->ia; 1036 break; 1037 } 1038 } 1039 if (ia == NULL) 1040 goto fail; 1041 } 1042 } 1043 /* 1044 * Don't do pcblookup call here; return interface in plocal_sin 1045 * and exit to caller, that will do the lookup. 1046 */ 1047 if (ia == NULL && jailed) { 1048 if ((jsin = prison_get_nonlocal( 1049 cred->cr_prison, AF_INET, NULL)) != NULL || 1050 (jsin = prison_get_local( 1051 cred->cr_prison, AF_INET, NULL)) != NULL) { 1052 *plocal_sin = satosin(jsin); 1053 } else { 1054 /* IPv6 only Jail */ 1055 goto fail; 1056 } 1057 } else { 1058 *plocal_sin = &ia->ia_addr; 1059 } 1060 } 1061 return (0); 1062 fail: 1063 if (alloc_route) 1064 in_pcbresetroute(inp); 1065 return (EADDRNOTAVAIL); 1066 } 1067 1068 int 1069 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam, 1070 struct sockaddr_in **plocal_sin, struct thread *td) 1071 { 1072 return in_pcbladdr_find(inp, nam, plocal_sin, td, 1073 (inp->inp_laddr.s_addr == INADDR_ANY)); 1074 } 1075 1076 /* 1077 * Outer subroutine: 1078 * Connect from a socket to a specified address. 1079 * Both address and port must be specified in argument sin. 1080 * If don't have a local address for this socket yet, 1081 * then pick one. 1082 */ 1083 int 1084 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 1085 { 1086 struct sockaddr_in *if_sin; 1087 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1088 int error; 1089 1090 /* Call inner routine to assign local interface address. */ 1091 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0) 1092 return (error); 1093 1094 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port, 1095 inp->inp_laddr.s_addr ? 1096 inp->inp_laddr : if_sin->sin_addr, 1097 inp->inp_lport, FALSE, NULL) != NULL) { 1098 return (EADDRINUSE); 1099 } 1100 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1101 if (inp->inp_lport == 0) { 1102 error = in_pcbbind(inp, NULL, td); 1103 if (error) 1104 return (error); 1105 } 1106 inp->inp_laddr = if_sin->sin_addr; 1107 } 1108 inp->inp_faddr = sin->sin_addr; 1109 inp->inp_fport = sin->sin_port; 1110 in_pcbinsconnhash(inp); 1111 return (0); 1112 } 1113 1114 void 1115 in_pcbdisconnect(struct inpcb *inp) 1116 { 1117 1118 in_pcbremconnhash(inp); 1119 inp->inp_faddr.s_addr = INADDR_ANY; 1120 inp->inp_fport = 0; 1121 } 1122 1123 void 1124 in_pcbdetach(struct inpcb *inp) 1125 { 1126 struct socket *so = inp->inp_socket; 1127 struct inpcbinfo *ipi = inp->inp_pcbinfo; 1128 1129 #ifdef IPSEC 1130 ipsec4_delete_pcbpolicy(inp); 1131 #endif /*IPSEC*/ 1132 inp->inp_gencnt = ++ipi->ipi_gencnt; 1133 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 1134 in_pcbremlists(inp); 1135 so->so_pcb = NULL; 1136 sofree(so); /* remove pcb ref */ 1137 if (inp->inp_options) 1138 m_free(inp->inp_options); 1139 if (inp->inp_route.ro_rt) 1140 rtfree(inp->inp_route.ro_rt); 1141 ip_freemoptions(inp->inp_moptions); 1142 kfree(inp, M_PCB); 1143 } 1144 1145 /* 1146 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP 1147 * socket received RST. 1148 */ 1149 static int 1150 in_setsockaddr(struct socket *so, struct sockaddr **nam) 1151 { 1152 struct inpcb *inp; 1153 struct sockaddr_in *sin; 1154 1155 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 1156 inp = so->so_pcb; 1157 if (!inp) 1158 return (ECONNRESET); 1159 1160 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1161 sin->sin_family = AF_INET; 1162 sin->sin_len = sizeof *sin; 1163 sin->sin_port = inp->inp_lport; 1164 sin->sin_addr = inp->inp_laddr; 1165 1166 *nam = (struct sockaddr *)sin; 1167 return (0); 1168 } 1169 1170 void 1171 in_setsockaddr_dispatch(netmsg_t msg) 1172 { 1173 int error; 1174 1175 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1176 lwkt_replymsg(&msg->lmsg, error); 1177 } 1178 1179 /* 1180 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP 1181 * socket received RST. 1182 */ 1183 int 1184 in_setpeeraddr(struct socket *so, struct sockaddr **nam) 1185 { 1186 struct inpcb *inp; 1187 struct sockaddr_in *sin; 1188 1189 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 1190 inp = so->so_pcb; 1191 if (!inp) 1192 return (ECONNRESET); 1193 1194 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1195 sin->sin_family = AF_INET; 1196 sin->sin_len = sizeof *sin; 1197 sin->sin_port = inp->inp_fport; 1198 sin->sin_addr = inp->inp_faddr; 1199 1200 *nam = (struct sockaddr *)sin; 1201 return (0); 1202 } 1203 1204 void 1205 in_setpeeraddr_dispatch(netmsg_t msg) 1206 { 1207 int error; 1208 1209 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1210 lwkt_replymsg(&msg->lmsg, error); 1211 } 1212 1213 void 1214 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err, 1215 inp_notify_t notify) 1216 { 1217 struct inpcb *inp, *marker; 1218 1219 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1220 ("not in the correct netisr")); 1221 marker = &in_pcbmarkers[mycpuid]; 1222 1223 /* 1224 * NOTE: 1225 * - If INP_PLACEMARKER is set we must ignore the rest of the 1226 * structure and skip it. 1227 * - It is safe to nuke inpcbs here, since we are in their own 1228 * netisr. 