1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1991, 1993, 1995 5 * The Regents of the University of California. 6 * Copyright (c) 2007-2009 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_ddb.h" 44 #include "opt_ipsec.h" 45 #include "opt_inet.h" 46 #include "opt_inet6.h" 47 #include "opt_ratelimit.h" 48 #include "opt_pcbgroup.h" 49 #include "opt_rss.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/callout.h> 57 #include <sys/eventhandler.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/rmlock.h> 61 #include <sys/smp.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <sys/sockio.h> 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/refcount.h> 68 #include <sys/jail.h> 69 #include <sys/kernel.h> 70 #include <sys/sysctl.h> 71 72 #ifdef DDB 73 #include <ddb/ddb.h> 74 #endif 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/if_types.h> 81 #include <net/if_llatbl.h> 82 #include <net/route.h> 83 #include <net/rss_config.h> 84 #include <net/vnet.h> 85 86 #if defined(INET) || defined(INET6) 87 #include <netinet/in.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/ip_var.h> 90 #include <netinet/tcp_var.h> 91 #ifdef TCPHPTS 92 #include <netinet/tcp_hpts.h> 93 #endif 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 #endif 97 #ifdef INET 98 #include <netinet/in_var.h> 99 #endif 100 #ifdef INET6 101 #include <netinet/ip6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #endif /* INET6 */ 106 107 #include <netipsec/ipsec_support.h> 108 109 #include <security/mac/mac_framework.h> 110 111 static struct callout ipport_tick_callout; 112 113 /* 114 * These configure the range of local port addresses assigned to 115 * "unspecified" outgoing connections/packets/whatever. 116 */ 117 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 118 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 119 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 120 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 121 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 122 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 123 124 /* 125 * Reserved ports accessible only to root. There are significant 126 * security considerations that must be accounted for when changing these, 127 * but the security benefits can be great. Please be careful. 128 */ 129 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 130 VNET_DEFINE(int, ipport_reservedlow); 131 132 /* Variables dealing with random ephemeral port allocation. */ 133 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ 134 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ 135 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ 136 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ 137 VNET_DEFINE(int, ipport_tcpallocs); 138 static VNET_DEFINE(int, ipport_tcplastcount); 139 140 #define V_ipport_tcplastcount VNET(ipport_tcplastcount) 141 142 static void in_pcbremlists(struct inpcb *inp); 143 #ifdef INET 144 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 145 struct in_addr faddr, u_int fport_arg, 146 struct in_addr laddr, u_int lport_arg, 147 int lookupflags, struct ifnet *ifp); 148 149 #define RANGECHK(var, min, max) \ 150 if ((var) < (min)) { (var) = (min); } \ 151 else if ((var) > (max)) { (var) = (max); } 152 153 static int 154 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 155 { 156 int error; 157 158 error = sysctl_handle_int(oidp, arg1, arg2, req); 159 if (error == 0) { 160 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 161 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 162 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 163 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 164 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 165 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 166 } 167 return (error); 168 } 169 170 #undef RANGECHK 171 172 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, 173 "IP Ports"); 174 175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 176 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 177 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", ""); 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 180 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", ""); 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 183 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", ""); 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 186 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", ""); 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 189 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", ""); 190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 192 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", ""); 193 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 194 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 195 &VNET_NAME(ipport_reservedhigh), 0, ""); 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 197 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 198 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 199 CTLFLAG_VNET | CTLFLAG_RW, 200 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, 202 CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " 204 "allocations before switching to a sequental one"); 205 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, 206 CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(ipport_randomtime), 0, 208 "Minimum time to keep sequental port " 209 "allocation before switching to a random one"); 210 #endif /* INET */ 211 212 /* 213 * in_pcb.c: manage the Protocol Control Blocks. 214 * 215 * NOTE: It is assumed that most of these functions will be called with 216 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 217 * functions often modify hash chains or addresses in pcbs. 218 */ 219 220 /* 221 * Different protocols initialize their inpcbs differently - giving 222 * different name to the lock. But they all are disposed the same. 223 */ 224 static void 225 inpcb_fini(void *mem, int size) 226 { 227 struct inpcb *inp = mem; 228 229 INP_LOCK_DESTROY(inp); 230 } 231 232 /* 233 * Initialize an inpcbinfo -- we should be able to reduce the number of 234 * arguments in time. 235 */ 236 void 237 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 238 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 239 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) 240 { 241 242 INP_INFO_LOCK_INIT(pcbinfo, name); 243 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 244 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 245 #ifdef VIMAGE 246 pcbinfo->ipi_vnet = curvnet; 247 #endif 248 pcbinfo->ipi_listhead = listhead; 249 LIST_INIT(pcbinfo->ipi_listhead); 250 pcbinfo->ipi_count = 0; 251 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 252 &pcbinfo->ipi_hashmask); 253 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 254 &pcbinfo->ipi_porthashmask); 255 #ifdef PCBGROUP 256 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 257 #endif 258 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 259 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); 260 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 261 uma_zone_set_warning(pcbinfo->ipi_zone, 262 "kern.ipc.maxsockets limit reached"); 263 } 264 265 /* 266 * Destroy an inpcbinfo. 267 */ 268 void 269 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 270 { 271 272 KASSERT(pcbinfo->ipi_count == 0, 273 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 274 275 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 276 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 277 pcbinfo->ipi_porthashmask); 278 #ifdef PCBGROUP 279 in_pcbgroup_destroy(pcbinfo); 280 #endif 281 uma_zdestroy(pcbinfo->ipi_zone); 282 INP_LIST_LOCK_DESTROY(pcbinfo); 283 INP_HASH_LOCK_DESTROY(pcbinfo); 284 INP_INFO_LOCK_DESTROY(pcbinfo); 285 } 286 287 /* 288 * Allocate a PCB and associate it with the socket. 289 * On success return with the PCB locked. 290 */ 291 int 292 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 293 { 294 struct inpcb *inp; 295 int error; 296 297 #ifdef INVARIANTS 298 if (pcbinfo == &V_tcbinfo) { 299 INP_INFO_RLOCK_ASSERT(pcbinfo); 300 } else { 301 INP_INFO_WLOCK_ASSERT(pcbinfo); 302 } 303 #endif 304 305 error = 0; 306 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 307 if (inp == NULL) 308 return (ENOBUFS); 309 bzero(&inp->inp_start_zero, inp_zero_size); 310 inp->inp_pcbinfo = pcbinfo; 311 inp->inp_socket = so; 312 inp->inp_cred = crhold(so->so_cred); 313 inp->inp_inc.inc_fibnum = so->so_fibnum; 314 #ifdef MAC 315 error = mac_inpcb_init(inp, M_NOWAIT); 316 if (error != 0) 317 goto out; 318 mac_inpcb_create(so, inp); 319 #endif 320 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 321 error = ipsec_init_pcbpolicy(inp); 322 if (error != 0) { 323 #ifdef MAC 324 mac_inpcb_destroy(inp); 325 #endif 326 goto out; 327 } 328 #endif /*IPSEC*/ 329 #ifdef INET6 330 if (INP_SOCKAF(so) == AF_INET6) { 331 inp->inp_vflag |= INP_IPV6PROTO; 332 if (V_ip6_v6only) 333 inp->inp_flags |= IN6P_IPV6_V6ONLY; 334 } 335 #endif 336 INP_WLOCK(inp); 337 INP_LIST_WLOCK(pcbinfo); 338 LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 339 pcbinfo->ipi_count++; 340 so->so_pcb = (caddr_t)inp; 341 #ifdef INET6 342 if (V_ip6_auto_flowlabel) 343 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 344 #endif 345 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 346 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 347 348 /* 349 * Routes in inpcb's can cache L2 as well; they are guaranteed 350 * to be cleaned up. 351 */ 352 inp->inp_route.ro_flags = RT_LLE_CACHE; 353 INP_LIST_WUNLOCK(pcbinfo); 354 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 355 out: 356 if (error != 0) { 357 crfree(inp->inp_cred); 358 uma_zfree(pcbinfo->ipi_zone, inp); 359 } 360 #endif 361 return (error); 362 } 363 364 #ifdef INET 365 int 366 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 367 { 368 int anonport, error; 369 370 INP_WLOCK_ASSERT(inp); 371 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 372 373 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 374 return (EINVAL); 375 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 376 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 377 &inp->inp_lport, cred); 378 if (error) 379 return (error); 380 if (in_pcbinshash(inp) != 0) { 381 inp->inp_laddr.s_addr = INADDR_ANY; 382 inp->inp_lport = 0; 383 return (EAGAIN); 384 } 385 if (anonport) 386 inp->inp_flags |= INP_ANONPORT; 387 return (0); 388 } 389 #endif 390 391 /* 392 * Select a local port (number) to use. 393 */ 394 #if defined(INET) || defined(INET6) 395 int 396 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 397 struct ucred *cred, int lookupflags) 398 { 399 struct inpcbinfo *pcbinfo; 400 struct inpcb *tmpinp; 401 unsigned short *lastport; 402 int count, dorandom, error; 403 u_short aux, first, last, lport; 404 #ifdef INET 405 struct in_addr laddr; 406 #endif 407 408 pcbinfo = inp->inp_pcbinfo; 409 410 /* 411 * Because no actual state changes occur here, a global write lock on 412 * the pcbinfo isn't required. 