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_inet6.h" 67 68 #include <sys/param.h> 69 #include <sys/systm.h> 70 #include <sys/malloc.h> 71 #include <sys/mbuf.h> 72 #include <sys/domain.h> 73 #include <sys/protosw.h> 74 #include <sys/socket.h> 75 #include <sys/socketvar.h> 76 #include <sys/proc.h> 77 #include <sys/caps.h> 78 #include <sys/jail.h> 79 #include <sys/kernel.h> 80 #include <sys/sysctl.h> 81 82 #include <sys/socketvar2.h> 83 #include <sys/msgport2.h> 84 85 #include <machine/limits.h> 86 87 #include <net/if.h> 88 #include <net/if_types.h> 89 #include <net/route.h> 90 #include <net/netisr2.h> 91 #include <net/toeplitz2.h> 92 93 #include <netinet/in.h> 94 #include <netinet/in_pcb.h> 95 #include <netinet/in_var.h> 96 #include <netinet/ip_var.h> 97 #ifdef INET6 98 #include <netinet/ip6.h> 99 #include <netinet6/ip6_var.h> 100 #endif /* INET6 */ 101 102 #define INP_LOCALGROUP_SIZMIN 8 103 #define INP_LOCALGROUP_SIZMAX 256 104 105 static struct inpcb *in_pcblookup_local(struct inpcbporthead *porthash, 106 struct in_addr laddr, u_int lport_arg, int wild_okay, 107 struct ucred *cred); 108 109 struct in_addr zeroin_addr; 110 111 /* 112 * These configure the range of local port addresses assigned to 113 * "unspecified" outgoing connections/packets/whatever. 114 */ 115 int ipport_lowfirstauto = IPPORT_RESERVED - 1; /* 1023 */ 116 int ipport_lowlastauto = IPPORT_RESERVEDSTART; /* 600 */ 117 118 int ipport_firstauto = IPPORT_RESERVED; /* 1024 */ 119 int ipport_lastauto = IPPORT_USERRESERVED; /* 5000 */ 120 121 int ipport_hifirstauto = IPPORT_HIFIRSTAUTO; /* 49152 */ 122 int ipport_hilastauto = IPPORT_HILASTAUTO; /* 65535 */ 123 124 #define RANGECHK(var, min, max) \ 125 if ((var) < (min)) { (var) = (min); } \ 126 else if ((var) > (max)) { (var) = (max); } 127 128 int udpencap_enable = 1; /* enabled by default */ 129 int udpencap_port = 4500; /* triggers decapsulation */ 130 131 /* 132 * Per-netisr inpcb markers. 133 * NOTE: they should only be used in netisrs. 134 */ 135 static struct inpcb *in_pcbmarkers; 136 static struct inpcontainer *in_pcbcontainer_markers; 137 138 static int 139 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 140 { 141 int error; 142 143 error = sysctl_handle_int(oidp, oidp->oid_arg1, oidp->oid_arg2, req); 144 if (!error) { 145 RANGECHK(ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 146 RANGECHK(ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 147 148 RANGECHK(ipport_firstauto, IPPORT_RESERVED, USHRT_MAX); 149 RANGECHK(ipport_lastauto, IPPORT_RESERVED, USHRT_MAX); 150 151 RANGECHK(ipport_hifirstauto, IPPORT_RESERVED, USHRT_MAX); 152 RANGECHK(ipport_hilastauto, IPPORT_RESERVED, USHRT_MAX); 153 } 154 return (error); 155 } 156 157 #undef RANGECHK 158 159 SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); 160 161 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLTYPE_INT|CTLFLAG_RW, 162 &ipport_lowfirstauto, 0, &sysctl_net_ipport_check, "I", ""); 163 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLTYPE_INT|CTLFLAG_RW, 164 &ipport_lowlastauto, 0, &sysctl_net_ipport_check, "I", ""); 165 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLTYPE_INT|CTLFLAG_RW, 166 &ipport_firstauto, 0, &sysctl_net_ipport_check, "I", ""); 167 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLTYPE_INT|CTLFLAG_RW, 168 &ipport_lastauto, 0, &sysctl_net_ipport_check, "I", ""); 169 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLTYPE_INT|CTLFLAG_RW, 170 &ipport_hifirstauto, 0, &sysctl_net_ipport_check, "I", ""); 171 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLTYPE_INT|CTLFLAG_RW, 172 &ipport_hilastauto, 0, &sysctl_net_ipport_check, "I", ""); 173 174 /* Initialized by ip_init() */ 175 int ip_porthash_trycount; 176 SYSCTL_INT(_net_inet_ip, OID_AUTO, porthash_trycount, CTLFLAG_RW, 177 &ip_porthash_trycount, 0, 178 "Number of tries to find local port matching hash of 4-tuple"); 179 180 /* 181 * in_pcb.c: manage the Protocol Control Blocks. 182 * 183 * NOTE: It is assumed that most of these functions will be called from 184 * a critical section. XXX - There are, unfortunately, a few exceptions 185 * to this rule that should be fixed. 186 * 187 * NOTE: The caller should initialize the cpu field to the cpu running the 188 * protocol stack associated with this inpcbinfo. 189 */ 190 191 void 192 in_pcbinfo_init(struct inpcbinfo *pcbinfo, int cpu, boolean_t shared) 193 { 194 KASSERT(cpu >= 0 && cpu < netisr_ncpus, ("invalid cpu%d", cpu)); 195 pcbinfo->cpu = cpu; 196 197 LIST_INIT(&pcbinfo->pcblisthead); 198 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), M_PCB, 199 M_WAITOK | M_ZERO); 200 201 if (shared) { 202 pcbinfo->infotoken = kmalloc(sizeof(struct lwkt_token), 203 M_PCB, M_WAITOK); 204 lwkt_token_init(pcbinfo->infotoken, "infotoken"); 205 } else { 206 pcbinfo->infotoken = NULL; 207 } 208 } 209 210 void 211 in_pcbportinfo_set(struct inpcbinfo *pcbinfo, struct inpcbportinfo *portinfo, 212 int portinfo_cnt) 213 { 214 215 KASSERT(portinfo_cnt > 0, ("invalid portinfo_cnt %d", portinfo_cnt)); 216 pcbinfo->portinfo = portinfo; 217 pcbinfo->portinfo_cnt = portinfo_cnt; 218 } 219 220 struct baddynamicports baddynamicports; 221 222 /* 223 * Check if the specified port is invalid for dynamic allocation. 224 */ 225 int 226 in_baddynamic(u_int16_t port, u_int16_t proto) 227 { 228 switch (proto) { 229 case IPPROTO_TCP: 230 return (DP_ISSET(baddynamicports.tcp, port)); 231 case IPPROTO_UDP: 232 return (DP_ISSET(baddynamicports.udp, port)); 233 default: 234 return (0); 235 } 236 } 237 238 void 239 in_pcbonlist(struct inpcb *inp) 240 { 241 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 242 243 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 244 ("not in the correct netisr")); 245 KASSERT((inp->inp_flags & INP_ONLIST) == 0, ("already on pcblist")); 246 inp->inp_flags |= INP_ONLIST; 247 248 GET_PCBINFO_TOKEN(pcbinfo); 249 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, inp, inp_list); 250 pcbinfo->ipi_count++; 251 REL_PCBINFO_TOKEN(pcbinfo); 252 } 253 254 void 255 in_pcbofflist(struct inpcb *inp) 256 { 257 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 258 259 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 260 ("not in the correct netisr")); 261 KASSERT(inp->inp_flags & INP_ONLIST, ("not on pcblist")); 262 inp->inp_flags &= ~INP_ONLIST; 263 264 GET_PCBINFO_TOKEN(pcbinfo); 265 LIST_REMOVE(inp, inp_list); 266 KASSERT(pcbinfo->ipi_count > 0, 267 ("invalid inpcb count %d", pcbinfo->ipi_count)); 268 pcbinfo->ipi_count--; 269 REL_PCBINFO_TOKEN(pcbinfo); 270 } 271 272 /* 273 * Allocate a PCB and associate it with the socket. 274 */ 275 int 276 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 277 { 278 struct inpcb *inp; 279 280 inp = kmalloc(pcbinfo->ipi_size, M_PCB, M_WAITOK|M_ZERO|M_NULLOK); 281 if (inp == NULL) 282 return (ENOMEM); 283 inp->inp_lgrpindex = -1; 284 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 285 inp->inp_pcbinfo = pcbinfo; 286 inp->inp_socket = so; 287 #ifdef INET6 288 if (INP_CHECK_SOCKAF(so, AF_INET6)) { 289 if (ip6_auto_flowlabel) 290 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 291 inp->inp_af = AF_INET6; 292 } else 293 #endif 294 inp->inp_af = AF_INET; 295 soreference(so); 296 so->so_pcb = inp; 297 298 in_pcbonlist(inp); 299 return (0); 300 } 301 302 /* 303 * Unlink a pcb with the intention of moving it to another cpu with a 304 * different pcbinfo. While unlinked nothing should attempt to dereference 305 * inp_pcbinfo, NULL it out so we assert if it does. 306 */ 307 void 308 in_pcbunlink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags) 309 { 310 KASSERT(inp->inp_pcbinfo == pcbinfo, ("pcbinfo mismatch")); 311 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0, 312 ("already linked")); 313 314 in_pcbofflist(inp); 315 inp->inp_pcbinfo = NULL; 316 } 317 318 void 319 in_pcbunlink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 320 { 321 in_pcbunlink_flags(inp, pcbinfo, INP_WILDCARD); 322 } 323 324 /* 325 * Relink a pcb into a new pcbinfo. 