1229 */ 1230 GET_PCBINFO_TOKEN(pcbinfo); 1231 1232 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1233 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) { 1234 LIST_REMOVE(marker, inp_list); 1235 LIST_INSERT_AFTER(inp, marker, inp_list); 1236 1237 if (inp->inp_flags & INP_PLACEMARKER) 1238 continue; 1239 #ifdef INET6 1240 if (!INP_ISIPV4(inp)) 1241 continue; 1242 #endif 1243 if (inp->inp_faddr.s_addr != faddr.s_addr || 1244 inp->inp_socket == NULL) 1245 continue; 1246 (*notify)(inp, err); /* can remove inp from list! */ 1247 } 1248 LIST_REMOVE(marker, inp_list); 1249 1250 REL_PCBINFO_TOKEN(pcbinfo); 1251 } 1252 1253 void 1254 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1255 { 1256 struct inpcb *inp, *marker; 1257 1258 /* 1259 * We only need to make sure that we are in netisr0, where all 1260 * multicast operation happen. We could check inpcbinfo which 1261 * does not belong to netisr0 by holding the inpcbinfo's token. 1262 * In this case, the pcbinfo must be able to be shared, i.e. 1263 * pcbinfo->infotoken is not NULL. 1264 */ 1265 ASSERT_IN_NETISR(0); 1266 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL, 1267 ("pcbinfo could not be shared")); 1268 1269 /* 1270 * Get a marker for the current netisr (netisr0). 1271 * 1272 * It is possible that the multicast address deletion blocks, 1273 * which could cause temporary token releasing. So we use 1274 * inpcb marker here to get a coherent view of the inpcb list. 1275 * 1276 * While, on the other hand, moptions are only added and deleted 1277 * in netisr0, so we would not see staled moption or miss moption 1278 * even if the token was released due to the blocking multicast 1279 * address deletion. 1280 */ 1281 marker = &in_pcbmarkers[mycpuid]; 1282 1283 GET_PCBINFO_TOKEN(pcbinfo); 1284 1285 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1286 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) { 1287 struct ip_moptions *imo; 1288 1289 LIST_REMOVE(marker, inp_list); 1290 LIST_INSERT_AFTER(inp, marker, inp_list); 1291 1292 if (inp->inp_flags & INP_PLACEMARKER) 1293 continue; 1294 imo = inp->inp_moptions; 1295 if (INP_ISIPV4(inp) && imo != NULL) { 1296 int i, gap; 1297 1298 /* 1299 * Unselect the outgoing interface if it is being 1300 * detached. 1301 */ 1302 if (imo->imo_multicast_ifp == ifp) 1303 imo->imo_multicast_ifp = NULL; 1304 1305 /* 1306 * Drop multicast group membership if we joined 1307 * through the interface being detached. 1308 */ 1309 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1310 i++) { 1311 if (imo->imo_membership[i]->inm_ifp == ifp) { 1312 /* 1313 * NOTE: 1314 * This could block and the pcbinfo 1315 * token could be passively released. 1316 */ 1317 in_delmulti(imo->imo_membership[i]); 1318 gap++; 1319 } else if (gap != 0) 1320 imo->imo_membership[i - gap] = 1321 imo->imo_membership[i]; 1322 } 1323 imo->imo_num_memberships -= gap; 1324 } 1325 } 1326 LIST_REMOVE(marker, inp_list); 1327 1328 REL_PCBINFO_TOKEN(pcbinfo); 1329 } 1330 1331 /* 1332 * Check for alternatives when higher level complains 1333 * about service problems. For now, invalidate cached 1334 * routing information. If the route was created dynamically 1335 * (by a redirect), time to try a default gateway again. 1336 */ 1337 void 1338 in_losing(struct inpcb *inp) 1339 { 1340 struct rtentry *rt; 1341 struct rt_addrinfo rtinfo; 1342 1343 if ((rt = inp->inp_route.ro_rt)) { 1344 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1345 rtinfo.rti_info[RTAX_DST] = rt_key(rt); 1346 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1347 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt); 1348 rtinfo.rti_flags = rt->rt_flags; 1349 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0); 1350 if (rt->rt_flags & RTF_DYNAMIC) { 1351 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1352 rt_mask(rt), rt->rt_flags, NULL); 1353 } 1354 inp->inp_route.ro_rt = NULL; 1355 rtfree(rt); 1356 /* 1357 * A new route can be allocated 1358 * the next time output is attempted. 1359 */ 1360 } 1361 } 1362 1363 /* 1364 * After a routing change, flush old routing 1365 * and allocate a (hopefully) better one. 1366 */ 1367 void 1368 in_rtchange(struct inpcb *inp, int err) 1369 { 1370 if (inp->inp_route.ro_rt) { 1371 rtfree(inp->inp_route.ro_rt); 1372 inp->inp_route.ro_rt = NULL; 1373 /* 1374 * A new route can be allocated the next time 1375 * output is attempted. 