413 */ 414 INP_LOCK_ASSERT(inp); 415 INP_HASH_LOCK_ASSERT(pcbinfo); 416 417 if (inp->inp_flags & INP_HIGHPORT) { 418 first = V_ipport_hifirstauto; /* sysctl */ 419 last = V_ipport_hilastauto; 420 lastport = &pcbinfo->ipi_lasthi; 421 } else if (inp->inp_flags & INP_LOWPORT) { 422 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0); 423 if (error) 424 return (error); 425 first = V_ipport_lowfirstauto; /* 1023 */ 426 last = V_ipport_lowlastauto; /* 600 */ 427 lastport = &pcbinfo->ipi_lastlow; 428 } else { 429 first = V_ipport_firstauto; /* sysctl */ 430 last = V_ipport_lastauto; 431 lastport = &pcbinfo->ipi_lastport; 432 } 433 /* 434 * For UDP(-Lite), use random port allocation as long as the user 435 * allows it. For TCP (and as of yet unknown) connections, 436 * use random port allocation only if the user allows it AND 437 * ipport_tick() allows it. 438 */ 439 if (V_ipport_randomized && 440 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 441 pcbinfo == &V_ulitecbinfo)) 442 dorandom = 1; 443 else 444 dorandom = 0; 445 /* 446 * It makes no sense to do random port allocation if 447 * we have the only port available. 448 */ 449 if (first == last) 450 dorandom = 0; 451 /* Make sure to not include UDP(-Lite) packets in the count. */ 452 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 453 V_ipport_tcpallocs++; 454 /* 455 * Instead of having two loops further down counting up or down 456 * make sure that first is always <= last and go with only one 457 * code path implementing all logic. 458 */ 459 if (first > last) { 460 aux = first; 461 first = last; 462 last = aux; 463 } 464 465 #ifdef INET 466 /* Make the compiler happy. */ 467 laddr.s_addr = 0; 468 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 469 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", 470 __func__, inp)); 471 laddr = *laddrp; 472 } 473 #endif 474 tmpinp = NULL; /* Make compiler happy. */ 475 lport = *lportp; 476 477 if (dorandom) 478 *lastport = first + (arc4random() % (last - first)); 479 480 count = last - first; 481 482 do { 483 if (count-- < 0) /* completely used? */ 484 return (EADDRNOTAVAIL); 485 ++*lastport; 486 if (*lastport < first || *lastport > last) 487 *lastport = first; 488 lport = htons(*lastport); 489 490 #ifdef INET6 491 if ((inp->inp_vflag & INP_IPV6) != 0) 492 tmpinp = in6_pcblookup_local(pcbinfo, 493 &inp->in6p_laddr, lport, lookupflags, cred); 494 #endif 495 #if defined(INET) && defined(INET6) 496 else 497 #endif 498 #ifdef INET 499 tmpinp = in_pcblookup_local(pcbinfo, laddr, 500 lport, lookupflags, cred); 501 #endif 502 } while (tmpinp != NULL); 503 504 #ifdef INET 505 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) 506 laddrp->s_addr = laddr.s_addr; 507 #endif 508 *lportp = lport; 509 510 return (0); 511 } 512 513 /* 514 * Return cached socket options. 515 */ 516 short 517 inp_so_options(const struct inpcb *inp) 518 { 519 short so_options; 520 521 so_options = 0; 522 523 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 524 so_options |= SO_REUSEPORT; 525 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 526 so_options |= SO_REUSEADDR; 527 return (so_options); 528 } 529 #endif /* INET || INET6 */ 530 531 /* 532 * Check if a new BINDMULTI socket is allowed to be created. 533 * 534 * ni points to the new inp. 535 * oi points to the exisitng inp. 536 * 537 * This checks whether the existing inp also has BINDMULTI and 538 * whether the credentials match. 539 */ 540 int 541 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 542 { 543 /* Check permissions match */ 544 if ((ni->inp_flags2 & INP_BINDMULTI) && 545 (ni->inp_cred->cr_uid != 546 oi->inp_cred->cr_uid)) 547 return (0); 548 549 /* Check the existing inp has BINDMULTI set */ 550 if ((ni->inp_flags2 & INP_BINDMULTI) && 551 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 552 return (0); 553 554 /* 555 * We're okay - either INP_BINDMULTI isn't set on ni, or 556 * it is and it matches the checks. 557 */ 558 return (1); 559 } 560 561 #ifdef INET 562 /* 563 * Set up a bind operation on a PCB, performing port allocation 564 * as required, but do not actually modify the PCB. Callers can 565 * either complete the bind by setting inp_laddr/inp_lport and 566 * calling in_pcbinshash(), or they can just use the resulting 567 * port and address to authorise the sending of a once-off packet. 568 * 569 * On error, the values of *laddrp and *lportp are not changed. 570 */ 571 int 572 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 573 u_short *lportp, struct ucred *cred) 574 { 575 struct socket *so = inp->inp_socket; 576 struct sockaddr_in *sin; 577 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 578 struct in_addr laddr; 579 u_short lport = 0; 580 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 581 int error; 582 583 /* 584 * No state changes, so read locks are sufficient here. 585 */ 586 INP_LOCK_ASSERT(inp); 587 INP_HASH_LOCK_ASSERT(pcbinfo); 588 589 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 590 return (EADDRNOTAVAIL); 591 laddr.s_addr = *laddrp; 592 if (nam != NULL && laddr.s_addr != INADDR_ANY) 593 return (EINVAL); 594 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0) 595 lookupflags = INPLOOKUP_WILDCARD; 596 if (nam == NULL) { 597 if ((error = prison_local_ip4(cred, &laddr)) != 0) 598 return (error); 599 } else { 600 sin = (struct sockaddr_in *)nam; 601 if (nam->sa_len != sizeof (*sin)) 602 return (EINVAL); 603 #ifdef notdef 604 /* 605 * We should check the family, but old programs 606 * incorrectly fail to initialize it. 607 */ 608 if (sin->sin_family != AF_INET) 609 return (EAFNOSUPPORT); 610 #endif 611 error = prison_local_ip4(cred, &sin->sin_addr); 612 if (error) 613 return (error); 614 if (sin->sin_port != *lportp) { 615 /* Don't allow the port to change. */ 616 if (*lportp != 0) 617 return (EINVAL); 618 lport = sin->sin_port; 619 } 620 /* NB: lport is left as 0 if the port isn't being changed. */ 621 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 622 /* 623 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 624 * allow complete duplication of binding if 625 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 626 * and a multicast address is bound on both 627 * new and duplicated sockets. 628 */ 629 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 630 reuseport = SO_REUSEADDR|SO_REUSEPORT; 631 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 632 sin->sin_port = 0; /* yech... */ 633 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 634 /* 635 * Is the address a local IP address? 636 * If INP_BINDANY is set, then the socket may be bound 637 * to any endpoint address, local or not. 638 */ 639 if ((inp->inp_flags & INP_BINDANY) == 0 && 640 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 641 return (EADDRNOTAVAIL); 642 } 643 laddr = sin->sin_addr; 644 if (lport) { 645 struct inpcb *t; 646 struct tcptw *tw; 647 648 /* GROSS */ 649 if (ntohs(lport) <= V_ipport_reservedhigh && 650 ntohs(lport) >= V_ipport_reservedlow && 651 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 652 0)) 653 return (EACCES); 654 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 655 priv_check_cred(inp->inp_cred, 656 PRIV_NETINET_REUSEPORT, 0) != 0) { 657 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 658 lport, INPLOOKUP_WILDCARD, cred); 659 /* 660 * XXX 661 * This entire block sorely needs a rewrite. 662 */ 663 if (t && 664 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 665 ((t->inp_flags & INP_TIMEWAIT) == 0) && 666 (so->so_type != SOCK_STREAM || 667 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 668 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 669 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 670 (t->inp_flags2 & INP_REUSEPORT) == 0) && 671 (inp->inp_cred->cr_uid != 672 t->inp_cred->cr_uid)) 673 return (EADDRINUSE); 674 675 /* 676 * If the socket is a BINDMULTI socket, then 677 * the credentials need to match and the 678 * original socket also has to have been bound 679 * with BINDMULTI. 680 */ 681 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 682 return (EADDRINUSE); 683 } 684 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 685 lport, lookupflags, cred); 686 if (t && (t->inp_flags & INP_TIMEWAIT)) { 687 /* 688 * XXXRW: If an incpb has had its timewait 689 * state recycled, we treat the address as 690 * being in use (for now). This is better 691 * than a panic, but not desirable. 692 */ 693 tw = intotw(t); 694 if (tw == NULL || 695 (reuseport & tw->tw_so_options) == 0) 696 return (EADDRINUSE); 697 } else if (t && 698 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 699 (reuseport & inp_so_options(t)) == 0) { 700 #ifdef INET6 701 if (ntohl(sin->sin_addr.s_addr) != 702 INADDR_ANY || 703 ntohl(t->inp_laddr.s_addr) != 704 INADDR_ANY || 705 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 706 (t->inp_vflag & INP_IPV6PROTO) == 0) 707 #endif 708 return (EADDRINUSE); 709 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 710 return (EADDRINUSE); 711 } 712 } 713 } 714 if (*lportp != 0) 715 lport = *lportp; 716 if (lport == 0) { 717 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 718 if (error != 0) 719 return (error); 720 721 } 722 *laddrp = laddr.s_addr; 723 *lportp = lport; 724 return (0); 725 } 726 727 /* 728 * Connect from a socket to a specified address. 729 * Both address and port must be specified in argument sin. 730 * If don't have a local address for this socket yet, 731 * then pick one. 732 */ 733 int 734 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 735 struct ucred *cred, struct mbuf *m) 736 { 737 u_short lport, fport; 738 in_addr_t laddr, faddr; 739 int anonport, error; 740 741 INP_WLOCK_ASSERT(inp); 742 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 743 744 lport = inp->inp_lport; 745 laddr = inp->inp_laddr.s_addr; 746 anonport = (lport == 0); 747 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 748 NULL, cred); 749 if (error) 750 return (error); 751 752 /* Do the initial binding of the local address if required. */ 753 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 754 inp->inp_lport = lport; 755 inp->inp_laddr.s_addr = laddr; 756 if (in_pcbinshash(inp) != 0) { 757 inp->inp_laddr.s_addr = INADDR_ANY; 758 inp->inp_lport = 0; 759 return (EAGAIN); 760 } 761 } 762 763 /* Commit the remaining changes. */ 764 inp->inp_lport = lport; 765 inp->inp_laddr.s_addr = laddr; 766 inp->inp_faddr.s_addr = faddr; 767 inp->inp_fport = fport; 768 in_pcbrehash_mbuf(inp, m); 769 770 if (anonport) 771 inp->inp_flags |= INP_ANONPORT; 772 return (0); 773 } 774 775 int 776 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 777 { 778 779 return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); 780 } 781 782 /* 783 * Do proper source address selection on an unbound socket in case 784 * of connect. Take jails into account as well. 785 */ 786 int 787 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 788 struct ucred *cred) 789 { 790 struct ifaddr *ifa; 791 struct sockaddr *sa; 792 struct sockaddr_in *sin; 793 struct route sro; 794 int error; 795 796 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 797 /* 798 * Bypass source address selection and use the primary jail IP 799 * if requested. 