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