1376 */ 1377 } 1378 } 1379 1380 /* 1381 * Lookup a PCB based on the local address and port. 1382 */ 1383 static struct inpcb * 1384 in_pcblookup_local(struct inpcbporthead *porthash, struct in_addr laddr, 1385 u_int lport_arg, int wild_okay, struct ucred *cred) 1386 { 1387 struct inpcb *inp; 1388 int matchwild = 3, wildcard; 1389 u_short lport = lport_arg; 1390 struct inpcbport *phd; 1391 struct inpcb *match = NULL; 1392 1393 /* 1394 * If the porthashbase is shared across several cpus, it must 1395 * have been locked. 1396 */ 1397 ASSERT_PORTHASH_TOKEN_HELD(porthash); 1398 1399 /* 1400 * Best fit PCB lookup. 1401 * 1402 * First see if this local port is in use by looking on the 1403 * port hash list. 1404 */ 1405 LIST_FOREACH(phd, porthash, phd_hash) { 1406 if (phd->phd_port == lport) 1407 break; 1408 } 1409 if (phd != NULL) { 1410 /* 1411 * Port is in use by one or more PCBs. Look for best 1412 * fit. 1413 */ 1414 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1415 wildcard = 0; 1416 #ifdef INET6 1417 if (!INP_ISIPV4(inp)) 1418 continue; 1419 #endif 1420 if (inp->inp_faddr.s_addr != INADDR_ANY) 1421 wildcard++; 1422 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1423 if (laddr.s_addr == INADDR_ANY) 1424 wildcard++; 1425 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1426 continue; 1427 } else { 1428 if (laddr.s_addr != INADDR_ANY) 1429 wildcard++; 1430 } 1431 if (wildcard && !wild_okay) 1432 continue; 1433 if (wildcard < matchwild && 1434 (cred == NULL || 1435 cred->cr_prison == 1436 inp->inp_socket->so_cred->cr_prison)) { 1437 match = inp; 1438 matchwild = wildcard; 1439 if (matchwild == 0) { 1440 break; 1441 } 1442 } 1443 } 1444 } 1445 return (match); 1446 } 1447 1448 struct inpcb * 1449 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo, 1450 const struct inpcb *inp) 1451 { 1452 const struct inp_localgrphead *hdr; 1453 const struct inp_localgroup *grp; 1454 int i; 1455 1456 if (pcbinfo->localgrphashbase == NULL) 1457 return NULL; 1458 1459 GET_PCBINFO_TOKEN(pcbinfo); 1460 1461 hdr = &pcbinfo->localgrphashbase[ 1462 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1463 1464 LIST_FOREACH(grp, hdr, il_list) { 1465 if (grp->il_af == inp->inp_af && 1466 grp->il_lport == inp->inp_lport && 1467 memcmp(&grp->il_dependladdr, 1468 &inp->inp_inc.inc_ie.ie_dependladdr, 1469 sizeof(grp->il_dependladdr)) == 0) { 1470 break; 1471 } 1472 } 1473 if (grp == NULL || grp->il_inpcnt == 1) { 1474 REL_PCBINFO_TOKEN(pcbinfo); 1475 return NULL; 1476 } 1477 1478 KASSERT(grp->il_inpcnt >= 2, 1479 ("invalid localgroup inp count %d", grp->il_inpcnt)); 1480 for (i = 0; i < grp->il_inpcnt; ++i) { 1481 if (grp->il_inp[i] == inp) { 1482 int last = grp->il_inpcnt - 1; 1483 1484 if (i == last) 1485 last = grp->il_inpcnt - 2; 1486 REL_PCBINFO_TOKEN(pcbinfo); 1487 return grp->il_inp[last]; 1488 } 1489 } 1490 REL_PCBINFO_TOKEN(pcbinfo); 1491 return NULL; 1492 } 1493 1494 static struct inpcb * 1495 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo, 1496 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash) 1497 { 1498 struct inpcb *local_wild = NULL; 1499 const struct inp_localgrphead *hdr; 1500 const struct inp_localgroup *grp; 1501 1502 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 1503 1504 hdr = &pcbinfo->localgrphashbase[ 1505 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)]; 1506 1507 /* 1508 * Order of socket selection: 1509 * 1. non-wild. 1510 * 2. wild. 1511 * 1512 * NOTE: Local group does not contain jailed sockets 1513 */ 1514 LIST_FOREACH(grp, hdr, il_list) { 1515 #ifdef INET6 1516 if (grp->il_af != AF_INET) 1517 continue; 1518 #endif 1519 if (grp->il_lport == lport) { 1520 int idx; 1521 1522 /* 1523 * Modulo-N is used here, which greatly reduces 1524 * completion queue token contention, thus more 1525 * cpu time is saved. 1526 */ 1527 idx = pkt_hash % grp->il_inpcnt; 1528 if (grp->il_laddr.s_addr == laddr.s_addr) 1529 return grp->il_inp[idx]; 1530 else if (grp->il_laddr.s_addr == INADDR_ANY) 1531 local_wild = grp->il_inp[idx]; 1532 } 1533 } 1534 if (local_wild != NULL) 1535 return local_wild; 1536 return NULL; 1537 } 1538 1539 /* 1540 * Lookup PCB in hash list. 1541 */ 1542 struct inpcb * 1543 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1544 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1545 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m) 1546 { 1547 struct inpcbhead *head; 1548 struct inpcb *inp, *jinp=NULL; 1549 u_short fport = fport_arg, lport = lport_arg; 1550 1551 /* 1552 * First look for an exact match. 