326 */ 327 void 328 in_pcblink_flags(struct inpcb *inp, struct inpcbinfo *pcbinfo, int flags) 329 { 330 KASSERT(inp->inp_pcbinfo == NULL, ("has pcbinfo")); 331 KASSERT((inp->inp_flags & (flags | INP_CONNECTED)) == 0, 332 ("already linked")); 333 334 inp->inp_pcbinfo = pcbinfo; 335 in_pcbonlist(inp); 336 } 337 338 void 339 in_pcblink(struct inpcb *inp, struct inpcbinfo *pcbinfo) 340 { 341 return in_pcblink_flags(inp, pcbinfo, INP_WILDCARD); 342 } 343 344 static boolean_t 345 in_pcbporthash_update(struct inpcbportinfo *portinfo, 346 struct inpcb *inp, u_short lport, struct ucred *cred, int wild) 347 { 348 struct inpcbporthead *porthash; 349 350 /* 351 * This has to be atomic. If the porthash is shared across multiple 352 * protocol threads, e.g. tcp and udp, then the token must be held. 353 */ 354 porthash = in_pcbporthash_head(portinfo, lport); 355 GET_PORTHASH_TOKEN(porthash); 356 357 if (in_pcblookup_local(porthash, inp->inp_laddr, lport, wild, cred)) { 358 REL_PORTHASH_TOKEN(porthash); 359 return FALSE; 360 } 361 inp->inp_lport = lport; 362 in_pcbinsporthash(porthash, inp); 363 364 REL_PORTHASH_TOKEN(porthash); 365 return TRUE; 366 } 367 368 static int 369 in_pcbsetlport(struct inpcb *inp, int wild, struct ucred *cred) 370 { 371 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 372 struct inpcbportinfo *portinfo; 373 u_short first, last, lport, step, first0, last0; 374 int count, error; 375 int portinfo_first, portinfo_idx; 376 uint32_t cut; 377 378 inp->inp_flags |= INP_ANONPORT; 379 380 step = pcbinfo->portinfo_cnt; 381 portinfo_first = mycpuid % pcbinfo->portinfo_cnt; 382 portinfo_idx = portinfo_first; 383 384 if (inp->inp_flags & INP_HIGHPORT) { 385 first0 = ipport_hifirstauto; /* sysctl */ 386 last0 = ipport_hilastauto; 387 } else if (inp->inp_flags & INP_LOWPORT) { 388 if (cred && 389 (error = caps_priv_check(cred, SYSCAP_NONET_RESPORT))) 390 { 391 inp->inp_laddr.s_addr = INADDR_ANY; 392 return error; 393 } 394 first0 = ipport_lowfirstauto; /* 1023 */ 395 last0 = ipport_lowlastauto; /* 600 */ 396 } else { 397 first0 = ipport_firstauto; /* sysctl */ 398 last0 = ipport_lastauto; 399 } 400 if (first0 > last0) { 401 lport = last0; 402 last0 = first0; 403 first0 = lport; 404 } 405 KKASSERT(last0 >= first0); 406 407 cut = karc4random(); 408 loop: 409 portinfo = &pcbinfo->portinfo[portinfo_idx]; 410 first = first0; 411 last = last0; 412 413 /* 414 * Simple check to ensure all ports are not used up causing 415 * a deadlock here. 416 */ 417 in_pcbportrange(&last, &first, portinfo->offset, step); 418 lport = last - first; 419 count = lport / step; 420 421 lport = rounddown(cut % lport, step) + first; 422 KKASSERT(lport % step == portinfo->offset); 423 424 for (;;) { 425 if (count-- < 0) { /* completely used? */ 426 error = EADDRNOTAVAIL; 427 break; 428 } 429 430 if (__predict_false(lport < first || lport > last)) { 431 lport = first; 432 KKASSERT(lport % step == portinfo->offset); 433 } 434 435 if (in_pcbporthash_update(portinfo, inp, htons(lport), 436 cred, wild)) { 437 error = 0; 438 break; 439 } 440 441 lport += step; 442 KKASSERT(lport % step == portinfo->offset); 443 } 444 445 if (error) { 446 /* Try next portinfo */ 447 portinfo_idx++; 448 portinfo_idx %= pcbinfo->portinfo_cnt; 449 if (portinfo_idx != portinfo_first) 450 goto loop; 451 inp->inp_laddr.s_addr = INADDR_ANY; 452 } 453 return error; 454 } 455 456 static __inline struct inpcbporthead * 457 OBTAIN_LPORTHASH_TOKEN(struct inpcbinfo *pcbinfo, u_short lport) 458 { 459 struct inpcbportinfo *portinfo; 460 struct inpcbporthead *porthash; 461 u_short lport_ho = ntohs(lport); 462 463 /* 464 * Locate the proper portinfo based on lport 465 */ 466 portinfo = &pcbinfo->portinfo[lport_ho % pcbinfo->portinfo_cnt]; 467 KKASSERT((lport_ho % pcbinfo->portinfo_cnt) == portinfo->offset); 468 469 porthash = in_pcbporthash_head(portinfo, lport); 470 GET_PORTHASH_TOKEN(porthash); 471 472 return porthash; 473 } 474 475 static int 476 in_pcbbind_laddr(struct sockaddr_in *sin, struct in_addr *laddr, 477 struct thread *td) 478 { 479 struct sockaddr_in jsin; 480 481 if (!prison_replace_wildcards(td, (struct sockaddr *)sin)) 482 return (EINVAL); 483 484 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 485 sin->sin_addr.s_addr != INADDR_ANY) { 486 sin->sin_port = 0; /* yech... */ 487 bzero(&sin->sin_zero, sizeof sin->sin_zero); 488 if (ifa_ifwithaddr((struct sockaddr *)sin) == NULL) 489 return (EADDRNOTAVAIL); 490 } 491 492 *laddr = sin->sin_addr; 493 494 jsin.sin_family = AF_INET; 495 jsin.sin_addr.s_addr = laddr->s_addr; 496 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 497 laddr->s_addr = INADDR_ANY; 498 return (EINVAL); 499 } 500 laddr->s_addr = jsin.sin_addr.s_addr; 501 502 return (0); 503 } 504 505 static int 506 in_pcbbind_laddrport_check(const struct socket *so, struct sockaddr_in *sin, 507 struct inpcbporthead *porthash, int wild, struct ucred *cred, 508 struct thread *td) 509 { 510 int reuseport = (so->so_options & SO_REUSEPORT); 511 struct inpcb *t; 512 513 ASSERT_PORTHASH_TOKEN_HELD(porthash); 514 515 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 516 /* 517 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 518 * allow complete duplication of binding if 519 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 520 * and a multicast address is bound on both 521 * new and duplicated sockets. 522 */ 523 if (so->so_options & SO_REUSEADDR) 524 reuseport = SO_REUSEADDR | SO_REUSEPORT; 525 } 526 527 if (so->so_cred->cr_uid != 0 && 528 !IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 529 t = in_pcblookup_local(porthash, sin->sin_addr, sin->sin_port, 530 INPLOOKUP_WILDCARD, cred); 531 if (t && t->inp_socket != so && 532 (so->so_cred->cr_uid != t->inp_socket->so_cred->cr_uid)) 533 return (EADDRINUSE); 534 } 535 if (cred && !prison_replace_wildcards(td, (struct sockaddr *)sin)) 536 return (EADDRNOTAVAIL); 537 538 /* 539 * When binding to a local port if the best match is against 540 * an accepted socket we generally want to allow the binding. 541 * This means that there is no longer any specific socket 542 * bound or bound for listening. 543 */ 544 t = in_pcblookup_local(porthash, sin->sin_addr, sin->sin_port, 545 wild, cred); 546 if (t && t->inp_socket != so && 547 (reuseport & t->inp_socket->so_options) == 0 && 548 (t->inp_socket->so_state & SS_ACCEPTMECH) == 0) 549 return (EADDRINUSE); 550 551 return (0); 552 } 553 554 int 555 in_pcbsrcaddr_check(const struct inpcb *inp, struct sockaddr_in *sin, 556 struct in_addr *laddr, struct thread *td) 557 { 558 const struct socket *so = inp->inp_socket; 559 struct inpcbporthead *porthash; 560 struct ucred *cred = NULL; 561 int wild = 0; 562 int error; 563 564 /* inp must be bound beforehand. */ 565 KKASSERT(inp->inp_lport != 0); 566 KKASSERT(sin->sin_len == sizeof(*sin)); 567 568 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT))) 569 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */ 570 if (td->td_proc) 571 cred = td->td_proc->p_ucred; 572 573 /* Always use inp_lport */ 574 sin->sin_port = inp->inp_lport; 575 576 error = in_pcbbind_laddr(sin, laddr, td); 577 if (error) 578 return (error); 579 580 if (IN_MULTICAST(ntohl(laddr->s_addr))) { 581 /* Unlike bind, multicast src address is not allowed. */ 582 return (EINVAL); 583 } 584 585 if (inp->inp_laddr.s_addr == laddr->s_addr) { 586 /* 587 * src address is same as what we bound to. 588 * 589 * inp_laddr == INADDR_ANY && srcaddr == INADDR_ANY 590 * is allowed, which does not really matter. 591 */ 592 return (0); 593 } else if (inp->inp_laddr.s_addr != INADDR_ANY && 594 !IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) { 595 /* Already bound to a specific address */ 596 return (EINVAL); 597 } 598 599 /* 600 * This has to be atomic. If the porthash is shared across 601 * multiple protocol threads, e.g. tcp and udp then the token 602 * must be held. 603 */ 604 porthash = OBTAIN_LPORTHASH_TOKEN(inp->inp_pcbinfo, inp->inp_lport); 605 606 /* 607 * Restore the sin_port whacked by in_pcbbind_ladddr(); 608 * sin->sin_port is checked by in_pcbbind_laddrport_check(). 609 */ 610 sin->sin_port = inp->inp_lport; 611 612 error = in_pcbbind_laddrport_check(so, sin, porthash, wild, cred, td); 613 if (error) 614 laddr->s_addr = INADDR_ANY; 615 616 REL_PORTHASH_TOKEN(porthash); 617 return (error); 618 } 619 620 int 621 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 622 { 623 const struct socket *so = inp->inp_socket; 624 struct ucred *cred = NULL; 625 int wild = 0; 626 627 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 628 return (EINVAL); /* already bound */ 629 630 if (!(so->so_options & (SO_REUSEADDR|SO_REUSEPORT))) 631 wild = 1; /* neither SO_REUSEADDR nor SO_REUSEPORT is set */ 632 if (td->td_proc) 633 cred = td->td_proc->p_ucred; 634 635 if (nam != NULL) { 636 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 637 struct inpcbporthead *porthash; 638 u_short lport; 639 int error; 640 641 if (nam->sa_len != sizeof *sin) 642 return (EINVAL); 643 #ifdef notdef 644 /* 645 * We should check the family, but old programs 646 * incorrectly fail to initialize it. 647 */ 648 if (sin->sin_family != AF_INET) 649 return (EAFNOSUPPORT); 650 #endif 651 652 /* 653 * Save sin_port for later use, since it will 654 * be whacked by in_pcbbind_laddr(). 