800 */ 801 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 802 return (0); 803 804 error = 0; 805 bzero(&sro, sizeof(sro)); 806 807 sin = (struct sockaddr_in *)&sro.ro_dst; 808 sin->sin_family = AF_INET; 809 sin->sin_len = sizeof(struct sockaddr_in); 810 sin->sin_addr.s_addr = faddr->s_addr; 811 812 /* 813 * If route is known our src addr is taken from the i/f, 814 * else punt. 815 * 816 * Find out route to destination. 817 */ 818 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 819 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 820 821 /* 822 * If we found a route, use the address corresponding to 823 * the outgoing interface. 824 * 825 * Otherwise assume faddr is reachable on a directly connected 826 * network and try to find a corresponding interface to take 827 * the source address from. 828 */ 829 NET_EPOCH_ENTER(); 830 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 831 struct in_ifaddr *ia; 832 struct ifnet *ifp; 833 834 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 835 inp->inp_socket->so_fibnum)); 836 if (ia == NULL) { 837 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 838 inp->inp_socket->so_fibnum)); 839 840 } 841 if (ia == NULL) { 842 printf("ifa_ifwithnet failed\n"); 843 error = ENETUNREACH; 844 goto done; 845 } 846 847 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 848 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 849 goto done; 850 } 851 852 ifp = ia->ia_ifp; 853 ia = NULL; 854 IF_ADDR_RLOCK(ifp); 855 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 856 857 sa = ifa->ifa_addr; 858 if (sa->sa_family != AF_INET) 859 continue; 860 sin = (struct sockaddr_in *)sa; 861 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 862 ia = (struct in_ifaddr *)ifa; 863 break; 864 } 865 } 866 if (ia != NULL) { 867 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 868 IF_ADDR_RUNLOCK(ifp); 869 goto done; 870 } 871 IF_ADDR_RUNLOCK(ifp); 872 873 /* 3. As a last resort return the 'default' jail address. */ 874 error = prison_get_ip4(cred, laddr); 875 goto done; 876 } 877 878 /* 879 * If the outgoing interface on the route found is not 880 * a loopback interface, use the address from that interface. 881 * In case of jails do those three steps: 882 * 1. check if the interface address belongs to the jail. If so use it. 883 * 2. check if we have any address on the outgoing interface 884 * belonging to this jail. If so use it. 885 * 3. as a last resort return the 'default' jail address. 886 */ 887 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 888 struct in_ifaddr *ia; 889 struct ifnet *ifp; 890 891 /* If not jailed, use the default returned. */ 892 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 893 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 894 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 895 goto done; 896 } 897 898 /* Jailed. */ 899 /* 1. Check if the iface address belongs to the jail. */ 900 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 901 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 902 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 903 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 904 goto done; 905 } 906 907 /* 908 * 2. Check if we have any address on the outgoing interface 909 * belonging to this jail. 910 */ 911 ia = NULL; 912 ifp = sro.ro_rt->rt_ifp; 913 IF_ADDR_RLOCK(ifp); 914 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 915 sa = ifa->ifa_addr; 916 if (sa->sa_family != AF_INET) 917 continue; 918 sin = (struct sockaddr_in *)sa; 919 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 920 ia = (struct in_ifaddr *)ifa; 921 break; 922 } 923 } 924 if (ia != NULL) { 925 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 926 IF_ADDR_RUNLOCK(ifp); 927 goto done; 928 } 929 IF_ADDR_RUNLOCK(ifp); 930 931 /* 3. As a last resort return the 'default' jail address. */ 932 error = prison_get_ip4(cred, laddr); 933 goto done; 934 } 935 936 /* 937 * The outgoing interface is marked with 'loopback net', so a route 938 * to ourselves is here. 939 * Try to find the interface of the destination address and then 940 * take the address from there. That interface is not necessarily 941 * a loopback interface. 942 * In case of jails, check that it is an address of the jail 943 * and if we cannot find, fall back to the 'default' jail address. 944 */ 945 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 946 struct sockaddr_in sain; 947 struct in_ifaddr *ia; 948 949 bzero(&sain, sizeof(struct sockaddr_in)); 950 sain.sin_family = AF_INET; 951 sain.sin_len = sizeof(struct sockaddr_in); 952 sain.sin_addr.s_addr = faddr->s_addr; 953 954 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), 955 inp->inp_socket->so_fibnum)); 956 if (ia == NULL) 957 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, 958 inp->inp_socket->so_fibnum)); 959 if (ia == NULL) 960 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 961 962 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 963 if (ia == NULL) { 964 error = ENETUNREACH; 965 goto done; 966 } 967 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 968 goto done; 969 } 970 971 /* Jailed. */ 972 if (ia != NULL) { 973 struct ifnet *ifp; 974 975 ifp = ia->ia_ifp; 976 ia = NULL; 977 IF_ADDR_RLOCK(ifp); 978 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 979 980 sa = ifa->ifa_addr; 981 if (sa->sa_family != AF_INET) 982 continue; 983 sin = (struct sockaddr_in *)sa; 984 if (prison_check_ip4(cred, 985 &sin->sin_addr) == 0) { 986 ia = (struct in_ifaddr *)ifa; 987 break; 988 } 989 } 990 if (ia != NULL) { 991 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 992 IF_ADDR_RUNLOCK(ifp); 993 goto done; 994 } 995 IF_ADDR_RUNLOCK(ifp); 996 } 997 998 /* 3. As a last resort return the 'default' jail address. */ 999 error = prison_get_ip4(cred, laddr); 1000 goto done; 1001 } 1002 1003 done: 1004 NET_EPOCH_EXIT(); 1005 if (sro.ro_rt != NULL) 1006 RTFREE(sro.ro_rt); 1007 return (error); 1008 } 1009 1010 /* 1011 * Set up for a connect from a socket to the specified address. 1012 * On entry, *laddrp and *lportp should contain the current local 1013 * address and port for the PCB; these are updated to the values 1014 * that should be placed in inp_laddr and inp_lport to complete 1015 * the connect. 1016 * 1017 * On success, *faddrp and *fportp will be set to the remote address 1018 * and port. These are not updated in the error case. 1019 * 1020 * If the operation fails because the connection already exists, 1021 * *oinpp will be set to the PCB of that connection so that the 1022 * caller can decide to override it. In all other cases, *oinpp 1023 * is set to NULL. 1024 */ 1025 int 1026 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1027 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1028 struct inpcb **oinpp, struct ucred *cred) 1029 { 1030 struct rm_priotracker in_ifa_tracker; 1031 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1032 struct in_ifaddr *ia; 1033 struct inpcb *oinp; 1034 struct in_addr laddr, faddr; 1035 u_short lport, fport; 1036 int error; 1037 1038 /* 1039 * Because a global state change doesn't actually occur here, a read 1040 * lock is sufficient. 1041 */ 1042 INP_LOCK_ASSERT(inp); 1043 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1044 1045 if (oinpp != NULL) 1046 *oinpp = NULL; 1047 if (nam->sa_len != sizeof (*sin)) 1048 return (EINVAL); 1049 if (sin->sin_family != AF_INET) 1050 return (EAFNOSUPPORT); 1051 if (sin->sin_port == 0) 1052 return (EADDRNOTAVAIL); 1053 laddr.s_addr = *laddrp; 1054 lport = *lportp; 1055 faddr = sin->sin_addr; 1056 fport = sin->sin_port; 1057 1058 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1059 /* 1060 * If the destination address is INADDR_ANY, 1061 * use the primary local address. 1062 * If the supplied address is INADDR_BROADCAST, 1063 * and the primary interface supports broadcast, 1064 * choose the broadcast address for that interface. 1065 */ 1066 if (faddr.s_addr == INADDR_ANY) { 1067 IN_IFADDR_RLOCK(&in_ifa_tracker); 1068 faddr = 1069 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1070 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1071 if (cred != NULL && 1072 (error = prison_get_ip4(cred, &faddr)) != 0) 1073 return (error); 1074 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1075 IN_IFADDR_RLOCK(&in_ifa_tracker); 1076 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1077 IFF_BROADCAST) 1078 faddr = satosin(&CK_STAILQ_FIRST( 1079 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1080 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1081 } 1082 } 1083 if (laddr.s_addr == INADDR_ANY) { 1084 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1085 /* 1086 * If the destination address is multicast and an outgoing 1087 * interface has been set as a multicast option, prefer the 1088 * address of that interface as our source address. 1089 */ 1090 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1091 inp->inp_moptions != NULL) { 1092 struct ip_moptions *imo; 1093 struct ifnet *ifp; 1094 1095 imo = inp->inp_moptions; 1096 if (imo->imo_multicast_ifp != NULL) { 1097 ifp = imo->imo_multicast_ifp; 1098 IN_IFADDR_RLOCK(&in_ifa_tracker); 1099 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1100 if ((ia->ia_ifp == ifp) && 1101 (cred == NULL || 1102 prison_check_ip4(cred, 1103 &ia->ia_addr.sin_addr) == 0)) 1104 break; 1105 } 1106 if (ia == NULL) 1107 error = EADDRNOTAVAIL; 1108 else { 1109 laddr = ia->ia_addr.sin_addr; 1110 error = 0; 1111 } 1112 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1113 } 1114 } 1115 if (error) 1116 return (error); 1117 } 1118 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1119 laddr, lport, 0, NULL); 1120 if (oinp != NULL) { 1121 if (oinpp != NULL) 1122 *oinpp = oinp; 1123 return (EADDRINUSE); 1124 } 1125 if (lport == 0) { 1126 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1127 cred); 1128 if (error) 1129 return (error); 1130 } 1131 *laddrp = laddr.s_addr; 1132 *lportp = lport; 1133 *faddrp = faddr.s_addr; 1134 *fportp = fport; 1135 return (0); 1136 } 1137 1138 void 1139 in_pcbdisconnect(struct inpcb *inp) 1140 { 1141 1142 INP_WLOCK_ASSERT(inp); 1143 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1144 1145 inp->inp_faddr.s_addr = INADDR_ANY; 1146 inp->inp_fport = 0; 1147 in_pcbrehash(inp); 1148 } 1149 #endif /* INET */ 1150 1151 /* 1152 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1153 * For most protocols, this will be invoked immediately prior to calling 1154 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1155 * socket, in which case in_pcbfree() is deferred. 