1553 */ 1554 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport, 1555 laddr.s_addr, lport, pcbinfo->hashmask)]; 1556 LIST_FOREACH(inp, head, inp_hash) { 1557 #ifdef INET6 1558 if (!INP_ISIPV4(inp)) 1559 continue; 1560 #endif 1561 if (in_hosteq(inp->inp_faddr, faddr) && 1562 in_hosteq(inp->inp_laddr, laddr) && 1563 inp->inp_fport == fport && inp->inp_lport == lport) { 1564 /* found */ 1565 if (inp->inp_socket == NULL || 1566 inp->inp_socket->so_cred->cr_prison == NULL) { 1567 return (inp); 1568 } else { 1569 if (jinp == NULL) 1570 jinp = inp; 1571 } 1572 } 1573 } 1574 if (jinp != NULL) 1575 return (jinp); 1576 1577 if (wildcard) { 1578 struct inpcb *local_wild = NULL; 1579 struct inpcb *jinp_wild = NULL; 1580 struct inpcontainer *ic; 1581 struct inpcontainerhead *chead; 1582 struct sockaddr_in jsin; 1583 struct ucred *cred; 1584 1585 GET_PCBINFO_TOKEN(pcbinfo); 1586 1587 /* 1588 * Check local group first 1589 */ 1590 if (pcbinfo->localgrphashbase != NULL && 1591 m != NULL && (m->m_flags & M_HASH) && 1592 !(ifp && ifp->if_type == IFT_FAITH)) { 1593 inp = inp_localgroup_lookup(pcbinfo, 1594 laddr, lport, m->m_pkthdr.hash); 1595 if (inp != NULL) { 1596 REL_PCBINFO_TOKEN(pcbinfo); 1597 return inp; 1598 } 1599 } 1600 1601 /* 1602 * Order of socket selection: 1603 * 1. non-jailed, non-wild. 1604 * 2. non-jailed, wild. 1605 * 3. jailed, non-wild. 1606 * 4. jailed, wild. 1607 */ 1608 jsin.sin_family = AF_INET; 1609 chead = &pcbinfo->wildcardhashbase[ 1610 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)]; 1611 LIST_FOREACH(ic, chead, ic_list) { 1612 inp = ic->ic_inp; 1613 if (inp->inp_flags & INP_PLACEMARKER) 1614 continue; 1615 1616 jsin.sin_addr.s_addr = laddr.s_addr; 1617 #ifdef INET6 1618 if (!INP_ISIPV4(inp)) 1619 continue; 1620 #endif 1621 if (inp->inp_socket != NULL) 1622 cred = inp->inp_socket->so_cred; 1623 else 1624 cred = NULL; 1625 if (cred != NULL && jailed(cred)) { 1626 if (jinp != NULL) 1627 continue; 1628 else 1629 if (!jailed_ip(cred->cr_prison, 1630 (struct sockaddr *)&jsin)) 1631 continue; 1632 } 1633 if (inp->inp_lport == lport) { 1634 if (ifp && ifp->if_type == IFT_FAITH && 1635 !(inp->inp_flags & INP_FAITH)) 1636 continue; 1637 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1638 if (cred != NULL && jailed(cred)) { 1639 jinp = inp; 1640 } else { 1641 REL_PCBINFO_TOKEN(pcbinfo); 1642 return (inp); 1643 } 1644 } 1645 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1646 if (cred != NULL && jailed(cred)) 1647 jinp_wild = inp; 1648 else 1649 local_wild = inp; 1650 } 1651 } 1652 } 1653 1654 REL_PCBINFO_TOKEN(pcbinfo); 1655 1656 if (local_wild != NULL) 1657 return (local_wild); 1658 if (jinp != NULL) 1659 return (jinp); 1660 return (jinp_wild); 1661 } 1662 1663 /* 1664 * Not found. 1665 */ 1666 return (NULL); 1667 } 1668 1669 struct inpcb * 1670 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1671 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1672 boolean_t wildcard, struct ifnet *ifp) 1673 { 1674 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg, 1675 laddr, lport_arg, wildcard, ifp, NULL); 1676 } 1677 1678 /* 1679 * Insert PCB into connection hash table. 1680 */ 1681 void 1682 in_pcbinsconnhash(struct inpcb *inp) 1683 { 1684 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1685 struct inpcbhead *bucket; 1686 u_int32_t hashkey_faddr, hashkey_laddr; 1687 1688 #ifdef INET6 1689 if (INP_ISIPV6(inp)) { 1690 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */; 1691 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */; 1692 } else { 1693 #endif 1694 hashkey_faddr = inp->inp_faddr.s_addr; 1695 hashkey_laddr = inp->inp_laddr.s_addr; 1696 #ifdef INET6 1697 } 1698 #endif 1699 1700 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1701 ("not in the correct netisr")); 1702 ASSERT_INP_NOTINHASH(inp); 1703 inp->inp_flags |= INP_CONNECTED; 1704 1705 /* 1706 * Insert into the connection hash table. 1707 */ 1708 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr, 1709 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)]; 1710 LIST_INSERT_HEAD(bucket, inp, inp_hash); 1711 } 1712 1713 /* 1714 * Remove PCB from connection hash table. 1715 */ 1716 void 1717 in_pcbremconnhash(struct inpcb *inp) 1718 { 1719 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo; 1720 1721 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1722 ("not in the correct netisr")); 1723 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected")); 1724 1725 LIST_REMOVE(inp, inp_hash); 1726 inp->inp_flags &= ~INP_CONNECTED; 1727 } 1728 1729 /* 1730 * Insert PCB into port hash table. 