655 */ 656 lport = sin->sin_port; 657 658 error = in_pcbbind_laddr(sin, &inp->inp_laddr, td); 659 if (error) 660 return (error); 661 662 if (lport == 0) { 663 /* Auto-select local port */ 664 return in_pcbsetlport(inp, wild, cred); 665 } 666 667 /* GROSS */ 668 if (ntohs(lport) < IPPORT_RESERVED && cred && 669 (error = caps_priv_check(cred, SYSCAP_NONET_RESPORT))) 670 { 671 inp->inp_laddr.s_addr = INADDR_ANY; 672 return (error); 673 } 674 675 /* 676 * This has to be atomic. If the porthash is shared across 677 * multiple protocol threads, e.g. tcp and udp then the token 678 * must be held. 679 */ 680 porthash = OBTAIN_LPORTHASH_TOKEN(inp->inp_pcbinfo, lport); 681 682 /* 683 * Restore the sin_port whacked by in_pcbbind_ladddr(); 684 * sin->sin_port is checked by in_pcbbind_laddrport_check(). 685 */ 686 sin->sin_port = lport; 687 688 error = in_pcbbind_laddrport_check(so, sin, porthash, 689 wild, cred, td); 690 if (error) { 691 inp->inp_laddr.s_addr = INADDR_ANY; 692 goto done; 693 } 694 695 inp->inp_lport = lport; 696 in_pcbinsporthash(porthash, inp); 697 error = 0; 698 done: 699 REL_PORTHASH_TOKEN(porthash); 700 return (error); 701 } else { 702 struct sockaddr_in jsin; 703 704 jsin.sin_family = AF_INET; 705 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 706 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 707 inp->inp_laddr.s_addr = INADDR_ANY; 708 return (EINVAL); 709 } 710 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 711 712 return in_pcbsetlport(inp, wild, cred); 713 } 714 } 715 716 /* 717 * Lookup a PCB based on the local and remote address and port. 718 * 719 * This function is only used when scanning for a free port. 720 */ 721 static struct inpcb * 722 in_pcblookup_localremote(struct inpcbporthead *porthash, struct in_addr laddr, 723 u_short lport, struct in_addr faddr, u_short fport, 724 struct ucred *cred) 725 { 726 struct inpcb *inp; 727 struct inpcbport *phd; 728 struct inpcb *match = NULL; 729 struct prison *pscan; 730 struct prison *pr; 731 732 /* 733 * If the porthashbase is shared across several cpus, it must 734 * have been locked. 735 */ 736 ASSERT_PORTHASH_TOKEN_HELD(porthash); 737 738 /* 739 * Best fit PCB lookup. 740 * 741 * First see if this local port is in use by looking on the 742 * port hash list. 743 */ 744 LIST_FOREACH(phd, porthash, phd_hash) { 745 if (phd->phd_port == lport) 746 break; 747 } 748 if (phd != NULL) { 749 pr = cred ? cred->cr_prison : NULL; 750 751 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 752 #ifdef INET6 753 if (!INP_ISIPV4(inp)) 754 continue; 755 #endif 756 if (inp->inp_laddr.s_addr == INADDR_ANY) { 757 if (inp->inp_socket && inp->inp_socket->so_cred) 758 pscan = inp->inp_socket->so_cred->cr_prison; 759 else 760 pscan = NULL; 761 if (pr != pscan) 762 continue; 763 } else { 764 if (inp->inp_laddr.s_addr != laddr.s_addr) 765 continue; 766 } 767 768 if (inp->inp_faddr.s_addr != INADDR_ANY && 769 inp->inp_faddr.s_addr != faddr.s_addr) 770 continue; 771 772 if (inp->inp_fport != 0 && inp->inp_fport != fport) 773 continue; 774 775 match = inp; 776 break; 777 } 778 } 779 return (match); 780 } 781 782 static boolean_t 783 in_pcbporthash_update4(struct inpcbportinfo *portinfo, struct inpcb *inp, 784 u_short lport, const struct sockaddr_in *sin, 785 struct ucred *cred) 786 { 787 struct inpcbporthead *porthash; 788 789 /* 790 * This has to be atomic. If the porthash is shared across multiple 791 * protocol threads, e.g. tcp and udp, then the token must be held. 792 */ 793 porthash = in_pcbporthash_head(portinfo, lport); 794 GET_PORTHASH_TOKEN(porthash); 795 796 if (in_pcblookup_localremote(porthash, inp->inp_laddr, lport, 797 sin->sin_addr, sin->sin_port, cred)) { 798 REL_PORTHASH_TOKEN(porthash); 799 return FALSE; 800 } 801 inp->inp_lport = lport; 802 in_pcbinsporthash(porthash, inp); 803 804 REL_PORTHASH_TOKEN(porthash); 805 return TRUE; 806 } 807 808 int 809 in_pcbbind_remote(struct inpcb *inp, const struct sockaddr *remote, 810 struct thread *td) 811 { 812 struct proc *p = td->td_proc; 813 const struct sockaddr_in *sin = (const struct sockaddr_in *)remote; 814 struct sockaddr_in jsin; 815 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 816 struct ucred *cred = NULL; 817 u_short first, last, lport; 818 int count, hash_count; 819 int error, selfconn = 0; 820 int cpuid = mycpuid; 821 uint32_t hash_base = 0, hash; 822 823 ASSERT_NETISR_NCPUS(cpuid); 824 825 if (TAILQ_EMPTY(&in_ifaddrheads[cpuid])) /* XXX broken! */ 826 return (EADDRNOTAVAIL); 827 828 KKASSERT(inp->inp_laddr.s_addr != INADDR_ANY); 829 if (inp->inp_lport != 0) 830 return (EINVAL); /* already bound */ 831 832 KKASSERT(p); 833 cred = p->p_ucred; 834 835 jsin.sin_family = AF_INET; 836 jsin.sin_addr.s_addr = inp->inp_laddr.s_addr; 837 if (!prison_replace_wildcards(td, (struct sockaddr *)&jsin)) { 838 inp->inp_laddr.s_addr = INADDR_ANY; 839 return (EINVAL); 840 } 841 inp->inp_laddr.s_addr = jsin.sin_addr.s_addr; 842 843 hash_count = ip_porthash_trycount; 844 if (hash_count > 0) { 845 hash_base = toeplitz_piecemeal_addr(sin->sin_addr.s_addr) ^ 846 toeplitz_piecemeal_addr(inp->inp_laddr.s_addr) ^ 847 toeplitz_piecemeal_port(sin->sin_port); 848 } else { 849 hash_count = 0; 850 } 851 852 inp->inp_flags |= INP_ANONPORT; 853 854 if (inp->inp_flags & INP_HIGHPORT) { 855 first = ipport_hifirstauto; /* sysctl */ 856 last = ipport_hilastauto; 857 } else if (inp->inp_flags & INP_LOWPORT) { 858 if (cred && 859 (error = caps_priv_check(cred, SYSCAP_NONET_RESPORT))) 860 { 861 inp->inp_laddr.s_addr = INADDR_ANY; 862 return (error); 863 } 864 first = ipport_lowfirstauto; /* 1023 */ 865 last = ipport_lowlastauto; /* 600 */ 866 } else { 867 first = ipport_firstauto; /* sysctl */ 868 last = ipport_lastauto; 869 } 870 if (first > last) { 871 lport = last; 872 last = first; 873 first = lport; 874 } 875 KKASSERT(last >= first); 876 877 count = last - first; 878 lport = (karc4random() % count) + first; 879 count += hash_count; 880 881 /* 882 * Simple check to ensure all ports are not used up causing 883 * a deadlock here. 884 */ 885 for (;;) { 886 u_short lport_no; 887 888 if (count-- < 0) { /* completely used? */ 889 error = EADDRNOTAVAIL; 890 break; 891 } 892 893 if (__predict_false(lport < first || lport > last)) 894 lport = first; 895 lport_no = htons(lport); 896 897 /* This could happen on loopback interface */ 898 if (__predict_false(sin->sin_port == lport_no && 899 sin->sin_addr.s_addr == inp->inp_laddr.s_addr)) { 900 if (!selfconn) { 901 ++count; /* don't count this try */ 902 selfconn = 1; 903 } 904 goto next; 905 } 906 907 if (hash_count) { 908 --hash_count; 909 hash = hash_base ^ 910 toeplitz_piecemeal_port(lport_no); 911 if (netisr_hashcpu(hash) != cpuid && hash_count) 912 goto next; 913 } 914 915 if (in_pcbporthash_update4( 916 &pcbinfo->portinfo[lport % pcbinfo->portinfo_cnt], 917 inp, lport_no, sin, cred)) { 918 error = 0; 919 break; 920 } 921 next: 922 ++lport; 923 } 924 925 if (error) 926 inp->inp_laddr.s_addr = INADDR_ANY; 927 return (error); 928 } 929 930 /* 931 * Figure out the local interface address to pair against the requested 932 * target address, as well as validate the target address. 933 */ 934 int 935 in_pcbladdr_find(struct inpcb *inp, struct sockaddr *nam, 936 struct sockaddr_in **plocal_sin, struct thread *td, int find) 937 { 938 struct in_ifaddr_container *iac; 939 struct in_ifaddr *ia; 940 struct ucred *cred = NULL; 941 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 942 struct sockaddr *jsin; 943 struct prison *pr; 944 struct route *ro; 945 int alloc_route = 0; 946 947 if (nam->sa_len != sizeof *sin) 948 return (EINVAL); 949 if (sin->sin_family != AF_INET) 950 return (EAFNOSUPPORT); 951 if (sin->sin_port == 0) 952 return (EADDRNOTAVAIL); 953 954 /* 955 * Are we in a jail? 956 */ 957 pr = NULL; 958 if (td && td->td_proc && td->td_proc->p_ucred) 959 cred = td->td_proc->p_ucred; 960 if (cred) 961 pr = cred->cr_prison; 962 963 /* 964 * If the destination address is INADDR_ANY then use the primary 965 * local address. 966 * 967 * If the supplied address is INADDR_BROADCAST, and the primary 968 * interface supports broadcast, choose the broadcast address for 969 * that interface. 970 * 971 * If jailed, locate an interface address acceptable to the jail. 972 */ 973 if (sin->sin_addr.s_addr == INADDR_ANY) { 974 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) { 975 ia = iac->ia; 976 if (pr == NULL || 977 jailed_ip(pr, sintosa(&ia->ia_addr))) { 978 sin->sin_addr = IA_SIN(ia)->sin_addr; 979 break; 980 } 981 } 982 } else if (sin->sin_addr.s_addr == (u_long)INADDR_BROADCAST) { 983 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) { 984 ia = iac->ia; 985 if ((pr == NULL || 986 jailed_ip(pr, sintosa(&ia->ia_addr))) && 987 (iac->ia->ia_ifp->if_flags & IFF_BROADCAST)) { 988 sin->sin_addr = 989 satosin(&ia->ia_broadaddr)->sin_addr; 990 break; 991 } 992 } 993 } 994 995 /* 996 * If asked to do a search, use the cached route or do a route table 997 * lookup to try to find an acceptable local interface IP. 998 */ 999 if (find == 0) 1000 return 0; 1001 1002 ia = NULL; 1003 1004 /* 1005 * If we have a cached route, check to see if it is acceptable. 