1156 */ 1157 void 1158 in_pcbdetach(struct inpcb *inp) 1159 { 1160 1161 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1162 1163 #ifdef RATELIMIT 1164 if (inp->inp_snd_tag != NULL) 1165 in_pcbdetach_txrtlmt(inp); 1166 #endif 1167 inp->inp_socket->so_pcb = NULL; 1168 inp->inp_socket = NULL; 1169 } 1170 1171 /* 1172 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1173 * stability of an inpcb pointer despite the inpcb lock being released. This 1174 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1175 * but where the inpcb lock may already held, or when acquiring a reference 1176 * via a pcbgroup. 1177 * 1178 * in_pcbref() should be used only to provide brief memory stability, and 1179 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1180 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1181 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1182 * lock and rele are the *only* safe operations that may be performed on the 1183 * inpcb. 1184 * 1185 * While the inpcb will not be freed, releasing the inpcb lock means that the 1186 * connection's state may change, so the caller should be careful to 1187 * revalidate any cached state on reacquiring the lock. Drop the reference 1188 * using in_pcbrele(). 1189 */ 1190 void 1191 in_pcbref(struct inpcb *inp) 1192 { 1193 1194 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1195 1196 refcount_acquire(&inp->inp_refcount); 1197 } 1198 1199 /* 1200 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1201 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1202 * return a flag indicating whether or not the inpcb remains valid. If it is 1203 * valid, we return with the inpcb lock held. 1204 * 1205 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1206 * reference on an inpcb. Historically more work was done here (actually, in 1207 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1208 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1209 * about memory stability (and continued use of the write lock). 1210 */ 1211 int 1212 in_pcbrele_rlocked(struct inpcb *inp) 1213 { 1214 struct inpcbinfo *pcbinfo; 1215 1216 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1217 1218 INP_RLOCK_ASSERT(inp); 1219 1220 if (refcount_release(&inp->inp_refcount) == 0) { 1221 /* 1222 * If the inpcb has been freed, let the caller know, even if 1223 * this isn't the last reference. 1224 */ 1225 if (inp->inp_flags2 & INP_FREED) { 1226 INP_RUNLOCK(inp); 1227 return (1); 1228 } 1229 return (0); 1230 } 1231 1232 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1233 #ifdef TCPHPTS 1234 if (inp->inp_in_hpts || inp->inp_in_input) { 1235 struct tcp_hpts_entry *hpts; 1236 /* 1237 * We should not be on the hpts at 1238 * this point in any form. we must 1239 * get the lock to be sure. 1240 */ 1241 hpts = tcp_hpts_lock(inp); 1242 if (inp->inp_in_hpts) 1243 panic("Hpts:%p inp:%p at free still on hpts", 1244 hpts, inp); 1245 mtx_unlock(&hpts->p_mtx); 1246 hpts = tcp_input_lock(inp); 1247 if (inp->inp_in_input) 1248 panic("Hpts:%p inp:%p at free still on input hpts", 1249 hpts, inp); 1250 mtx_unlock(&hpts->p_mtx); 1251 } 1252 #endif 1253 INP_RUNLOCK(inp); 1254 pcbinfo = inp->inp_pcbinfo; 1255 uma_zfree(pcbinfo->ipi_zone, inp); 1256 return (1); 1257 } 1258 1259 int 1260 in_pcbrele_wlocked(struct inpcb *inp) 1261 { 1262 struct inpcbinfo *pcbinfo; 1263 1264 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1265 1266 INP_WLOCK_ASSERT(inp); 1267 1268 if (refcount_release(&inp->inp_refcount) == 0) { 1269 /* 1270 * If the inpcb has been freed, let the caller know, even if 1271 * this isn't the last reference. 1272 */ 1273 if (inp->inp_flags2 & INP_FREED) { 1274 INP_WUNLOCK(inp); 1275 return (1); 1276 } 1277 return (0); 1278 } 1279 1280 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1281 #ifdef TCPHPTS 1282 if (inp->inp_in_hpts || inp->inp_in_input) { 1283 struct tcp_hpts_entry *hpts; 1284 /* 1285 * We should not be on the hpts at 1286 * this point in any form. we must 1287 * get the lock to be sure. 1288 */ 1289 hpts = tcp_hpts_lock(inp); 1290 if (inp->inp_in_hpts) 1291 panic("Hpts:%p inp:%p at free still on hpts", 1292 hpts, inp); 1293 mtx_unlock(&hpts->p_mtx); 1294 hpts = tcp_input_lock(inp); 1295 if (inp->inp_in_input) 1296 panic("Hpts:%p inp:%p at free still on input hpts", 1297 hpts, inp); 1298 mtx_unlock(&hpts->p_mtx); 1299 } 1300 #endif 1301 INP_WUNLOCK(inp); 1302 pcbinfo = inp->inp_pcbinfo; 1303 uma_zfree(pcbinfo->ipi_zone, inp); 1304 return (1); 1305 } 1306 1307 /* 1308 * Temporary wrapper. 1309 */ 1310 int 1311 in_pcbrele(struct inpcb *inp) 1312 { 1313 1314 return (in_pcbrele_wlocked(inp)); 1315 } 1316 1317 void 1318 in_pcblist_rele_rlocked(epoch_context_t ctx) 1319 { 1320 struct in_pcblist *il; 1321 struct inpcb *inp; 1322 struct inpcbinfo *pcbinfo; 1323 int i, n; 1324 1325 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1326 pcbinfo = il->il_pcbinfo; 1327 n = il->il_count; 1328 INP_INFO_WLOCK(pcbinfo); 1329 for (i = 0; i < n; i++) { 1330 inp = il->il_inp_list[i]; 1331 INP_RLOCK(inp); 1332 if (!in_pcbrele_rlocked(inp)) 1333 INP_RUNLOCK(inp); 1334 } 1335 INP_INFO_WUNLOCK(pcbinfo); 1336 free(il, M_TEMP); 1337 } 1338 1339 /* 1340 * Unconditionally schedule an inpcb to be freed by decrementing its 1341 * reference count, which should occur only after the inpcb has been detached 1342 * from its socket. If another thread holds a temporary reference (acquired 1343 * using in_pcbref()) then the free is deferred until that reference is 1344 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1345 * work, including removal from global lists, is done in this context, where 1346 * the pcbinfo lock is held. 1347 */ 1348 void 1349 in_pcbfree(struct inpcb *inp) 1350 { 1351 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1352 1353 #ifdef INET6 1354 struct ip6_moptions *im6o = NULL; 1355 #endif 1356 #ifdef INET 1357 struct ip_moptions *imo = NULL; 1358 #endif 1359 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1360 1361 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1362 ("%s: called twice for pcb %p", __func__, inp)); 1363 if (inp->inp_flags2 & INP_FREED) { 1364 INP_WUNLOCK(inp); 1365 return; 1366 } 1367 1368 #ifdef INVARIANTS 1369 if (pcbinfo == &V_tcbinfo) { 1370 INP_INFO_LOCK_ASSERT(pcbinfo); 1371 } else { 1372 INP_INFO_WLOCK_ASSERT(pcbinfo); 1373 } 1374 #endif 1375 INP_WLOCK_ASSERT(inp); 1376 1377 #ifdef INET 1378 imo = inp->inp_moptions; 1379 inp->inp_moptions = NULL; 1380 #endif 1381 /* XXXRW: Do as much as possible here. */ 1382 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1383 if (inp->inp_sp != NULL) 1384 ipsec_delete_pcbpolicy(inp); 1385 #endif 1386 INP_LIST_WLOCK(pcbinfo); 1387 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1388 in_pcbremlists(inp); 1389 INP_LIST_WUNLOCK(pcbinfo); 1390 #ifdef INET6 1391 if (inp->inp_vflag & INP_IPV6PROTO) { 1392 ip6_freepcbopts(inp->in6p_outputopts); 1393 im6o = inp->in6p_moptions; 1394 inp->in6p_moptions = NULL; 1395 } 1396 #endif 1397 if (inp->inp_options) 1398 (void)m_free(inp->inp_options); 1399 RO_INVALIDATE_CACHE(&inp->inp_route); 1400 1401 inp->inp_vflag = 0; 1402 inp->inp_flags2 |= INP_FREED; 1403 crfree(inp->inp_cred); 1404 #ifdef MAC 1405 mac_inpcb_destroy(inp); 1406 #endif 1407 #ifdef INET6 1408 ip6_freemoptions(im6o); 1409 #endif 1410 #ifdef INET 1411 inp_freemoptions(imo); 1412 #endif 1413 if (!in_pcbrele_wlocked(inp)) 1414 INP_WUNLOCK(inp); 1415 } 1416 1417 /* 1418 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1419 * port reservation, and preventing it from being returned by inpcb lookups. 1420 * 1421 * It is used by TCP to mark an inpcb as unused and avoid future packet 1422 * delivery or event notification when a socket remains open but TCP has 1423 * closed. This might occur as a result of a shutdown()-initiated TCP close 1424 * or a RST on the wire, and allows the port binding to be reused while still 1425 * maintaining the invariant that so_pcb always points to a valid inpcb until 1426 * in_pcbdetach(). 1427 * 1428 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1429 * in_pcbnotifyall() and in_pcbpurgeif0()? 1430 */ 1431 void 1432 in_pcbdrop(struct inpcb *inp) 1433 { 1434 1435 INP_WLOCK_ASSERT(inp); 1436 1437 /* 1438 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1439 * the hash lock...? 1440 */ 1441 inp->inp_flags |= INP_DROPPED; 1442 if (inp->inp_flags & INP_INHASHLIST) { 1443 struct inpcbport *phd = inp->inp_phd; 1444 1445 INP_HASH_WLOCK(inp->inp_pcbinfo); 1446 LIST_REMOVE(inp, inp_hash); 1447 LIST_REMOVE(inp, inp_portlist); 1448 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 1449 LIST_REMOVE(phd, phd_hash); 1450 free(phd, M_PCB); 1451 } 1452 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1453 inp->inp_flags &= ~INP_INHASHLIST; 1454 #ifdef PCBGROUP 1455 in_pcbgroup_remove(inp); 1456 #endif 1457 } 1458 } 1459 1460 #ifdef INET 1461 /* 1462 * Common routines to return the socket addresses associated with inpcbs. 1463 */ 1464 struct sockaddr * 1465 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1466 { 1467 struct sockaddr_in *sin; 1468 1469 sin = malloc(sizeof *sin, M_SONAME, 1470 M_WAITOK | M_ZERO); 1471 sin->sin_family = AF_INET; 1472 sin->sin_len = sizeof(*sin); 1473 sin->sin_addr = *addr_p; 1474 sin->sin_port = port; 1475 1476 return (struct sockaddr *)sin; 1477 } 1478 1479 int 1480 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1481 { 1482 struct inpcb *inp; 1483 struct in_addr addr; 1484 in_port_t port; 1485 1486 inp = sotoinpcb(so); 1487 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1488 1489 INP_RLOCK(inp); 1490 port = inp->inp_lport; 1491 addr = inp->inp_laddr; 1492 INP_RUNLOCK(inp); 1493 1494 *nam = in_sockaddr(port, &addr); 1495 return 0; 1496 } 1497 1498 int 1499 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1500 { 1501 struct inpcb *inp; 1502 struct in_addr addr; 1503 in_port_t port; 1504 1505 inp = sotoinpcb(so); 1506 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1507 1508 INP_RLOCK(inp); 1509 port = inp->inp_fport; 1510 addr = inp->inp_faddr; 1511 INP_RUNLOCK(inp); 1512 1513 *nam = in_sockaddr(port, &addr); 1514 return 0; 1515 } 1516 1517 void 1518 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1519 struct inpcb *(*notify)(struct inpcb *, int)) 1520 { 1521 struct inpcb *inp, *inp_temp; 1522 1523 INP_INFO_WLOCK(pcbinfo); 1524 LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1525 INP_WLOCK(inp); 1526 #ifdef INET6 1527 if ((inp->inp_vflag & INP_IPV4) == 0) { 1528 INP_WUNLOCK(inp); 1529 continue; 1530 } 1531 #endif 1532 if (inp->inp_faddr.s_addr != faddr.s_addr || 1533 inp->inp_socket == NULL) { 1534 INP_WUNLOCK(inp); 1535 continue; 1536 } 1537 if ((*notify)(inp, errno)) 1538 INP_WUNLOCK(inp); 1539 } 1540 INP_INFO_WUNLOCK(pcbinfo); 1541 } 1542 1543 void 1544 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1545 { 1546 struct inpcb *inp; 1547 struct ip_moptions *imo; 1548 int i, gap; 1549 1550 INP_INFO_WLOCK(pcbinfo); 1551 LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1552 INP_WLOCK(inp); 1553 imo = inp->inp_moptions; 1554 if ((inp->inp_vflag & INP_IPV4) && 1555 imo != NULL) { 1556 /* 1557 * Unselect the outgoing interface if it is being 1558 * detached. 