1731 */ 1732 void 1733 in_pcbinsporthash(struct inpcbporthead *pcbporthash, struct inpcb *inp) 1734 { 1735 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1736 struct inpcbport *phd; 1737 1738 /* 1739 * If the porthashbase is shared across several cpus, it must 1740 * have been locked. 1741 */ 1742 ASSERT_PORTHASH_TOKEN_HELD(pcbporthash); 1743 1744 /* 1745 * Insert into the port hash table. 1746 */ 1747 1748 /* Go through port list and look for a head for this lport. */ 1749 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1750 if (phd->phd_port == inp->inp_lport) 1751 break; 1752 } 1753 1754 /* If none exists, use saved one and tack it on. */ 1755 if (phd == NULL) { 1756 KKASSERT(pcbinfo->portsave != NULL); 1757 phd = pcbinfo->portsave; 1758 pcbinfo->portsave = NULL; 1759 phd->phd_port = inp->inp_lport; 1760 LIST_INIT(&phd->phd_pcblist); 1761 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1762 } 1763 1764 inp->inp_porthash = pcbporthash; 1765 inp->inp_phd = phd; 1766 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1767 1768 /* 1769 * Malloc one inpcbport for later use. It is safe to use 1770 * "wait" malloc here (port token would be released, if 1771 * malloc ever blocked), since all changes to the porthash 1772 * are done. 1773 */ 1774 if (pcbinfo->portsave == NULL) { 1775 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), 1776 M_PCB, M_INTWAIT | M_ZERO); 1777 } 1778 } 1779 1780 void 1781 in_pcbinsporthash_lport(struct inpcb *inp) 1782 { 1783 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1784 struct inpcbportinfo *portinfo; 1785 struct inpcbporthead *porthash; 1786 u_short lport_ho; 1787 1788 /* Locate the proper portinfo based on lport */ 1789 lport_ho = ntohs(inp->inp_lport); 1790 portinfo = &pcbinfo->portinfo[lport_ho & pcbinfo->portinfo_mask]; 1791 KKASSERT((lport_ho & pcbinfo->portinfo_mask) == portinfo->offset); 1792 1793 porthash = in_pcbporthash_head(portinfo, inp->inp_lport); 1794 GET_PORTHASH_TOKEN(porthash); 1795 in_pcbinsporthash(porthash, inp); 1796 REL_PORTHASH_TOKEN(porthash); 1797 } 1798 1799 void 1800 in_pcbremporthash(struct inpcb *inp) 1801 { 1802 struct inpcbporthead *porthash; 1803 struct inpcbport *phd; 1804 1805 if (inp->inp_phd == NULL) 1806 return; 1807 KASSERT(inp->inp_lport != 0, ("inpcb has no lport")); 1808 1809 porthash = inp->inp_porthash; 1810 GET_PORTHASH_TOKEN(porthash); 1811 1812 phd = inp->inp_phd; 1813 LIST_REMOVE(inp, inp_portlist); 1814 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1815 LIST_REMOVE(phd, phd_hash); 1816 kfree(phd, M_PCB); 1817 } 1818 1819 REL_PORTHASH_TOKEN(porthash); 1820 1821 inp->inp_phd = NULL; 1822 /* NOTE: Don't whack inp_lport, which may be used later */ 1823 } 1824 1825 static struct inp_localgroup * 1826 inp_localgroup_alloc(u_char af, uint16_t port, 1827 const union in_dependaddr *addr, int size) 1828 { 1829 struct inp_localgroup *grp; 1830 1831 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]), 1832 M_TEMP, M_INTWAIT | M_ZERO); 1833 grp->il_af = af; 1834 grp->il_lport = port; 1835 grp->il_dependladdr = *addr; 1836 grp->il_inpsiz = size; 1837 1838 return grp; 1839 } 1840 1841 static void 1842 inp_localgroup_free(struct inp_localgroup *grp) 1843 { 1844 kfree(grp, M_TEMP); 1845 } 1846 1847 static void 1848 inp_localgroup_destroy(struct inp_localgroup *grp) 1849 { 1850 LIST_REMOVE(grp, il_list); 1851 inp_localgroup_free(grp); 1852 } 1853 1854 static void 1855 inp_localgroup_copy(struct inp_localgroup *grp, 1856 const struct inp_localgroup *old_grp) 1857 { 1858 int i; 1859 1860 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 1861 ("invalid new local group size %d and old local group count %d", 1862 grp->il_inpsiz, old_grp->il_inpcnt)); 1863 for (i = 0; i < old_grp->il_inpcnt; ++i) 1864 grp->il_inp[i] = old_grp->il_inp[i]; 1865 grp->il_inpcnt = old_grp->il_inpcnt; 1866 } 1867 1868 static void 1869 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 1870 { 1871 struct inp_localgrphead *hdr; 1872 struct inp_localgroup *grp, *grp_alloc = NULL; 1873 struct ucred *cred; 1874 int i, idx; 1875 1876 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 1877 1878 if (pcbinfo->localgrphashbase == NULL) 1879 return; 1880 1881 /* 1882 * XXX don't allow jailed socket to join local group 1883 */ 1884 if (inp->inp_socket != NULL) 1885 cred = inp->inp_socket->so_cred; 1886 else 1887 cred = NULL; 1888 if (cred != NULL && jailed(cred)) 1889 return; 1890 1891 hdr = &pcbinfo->localgrphashbase[ 1892 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1893 1894 again: 1895 LIST_FOREACH(grp, hdr, il_list) { 1896 if (grp->il_af == inp->inp_af && 1897 grp->il_lport == inp->inp_lport && 1898 memcmp(&grp->il_dependladdr, 1899 &inp->inp_inc.inc_ie.ie_dependladdr, 1900 sizeof(grp->il_dependladdr)) == 0) { 1901 break; 1902 } 1903 } 1904 if (grp == NULL) { 1905 /* 1906 * Create a new local group 1907 */ 1908 if (grp_alloc == NULL) { 1909 grp_alloc = inp_localgroup_alloc(inp->inp_af, 1910 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 1911 INP_LOCALGROUP_SIZMIN); 1912 /* 1913 * Local group allocation could block and the 1914 * local group w/ the same property might have 1915 * been added by others when we were blocked; 1916 * check again. 