1006 * If not, free it. 1007 */ 1008 ro = &inp->inp_route; 1009 if (ro->ro_rt && 1010 (!(ro->ro_rt->rt_flags & RTF_UP) || 1011 ro->ro_dst.sa_family != AF_INET || 1012 satosin(&ro->ro_dst)->sin_addr.s_addr != 1013 sin->sin_addr.s_addr || 1014 inp->inp_socket->so_options & SO_DONTROUTE)) { 1015 RTFREE(ro->ro_rt); 1016 ro->ro_rt = NULL; 1017 } 1018 1019 /* 1020 * If we do not have a route, construct one and do a lookup, 1021 * unless we are forbidden to do so. 1022 * 1023 * Note that we should check the address family of the cached 1024 * destination, in case of sharing the cache with IPv6. 1025 */ 1026 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/ 1027 (ro->ro_rt == NULL || ro->ro_rt->rt_ifp == NULL)) { 1028 bzero(&ro->ro_dst, sizeof(struct sockaddr_in)); 1029 ro->ro_dst.sa_family = AF_INET; 1030 ro->ro_dst.sa_len = sizeof(struct sockaddr_in); 1031 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sin->sin_addr; 1032 rtalloc(ro); 1033 alloc_route = 1; 1034 } 1035 1036 /* 1037 * If we found a route, use the address corresponding to the 1038 * outgoing interface. 1039 * 1040 * If jailed, try to find a compatible address on the outgoing 1041 * interface. 1042 */ 1043 if (ro->ro_rt) { 1044 ia = ifatoia(ro->ro_rt->rt_ifa); 1045 if (pr == NULL) 1046 goto skip; 1047 if (jailed_ip(pr, sintosa(&ia->ia_addr))) 1048 goto skip; 1049 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) { 1050 if (iac->ia->ia_ifp != ia->ia_ifp) 1051 continue; 1052 ia = iac->ia; 1053 if (jailed_ip(pr, sintosa(&ia->ia_addr))) 1054 goto skip; 1055 } 1056 ia = NULL; 1057 } 1058 skip: 1059 1060 /* 1061 * If the route didn't work or there was no route, 1062 * fall-back to the first address in in_ifaddrheads[]. 1063 * 1064 * If jailed and this address is not available for 1065 * the jail, leave ia set to NULL. 1066 */ 1067 if (ia == NULL) { 1068 u_short fport = sin->sin_port; 1069 1070 sin->sin_port = 0; 1071 ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin))); 1072 if (ia && pr && !jailed_ip(pr, sintosa(&ia->ia_addr))) 1073 ia = NULL; 1074 1075 if (ia == NULL) 1076 ia = ifatoia(ifa_ifwithnet(sintosa(sin))); 1077 if (ia && pr && !jailed_ip(pr, sintosa(&ia->ia_addr))) 1078 ia = NULL; 1079 1080 sin->sin_port = fport; 1081 if (ia == NULL && !TAILQ_EMPTY(&in_ifaddrheads[mycpuid])) 1082 ia = TAILQ_FIRST(&in_ifaddrheads[mycpuid])->ia; 1083 1084 if (ia && pr && !jailed_ip(pr, sintosa(&ia->ia_addr))) 1085 ia = NULL; 1086 1087 if (pr == NULL && ia == NULL) 1088 goto fail; 1089 } 1090 1091 /* 1092 * If the destination address is multicast and an outgoing 1093 * interface has been set as a multicast option, use the 1094 * address of that interface as our source address. 1095 */ 1096 if (pr == NULL && IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 1097 inp->inp_moptions != NULL) { 1098 struct ip_moptions *imo; 1099 struct ifnet *ifp; 1100 1101 imo = inp->inp_moptions; 1102 if ((ifp = imo->imo_multicast_ifp) != NULL) { 1103 struct in_ifaddr_container *iac; 1104 1105 ia = NULL; 1106 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) { 1107 if (iac->ia->ia_ifp == ifp) { 1108 ia = iac->ia; 1109 break; 1110 } 1111 } 1112 if (ia == NULL) 1113 goto fail; 1114 } 1115 } 1116 1117 /* 1118 * If we still don't have a local address, and are jailed, 1119 * use the jail's first non-localhost IP. If there isn't 1120 * one, use the jail's first localhost IP. 1121 * 1122 * Don't do pcblookup call here; return interface in plocal_sin 1123 * and exit to caller, that will do the lookup. 1124 */ 1125 if (ia == NULL && pr) { 1126 jsin = prison_get_nonlocal(cred->cr_prison, AF_INET, NULL); 1127 if (jsin == NULL) 1128 jsin = prison_get_local(cred->cr_prison, AF_INET, NULL); 1129 if (jsin) 1130 *plocal_sin = satosin(jsin); 1131 else 1132 goto fail; 1133 } else if (ia) { 1134 *plocal_sin = &ia->ia_addr; 1135 } else { 1136 goto fail; 1137 } 1138 return (0); 1139 fail: 1140 if (alloc_route) 1141 in_pcbresetroute(inp); 1142 return (EADDRNOTAVAIL); 1143 } 1144 1145 int 1146 in_pcbladdr(struct inpcb *inp, struct sockaddr *nam, 1147 struct sockaddr_in **plocal_sin, struct thread *td) 1148 { 1149 return in_pcbladdr_find(inp, nam, plocal_sin, td, 1150 (inp->inp_laddr.s_addr == INADDR_ANY)); 1151 } 1152 1153 /* 1154 * Outer subroutine: 1155 * Connect from a socket to a specified address. 1156 * Both address and port must be specified in argument sin. 1157 * If don't have a local address for this socket yet, 1158 * then pick one. 1159 */ 1160 int 1161 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct thread *td) 1162 { 1163 struct sockaddr_in *if_sin; 1164 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1165 int error; 1166 1167 if_sin = NULL; /* avoid gcc warnings */ 1168 1169 /* Call inner routine to assign local interface address. */ 1170 if ((error = in_pcbladdr(inp, nam, &if_sin, td)) != 0) 1171 return (error); 1172 1173 if (in_pcblookup_hash(inp->inp_pcbinfo, sin->sin_addr, sin->sin_port, 1174 inp->inp_laddr.s_addr ? 1175 inp->inp_laddr : if_sin->sin_addr, 1176 inp->inp_lport, FALSE, NULL) != NULL) { 1177 return (EADDRINUSE); 1178 } 1179 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1180 if (inp->inp_lport == 0) { 1181 error = in_pcbbind(inp, NULL, td); 1182 if (error) 1183 return (error); 1184 } 1185 inp->inp_laddr = if_sin->sin_addr; 1186 } 1187 inp->inp_faddr = sin->sin_addr; 1188 inp->inp_fport = sin->sin_port; 1189 in_pcbinsconnhash(inp); 1190 return (0); 1191 } 1192 1193 void 1194 in_pcbdisconnect(struct inpcb *inp) 1195 { 1196 1197 in_pcbremconnhash(inp); 1198 inp->inp_faddr.s_addr = INADDR_ANY; 1199 inp->inp_fport = 0; 1200 } 1201 1202 void 1203 in_pcbdetach(struct inpcb *inp) 1204 { 1205 struct socket *so = inp->inp_socket; 1206 struct inpcbinfo *ipi = inp->inp_pcbinfo; 1207 1208 inp->inp_gencnt = ++ipi->ipi_gencnt; 1209 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0); 1210 in_pcbremlists(inp); 1211 so->so_pcb = NULL; 1212 sofree(so); /* remove pcb ref */ 1213 if (inp->inp_options) 1214 m_free(inp->inp_options); 1215 if (inp->inp_route.ro_rt) 1216 rtfree(inp->inp_route.ro_rt); 1217 ip_freemoptions(inp->inp_moptions); 1218 kfree(inp, M_PCB); 1219 } 1220 1221 /* 1222 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP 1223 * socket received RST. 1224 */ 1225 static int 1226 in_setsockaddr(struct socket *so, struct sockaddr **nam) 1227 { 1228 struct inpcb *inp; 1229 struct sockaddr_in *sin; 1230 1231 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 1232 inp = so->so_pcb; 1233 if (!inp) 1234 return (ECONNRESET); 1235 1236 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1237 sin->sin_family = AF_INET; 1238 sin->sin_len = sizeof *sin; 1239 sin->sin_port = inp->inp_lport; 1240 sin->sin_addr = inp->inp_laddr; 1241 1242 *nam = (struct sockaddr *)sin; 1243 return (0); 1244 } 1245 1246 void 1247 in_setsockaddr_dispatch(netmsg_t msg) 1248 { 1249 int error; 1250 1251 error = in_setsockaddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1252 lwkt_replymsg(&msg->lmsg, error); 1253 } 1254 1255 /* 1256 * The socket may have an invalid PCB, i.e. NULL. For example, a TCP 1257 * socket received RST. 1258 */ 1259 int 1260 in_setpeeraddr(struct socket *so, struct sockaddr **nam) 1261 { 1262 struct inpcb *inp; 1263 struct sockaddr_in *sin; 1264 1265 KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr")); 1266 inp = so->so_pcb; 1267 if (!inp) 1268 return (ECONNRESET); 1269 1270 sin = kmalloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); 1271 sin->sin_family = AF_INET; 1272 sin->sin_len = sizeof *sin; 1273 sin->sin_port = inp->inp_fport; 1274 sin->sin_addr = inp->inp_faddr; 1275 1276 *nam = (struct sockaddr *)sin; 1277 return (0); 1278 } 1279 1280 void 1281 in_setpeeraddr_dispatch(netmsg_t msg) 1282 { 1283 int error; 1284 1285 error = in_setpeeraddr(msg->base.nm_so, msg->peeraddr.nm_nam); 1286 lwkt_replymsg(&msg->lmsg, error); 1287 } 1288 1289 void 1290 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int err, 1291 inp_notify_t notify) 1292 { 1293 struct inpcb *inp, *marker; 1294 1295 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1296 ("not in the correct netisr")); 1297 marker = in_pcbmarker(); 1298 1299 /* 1300 * NOTE: 1301 * - If INP_PLACEMARKER is set we must ignore the rest of the 1302 * structure and skip it. 1303 * - It is safe to nuke inpcbs here, since we are in their own 1304 * netisr. 