1559 */ 1560 if (imo->imo_multicast_ifp == ifp) 1561 imo->imo_multicast_ifp = NULL; 1562 1563 /* 1564 * Drop multicast group membership if we joined 1565 * through the interface being detached. 1566 * 1567 * XXX This can all be deferred to an epoch_call 1568 */ 1569 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1570 i++) { 1571 if (imo->imo_membership[i]->inm_ifp == ifp) { 1572 IN_MULTI_LOCK_ASSERT(); 1573 in_leavegroup_locked(imo->imo_membership[i], NULL); 1574 gap++; 1575 } else if (gap != 0) 1576 imo->imo_membership[i - gap] = 1577 imo->imo_membership[i]; 1578 } 1579 imo->imo_num_memberships -= gap; 1580 } 1581 INP_WUNLOCK(inp); 1582 } 1583 INP_INFO_WUNLOCK(pcbinfo); 1584 } 1585 1586 /* 1587 * Lookup a PCB based on the local address and port. Caller must hold the 1588 * hash lock. No inpcb locks or references are acquired. 1589 */ 1590 #define INP_LOOKUP_MAPPED_PCB_COST 3 1591 struct inpcb * 1592 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1593 u_short lport, int lookupflags, struct ucred *cred) 1594 { 1595 struct inpcb *inp; 1596 #ifdef INET6 1597 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1598 #else 1599 int matchwild = 3; 1600 #endif 1601 int wildcard; 1602 1603 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1604 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1605 1606 INP_HASH_LOCK_ASSERT(pcbinfo); 1607 1608 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1609 struct inpcbhead *head; 1610 /* 1611 * Look for an unconnected (wildcard foreign addr) PCB that 1612 * matches the local address and port we're looking for. 1613 */ 1614 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1615 0, pcbinfo->ipi_hashmask)]; 1616 LIST_FOREACH(inp, head, inp_hash) { 1617 #ifdef INET6 1618 /* XXX inp locking */ 1619 if ((inp->inp_vflag & INP_IPV4) == 0) 1620 continue; 1621 #endif 1622 if (inp->inp_faddr.s_addr == INADDR_ANY && 1623 inp->inp_laddr.s_addr == laddr.s_addr && 1624 inp->inp_lport == lport) { 1625 /* 1626 * Found? 1627 */ 1628 if (cred == NULL || 1629 prison_equal_ip4(cred->cr_prison, 1630 inp->inp_cred->cr_prison)) 1631 return (inp); 1632 } 1633 } 1634 /* 1635 * Not found. 1636 */ 1637 return (NULL); 1638 } else { 1639 struct inpcbporthead *porthash; 1640 struct inpcbport *phd; 1641 struct inpcb *match = NULL; 1642 /* 1643 * Best fit PCB lookup. 1644 * 1645 * First see if this local port is in use by looking on the 1646 * port hash list. 1647 */ 1648 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1649 pcbinfo->ipi_porthashmask)]; 1650 LIST_FOREACH(phd, porthash, phd_hash) { 1651 if (phd->phd_port == lport) 1652 break; 1653 } 1654 if (phd != NULL) { 1655 /* 1656 * Port is in use by one or more PCBs. Look for best 1657 * fit. 1658 */ 1659 LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1660 wildcard = 0; 1661 if (cred != NULL && 1662 !prison_equal_ip4(inp->inp_cred->cr_prison, 1663 cred->cr_prison)) 1664 continue; 1665 #ifdef INET6 1666 /* XXX inp locking */ 1667 if ((inp->inp_vflag & INP_IPV4) == 0) 1668 continue; 1669 /* 1670 * We never select the PCB that has 1671 * INP_IPV6 flag and is bound to :: if 1672 * we have another PCB which is bound 1673 * to 0.0.0.0. If a PCB has the 1674 * INP_IPV6 flag, then we set its cost 1675 * higher than IPv4 only PCBs. 1676 * 1677 * Note that the case only happens 1678 * when a socket is bound to ::, under 1679 * the condition that the use of the 1680 * mapped address is allowed. 1681 */ 1682 if ((inp->inp_vflag & INP_IPV6) != 0) 1683 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1684 #endif 1685 if (inp->inp_faddr.s_addr != INADDR_ANY) 1686 wildcard++; 1687 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1688 if (laddr.s_addr == INADDR_ANY) 1689 wildcard++; 1690 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1691 continue; 1692 } else { 1693 if (laddr.s_addr != INADDR_ANY) 1694 wildcard++; 1695 } 1696 if (wildcard < matchwild) { 1697 match = inp; 1698 matchwild = wildcard; 1699 if (matchwild == 0) 1700 break; 1701 } 1702 } 1703 } 1704 return (match); 1705 } 1706 } 1707 #undef INP_LOOKUP_MAPPED_PCB_COST 1708 1709 #ifdef PCBGROUP 1710 /* 1711 * Lookup PCB in hash list, using pcbgroup tables. 1712 */ 1713 static struct inpcb * 1714 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 1715 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 1716 u_int lport_arg, int lookupflags, struct ifnet *ifp) 1717 { 1718 struct inpcbhead *head; 1719 struct inpcb *inp, *tmpinp; 1720 u_short fport = fport_arg, lport = lport_arg; 1721 bool locked; 1722 1723 /* 1724 * First look for an exact match. 1725 */ 1726 tmpinp = NULL; 1727 INP_GROUP_LOCK(pcbgroup); 1728 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 1729 pcbgroup->ipg_hashmask)]; 1730 LIST_FOREACH(inp, head, inp_pcbgrouphash) { 1731 #ifdef INET6 1732 /* XXX inp locking */ 1733 if ((inp->inp_vflag & INP_IPV4) == 0) 1734 continue; 1735 #endif 1736 if (inp->inp_faddr.s_addr == faddr.s_addr && 1737 inp->inp_laddr.s_addr == laddr.s_addr && 1738 inp->inp_fport == fport && 1739 inp->inp_lport == lport) { 1740 /* 1741 * XXX We should be able to directly return 1742 * the inp here, without any checks. 1743 * Well unless both bound with SO_REUSEPORT? 1744 */ 1745 if (prison_flag(inp->inp_cred, PR_IP4)) 1746 goto found; 1747 if (tmpinp == NULL) 1748 tmpinp = inp; 1749 } 1750 } 1751 if (tmpinp != NULL) { 1752 inp = tmpinp; 1753 goto found; 1754 } 1755 1756 #ifdef RSS 1757 /* 1758 * For incoming connections, we may wish to do a wildcard 1759 * match for an RSS-local socket. 1760 */ 1761 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 1762 struct inpcb *local_wild = NULL, *local_exact = NULL; 1763 #ifdef INET6 1764 struct inpcb *local_wild_mapped = NULL; 1765 #endif 1766 struct inpcb *jail_wild = NULL; 1767 struct inpcbhead *head; 1768 int injail; 1769 1770 /* 1771 * Order of socket selection - we always prefer jails. 1772 * 1. jailed, non-wild. 1773 * 2. jailed, wild. 1774 * 3. non-jailed, non-wild. 1775 * 4. non-jailed, wild. 1776 */ 1777 1778 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 1779 lport, 0, pcbgroup->ipg_hashmask)]; 1780 LIST_FOREACH(inp, head, inp_pcbgrouphash) { 1781 #ifdef INET6 1782 /* XXX inp locking */ 1783 if ((inp->inp_vflag & INP_IPV4) == 0) 1784 continue; 1785 #endif 1786 if (inp->inp_faddr.s_addr != INADDR_ANY || 1787 inp->inp_lport != lport) 1788 continue; 1789 1790 injail = prison_flag(inp->inp_cred, PR_IP4); 1791 if (injail) { 1792 if (prison_check_ip4(inp->inp_cred, 1793 &laddr) != 0) 1794 continue; 1795 } else { 1796 if (local_exact != NULL) 1797 continue; 1798 } 1799 1800 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1801 if (injail) 1802 goto found; 1803 else 1804 local_exact = inp; 1805 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 1806 #ifdef INET6 1807 /* XXX inp locking, NULL check */ 1808 if (inp->inp_vflag & INP_IPV6PROTO) 1809 local_wild_mapped = inp; 1810 else 1811 #endif 1812 if (injail) 1813 jail_wild = inp; 1814 else 1815 local_wild = inp; 1816 } 1817 } /* LIST_FOREACH */ 1818 1819 inp = jail_wild; 1820 if (inp == NULL) 1821 inp = local_exact; 1822 if (inp == NULL) 1823 inp = local_wild; 1824 #ifdef INET6 1825 if (inp == NULL) 1826 inp = local_wild_mapped; 1827 #endif 1828 if (inp != NULL) 1829 goto found; 1830 } 1831 #endif 1832 1833 /* 1834 * Then look for a wildcard match, if requested. 1835 */ 1836 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 1837 struct inpcb *local_wild = NULL, *local_exact = NULL; 1838 #ifdef INET6 1839 struct inpcb *local_wild_mapped = NULL; 1840 #endif 1841 struct inpcb *jail_wild = NULL; 1842 struct inpcbhead *head; 1843 int injail; 1844 1845 /* 1846 * Order of socket selection - we always prefer jails. 1847 * 1. jailed, non-wild. 1848 * 2. jailed, wild. 1849 * 3. non-jailed, non-wild. 1850 * 4. non-jailed, wild. 1851 */ 1852 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 1853 0, pcbinfo->ipi_wildmask)]; 1854 LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 1855 #ifdef INET6 1856 /* XXX inp locking */ 1857 if ((inp->inp_vflag & INP_IPV4) == 0) 1858 continue; 1859 #endif 1860 if (inp->inp_faddr.s_addr != INADDR_ANY || 1861 inp->inp_lport != lport) 1862 continue; 1863 1864 injail = prison_flag(inp->inp_cred, PR_IP4); 1865 if (injail) { 1866 if (prison_check_ip4(inp->inp_cred, 1867 &laddr) != 0) 1868 continue; 1869 } else { 1870 if (local_exact != NULL) 1871 continue; 1872 } 1873 1874 if (inp->inp_laddr.s_addr == laddr.s_addr) { 1875 if (injail) 1876 goto found; 1877 else 1878 local_exact = inp; 1879 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 1880 #ifdef INET6 1881 /* XXX inp locking, NULL check */ 1882 if (inp->inp_vflag & INP_IPV6PROTO) 1883 local_wild_mapped = inp; 1884 else 1885 #endif 1886 if (injail) 1887 jail_wild = inp; 1888 else 1889 local_wild = inp; 1890 } 1891 } /* LIST_FOREACH */ 1892 inp = jail_wild; 1893 if (inp == NULL) 1894 inp = local_exact; 1895 if (inp == NULL) 1896 inp = local_wild; 1897 #ifdef INET6 1898 if (inp == NULL) 1899 inp = local_wild_mapped; 1900 #endif 1901 if (inp != NULL) 1902 goto found; 1903 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 1904 INP_GROUP_UNLOCK(pcbgroup); 1905 return (NULL); 1906 1907 found: 1908 if (lookupflags & INPLOOKUP_WLOCKPCB) 1909 locked = INP_TRY_WLOCK(inp); 1910 else if (lookupflags & INPLOOKUP_RLOCKPCB) 1911 locked = INP_TRY_RLOCK(inp); 1912 else 1913 panic("%s: locking bug", __func__); 1914 if (!locked) 1915 in_pcbref(inp); 1916 INP_GROUP_UNLOCK(pcbgroup); 1917 if (!locked) { 1918 if (lookupflags & INPLOOKUP_WLOCKPCB) { 1919 INP_WLOCK(inp); 1920 if (in_pcbrele_wlocked(inp)) 1921 return (NULL); 1922 } else { 1923 INP_RLOCK(inp); 1924 if (in_pcbrele_rlocked(inp)) 1925 return (NULL); 1926 } 1927 } 1928 #ifdef INVARIANTS 1929 if (lookupflags & INPLOOKUP_WLOCKPCB) 1930 INP_WLOCK_ASSERT(inp); 1931 else 1932 INP_RLOCK_ASSERT(inp); 1933 #endif 1934 return (inp); 1935 } 1936 #endif /* PCBGROUP */ 1937 1938 /* 1939 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 1940 * that the caller has locked the hash list, and will not perform any further 1941 * locking or reference operations on either the hash list or the connection. 