1917 */ 1918 goto again; 1919 } else { 1920 /* Local group has been allocated; link it */ 1921 grp = grp_alloc; 1922 grp_alloc = NULL; 1923 LIST_INSERT_HEAD(hdr, grp, il_list); 1924 } 1925 } else if (grp->il_inpcnt == grp->il_inpsiz) { 1926 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) { 1927 static int limit_logged = 0; 1928 1929 if (!limit_logged) { 1930 limit_logged = 1; 1931 kprintf("local group port %d, " 1932 "limit reached\n", ntohs(grp->il_lport)); 1933 } 1934 if (grp_alloc != NULL) { 1935 /* 1936 * This would happen if the local group 1937 * w/ the same property was expanded when 1938 * our local group allocation blocked. 1939 */ 1940 inp_localgroup_free(grp_alloc); 1941 } 1942 return; 1943 } 1944 1945 /* 1946 * Expand this local group 1947 */ 1948 if (grp_alloc == NULL || 1949 grp->il_inpcnt >= grp_alloc->il_inpsiz) { 1950 if (grp_alloc != NULL) 1951 inp_localgroup_free(grp_alloc); 1952 grp_alloc = inp_localgroup_alloc(grp->il_af, 1953 grp->il_lport, &grp->il_dependladdr, 1954 grp->il_inpsiz * 2); 1955 /* 1956 * Local group allocation could block and the 1957 * local group w/ the same property might have 1958 * been expanded by others when we were blocked; 1959 * check again. 1960 */ 1961 goto again; 1962 } 1963 1964 /* 1965 * Save the old local group, link the new one, and then 1966 * destroy the old local group 1967 */ 1968 inp_localgroup_copy(grp_alloc, grp); 1969 LIST_INSERT_HEAD(hdr, grp_alloc, il_list); 1970 inp_localgroup_destroy(grp); 1971 1972 grp = grp_alloc; 1973 grp_alloc = NULL; 1974 } else { 1975 /* 1976 * Found the local group 1977 */ 1978 if (grp_alloc != NULL) { 1979 /* 1980 * This would happen if the local group w/ the 1981 * same property was added or expanded when our 1982 * local group allocation blocked. 1983 */ 1984 inp_localgroup_free(grp_alloc); 1985 grp_alloc = NULL; 1986 } 1987 } 1988 1989 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 1990 ("invalid local group size %d and count %d", 1991 grp->il_inpsiz, grp->il_inpcnt)); 1992 1993 /* 1994 * Keep the local group sorted by the inpcb local group index 1995 * in ascending order. 1996 * 1997 * This eases the multi-process userland application which uses 1998 * SO_REUSEPORT sockets and binds process to the owner cpu of 1999 * the SO_REUSEPORT socket: 2000 * If we didn't sort the local group by the inpcb local group 2001 * index and one of the process owning an inpcb in this local 2002 * group restarted, e.g. crashed and restarted by watchdog, 2003 * other processes owning a inpcb in this local group would have 2004 * to detect that event, refetch its socket's owner cpu, and 2005 * re-bind. 2006 */ 2007 idx = grp->il_inpcnt; 2008 for (i = 0; i < idx; ++i) { 2009 struct inpcb *oinp = grp->il_inp[i]; 2010 2011 if (oinp->inp_lgrpindex > i) { 2012 if (inp->inp_lgrpindex < 0) { 2013 inp->inp_lgrpindex = i; 2014 } else if (inp->inp_lgrpindex != i) { 2015 if (bootverbose) { 2016 kprintf("inp %p: grpidx %d, " 2017 "assigned to %d, cpu%d\n", 2018 inp, inp->inp_lgrpindex, i, 2019 mycpuid); 2020 } 2021 } 2022 grp->il_inp[i] = inp; 2023 2024 /* Pull down inpcbs */ 2025 for (; i < grp->il_inpcnt; ++i) { 2026 struct inpcb *oinp1 = grp->il_inp[i + 1]; 2027 2028 grp->il_inp[i + 1] = oinp; 2029 oinp = oinp1; 2030 } 2031 grp->il_inpcnt++; 2032 return; 2033 } 2034 } 2035 2036 if (inp->inp_lgrpindex < 0) { 2037 inp->inp_lgrpindex = idx; 2038 } else if (inp->inp_lgrpindex != idx) { 2039 if (bootverbose) { 2040 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n", 2041 inp, inp->inp_lgrpindex, idx, mycpuid); 2042 } 2043 } 2044 grp->il_inp[idx] = inp; 2045 grp->il_inpcnt++; 2046 } 2047 2048 void 2049 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2050 { 2051 struct inpcontainer *ic; 2052 struct inpcontainerhead *bucket; 2053 2054 GET_PCBINFO_TOKEN(pcbinfo); 2055 2056 in_pcbinslocalgrphash_oncpu(inp, pcbinfo); 2057 2058 bucket = &pcbinfo->wildcardhashbase[ 2059 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 2060 2061 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT); 2062 ic->ic_inp = inp; 2063 LIST_INSERT_HEAD(bucket, ic, ic_list); 2064 2065 REL_PCBINFO_TOKEN(pcbinfo); 2066 } 2067 2068 /* 2069 * Insert PCB into wildcard hash table. 