1305 */ 1306 GET_PCBINFO_TOKEN(pcbinfo); 1307 1308 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1309 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) { 1310 LIST_REMOVE(marker, inp_list); 1311 LIST_INSERT_AFTER(inp, marker, inp_list); 1312 1313 if (inp->inp_flags & INP_PLACEMARKER) 1314 continue; 1315 #ifdef INET6 1316 if (!INP_ISIPV4(inp)) 1317 continue; 1318 #endif 1319 if (inp->inp_faddr.s_addr != faddr.s_addr || 1320 inp->inp_socket == NULL) 1321 continue; 1322 (*notify)(inp, err); /* can remove inp from list! */ 1323 } 1324 LIST_REMOVE(marker, inp_list); 1325 1326 REL_PCBINFO_TOKEN(pcbinfo); 1327 } 1328 1329 void 1330 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1331 { 1332 struct inpcb *inp, *marker; 1333 1334 /* 1335 * We only need to make sure that we are in netisr0, where all 1336 * multicast operation happen. We could check inpcbinfo which 1337 * does not belong to netisr0 by holding the inpcbinfo's token. 1338 * In this case, the pcbinfo must be able to be shared, i.e. 1339 * pcbinfo->infotoken is not NULL. 1340 */ 1341 ASSERT_NETISR0; 1342 KASSERT(pcbinfo->cpu == 0 || pcbinfo->infotoken != NULL, 1343 ("pcbinfo could not be shared")); 1344 1345 /* 1346 * Get a marker for the current netisr (netisr0). 1347 * 1348 * It is possible that the multicast address deletion blocks, 1349 * which could cause temporary token releasing. So we use 1350 * inpcb marker here to get a coherent view of the inpcb list. 1351 * 1352 * While, on the other hand, moptions are only added and deleted 1353 * in netisr0, so we would not see staled moption or miss moption 1354 * even if the token was released due to the blocking multicast 1355 * address deletion. 1356 */ 1357 marker = in_pcbmarker(); 1358 1359 GET_PCBINFO_TOKEN(pcbinfo); 1360 1361 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 1362 while ((inp = LIST_NEXT(marker, inp_list)) != NULL) { 1363 struct ip_moptions *imo; 1364 1365 LIST_REMOVE(marker, inp_list); 1366 LIST_INSERT_AFTER(inp, marker, inp_list); 1367 1368 if (inp->inp_flags & INP_PLACEMARKER) 1369 continue; 1370 imo = inp->inp_moptions; 1371 if (INP_ISIPV4(inp) && imo != NULL) { 1372 int i, gap; 1373 1374 /* 1375 * Unselect the outgoing interface if it is being 1376 * detached. 1377 */ 1378 if (imo->imo_multicast_ifp == ifp) 1379 imo->imo_multicast_ifp = NULL; 1380 1381 /* 1382 * Drop multicast group membership if we joined 1383 * through the interface being detached. 1384 */ 1385 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1386 i++) { 1387 if (imo->imo_membership[i]->inm_ifp == ifp) { 1388 /* 1389 * NOTE: 1390 * This could block and the pcbinfo 1391 * token could be passively released. 1392 */ 1393 in_delmulti(imo->imo_membership[i]); 1394 gap++; 1395 } else if (gap != 0) 1396 imo->imo_membership[i - gap] = 1397 imo->imo_membership[i]; 1398 } 1399 imo->imo_num_memberships -= gap; 1400 } 1401 } 1402 LIST_REMOVE(marker, inp_list); 1403 1404 REL_PCBINFO_TOKEN(pcbinfo); 1405 } 1406 1407 /* 1408 * Check for alternatives when higher level complains 1409 * about service problems. For now, invalidate cached 1410 * routing information. If the route was created dynamically 1411 * (by a redirect), time to try a default gateway again. 1412 */ 1413 void 1414 in_losing(struct inpcb *inp) 1415 { 1416 struct rtentry *rt; 1417 struct rt_addrinfo rtinfo; 1418 1419 if ((rt = inp->inp_route.ro_rt)) { 1420 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1421 rtinfo.rti_info[RTAX_DST] = rt_key(rt); 1422 rtinfo.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 1423 rtinfo.rti_info[RTAX_NETMASK] = rt_mask(rt); 1424 rtinfo.rti_flags = rt->rt_flags; 1425 rt_missmsg(RTM_LOSING, &rtinfo, rt->rt_flags, 0); 1426 if (rt->rt_flags & RTF_DYNAMIC) { 1427 rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, 1428 rt_mask(rt), rt->rt_flags, NULL); 1429 } 1430 inp->inp_route.ro_rt = NULL; 1431 rtfree(rt); 1432 /* 1433 * A new route can be allocated 1434 * the next time output is attempted. 1435 */ 1436 } 1437 } 1438 1439 /* 1440 * After a routing change, flush old routing 1441 * and allocate a (hopefully) better one. 1442 */ 1443 void 1444 in_rtchange(struct inpcb *inp, int err) 1445 { 1446 if (inp->inp_route.ro_rt) { 1447 rtfree(inp->inp_route.ro_rt); 1448 inp->inp_route.ro_rt = NULL; 1449 /* 1450 * A new route can be allocated the next time 1451 * output is attempted. 1452 */ 1453 } 1454 } 1455 1456 /* 1457 * Lookup a PCB based on the local address and port. 1458 * 1459 * This function is only used when scanning for a free port. 1460 */ 1461 static struct inpcb * 1462 in_pcblookup_local(struct inpcbporthead *porthash, struct in_addr laddr, 1463 u_int lport_arg, int wild_okay, struct ucred *cred) 1464 { 1465 struct prison *pscan; 1466 struct prison *pr; 1467 struct inpcb *inp; 1468 int matchwild = 3, wildcard; 1469 u_short lport = lport_arg; 1470 struct inpcbport *phd; 1471 struct inpcb *match = NULL; 1472 1473 /* 1474 * If the porthashbase is shared across several cpus, it must 1475 * have been locked. 1476 */ 1477 ASSERT_PORTHASH_TOKEN_HELD(porthash); 1478 1479 /* 1480 * Best fit PCB lookup. 1481 * 1482 * First see if this local port is in use by looking on the 1483 * port hash list. 1484 */ 1485 LIST_FOREACH(phd, porthash, phd_hash) { 1486 if (phd->phd_port == lport) 1487 break; 1488 } 1489 if (phd != NULL) { 1490 pr = cred ? cred->cr_prison : NULL; 1491 1492 /* 1493 * Port is in use by one or more PCBs. Look for best 1494 * fit. 1495 * 1496 * If in a prison we may wish to allow the jail to override 1497 * a wildcard listen on the host. Since the jail forces its 1498 * own wildcard listens to a specific set of jail IPs, this 1499 * override allows most services on the host to remain as 1500 * they were and still be 'jail friendly'. 1501 */ 1502 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1503 wildcard = 0; 1504 #ifdef INET6 1505 if (!INP_ISIPV4(inp)) 1506 continue; 1507 #endif 1508 if (inp->inp_faddr.s_addr != INADDR_ANY) 1509 wildcard++; 1510 1511 /* 1512 * Prison are independent of each other in terms 1513 * of allowing bindings. This can result in multiple 1514 * overloaded bindings which in_pcblookup_pkthash() 1515 * will have to sort out. 1516 * 1517 * Allow wildcarded entries to co-exist with specific 1518 * entries. Specific entries override wildcarded 1519 * entries. 1520 */ 1521 if (inp->inp_socket && inp->inp_socket->so_cred) 1522 pscan = inp->inp_socket->so_cred->cr_prison; 1523 else 1524 pscan = NULL; 1525 if (pr != pscan) 1526 continue; 1527 if (inp->inp_laddr.s_addr == INADDR_ANY) { 1528 if (laddr.s_addr != INADDR_ANY) 1529 wildcard++; 1530 } else { 1531 if (laddr.s_addr == INADDR_ANY) 1532 wildcard++; 1533 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1534 continue; 1535 } 1536 if (wildcard && !wild_okay) 1537 continue; 1538 if (wildcard < matchwild) { 1539 match = inp; 1540 matchwild = wildcard; 1541 if (matchwild == 0) 1542 break; 1543 } 1544 } 1545 } 1546 return (match); 1547 } 1548 1549 struct inpcb * 1550 in_pcblocalgroup_last(const struct inpcbinfo *pcbinfo, 1551 const struct inpcb *inp) 1552 { 1553 const struct inp_localgrphead *hdr; 1554 const struct inp_localgroup *grp; 1555 int i; 1556 1557 if (pcbinfo->localgrphashbase == NULL) 1558 return NULL; 1559 1560 GET_PCBINFO_TOKEN(pcbinfo); 1561 1562 hdr = &pcbinfo->localgrphashbase[ 1563 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 1564 1565 LIST_FOREACH(grp, hdr, il_list) { 1566 if (grp->il_af == inp->inp_af && 1567 grp->il_lport == inp->inp_lport && 1568 memcmp(&grp->il_dependladdr, 1569 &inp->inp_inc.inc_ie.ie_dependladdr, 1570 sizeof(grp->il_dependladdr)) == 0) { 1571 break; 1572 } 1573 } 1574 if (grp == NULL || grp->il_inpcnt == 1) { 1575 REL_PCBINFO_TOKEN(pcbinfo); 1576 return NULL; 1577 } 1578 1579 KASSERT(grp->il_inpcnt >= 2, 1580 ("invalid localgroup inp count %d", grp->il_inpcnt)); 1581 for (i = 0; i < grp->il_inpcnt; ++i) { 1582 if (grp->il_inp[i] == inp) { 1583 int last = grp->il_inpcnt - 1; 1584 1585 if (i == last) 1586 last = grp->il_inpcnt - 2; 1587 REL_PCBINFO_TOKEN(pcbinfo); 1588 return grp->il_inp[last]; 1589 } 1590 } 1591 REL_PCBINFO_TOKEN(pcbinfo); 1592 return NULL; 1593 } 1594 1595 static struct inpcb * 1596 inp_localgroup_lookup(const struct inpcbinfo *pcbinfo, 1597 struct in_addr laddr, uint16_t lport, uint32_t pkt_hash) 1598 { 1599 struct inpcb *local_wild; 1600 struct inpcb *jinp; 1601 struct inpcb *jinp_wild; 1602 struct inpcb *inp; 1603 const struct inp_localgrphead *hdr; 1604 const struct inp_localgroup *grp; 1605 struct sockaddr_in jsin; 1606 struct prison *pr; 1607 struct ucred *cred; 1608 int idx; 1609 int net_listen_ov_local; 1610 int net_listen_ov_wild; 1611 1612 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 1613 1614 hdr = &pcbinfo->localgrphashbase[ 1615 INP_PCBLOCALGRPHASH(lport, pcbinfo->localgrphashmask)]; 1616 1617 /* 1618 * Order of socket selection: 1619 * 1. non-wild. 1620 * 2. wild. 1621 * 1622 * NOTE: Local group does not contain jailed sockets 1623 */ 1624 jsin.sin_family = AF_INET; 1625 jsin.sin_addr.s_addr = laddr.s_addr; 1626 1627 jinp = NULL; 1628 jinp_wild = NULL; 1629 local_wild = NULL; 1630 net_listen_ov_local = 0; 1631 net_listen_ov_wild = 0; 1632 1633 LIST_FOREACH(grp, hdr, il_list) { 1634 #ifdef INET6 1635 if (grp->il_af != AF_INET) 1636 continue; 1637 #endif 1638 if (grp->il_lport != lport) 1639 continue; 1640 1641 /* 1642 * look for a match 1643 */ 1644 idx = netisr_hashlsb(pkt_hash) % grp->il_inpcnt; 1645 inp = grp->il_inp[idx]; 1646 1647 /* 1648 * Modulo-N is used here, which greatly reduces 1649 * completion queue token contention, thus more 1650 * cpu time is saved. 