1942 */ 1943 static struct inpcb * 1944 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 1945 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 1946 struct ifnet *ifp) 1947 { 1948 struct inpcbhead *head; 1949 struct inpcb *inp, *tmpinp; 1950 u_short fport = fport_arg, lport = lport_arg; 1951 1952 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1953 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1954 1955 INP_HASH_LOCK_ASSERT(pcbinfo); 1956 1957 /* 1958 * First look for an exact match. 1959 */ 1960 tmpinp = NULL; 1961 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 1962 pcbinfo->ipi_hashmask)]; 1963 LIST_FOREACH(inp, head, inp_hash) { 1964 #ifdef INET6 1965 /* XXX inp locking */ 1966 if ((inp->inp_vflag & INP_IPV4) == 0) 1967 continue; 1968 #endif 1969 if (inp->inp_faddr.s_addr == faddr.s_addr && 1970 inp->inp_laddr.s_addr == laddr.s_addr && 1971 inp->inp_fport == fport && 1972 inp->inp_lport == lport) { 1973 /* 1974 * XXX We should be able to directly return 1975 * the inp here, without any checks. 1976 * Well unless both bound with SO_REUSEPORT? 1977 */ 1978 if (prison_flag(inp->inp_cred, PR_IP4)) 1979 return (inp); 1980 if (tmpinp == NULL) 1981 tmpinp = inp; 1982 } 1983 } 1984 if (tmpinp != NULL) 1985 return (tmpinp); 1986 1987 /* 1988 * Then look for a wildcard match, if requested. 1989 */ 1990 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 1991 struct inpcb *local_wild = NULL, *local_exact = NULL; 1992 #ifdef INET6 1993 struct inpcb *local_wild_mapped = NULL; 1994 #endif 1995 struct inpcb *jail_wild = NULL; 1996 int injail; 1997 1998 /* 1999 * Order of socket selection - we always prefer jails. 2000 * 1. jailed, non-wild. 2001 * 2. jailed, wild. 2002 * 3. non-jailed, non-wild. 2003 * 4. non-jailed, wild. 2004 */ 2005 2006 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2007 0, pcbinfo->ipi_hashmask)]; 2008 LIST_FOREACH(inp, head, inp_hash) { 2009 #ifdef INET6 2010 /* XXX inp locking */ 2011 if ((inp->inp_vflag & INP_IPV4) == 0) 2012 continue; 2013 #endif 2014 if (inp->inp_faddr.s_addr != INADDR_ANY || 2015 inp->inp_lport != lport) 2016 continue; 2017 2018 injail = prison_flag(inp->inp_cred, PR_IP4); 2019 if (injail) { 2020 if (prison_check_ip4(inp->inp_cred, 2021 &laddr) != 0) 2022 continue; 2023 } else { 2024 if (local_exact != NULL) 2025 continue; 2026 } 2027 2028 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2029 if (injail) 2030 return (inp); 2031 else 2032 local_exact = inp; 2033 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2034 #ifdef INET6 2035 /* XXX inp locking, NULL check */ 2036 if (inp->inp_vflag & INP_IPV6PROTO) 2037 local_wild_mapped = inp; 2038 else 2039 #endif 2040 if (injail) 2041 jail_wild = inp; 2042 else 2043 local_wild = inp; 2044 } 2045 } /* LIST_FOREACH */ 2046 if (jail_wild != NULL) 2047 return (jail_wild); 2048 if (local_exact != NULL) 2049 return (local_exact); 2050 if (local_wild != NULL) 2051 return (local_wild); 2052 #ifdef INET6 2053 if (local_wild_mapped != NULL) 2054 return (local_wild_mapped); 2055 #endif 2056 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2057 2058 return (NULL); 2059 } 2060 2061 /* 2062 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2063 * hash list lock, and will return the inpcb locked (i.e., requires 2064 * INPLOOKUP_LOCKPCB). 2065 */ 2066 static struct inpcb * 2067 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2068 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2069 struct ifnet *ifp) 2070 { 2071 struct inpcb *inp; 2072 bool locked; 2073 2074 INP_HASH_RLOCK(pcbinfo); 2075 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2076 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 2077 if (inp != NULL) { 2078 if (lookupflags & INPLOOKUP_WLOCKPCB) 2079 locked = INP_TRY_WLOCK(inp); 2080 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2081 locked = INP_TRY_RLOCK(inp); 2082 else 2083 panic("%s: locking bug", __func__); 2084 if (!locked) 2085 in_pcbref(inp); 2086 INP_HASH_RUNLOCK(pcbinfo); 2087 if (!locked) { 2088 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2089 INP_WLOCK(inp); 2090 if (in_pcbrele_wlocked(inp)) 2091 return (NULL); 2092 } else { 2093 INP_RLOCK(inp); 2094 if (in_pcbrele_rlocked(inp)) 2095 return (NULL); 2096 } 2097 } 2098 #ifdef INVARIANTS 2099 if (lookupflags & INPLOOKUP_WLOCKPCB) 2100 INP_WLOCK_ASSERT(inp); 2101 else 2102 INP_RLOCK_ASSERT(inp); 2103 #endif 2104 } else 2105 INP_HASH_RUNLOCK(pcbinfo); 2106 return (inp); 2107 } 2108 2109 /* 2110 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2111 * from which a pre-calculated hash value may be extracted. 2112 * 2113 * Possibly more of this logic should be in in_pcbgroup.c. 2114 */ 2115 struct inpcb * 2116 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2117 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2118 { 2119 #if defined(PCBGROUP) && !defined(RSS) 2120 struct inpcbgroup *pcbgroup; 2121 #endif 2122 2123 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2124 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2125 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2126 ("%s: LOCKPCB not set", __func__)); 2127 2128 /* 2129 * When not using RSS, use connection groups in preference to the 2130 * reservation table when looking up 4-tuples. When using RSS, just 2131 * use the reservation table, due to the cost of the Toeplitz hash 2132 * in software. 2133 * 2134 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2135 * we could be doing RSS with a non-Toeplitz hash that is affordable 2136 * in software. 2137 */ 2138 #if defined(PCBGROUP) && !defined(RSS) 2139 if (in_pcbgroup_enabled(pcbinfo)) { 2140 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2141 fport); 2142 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2143 laddr, lport, lookupflags, ifp)); 2144 } 2145 #endif 2146 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2147 lookupflags, ifp)); 2148 } 2149 2150 struct inpcb * 2151 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2152 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2153 struct ifnet *ifp, struct mbuf *m) 2154 { 2155 #ifdef PCBGROUP 2156 struct inpcbgroup *pcbgroup; 2157 #endif 2158 2159 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2160 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2161 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2162 ("%s: LOCKPCB not set", __func__)); 2163 2164 #ifdef PCBGROUP 2165 /* 2166 * If we can use a hardware-generated hash to look up the connection 2167 * group, use that connection group to find the inpcb. Otherwise 2168 * fall back on a software hash -- or the reservation table if we're 2169 * using RSS. 2170 * 2171 * XXXRW: As above, that policy belongs in the pcbgroup code. 2172 */ 2173 if (in_pcbgroup_enabled(pcbinfo) && 2174 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2175 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2176 m->m_pkthdr.flowid); 2177 if (pcbgroup != NULL) 2178 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2179 fport, laddr, lport, lookupflags, ifp)); 2180 #ifndef RSS 2181 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2182 fport); 2183 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2184 laddr, lport, lookupflags, ifp)); 2185 #endif 2186 } 2187 #endif 2188 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2189 lookupflags, ifp)); 2190 } 2191 #endif /* INET */ 2192 2193 /* 2194 * Insert PCB onto various hash lists. 2195 */ 2196 static int 2197 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 2198 { 2199 struct inpcbhead *pcbhash; 2200 struct inpcbporthead *pcbporthash; 2201 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2202 struct inpcbport *phd; 2203 u_int32_t hashkey_faddr; 2204 2205 INP_WLOCK_ASSERT(inp); 2206 INP_HASH_WLOCK_ASSERT(pcbinfo); 2207 2208 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2209 ("in_pcbinshash: INP_INHASHLIST")); 2210 2211 #ifdef INET6 2212 if (inp->inp_vflag & INP_IPV6) 2213 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2214 else 2215 #endif 2216 hashkey_faddr = inp->inp_faddr.s_addr; 2217 2218 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2219 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2220 2221 pcbporthash = &pcbinfo->ipi_porthashbase[ 2222 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2223 2224 /* 2225 * Go through port list and look for a head for this lport. 2226 */ 2227 LIST_FOREACH(phd, pcbporthash, phd_hash) { 2228 if (phd->phd_port == inp->inp_lport) 2229 break; 2230 } 2231 /* 2232 * If none exists, malloc one and tack it on. 2233 */ 2234 if (phd == NULL) { 2235 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2236 if (phd == NULL) { 2237 return (ENOBUFS); /* XXX */ 2238 } 2239 phd->phd_port = inp->inp_lport; 2240 LIST_INIT(&phd->phd_pcblist); 2241 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2242 } 2243 inp->inp_phd = phd; 2244 LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2245 LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2246 inp->inp_flags |= INP_INHASHLIST; 2247 #ifdef PCBGROUP 2248 if (do_pcbgroup_update) 2249 in_pcbgroup_update(inp); 2250 #endif 2251 return (0); 2252 } 2253 2254 /* 2255 * For now, there are two public interfaces to insert an inpcb into the hash 2256 * lists -- one that does update pcbgroups, and one that doesn't. The latter 2257 * is used only in the TCP syncache, where in_pcbinshash is called before the 2258 * full 4-tuple is set for the inpcb, and we don't want to install in the 2259 * pcbgroup until later. 2260 * 2261 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 2262 * connection groups, and partially initialised inpcbs should not be exposed 2263 * to either reservation hash tables or pcbgroups. 2264 */ 2265 int 2266 in_pcbinshash(struct inpcb *inp) 2267 { 2268 2269 return (in_pcbinshash_internal(inp, 1)); 2270 } 2271 2272 int 2273 in_pcbinshash_nopcbgroup(struct inpcb *inp) 2274 { 2275 2276 return (in_pcbinshash_internal(inp, 0)); 2277 } 2278 2279 /* 2280 * Move PCB to the proper hash bucket when { faddr, fport } have been 2281 * changed. NOTE: This does not handle the case of the lport changing (the 2282 * hashed port list would have to be updated as well), so the lport must 2283 * not change after in_pcbinshash() has been called. 