2070 */ 2071 void 2072 in_pcbinswildcardhash(struct inpcb *inp) 2073 { 2074 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2075 2076 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 2077 ("not in correct netisr")); 2078 ASSERT_INP_NOTINHASH(inp); 2079 inp->inp_flags |= INP_WILDCARD; 2080 2081 in_pcbinswildcardhash_oncpu(inp, pcbinfo); 2082 } 2083 2084 static void 2085 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2086 { 2087 struct inp_localgrphead *hdr; 2088 struct inp_localgroup *grp; 2089 2090 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 2091 2092 if (pcbinfo->localgrphashbase == NULL) 2093 return; 2094 2095 hdr = &pcbinfo->localgrphashbase[ 2096 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 2097 2098 LIST_FOREACH(grp, hdr, il_list) { 2099 int i; 2100 2101 for (i = 0; i < grp->il_inpcnt; ++i) { 2102 if (grp->il_inp[i] != inp) 2103 continue; 2104 2105 if (grp->il_inpcnt == 1) { 2106 /* Destroy this local group */ 2107 inp_localgroup_destroy(grp); 2108 } else { 2109 /* Pull up inpcbs */ 2110 for (; i + 1 < grp->il_inpcnt; ++i) 2111 grp->il_inp[i] = grp->il_inp[i + 1]; 2112 grp->il_inpcnt--; 2113 } 2114 return; 2115 } 2116 } 2117 } 2118 2119 void 2120 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2121 { 2122 struct inpcontainer *ic; 2123 struct inpcontainerhead *head; 2124 2125 GET_PCBINFO_TOKEN(pcbinfo); 2126 2127 in_pcbremlocalgrphash_oncpu(inp, pcbinfo); 2128 2129 /* find bucket */ 2130 head = &pcbinfo->wildcardhashbase[ 2131 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 2132 2133 LIST_FOREACH(ic, head, ic_list) { 2134 if (ic->ic_inp == inp) 2135 goto found; 2136 } 2137 REL_PCBINFO_TOKEN(pcbinfo); 2138 return; /* not found! */ 2139 2140 found: 2141 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */ 2142 REL_PCBINFO_TOKEN(pcbinfo); 2143 kfree(ic, M_TEMP); /* deallocate container */ 2144 } 2145 2146 /* 2147 * Remove PCB from wildcard hash table. 2148 */ 2149 void 2150 in_pcbremwildcardhash(struct inpcb *inp) 2151 { 2152 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2153 2154 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 2155 ("not in correct netisr")); 2156 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard")); 2157 2158 in_pcbremwildcardhash_oncpu(inp, pcbinfo); 2159 inp->inp_lgrpindex = -1; 2160 inp->inp_flags &= ~INP_WILDCARD; 2161 } 2162 2163 /* 2164 * Remove PCB from various lists. 2165 */ 2166 void 2167 in_pcbremlists(struct inpcb *inp) 2168 { 2169 in_pcbremporthash(inp); 2170 if (inp->inp_flags & INP_WILDCARD) { 2171 in_pcbremwildcardhash(inp); 2172 } else if (inp->inp_flags & INP_CONNECTED) { 2173 in_pcbremconnhash(inp); 2174 } 2175 2176 if (inp->inp_flags & INP_ONLIST) 2177 in_pcbofflist(inp); 2178 } 2179 2180 int 2181 prison_xinpcb(struct thread *td, struct inpcb *inp) 2182 { 2183 struct ucred *cr; 2184 2185 if (td->td_proc == NULL) 2186 return (0); 2187 cr = td->td_proc->p_ucred; 2188 if (cr->cr_prison == NULL) 2189 return (0); 2190 if (inp->inp_socket && inp->inp_socket->so_cred && 2191 inp->inp_socket->so_cred->cr_prison && 2192 cr->cr_prison == inp->inp_socket->so_cred->cr_prison) 2193 return (0); 2194 return (1); 2195 } 2196 2197 int 2198 in_pcblist_global(SYSCTL_HANDLER_ARGS) 2199 { 2200 struct inpcbinfo *pcbinfo_arr = arg1; 2201 int pcbinfo_arrlen = arg2; 2202 struct inpcb *marker; 2203 int cpu, origcpu; 2204 int error, n; 2205 2206 KASSERT(pcbinfo_arrlen <= ncpus && pcbinfo_arrlen >= 1, 2207 ("invalid pcbinfo count %d", pcbinfo_arrlen)); 2208 2209 /* 2210 * The process of preparing the TCB list is too time-consuming and 2211 * resource-intensive to repeat twice on every request. 2212 */ 2213 n = 0; 2214 if (req->oldptr == NULL) { 2215 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu) 2216 n += pcbinfo_arr[cpu].ipi_count; 2217 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 2218 return 0; 2219 } 2220 2221 if (req->newptr != NULL) 2222 return EPERM; 2223 2224 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 2225 marker->inp_flags |= INP_PLACEMARKER; 2226 2227 /* 2228 * OK, now we're committed to doing something. Re-fetch ipi_count 2229 * after obtaining the generation count. 