1651 */ 1652 if (grp->il_jailed) { 1653 if (inp->inp_socket == NULL) 1654 continue; 1655 cred = inp->inp_socket->so_cred; 1656 if (cred == NULL) 1657 continue; 1658 pr = cred->cr_prison; 1659 if (pr == NULL) 1660 continue; 1661 if (!jailed_ip(pr, (struct sockaddr *)&jsin)) 1662 continue; 1663 if (grp->il_laddr.s_addr == laddr.s_addr) { 1664 jinp = inp; 1665 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE)) 1666 net_listen_ov_local = 1; 1667 1668 } else if (grp->il_laddr.s_addr == INADDR_ANY && 1669 jinp_wild == NULL) { 1670 jinp_wild = inp; 1671 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE)) 1672 net_listen_ov_wild = 1; 1673 } 1674 } else { 1675 if (grp->il_laddr.s_addr == laddr.s_addr) { 1676 return inp; 1677 } else if (grp->il_laddr.s_addr == INADDR_ANY) { 1678 local_wild = inp; 1679 } 1680 } 1681 } 1682 1683 if (net_listen_ov_local) 1684 return jinp; 1685 if (net_listen_ov_wild) 1686 return jinp_wild; 1687 if (local_wild) 1688 return (local_wild); 1689 if (jinp) 1690 return (jinp); 1691 return (jinp_wild); 1692 } 1693 1694 /* 1695 * Lookup PCB in hash list. 1696 * 1697 * This is used to match incoming packets to a pcb 1698 */ 1699 struct inpcb * 1700 in_pcblookup_pkthash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1701 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1702 boolean_t wildcard, struct ifnet *ifp, const struct mbuf *m) 1703 { 1704 struct inpcbhead *head; 1705 struct inpcb *inp, *jinp=NULL; 1706 u_short fport = fport_arg, lport = lport_arg; 1707 1708 /* 1709 * First look for an exact match. 1710 */ 1711 head = &pcbinfo->hashbase[INP_PCBCONNHASH(faddr.s_addr, fport, 1712 laddr.s_addr, lport, 1713 pcbinfo->hashmask)]; 1714 LIST_FOREACH(inp, head, inp_hash) { 1715 #ifdef INET6 1716 if (!INP_ISIPV4(inp)) 1717 continue; 1718 #endif 1719 if (in_hosteq(inp->inp_faddr, faddr) && 1720 in_hosteq(inp->inp_laddr, laddr) && 1721 inp->inp_fport == fport && inp->inp_lport == lport) { 1722 /* 1723 * Found specific address, host overrides jailed 1724 * inpcb. 1725 */ 1726 if (inp->inp_socket == NULL || 1727 inp->inp_socket->so_cred->cr_prison == NULL) { 1728 return (inp); 1729 } 1730 if (jinp == NULL) 1731 jinp = inp; 1732 } 1733 } 1734 if (jinp != NULL) 1735 return (jinp); 1736 1737 /* 1738 * We generally get here for connections to wildcarded listeners. 1739 * Any wildcarded listeners in jails must be restricted to the 1740 * jailed IPs only. 1741 */ 1742 if (wildcard) { 1743 struct inpcb *local_wild = NULL; 1744 struct inpcb *jinp_wild = NULL; 1745 struct inpcontainer *ic; 1746 struct inpcontainerhead *chead; 1747 struct sockaddr_in jsin; 1748 struct ucred *cred; 1749 struct prison *pr; 1750 int net_listen_ov_local = 0; 1751 int net_listen_ov_wild = 0; 1752 1753 GET_PCBINFO_TOKEN(pcbinfo); 1754 1755 /* 1756 * Check local group first. When present, the localgroup 1757 * hash utilizes the same non-jailed-vs/jailed priortization 1758 * that the normal wildcardhash does. 1759 */ 1760 if (pcbinfo->localgrphashbase != NULL && 1761 m != NULL && (m->m_flags & M_HASH)) { 1762 inp = inp_localgroup_lookup(pcbinfo, laddr, lport, 1763 m->m_pkthdr.hash); 1764 if (inp != NULL) { 1765 REL_PCBINFO_TOKEN(pcbinfo); 1766 return inp; 1767 } 1768 } 1769 1770 /* 1771 * Order of socket selection: 1772 * 1773 * 1. non-jailed, non-wild. 1774 * 2. non-jailed, wild. (allow_listen_override on) 1775 * 3. jailed, non-wild. 1776 * 4. jailed, wild. 1777 * 5. non-jailed, wild. (allow_listen_override off) 1778 * 1779 * NOTE: jailed wildcards are still restricted to the jail 1780 * IPs. 1781 * 1782 * NOTE: (1) and (3) already handled above. 1783 */ 1784 jsin.sin_family = AF_INET; 1785 chead = &pcbinfo->wildcardhashbase[ 1786 INP_PCBWILDCARDHASH(lport, pcbinfo->wildcardhashmask)]; 1787 1788 LIST_FOREACH(ic, chead, ic_list) { 1789 inp = ic->ic_inp; 1790 if (inp->inp_flags & INP_PLACEMARKER) 1791 continue; 1792 1793 /* 1794 * Basic validation 1795 */ 1796 #ifdef INET6 1797 if (!INP_ISIPV4(inp)) 1798 continue; 1799 #endif 1800 if (inp->inp_lport != lport) 1801 continue; 1802 1803 /* 1804 * Calculate prison, setup jsin for jailed_ip() 1805 * check. 1806 */ 1807 jsin.sin_addr.s_addr = laddr.s_addr; 1808 pr = NULL; 1809 cred = NULL; 1810 if (inp->inp_socket) { 1811 cred = inp->inp_socket->so_cred; 1812 if (cred) 1813 pr = cred->cr_prison; 1814 } 1815 1816 /* 1817 * Assign jinp, jinp_wild, and local_wild as 1818 * appropriate, track whether the jail supports 1819 * listen overrides. 1820 */ 1821 if (pr) { 1822 if (!jailed_ip(pr, (struct sockaddr *)&jsin)) 1823 continue; 1824 if (inp->inp_laddr.s_addr == laddr.s_addr && 1825 jinp == NULL) { 1826 jinp = inp; 1827 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE)) 1828 net_listen_ov_local = 1; 1829 } 1830 if (inp->inp_laddr.s_addr == INADDR_ANY && 1831 jinp_wild == NULL) { 1832 jinp_wild = inp; 1833 if (PRISON_CAP_ISSET(pr->pr_caps, PRISON_CAP_NET_LISTEN_OVERRIDE)) 1834 net_listen_ov_wild = 1; 1835 } 1836 } else { 1837 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1838 REL_PCBINFO_TOKEN(pcbinfo); 1839 return (inp); 1840 } 1841 if (inp->inp_laddr.s_addr == INADDR_ANY) 1842 local_wild = inp; 1843 } 1844 } 1845 1846 REL_PCBINFO_TOKEN(pcbinfo); 1847 1848 if (net_listen_ov_local) 1849 return jinp; 1850 if (net_listen_ov_wild) 1851 return jinp_wild; 1852 if (local_wild) 1853 return (local_wild); 1854 if (jinp) 1855 return (jinp); 1856 return (jinp_wild); 1857 } 1858 1859 /* 1860 * Not found. 1861 */ 1862 return (NULL); 1863 } 1864 1865 struct inpcb * 1866 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1867 u_int fport_arg, struct in_addr laddr, u_int lport_arg, 1868 boolean_t wildcard, struct ifnet *ifp) 1869 { 1870 return in_pcblookup_pkthash(pcbinfo, faddr, fport_arg, 1871 laddr, lport_arg, wildcard, ifp, NULL); 1872 } 1873 1874 /* 1875 * Insert PCB into connection hash table. 1876 */ 1877 void 1878 in_pcbinsconnhash(struct inpcb *inp) 1879 { 1880 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1881 struct inpcbhead *bucket; 1882 u_int32_t hashkey_faddr, hashkey_laddr; 1883 1884 #ifdef INET6 1885 if (INP_ISIPV6(inp)) { 1886 hashkey_faddr = inp->in6p_faddr.s6_addr32[3] /* XXX JH */; 1887 hashkey_laddr = inp->in6p_laddr.s6_addr32[3] /* XXX JH */; 1888 } else { 1889 #endif 1890 hashkey_faddr = inp->inp_faddr.s_addr; 1891 hashkey_laddr = inp->inp_laddr.s_addr; 1892 #ifdef INET6 1893 } 1894 #endif 1895 1896 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1897 ("not in the correct netisr")); 1898 ASSERT_INP_NOTINHASH(inp); 1899 inp->inp_flags |= INP_CONNECTED; 1900 1901 /* 1902 * Insert into the connection hash table. 1903 */ 1904 bucket = &pcbinfo->hashbase[INP_PCBCONNHASH(hashkey_faddr, 1905 inp->inp_fport, hashkey_laddr, inp->inp_lport, pcbinfo->hashmask)]; 1906 LIST_INSERT_HEAD(bucket, inp, inp_hash); 1907 } 1908 1909 /* 1910 * Remove PCB from connection hash table. 1911 */ 1912 void 1913 in_pcbremconnhash(struct inpcb *inp) 1914 { 1915 struct inpcbinfo *pcbinfo __debugvar = inp->inp_pcbinfo; 1916 1917 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 1918 ("not in the correct netisr")); 1919 KASSERT(inp->inp_flags & INP_CONNECTED, ("inp not connected")); 1920 1921 LIST_REMOVE(inp, inp_hash); 1922 inp->inp_flags &= ~INP_CONNECTED; 1923 } 1924 1925 /* 1926 * Insert PCB into port hash table. 1927 */ 1928 void 1929 in_pcbinsporthash(struct inpcbporthead *pcbporthash, struct inpcb *inp) 1930 { 1931 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1932 struct inpcbport *phd; 1933 1934 /* 1935 * If the porthashbase is shared across several cpus, it must 1936 * have been locked. 1937 */ 1938 ASSERT_PORTHASH_TOKEN_HELD(pcbporthash); 1939 1940 /* 1941 * Insert into the port hash table. 