2284 */ 2285 void 2286 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2287 { 2288 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2289 struct inpcbhead *head; 2290 u_int32_t hashkey_faddr; 2291 2292 INP_WLOCK_ASSERT(inp); 2293 INP_HASH_WLOCK_ASSERT(pcbinfo); 2294 2295 KASSERT(inp->inp_flags & INP_INHASHLIST, 2296 ("in_pcbrehash: !INP_INHASHLIST")); 2297 2298 #ifdef INET6 2299 if (inp->inp_vflag & INP_IPV6) 2300 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2301 else 2302 #endif 2303 hashkey_faddr = inp->inp_faddr.s_addr; 2304 2305 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2306 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2307 2308 LIST_REMOVE(inp, inp_hash); 2309 LIST_INSERT_HEAD(head, inp, inp_hash); 2310 2311 #ifdef PCBGROUP 2312 if (m != NULL) 2313 in_pcbgroup_update_mbuf(inp, m); 2314 else 2315 in_pcbgroup_update(inp); 2316 #endif 2317 } 2318 2319 void 2320 in_pcbrehash(struct inpcb *inp) 2321 { 2322 2323 in_pcbrehash_mbuf(inp, NULL); 2324 } 2325 2326 /* 2327 * Remove PCB from various lists. 2328 */ 2329 static void 2330 in_pcbremlists(struct inpcb *inp) 2331 { 2332 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2333 2334 #ifdef INVARIANTS 2335 if (pcbinfo == &V_tcbinfo) { 2336 INP_INFO_RLOCK_ASSERT(pcbinfo); 2337 } else { 2338 INP_INFO_WLOCK_ASSERT(pcbinfo); 2339 } 2340 #endif 2341 2342 INP_WLOCK_ASSERT(inp); 2343 INP_LIST_WLOCK_ASSERT(pcbinfo); 2344 2345 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2346 if (inp->inp_flags & INP_INHASHLIST) { 2347 struct inpcbport *phd = inp->inp_phd; 2348 2349 INP_HASH_WLOCK(pcbinfo); 2350 LIST_REMOVE(inp, inp_hash); 2351 LIST_REMOVE(inp, inp_portlist); 2352 if (LIST_FIRST(&phd->phd_pcblist) == NULL) { 2353 LIST_REMOVE(phd, phd_hash); 2354 free(phd, M_PCB); 2355 } 2356 INP_HASH_WUNLOCK(pcbinfo); 2357 inp->inp_flags &= ~INP_INHASHLIST; 2358 } 2359 LIST_REMOVE(inp, inp_list); 2360 pcbinfo->ipi_count--; 2361 #ifdef PCBGROUP 2362 in_pcbgroup_remove(inp); 2363 #endif 2364 } 2365 2366 /* 2367 * Check for alternatives when higher level complains 2368 * about service problems. For now, invalidate cached 2369 * routing information. If the route was created dynamically 2370 * (by a redirect), time to try a default gateway again. 2371 */ 2372 void 2373 in_losing(struct inpcb *inp) 2374 { 2375 2376 RO_INVALIDATE_CACHE(&inp->inp_route); 2377 return; 2378 } 2379 2380 /* 2381 * A set label operation has occurred at the socket layer, propagate the 2382 * label change into the in_pcb for the socket. 2383 */ 2384 void 2385 in_pcbsosetlabel(struct socket *so) 2386 { 2387 #ifdef MAC 2388 struct inpcb *inp; 2389 2390 inp = sotoinpcb(so); 2391 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2392 2393 INP_WLOCK(inp); 2394 SOCK_LOCK(so); 2395 mac_inpcb_sosetlabel(so, inp); 2396 SOCK_UNLOCK(so); 2397 INP_WUNLOCK(inp); 2398 #endif 2399 } 2400 2401 /* 2402 * ipport_tick runs once per second, determining if random port allocation 2403 * should be continued. If more than ipport_randomcps ports have been 2404 * allocated in the last second, then we return to sequential port 2405 * allocation. We return to random allocation only once we drop below 2406 * ipport_randomcps for at least ipport_randomtime seconds. 2407 */ 2408 static void 2409 ipport_tick(void *xtp) 2410 { 2411 VNET_ITERATOR_DECL(vnet_iter); 2412 2413 VNET_LIST_RLOCK_NOSLEEP(); 2414 VNET_FOREACH(vnet_iter) { 2415 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2416 if (V_ipport_tcpallocs <= 2417 V_ipport_tcplastcount + V_ipport_randomcps) { 2418 if (V_ipport_stoprandom > 0) 2419 V_ipport_stoprandom--; 2420 } else 2421 V_ipport_stoprandom = V_ipport_randomtime; 2422 V_ipport_tcplastcount = V_ipport_tcpallocs; 2423 CURVNET_RESTORE(); 2424 } 2425 VNET_LIST_RUNLOCK_NOSLEEP(); 2426 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2427 } 2428 2429 static void 2430 ip_fini(void *xtp) 2431 { 2432 2433 callout_stop(&ipport_tick_callout); 2434 } 2435 2436 /* 2437 * The ipport_callout should start running at about the time we attach the 2438 * inet or inet6 domains. 2439 */ 2440 static void 2441 ipport_tick_init(const void *unused __unused) 2442 { 2443 2444 /* Start ipport_tick. */ 2445 callout_init(&ipport_tick_callout, 1); 2446 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2447 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2448 SHUTDOWN_PRI_DEFAULT); 2449 } 2450 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2451 ipport_tick_init, NULL); 2452 2453 void 2454 inp_wlock(struct inpcb *inp) 2455 { 2456 2457 INP_WLOCK(inp); 2458 } 2459 2460 void 2461 inp_wunlock(struct inpcb *inp) 2462 { 2463 2464 INP_WUNLOCK(inp); 2465 } 2466 2467 void 2468 inp_rlock(struct inpcb *inp) 2469 { 2470 2471 INP_RLOCK(inp); 2472 } 2473 2474 void 2475 inp_runlock(struct inpcb *inp) 2476 { 2477 2478 INP_RUNLOCK(inp); 2479 } 2480 2481 #ifdef INVARIANT_SUPPORT 2482 void 2483 inp_lock_assert(struct inpcb *inp) 2484 { 2485 2486 INP_WLOCK_ASSERT(inp); 2487 } 2488 2489 void 2490 inp_unlock_assert(struct inpcb *inp) 2491 { 2492 2493 INP_UNLOCK_ASSERT(inp); 2494 } 2495 #endif 2496 2497 void 2498 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2499 { 2500 struct inpcb *inp; 2501 2502 INP_INFO_WLOCK(&V_tcbinfo); 2503 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2504 INP_WLOCK(inp); 2505 func(inp, arg); 2506 INP_WUNLOCK(inp); 2507 } 2508 INP_INFO_WUNLOCK(&V_tcbinfo); 2509 } 2510 2511 struct socket * 2512 inp_inpcbtosocket(struct inpcb *inp) 2513 { 2514 2515 INP_WLOCK_ASSERT(inp); 2516 return (inp->inp_socket); 2517 } 2518 2519 struct tcpcb * 2520 inp_inpcbtotcpcb(struct inpcb *inp) 2521 { 2522 2523 INP_WLOCK_ASSERT(inp); 2524 return ((struct tcpcb *)inp->inp_ppcb); 2525 } 2526 2527 int 2528 inp_ip_tos_get(const struct inpcb *inp) 2529 { 2530 2531 return (inp->inp_ip_tos); 2532 } 2533 2534 void 2535 inp_ip_tos_set(struct inpcb *inp, int val) 2536 { 2537 2538 inp->inp_ip_tos = val; 2539 } 2540 2541 void 2542 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2543 uint32_t *faddr, uint16_t *fp) 2544 { 2545 2546 INP_LOCK_ASSERT(inp); 2547 *laddr = inp->inp_laddr.s_addr; 2548 *faddr = inp->inp_faddr.s_addr; 2549 *lp = inp->inp_lport; 2550 *fp = inp->inp_fport; 2551 } 2552 2553 struct inpcb * 2554 so_sotoinpcb(struct socket *so) 2555 { 2556 2557 return (sotoinpcb(so)); 2558 } 2559 2560 struct tcpcb * 2561 so_sototcpcb(struct socket *so) 2562 { 2563 2564 return (sototcpcb(so)); 2565 } 2566 2567 /* 2568 * Create an external-format (``xinpcb'') structure using the information in 2569 * the kernel-format in_pcb structure pointed to by inp. This is done to 2570 * reduce the spew of irrelevant information over this interface, to isolate 2571 * user code from changes in the kernel structure, and potentially to provide 2572 * information-hiding if we decide that some of this information should be 2573 * hidden from users. 2574 */ 2575 void 2576 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2577 { 2578 2579 xi->xi_len = sizeof(struct xinpcb); 2580 if (inp->inp_socket) 2581 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2582 else 2583 bzero(&xi->xi_socket, sizeof(struct xsocket)); 2584 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2585 xi->inp_gencnt = inp->inp_gencnt; 2586 xi->inp_ppcb = inp->inp_ppcb; 2587 xi->inp_flow = inp->inp_flow; 2588 xi->inp_flowid = inp->inp_flowid; 2589 xi->inp_flowtype = inp->inp_flowtype; 2590 xi->inp_flags = inp->inp_flags; 2591 xi->inp_flags2 = inp->inp_flags2; 2592 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2593 xi->in6p_cksum = inp->in6p_cksum; 2594 xi->in6p_hops = inp->in6p_hops; 2595 xi->inp_ip_tos = inp->inp_ip_tos; 2596 xi->inp_vflag = inp->inp_vflag; 2597 xi->inp_ip_ttl = inp->inp_ip_ttl; 2598 xi->inp_ip_p = inp->inp_ip_p; 2599 xi->inp_ip_minttl = inp->inp_ip_minttl; 2600 } 2601 2602 #ifdef DDB 2603 static void 2604 db_print_indent(int indent) 2605 { 2606 int i; 2607 2608 for (i = 0; i < indent; i++) 2609 db_printf(" "); 2610 } 2611 2612 static void 2613 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2614 { 2615 char faddr_str[48], laddr_str[48]; 2616 2617 db_print_indent(indent); 2618 db_printf("%s at %p\n", name, inc); 2619 2620 indent += 2; 2621 2622 #ifdef INET6 2623 if (inc->inc_flags & INC_ISIPV6) { 2624 /* IPv6. */ 2625 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2626 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2627 } else 2628 #endif 2629 { 2630 /* IPv4. */ 2631 inet_ntoa_r(inc->inc_laddr, laddr_str); 2632 inet_ntoa_r(inc->inc_faddr, faddr_str); 2633 } 2634 db_print_indent(indent); 2635 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2636 ntohs(inc->inc_lport)); 2637 db_print_indent(indent); 2638 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2639 ntohs(inc->inc_fport)); 2640 } 2641 2642 static void 2643 db_print_inpflags(int inp_flags) 2644 { 2645 int comma; 2646 2647 comma = 0; 2648 if (inp_flags & INP_RECVOPTS) { 2649 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2650 comma = 1; 2651 } 2652 if (inp_flags & INP_RECVRETOPTS) { 2653 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2654 comma = 1; 2655 } 2656 if (inp_flags & INP_RECVDSTADDR) { 2657 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2658 comma = 1; 2659 } 2660 if (inp_flags & INP_ORIGDSTADDR) { 2661 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2662 comma = 1; 2663 } 2664 if (inp_flags & INP_HDRINCL) { 2665 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2666 comma = 1; 2667 } 2668 if (inp_flags & INP_HIGHPORT) { 2669 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2670 comma = 1; 2671 } 2672 if (inp_flags & INP_LOWPORT) { 2673 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 2674 comma = 1; 2675 } 2676 if (inp_flags & INP_ANONPORT) { 2677 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 2678 comma = 1; 2679 } 2680 if (inp_flags & INP_RECVIF) { 2681 db_printf("%sINP_RECVIF", comma ? ", " : ""); 2682 comma = 1; 2683 } 2684 if (inp_flags & INP_MTUDISC) { 2685 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 2686 comma = 1; 2687 } 2688 if (inp_flags & INP_RECVTTL) { 2689 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 2690 comma = 1; 2691 } 2692 if (inp_flags & INP_DONTFRAG) { 2693 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 2694 comma = 1; 2695 } 2696 if (inp_flags & INP_RECVTOS) { 2697 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 2698 comma = 1; 2699 } 2700 if (inp_flags & IN6P_IPV6_V6ONLY) { 2701 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 2702 comma = 1; 2703 } 2704 if (inp_flags & IN6P_PKTINFO) { 2705 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 2706 comma = 1; 2707 } 2708 if (inp_flags & IN6P_HOPLIMIT) { 2709 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 2710 comma = 1; 2711 } 2712 if (inp_flags & IN6P_HOPOPTS) { 2713 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 2714 comma = 1; 2715 } 2716 if (inp_flags & IN6P_DSTOPTS) { 2717 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 2718 comma = 1; 2719 } 2720 if (inp_flags & IN6P_RTHDR) { 2721 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 2722 comma = 1; 2723 } 2724 if (inp_flags & IN6P_RTHDRDSTOPTS) { 2725 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 2726 comma = 1; 2727 } 2728 if (inp_flags & IN6P_TCLASS) { 2729 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 2730 comma = 1; 2731 } 2732 if (inp_flags & IN6P_AUTOFLOWLABEL) { 2733 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 2734 comma = 1; 2735 } 2736 if (inp_flags & INP_TIMEWAIT) { 2737 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 2738 comma = 1; 2739 } 2740 if (inp_flags & INP_ONESBCAST) { 2741 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 2742 comma = 1; 2743 } 2744 if (inp_flags & INP_DROPPED) { 2745 db_printf("%sINP_DROPPED", comma ? ", " : ""); 2746 comma = 1; 2747 } 2748 if (inp_flags & INP_SOCKREF) { 2749 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 2750 comma = 1; 2751 } 2752 if (inp_flags & IN6P_RFC2292) { 2753 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 2754 comma = 1; 2755 } 2756 if (inp_flags & IN6P_MTU) { 2757 db_printf("IN6P_MTU%s", comma ? ", " : ""); 2758 comma = 1; 2759 } 2760 } 2761 2762 static void 2763 db_print_inpvflag(u_char inp_vflag) 2764 { 2765 int comma; 2766 2767 comma = 0; 2768 if (inp_vflag & INP_IPV4) { 2769 db_printf("%sINP_IPV4", comma ? ", " : ""); 2770 comma = 1; 2771 } 2772 if (inp_vflag & INP_IPV6) { 2773 db_printf("%sINP_IPV6", comma ? ", " : ""); 2774 comma = 1; 2775 } 2776 if (inp_vflag & INP_IPV6PROTO) { 2777 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 2778 comma = 1; 2779 } 2780 } 2781 2782 static void 2783 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 2784 { 2785 2786 db_print_indent(indent); 2787 db_printf("%s at %p\n", name, inp); 2788 2789 indent += 2; 2790 2791 db_print_indent(indent); 2792 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 2793 2794 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 2795 2796 db_print_indent(indent); 2797 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 2798 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 2799 2800 db_print_indent(indent); 2801 db_printf("inp_label: %p inp_flags: 0x%x (", 2802 inp->inp_label, inp->inp_flags); 2803 db_print_inpflags(inp->inp_flags); 2804 db_printf(")\n"); 2805 2806 db_print_indent(indent); 2807 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 2808 inp->inp_vflag); 2809 db_print_inpvflag(inp->inp_vflag); 2810 db_printf(")\n"); 2811 2812 db_print_indent(indent); 2813 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 2814 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 2815 2816 db_print_indent(indent); 2817 #ifdef INET6 2818 if (inp->inp_vflag & INP_IPV6) { 2819 db_printf("in6p_options: %p in6p_outputopts: %p " 2820 "in6p_moptions: %p\n", inp->in6p_options, 2821 inp->in6p_outputopts, inp->in6p_moptions); 2822 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 2823 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 2824 inp->in6p_hops); 2825 } else 2826 #endif 2827 { 2828 db_printf("inp_ip_tos: %d inp_ip_options: %p " 2829 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 2830 inp->inp_options, inp->inp_moptions); 2831 } 2832 2833 db_print_indent(indent); 2834 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 2835 (uintmax_t)inp->inp_gencnt); 2836 } 2837 2838 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 2839 { 2840 struct inpcb *inp; 2841 2842 if (!have_addr) { 2843 db_printf("usage: show inpcb <addr>\n"); 2844 return; 2845 } 2846 inp = (struct inpcb *)addr; 2847 2848 db_print_inpcb(inp, "inpcb", 0); 2849 } 2850 #endif /* DDB */ 2851 2852 #ifdef RATELIMIT 2853 /* 2854 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 2855 * if any. 2856 */ 2857 int 2858 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 2859 { 2860 union if_snd_tag_modify_params params = { 2861 .rate_limit.max_rate = max_pacing_rate, 2862 }; 2863 struct m_snd_tag *mst; 2864 struct ifnet *ifp; 2865 int error; 2866 2867 mst = inp->inp_snd_tag; 2868 if (mst == NULL) 2869 return (EINVAL); 2870 2871 ifp = mst->ifp; 2872 if (ifp == NULL) 2873 return (EINVAL); 2874 2875 if (ifp->if_snd_tag_modify == NULL) { 2876 error = EOPNOTSUPP; 2877 } else { 2878 error = ifp->if_snd_tag_modify(mst, ¶ms); 2879 } 2880 return (error); 2881 } 2882 2883 /* 2884 * Query existing TX rate limit based on the existing 2885 * "inp->inp_snd_tag", if any. 2886 */ 2887 int 2888 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 2889 { 2890 union if_snd_tag_query_params params = { }; 2891 struct m_snd_tag *mst; 2892 struct ifnet *ifp; 2893 int error; 2894 2895 mst = inp->inp_snd_tag; 2896 if (mst == NULL) 2897 return (EINVAL); 2898 2899 ifp = mst->ifp; 2900 if (ifp == NULL) 2901 return (EINVAL); 2902 2903 if (ifp->if_snd_tag_query == NULL) { 2904 error = EOPNOTSUPP; 2905 } else { 2906 error = ifp->if_snd_tag_query(mst, ¶ms); 2907 if (error == 0 && p_max_pacing_rate != NULL) 2908 *p_max_pacing_rate = params.rate_limit.max_rate; 2909 } 2910 return (error); 2911 } 2912 2913 /* 2914 * Query existing TX queue level based on the existing 2915 * "inp->inp_snd_tag", if any. 2916 */ 2917 int 2918 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 2919 { 2920 union if_snd_tag_query_params params = { }; 2921 struct m_snd_tag *mst; 2922 struct ifnet *ifp; 2923 int error; 2924 2925 mst = inp->inp_snd_tag; 2926 if (mst == NULL) 2927 return (EINVAL); 2928 2929 ifp = mst->ifp; 2930 if (ifp == NULL) 2931 return (EINVAL); 2932 2933 if (ifp->if_snd_tag_query == NULL) 2934 return (EOPNOTSUPP); 2935 2936 error = ifp->if_snd_tag_query(mst, ¶ms); 2937 if (error == 0 && p_txqueue_level != NULL) 2938 *p_txqueue_level = params.rate_limit.queue_level; 2939 return (error); 2940 } 2941 2942 /* 2943 * Allocate a new TX rate limit send tag from the network interface 2944 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 2945 */ 2946 int 2947 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 2948 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate) 2949 { 2950 union if_snd_tag_alloc_params params = { 2951 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 2952 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 2953 .rate_limit.hdr.flowid = flowid, 2954 .rate_limit.hdr.flowtype = flowtype, 2955 .rate_limit.max_rate = max_pacing_rate, 2956 }; 2957 int error; 2958 2959 INP_WLOCK_ASSERT(inp); 2960 2961 if (inp->inp_snd_tag != NULL) 2962 return (EINVAL); 2963 2964 if (ifp->if_snd_tag_alloc == NULL) { 2965 error = EOPNOTSUPP; 2966 } else { 2967 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); 2968 2969 /* 2970 * At success increment the refcount on 2971 * the send tag's network interface: 2972 */ 2973 if (error == 0) 2974 if_ref(inp->inp_snd_tag->ifp); 2975 } 2976 return (error); 2977 } 2978 2979 /* 2980 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 2981 * if any: 2982 */ 2983 void 2984 in_pcbdetach_txrtlmt(struct inpcb *inp) 2985 { 2986 struct m_snd_tag *mst; 2987 struct ifnet *ifp; 2988 2989 INP_WLOCK_ASSERT(inp); 2990 2991 mst = inp->inp_snd_tag; 2992 inp->inp_snd_tag = NULL; 2993 2994 if (mst == NULL) 2995 return; 2996 2997 ifp = mst->ifp; 2998 if (ifp == NULL) 2999 return; 3000 3001 /* 3002 * If the device was detached while we still had reference(s) 3003 * on the ifp, we assume if_snd_tag_free() was replaced with 3004 * stubs. 3005 */ 3006 ifp->if_snd_tag_free(mst); 3007 3008 /* release reference count on network interface */ 3009 if_rele(ifp); 3010 } 3011 3012 /* 3013 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3014 * is set in the fast path and will attach/detach/modify the TX rate 3015 * limit send tag based on the socket's so_max_pacing_rate value. 3016 */ 3017 void 3018 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3019 { 3020 struct socket *socket; 3021 uint32_t max_pacing_rate; 3022 bool did_upgrade; 3023 int error; 3024 3025 if (inp == NULL) 3026 return; 3027 3028 socket = inp->inp_socket; 3029 if (socket == NULL) 3030 return; 3031 3032 if (!INP_WLOCKED(inp)) { 3033 /* 3034 * NOTE: If the write locking fails, we need to bail 3035 * out and use the non-ratelimited ring for the 3036 * transmit until there is a new chance to get the 3037 * write lock. 3038 */ 3039 if (!INP_TRY_UPGRADE(inp)) 3040 return; 3041 did_upgrade = 1; 3042 } else { 3043 did_upgrade = 0; 3044 } 3045 3046 /* 3047 * NOTE: The so_max_pacing_rate value is read unlocked, 3048 * because atomic updates are not required since the variable 3049 * is checked at every mbuf we send. It is assumed that the 3050 * variable read itself will be atomic. 3051 */ 3052 max_pacing_rate = socket->so_max_pacing_rate; 3053 3054 /* 3055 * NOTE: When attaching to a network interface a reference is 3056 * made to ensure the network interface doesn't go away until 3057 * all ratelimit connections are gone. The network interface 3058 * pointers compared below represent valid network interfaces, 3059 * except when comparing towards NULL. 3060 */ 3061 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3062 error = 0; 3063 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3064 if (inp->inp_snd_tag != NULL) 3065 in_pcbdetach_txrtlmt(inp); 3066 error = 0; 3067 } else if (inp->inp_snd_tag == NULL) { 3068 /* 3069 * In order to utilize packet pacing with RSS, we need 3070 * to wait until there is a valid RSS hash before we 3071 * can proceed: 3072 */ 3073 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3074 error = EAGAIN; 3075 } else { 3076 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3077 mb->m_pkthdr.flowid, max_pacing_rate); 3078 } 3079 } else { 3080 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3081 } 3082 if (error == 0 || error == EOPNOTSUPP) 3083 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3084 if (did_upgrade) 3085 INP_DOWNGRADE(inp); 3086 } 3087 3088 /* 3089 * Track route changes for TX rate limiting. 3090 */ 3091 void 3092 in_pcboutput_eagain(struct inpcb *inp) 3093 { 3094 struct socket *socket; 3095 bool did_upgrade; 3096 3097 if (inp == NULL) 3098 return; 3099 3100 socket = inp->inp_socket; 3101 if (socket == NULL) 3102 return; 3103 3104 if (inp->inp_snd_tag == NULL) 3105 return; 3106 3107 if (!INP_WLOCKED(inp)) { 3108 /* 3109 * NOTE: If the write locking fails, we need to bail 3110 * out and use the non-ratelimited ring for the 3111 * transmit until there is a new chance to get the 3112 * write lock. 3113 */ 3114 if (!INP_TRY_UPGRADE(inp)) 3115 return; 3116 did_upgrade = 1; 3117 } else { 3118 did_upgrade = 0; 3119 } 3120 3121 /* detach rate limiting */ 3122 in_pcbdetach_txrtlmt(inp); 3123 3124 /* make sure new mbuf send tag allocation is made */ 3125 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3126 3127 if (did_upgrade) 3128 INP_DOWNGRADE(inp); 3129 } 3130 #endif /* RATELIMIT */ 3131