2230 */ 2231 error = 0; 2232 origcpu = mycpuid; 2233 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) { 2234 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu]; 2235 struct inpcb *inp; 2236 struct xinpcb xi; 2237 int i; 2238 2239 lwkt_migratecpu(cpu); 2240 2241 GET_PCBINFO_TOKEN(pcbinfo); 2242 2243 n = pcbinfo->ipi_count; 2244 2245 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 2246 i = 0; 2247 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 2248 LIST_REMOVE(marker, inp_list); 2249 LIST_INSERT_AFTER(inp, marker, inp_list); 2250 2251 if (inp->inp_flags & INP_PLACEMARKER) 2252 continue; 2253 if (prison_xinpcb(req->td, inp)) 2254 continue; 2255 2256 bzero(&xi, sizeof xi); 2257 xi.xi_len = sizeof xi; 2258 bcopy(inp, &xi.xi_inp, sizeof *inp); 2259 if (inp->inp_socket) 2260 sotoxsocket(inp->inp_socket, &xi.xi_socket); 2261 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0) 2262 break; 2263 ++i; 2264 } 2265 LIST_REMOVE(marker, inp_list); 2266 2267 REL_PCBINFO_TOKEN(pcbinfo); 2268 2269 if (error == 0 && i < n) { 2270 bzero(&xi, sizeof xi); 2271 xi.xi_len = sizeof xi; 2272 while (i < n) { 2273 error = SYSCTL_OUT(req, &xi, sizeof xi); 2274 if (error) 2275 break; 2276 ++i; 2277 } 2278 } 2279 } 2280 2281 lwkt_migratecpu(origcpu); 2282 kfree(marker, M_TEMP); 2283 return error; 2284 } 2285 2286 int 2287 in_pcblist_global_ncpus2(SYSCTL_HANDLER_ARGS) 2288 { 2289 return in_pcblist_global(oidp, arg1, ncpus2, req); 2290 } 2291 2292 void 2293 in_savefaddr(struct socket *so, const struct sockaddr *faddr) 2294 { 2295 struct sockaddr_in *sin; 2296 2297 KASSERT(faddr->sa_family == AF_INET, 2298 ("not AF_INET faddr %d", faddr->sa_family)); 2299 2300 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO); 2301 sin->sin_family = AF_INET; 2302 sin->sin_len = sizeof(*sin); 2303 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port; 2304 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr; 2305 2306 so->so_faddr = (struct sockaddr *)sin; 2307 } 2308 2309 void 2310 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize, 2311 u_short offset) 2312 { 2313 memset(portinfo, 0, sizeof(*portinfo)); 2314 2315 portinfo->offset = offset; 2316 portinfo->lastport = offset; 2317 portinfo->lastlow = offset; 2318 portinfo->lasthi = offset; 2319 2320 portinfo->porthashbase = hashinit(hashsize, M_PCB, 2321 &portinfo->porthashmask); 2322 } 2323 2324 void 2325 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step) 2326 { 2327 int hi, lo; 2328 2329 if (step == 1) 2330 return; 2331 2332 hi = *hi0; 2333 lo = *lo0; 2334 2335 hi = rounddown2(hi, step); 2336 hi += ofs; 2337 if (hi > (int)*hi0) 2338 hi -= step; 2339 2340 lo = roundup2(lo, step); 2341 lo -= (step - ofs); 2342 if (lo < (int)*lo0) 2343 lo += step; 2344 2345 *hi0 = hi; 2346 *lo0 = lo; 2347 } 2348 2349 void 2350 in_pcbglobalinit(void) 2351 { 2352 int cpu; 2353 2354 in_pcbmarkers = kmalloc(ncpus * sizeof(struct inpcb), M_PCB, 2355 M_WAITOK | M_ZERO); 2356 in_pcbcontainer_markers = kmalloc(ncpus * sizeof(struct inpcontainer), 2357 M_PCB, M_WAITOK | M_ZERO); 2358 2359 for (cpu = 0; cpu < ncpus; ++cpu) { 2360 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu]; 2361 struct inpcb *marker = &in_pcbmarkers[cpu]; 2362 2363 marker->inp_flags |= INP_PLACEMARKER; 2364 ic->ic_inp = marker; 2365 } 2366 } 2367 2368 struct inpcb * 2369 in_pcbmarker(int cpuid) 2370 { 2371 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid)); 2372 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 2373 2374 return &in_pcbmarkers[cpuid]; 2375 } 2376 2377 struct inpcontainer * 2378 in_pcbcontainer_marker(int cpuid) 2379 { 2380 KASSERT(cpuid >= 0 && cpuid < ncpus, ("invalid cpuid %d", cpuid)); 2381 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 2382 2383 return &in_pcbcontainer_markers[cpuid]; 2384 } 2385 2386 void 2387 in_pcbresetroute(struct inpcb *inp) 2388 { 2389 struct route *ro = &inp->inp_route; 2390 2391 if (ro->ro_rt != NULL) 2392 RTFREE(ro->ro_rt); 2393 bzero(ro, sizeof(*ro)); 2394 } 2395 2396 u_short 2397 in_pcblastport_down(volatile u_short *lastport, u_short first, u_short last, 2398 u_short step) 2399 { 2400 u_short lport; 2401 2402 for (;;) { 2403 u_short olport; 2404 2405 olport = *lastport; 2406 lport = olport - step; 2407 if (__predict_false(lport > first || lport < last)) 2408 lport = first; 2409 if (atomic_cmpset_short(lastport, olport, lport)) 2410 break; 2411 } 2412 return lport; 2413 } 2414 2415 u_short 2416 in_pcblastport_up(volatile u_short *lastport, u_short first, u_short last, 2417 u_short step) 2418 { 2419 u_short lport; 2420 2421 for (;;) { 2422 u_short olport; 2423 2424 olport = *lastport; 2425 lport = olport + step; 2426 if (__predict_false(lport < first || lport > last)) 2427 lport = first; 2428 if (atomic_cmpset_short(lastport, olport, lport)) 2429 break; 2430 } 2431 return lport; 2432 } 2433