1942 */ 1943 1944 /* Go through port list and look for a head for this lport. */ 1945 LIST_FOREACH(phd, pcbporthash, phd_hash) { 1946 if (phd->phd_port == inp->inp_lport) 1947 break; 1948 } 1949 1950 /* If none exists, use saved one and tack it on. */ 1951 if (phd == NULL) { 1952 KKASSERT(pcbinfo->portsave != NULL); 1953 phd = pcbinfo->portsave; 1954 pcbinfo->portsave = NULL; 1955 phd->phd_port = inp->inp_lport; 1956 LIST_INIT(&phd->phd_pcblist); 1957 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 1958 } 1959 1960 inp->inp_porthash = pcbporthash; 1961 inp->inp_phd = phd; 1962 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 1963 1964 /* 1965 * Malloc one inpcbport for later use. It is safe to use 1966 * "wait" malloc here (port token would be released, if 1967 * malloc ever blocked), since all changes to the porthash 1968 * are done. 1969 */ 1970 if (pcbinfo->portsave == NULL) { 1971 pcbinfo->portsave = kmalloc(sizeof(*pcbinfo->portsave), 1972 M_PCB, M_INTWAIT | M_ZERO); 1973 } 1974 } 1975 1976 void 1977 in_pcbinsporthash_lport(struct inpcb *inp) 1978 { 1979 struct inpcbporthead *porthash; 1980 1981 porthash = OBTAIN_LPORTHASH_TOKEN(inp->inp_pcbinfo, inp->inp_lport); 1982 in_pcbinsporthash(porthash, inp); 1983 REL_PORTHASH_TOKEN(porthash); 1984 } 1985 1986 void 1987 in_pcbremporthash(struct inpcb *inp) 1988 { 1989 struct inpcbporthead *porthash; 1990 struct inpcbport *phd; 1991 1992 if (inp->inp_phd == NULL) 1993 return; 1994 KASSERT(inp->inp_lport != 0, ("inpcb has no lport")); 1995 1996 porthash = inp->inp_porthash; 1997 KASSERT(porthash != NULL, ("no porthash")); 1998 1999 GET_PORTHASH_TOKEN(porthash); 2000 2001 phd = inp->inp_phd; 2002 LIST_REMOVE(inp, inp_portlist); 2003 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 2004 LIST_REMOVE(phd, phd_hash); 2005 kfree(phd, M_PCB); 2006 } 2007 2008 REL_PORTHASH_TOKEN(porthash); 2009 2010 inp->inp_phd = NULL; 2011 /* NOTE: Don't whack inp_lport, which may be used later */ 2012 } 2013 2014 static struct inp_localgroup * 2015 inp_localgroup_alloc(u_char af, uint16_t port, 2016 const union in_dependaddr *addr, int size) 2017 { 2018 struct inp_localgroup *grp; 2019 2020 grp = kmalloc(__offsetof(struct inp_localgroup, il_inp[size]), 2021 M_TEMP, M_INTWAIT | M_ZERO); 2022 grp->il_af = af; 2023 grp->il_lport = port; 2024 grp->il_dependladdr = *addr; 2025 grp->il_inpsiz = size; 2026 2027 return grp; 2028 } 2029 2030 static void 2031 inp_localgroup_free(struct inp_localgroup *grp) 2032 { 2033 kfree(grp, M_TEMP); 2034 } 2035 2036 static void 2037 inp_localgroup_destroy(struct inp_localgroup *grp) 2038 { 2039 LIST_REMOVE(grp, il_list); 2040 inp_localgroup_free(grp); 2041 } 2042 2043 static void 2044 inp_localgroup_copy(struct inp_localgroup *grp, 2045 const struct inp_localgroup *old_grp) 2046 { 2047 int i; 2048 2049 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 2050 ("invalid new local group size %d and old local group count %d", 2051 grp->il_inpsiz, old_grp->il_inpcnt)); 2052 for (i = 0; i < old_grp->il_inpcnt; ++i) 2053 grp->il_inp[i] = old_grp->il_inp[i]; 2054 grp->il_inpcnt = old_grp->il_inpcnt; 2055 } 2056 2057 static void 2058 in_pcbinslocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2059 { 2060 struct inp_localgrphead *hdr; 2061 struct inp_localgroup *grp, *grp_alloc = NULL; 2062 u_char isjailed; 2063 int i, idx; 2064 2065 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 2066 2067 if (pcbinfo->localgrphashbase == NULL) 2068 return; 2069 2070 /* 2071 * Further separate groups by whether the inp is jailed or not. 2072 * This allows the inp_localgroup_lookup() code to manage port 2073 * overloading between jails and non-jails. 2074 * 2075 * XXX all jails are collected into one group, which works fine 2076 * as we expect the jails to be listening on different addresses. 2077 * If this changes in the future we may have to break the groups 2078 * up by prison pointer as well. 2079 */ 2080 if (inp->inp_socket && inp->inp_socket->so_cred) 2081 isjailed = jailed(inp->inp_socket->so_cred); 2082 else 2083 isjailed = 0; 2084 2085 hdr = &pcbinfo->localgrphashbase[ 2086 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 2087 2088 again: 2089 LIST_FOREACH(grp, hdr, il_list) { 2090 if (grp->il_af == inp->inp_af && 2091 grp->il_lport == inp->inp_lport && 2092 grp->il_jailed == isjailed && 2093 memcmp(&grp->il_dependladdr, 2094 &inp->inp_inc.inc_ie.ie_dependladdr, 2095 sizeof(grp->il_dependladdr)) == 0) { 2096 break; 2097 } 2098 } 2099 if (grp == NULL) { 2100 /* 2101 * Create a new local group 2102 */ 2103 if (grp_alloc == NULL) { 2104 grp_alloc = inp_localgroup_alloc(inp->inp_af, 2105 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 2106 INP_LOCALGROUP_SIZMIN); 2107 /* 2108 * Local group allocation could block and the 2109 * local group w/ the same property might have 2110 * been added by others when we were blocked; 2111 * check again. 2112 */ 2113 goto again; 2114 } else { 2115 /* Local group has been allocated; link it */ 2116 grp = grp_alloc; 2117 grp->il_jailed = isjailed; 2118 grp_alloc = NULL; 2119 LIST_INSERT_HEAD(hdr, grp, il_list); 2120 } 2121 } else if (grp->il_inpcnt == grp->il_inpsiz) { 2122 #if 0 2123 /* 2124 * REMOVED - Ensure that all entries are placed in the 2125 * localgroup so jail operations can be 2126 * deterministic on a il_lport basis. 2127 */ 2128 if (grp->il_inpsiz >= INP_LOCALGROUP_SIZMAX) { 2129 static int limit_logged = 0; 2130 2131 if (!limit_logged) { 2132 limit_logged = 1; 2133 kprintf("local group port %d, " 2134 "limit reached\n", ntohs(grp->il_lport)); 2135 } 2136 if (grp_alloc != NULL) { 2137 /* 2138 * This would happen if the local group 2139 * w/ the same property was expanded when 2140 * our local group allocation blocked. 2141 */ 2142 inp_localgroup_free(grp_alloc); 2143 } 2144 return; 2145 } 2146 #endif 2147 2148 /* 2149 * Expand this local group 2150 */ 2151 if (grp_alloc == NULL || 2152 grp->il_inpcnt >= grp_alloc->il_inpsiz) { 2153 if (grp_alloc != NULL) 2154 inp_localgroup_free(grp_alloc); 2155 grp_alloc = inp_localgroup_alloc(grp->il_af, 2156 grp->il_lport, &grp->il_dependladdr, 2157 grp->il_inpsiz * 2); 2158 /* 2159 * Local group allocation could block and the 2160 * local group w/ the same property might have 2161 * been expanded by others when we were blocked; 2162 * check again. 2163 */ 2164 goto again; 2165 } 2166 2167 /* 2168 * Save the old local group, link the new one, and then 2169 * destroy the old local group 2170 */ 2171 inp_localgroup_copy(grp_alloc, grp); 2172 LIST_INSERT_HEAD(hdr, grp_alloc, il_list); 2173 inp_localgroup_destroy(grp); 2174 2175 grp = grp_alloc; 2176 grp->il_jailed = isjailed; 2177 grp_alloc = NULL; 2178 } else { 2179 /* 2180 * Found the local group 2181 */ 2182 if (grp_alloc != NULL) { 2183 /* 2184 * This would happen if the local group w/ the 2185 * same property was added or expanded when our 2186 * local group allocation blocked. 2187 */ 2188 inp_localgroup_free(grp_alloc); 2189 grp_alloc = NULL; 2190 } 2191 } 2192 2193 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 2194 ("invalid local group size %d and count %d", 2195 grp->il_inpsiz, grp->il_inpcnt)); 2196 2197 /* 2198 * Keep the local group sorted by the inpcb local group index 2199 * in ascending order. 2200 * 2201 * This eases the multi-process userland application which uses 2202 * SO_REUSEPORT sockets and binds process to the owner cpu of 2203 * the SO_REUSEPORT socket: 2204 * If we didn't sort the local group by the inpcb local group 2205 * index and one of the process owning an inpcb in this local 2206 * group restarted, e.g. crashed and restarted by watchdog, 2207 * other processes owning a inpcb in this local group would have 2208 * to detect that event, refetch its socket's owner cpu, and 2209 * re-bind. 2210 */ 2211 idx = grp->il_inpcnt; 2212 for (i = 0; i < idx; ++i) { 2213 struct inpcb *oinp = grp->il_inp[i]; 2214 2215 if (oinp->inp_lgrpindex > i) { 2216 if (inp->inp_lgrpindex < 0) { 2217 inp->inp_lgrpindex = i; 2218 } else if (inp->inp_lgrpindex != i) { 2219 if (bootverbose) { 2220 kprintf("inp %p: grpidx %d, " 2221 "assigned to %d, cpu%d\n", 2222 inp, inp->inp_lgrpindex, i, 2223 mycpuid); 2224 } 2225 } 2226 grp->il_inp[i] = inp; 2227 2228 /* Pull down inpcbs */ 2229 for (; i < grp->il_inpcnt; ++i) { 2230 struct inpcb *oinp1 = grp->il_inp[i + 1]; 2231 2232 grp->il_inp[i + 1] = oinp; 2233 oinp = oinp1; 2234 } 2235 grp->il_inpcnt++; 2236 return; 2237 } 2238 } 2239 2240 if (inp->inp_lgrpindex < 0) { 2241 inp->inp_lgrpindex = idx; 2242 } else if (inp->inp_lgrpindex != idx) { 2243 if (bootverbose) { 2244 kprintf("inp %p: grpidx %d, assigned to %d, cpu%d\n", 2245 inp, inp->inp_lgrpindex, idx, mycpuid); 2246 } 2247 } 2248 grp->il_inp[idx] = inp; 2249 grp->il_inpcnt++; 2250 } 2251 2252 void 2253 in_pcbinswildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2254 { 2255 struct inpcontainer *ic; 2256 struct inpcontainerhead *bucket; 2257 2258 GET_PCBINFO_TOKEN(pcbinfo); 2259 2260 in_pcbinslocalgrphash_oncpu(inp, pcbinfo); 2261 2262 bucket = &pcbinfo->wildcardhashbase[ 2263 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 2264 2265 ic = kmalloc(sizeof(struct inpcontainer), M_TEMP, M_INTWAIT); 2266 ic->ic_inp = inp; 2267 LIST_INSERT_HEAD(bucket, ic, ic_list); 2268 2269 REL_PCBINFO_TOKEN(pcbinfo); 2270 } 2271 2272 /* 2273 * Insert PCB into wildcard hash table. 2274 */ 2275 void 2276 in_pcbinswildcardhash(struct inpcb *inp) 2277 { 2278 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2279 2280 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 2281 ("not in correct netisr")); 2282 ASSERT_INP_NOTINHASH(inp); 2283 inp->inp_flags |= INP_WILDCARD; 2284 2285 in_pcbinswildcardhash_oncpu(inp, pcbinfo); 2286 } 2287 2288 static void 2289 in_pcbremlocalgrphash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2290 { 2291 struct inp_localgrphead *hdr; 2292 struct inp_localgroup *grp; 2293 2294 ASSERT_PCBINFO_TOKEN_HELD(pcbinfo); 2295 2296 if (pcbinfo->localgrphashbase == NULL) 2297 return; 2298 2299 hdr = &pcbinfo->localgrphashbase[ 2300 INP_PCBLOCALGRPHASH(inp->inp_lport, pcbinfo->localgrphashmask)]; 2301 2302 LIST_FOREACH(grp, hdr, il_list) { 2303 int i; 2304 2305 for (i = 0; i < grp->il_inpcnt; ++i) { 2306 if (grp->il_inp[i] != inp) 2307 continue; 2308 2309 if (grp->il_inpcnt == 1) { 2310 /* Destroy this local group */ 2311 inp_localgroup_destroy(grp); 2312 } else { 2313 /* Pull up inpcbs */ 2314 for (; i + 1 < grp->il_inpcnt; ++i) 2315 grp->il_inp[i] = grp->il_inp[i + 1]; 2316 grp->il_inpcnt--; 2317 } 2318 return; 2319 } 2320 } 2321 } 2322 2323 void 2324 in_pcbremwildcardhash_oncpu(struct inpcb *inp, struct inpcbinfo *pcbinfo) 2325 { 2326 struct inpcontainer *ic; 2327 struct inpcontainerhead *head; 2328 2329 GET_PCBINFO_TOKEN(pcbinfo); 2330 2331 in_pcbremlocalgrphash_oncpu(inp, pcbinfo); 2332 2333 /* find bucket */ 2334 head = &pcbinfo->wildcardhashbase[ 2335 INP_PCBWILDCARDHASH(inp->inp_lport, pcbinfo->wildcardhashmask)]; 2336 2337 LIST_FOREACH(ic, head, ic_list) { 2338 if (ic->ic_inp == inp) 2339 goto found; 2340 } 2341 REL_PCBINFO_TOKEN(pcbinfo); 2342 return; /* not found! */ 2343 2344 found: 2345 LIST_REMOVE(ic, ic_list); /* remove container from bucket chain */ 2346 REL_PCBINFO_TOKEN(pcbinfo); 2347 kfree(ic, M_TEMP); /* deallocate container */ 2348 } 2349 2350 /* 2351 * Remove PCB from wildcard hash table. 2352 */ 2353 void 2354 in_pcbremwildcardhash(struct inpcb *inp) 2355 { 2356 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2357 2358 KASSERT(&curthread->td_msgport == netisr_cpuport(pcbinfo->cpu), 2359 ("not in correct netisr")); 2360 KASSERT(inp->inp_flags & INP_WILDCARD, ("inp not wildcard")); 2361 2362 in_pcbremwildcardhash_oncpu(inp, pcbinfo); 2363 inp->inp_lgrpindex = -1; 2364 inp->inp_flags &= ~INP_WILDCARD; 2365 } 2366 2367 /* 2368 * Remove PCB from various lists. 2369 */ 2370 void 2371 in_pcbremlists(struct inpcb *inp) 2372 { 2373 in_pcbremporthash(inp); 2374 if (inp->inp_flags & INP_WILDCARD) { 2375 in_pcbremwildcardhash(inp); 2376 } else if (inp->inp_flags & INP_CONNECTED) { 2377 in_pcbremconnhash(inp); 2378 } 2379 2380 if (inp->inp_flags & INP_ONLIST) 2381 in_pcbofflist(inp); 2382 } 2383 2384 int 2385 prison_xinpcb(struct thread *td, struct inpcb *inp) 2386 { 2387 struct ucred *cr; 2388 2389 if (td->td_proc == NULL) 2390 return (0); 2391 cr = td->td_proc->p_ucred; 2392 if (cr->cr_prison == NULL) 2393 return (0); 2394 if (inp->inp_socket && inp->inp_socket->so_cred && 2395 inp->inp_socket->so_cred->cr_prison && 2396 cr->cr_prison == inp->inp_socket->so_cred->cr_prison) 2397 return (0); 2398 return (1); 2399 } 2400 2401 int 2402 in_pcblist_range(SYSCTL_HANDLER_ARGS) 2403 { 2404 struct inpcbinfo *pcbinfo_arr = arg1; 2405 int pcbinfo_arrlen = arg2; 2406 struct inpcb *marker; 2407 int cpu, origcpu; 2408 int error, n; 2409 2410 KASSERT(pcbinfo_arrlen <= netisr_ncpus && pcbinfo_arrlen >= 1, 2411 ("invalid pcbinfo count %d", pcbinfo_arrlen)); 2412 2413 /* 2414 * The process of preparing the TCB list is too time-consuming and 2415 * resource-intensive to repeat twice on every request. 2416 */ 2417 n = 0; 2418 if (req->oldptr == NULL) { 2419 for (cpu = 0; cpu < pcbinfo_arrlen; ++cpu) 2420 n += pcbinfo_arr[cpu].ipi_count; 2421 req->oldidx = (n + n/8 + 10) * sizeof(struct xinpcb); 2422 return 0; 2423 } 2424 2425 if (req->newptr != NULL) 2426 return EPERM; 2427 2428 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO); 2429 marker->inp_flags |= INP_PLACEMARKER; 2430 2431 /* 2432 * OK, now we're committed to doing something. Re-fetch ipi_count 2433 * after obtaining the generation count. 2434 */ 2435 error = 0; 2436 origcpu = mycpuid; 2437 for (cpu = 0; cpu < pcbinfo_arrlen && error == 0; ++cpu) { 2438 struct inpcbinfo *pcbinfo = &pcbinfo_arr[cpu]; 2439 struct inpcb *inp; 2440 struct xinpcb xi; 2441 int i; 2442 2443 lwkt_migratecpu(cpu); 2444 2445 GET_PCBINFO_TOKEN(pcbinfo); 2446 2447 n = pcbinfo->ipi_count; 2448 2449 LIST_INSERT_HEAD(&pcbinfo->pcblisthead, marker, inp_list); 2450 i = 0; 2451 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) { 2452 LIST_REMOVE(marker, inp_list); 2453 LIST_INSERT_AFTER(inp, marker, inp_list); 2454 2455 if (inp->inp_flags & INP_PLACEMARKER) 2456 continue; 2457 if (prison_xinpcb(req->td, inp)) 2458 continue; 2459 2460 bzero(&xi, sizeof xi); 2461 xi.xi_len = sizeof xi; 2462 bcopy(inp, &xi.xi_inp, sizeof *inp); 2463 if (inp->inp_socket) 2464 sotoxsocket(inp->inp_socket, &xi.xi_socket); 2465 if ((error = SYSCTL_OUT(req, &xi, sizeof xi)) != 0) 2466 break; 2467 ++i; 2468 } 2469 LIST_REMOVE(marker, inp_list); 2470 2471 REL_PCBINFO_TOKEN(pcbinfo); 2472 2473 if (error == 0 && i < n) { 2474 bzero(&xi, sizeof xi); 2475 xi.xi_len = sizeof xi; 2476 while (i < n) { 2477 error = SYSCTL_OUT(req, &xi, sizeof xi); 2478 if (error) 2479 break; 2480 ++i; 2481 } 2482 } 2483 } 2484 2485 lwkt_migratecpu(origcpu); 2486 kfree(marker, M_TEMP); 2487 return error; 2488 } 2489 2490 int 2491 in_pcblist_ncpus(SYSCTL_HANDLER_ARGS) 2492 { 2493 2494 return (in_pcblist_range(oidp, arg1, netisr_ncpus, req)); 2495 } 2496 2497 void 2498 in_savefaddr(struct socket *so, const struct sockaddr *faddr) 2499 { 2500 struct sockaddr_in *sin; 2501 2502 KASSERT(faddr->sa_family == AF_INET, 2503 ("not AF_INET faddr %d", faddr->sa_family)); 2504 2505 sin = kmalloc(sizeof(*sin), M_SONAME, M_WAITOK | M_ZERO); 2506 sin->sin_family = AF_INET; 2507 sin->sin_len = sizeof(*sin); 2508 sin->sin_port = ((const struct sockaddr_in *)faddr)->sin_port; 2509 sin->sin_addr = ((const struct sockaddr_in *)faddr)->sin_addr; 2510 2511 so->so_faddr = (struct sockaddr *)sin; 2512 } 2513 2514 void 2515 in_pcbportinfo_init(struct inpcbportinfo *portinfo, int hashsize, 2516 u_short offset) 2517 { 2518 memset(portinfo, 0, sizeof(*portinfo)); 2519 2520 portinfo->offset = offset; 2521 portinfo->porthashbase = phashinit(hashsize, M_PCB, 2522 &portinfo->porthashcnt); 2523 } 2524 2525 void 2526 in_pcbportrange(u_short *hi0, u_short *lo0, u_short ofs, u_short step) 2527 { 2528 int hi, lo; 2529 2530 if (step == 1) 2531 return; 2532 2533 hi = *hi0; 2534 lo = *lo0; 2535 2536 hi = rounddown(hi, step); 2537 hi += ofs; 2538 if (hi > (int)*hi0) 2539 hi -= step; 2540 2541 lo = roundup(lo, step); 2542 lo -= (step - ofs); 2543 if (lo < (int)*lo0) 2544 lo += step; 2545 2546 *hi0 = hi; 2547 *lo0 = lo; 2548 } 2549 2550 void 2551 in_pcbglobalinit(void) 2552 { 2553 int cpu; 2554 2555 in_pcbmarkers = kmalloc(netisr_ncpus * sizeof(struct inpcb), M_PCB, 2556 M_WAITOK | M_ZERO); 2557 in_pcbcontainer_markers = 2558 kmalloc(netisr_ncpus * sizeof(struct inpcontainer), M_PCB, 2559 M_WAITOK | M_ZERO); 2560 2561 for (cpu = 0; cpu < netisr_ncpus; ++cpu) { 2562 struct inpcontainer *ic = &in_pcbcontainer_markers[cpu]; 2563 struct inpcb *marker = &in_pcbmarkers[cpu]; 2564 2565 marker->inp_flags |= INP_PLACEMARKER; 2566 ic->ic_inp = marker; 2567 } 2568 } 2569 2570 struct inpcb * 2571 in_pcbmarker(void) 2572 { 2573 2574 ASSERT_NETISR_NCPUS(mycpuid); 2575 return &in_pcbmarkers[mycpuid]; 2576 } 2577 2578 struct inpcontainer * 2579 in_pcbcontainer_marker(void) 2580 { 2581 2582 ASSERT_NETISR_NCPUS(mycpuid); 2583 return &in_pcbcontainer_markers[mycpuid]; 2584 } 2585 2586 void 2587 in_pcbresetroute(struct inpcb *inp) 2588 { 2589 struct route *ro = &inp->inp_route; 2590 2591 if (ro->ro_rt != NULL) 2592 RTFREE(ro->ro_rt); 2593 bzero(ro, sizeof(*ro)); 2594 } 2595