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 * Copyright (c) 2021-2022 Gleb Smirnoff <glebius@FreeBSD.org> 9 * All rights reserved. 10 * 11 * Portions of this software were developed by Robert N. M. Watson under 12 * contract to Juniper Networks, Inc. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 */ 38 39 #include <sys/cdefs.h> 40 #include "opt_ddb.h" 41 #include "opt_ipsec.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_ratelimit.h" 45 #include "opt_route.h" 46 #include "opt_rss.h" 47 48 #include <sys/param.h> 49 #include <sys/hash.h> 50 #include <sys/systm.h> 51 #include <sys/libkern.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mbuf.h> 55 #include <sys/eventhandler.h> 56 #include <sys/domain.h> 57 #include <sys/proc.h> 58 #include <sys/protosw.h> 59 #include <sys/smp.h> 60 #include <sys/smr.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 #include <vm/vm.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/if_private.h> 81 #include <net/if_types.h> 82 #include <net/if_llatbl.h> 83 #include <net/route.h> 84 #include <net/rss_config.h> 85 #include <net/vnet.h> 86 87 #if defined(INET) || defined(INET6) 88 #include <netinet/in.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/in_pcb_var.h> 91 #include <netinet/tcp.h> 92 #ifdef INET 93 #include <netinet/in_var.h> 94 #include <netinet/in_fib.h> 95 #endif 96 #include <netinet/ip_var.h> 97 #ifdef INET6 98 #include <netinet/ip6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_var.h> 101 #include <netinet6/ip6_var.h> 102 #endif /* INET6 */ 103 #include <net/route/nhop.h> 104 #endif 105 106 #include <netipsec/ipsec_support.h> 107 108 #include <security/mac/mac_framework.h> 109 110 #define INPCBLBGROUP_SIZMIN 8 111 #define INPCBLBGROUP_SIZMAX 256 112 113 #define INP_FREED 0x00000200 /* Went through in_pcbfree(). */ 114 #define INP_INLBGROUP 0x01000000 /* Inserted into inpcblbgroup. */ 115 116 /* 117 * These configure the range of local port addresses assigned to 118 * "unspecified" outgoing connections/packets/whatever. 119 */ 120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 126 127 /* 128 * Reserved ports accessible only to root. There are significant 129 * security considerations that must be accounted for when changing these, 130 * but the security benefits can be great. Please be careful. 131 */ 132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 133 VNET_DEFINE(int, ipport_reservedlow); 134 135 /* Enable random ephemeral port allocation by default. */ 136 VNET_DEFINE(int, ipport_randomized) = 1; 137 138 #ifdef INET 139 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 140 struct in_addr faddr, u_int fport_arg, 141 struct in_addr laddr, u_int lport_arg, 142 int lookupflags, uint8_t numa_domain, int fib); 143 144 #define RANGECHK(var, min, max) \ 145 if ((var) < (min)) { (var) = (min); } \ 146 else if ((var) > (max)) { (var) = (max); } 147 148 static int 149 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 150 { 151 int error; 152 153 error = sysctl_handle_int(oidp, arg1, arg2, req); 154 if (error == 0) { 155 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 156 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 157 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 158 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 159 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 160 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 161 } 162 return (error); 163 } 164 165 #undef RANGECHK 166 167 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, 168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 169 "IP Ports"); 170 171 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 172 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 173 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", 174 ""); 175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 176 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 177 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", 178 ""); 179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 180 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 181 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", 182 ""); 183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 184 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 185 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", 186 ""); 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 189 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", 190 ""); 191 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 192 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 193 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", 194 ""); 195 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 196 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 197 &VNET_NAME(ipport_reservedhigh), 0, ""); 198 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 199 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 201 CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 203 204 #ifdef RATELIMIT 205 counter_u64_t rate_limit_new; 206 counter_u64_t rate_limit_chg; 207 counter_u64_t rate_limit_active; 208 counter_u64_t rate_limit_alloc_fail; 209 counter_u64_t rate_limit_set_ok; 210 211 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 212 "IP Rate Limiting"); 213 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, 214 &rate_limit_active, "Active rate limited connections"); 215 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, 216 &rate_limit_alloc_fail, "Rate limited connection failures"); 217 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, 218 &rate_limit_set_ok, "Rate limited setting succeeded"); 219 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD, 220 &rate_limit_new, "Total Rate limit new attempts"); 221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD, 222 &rate_limit_chg, "Total Rate limited change attempts"); 223 #endif /* RATELIMIT */ 224 225 #endif /* INET */ 226 227 VNET_DEFINE(uint32_t, in_pcbhashseed); 228 static void 229 in_pcbhashseed_init(void) 230 { 231 232 V_in_pcbhashseed = arc4random(); 233 } 234 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, 235 in_pcbhashseed_init, NULL); 236 237 #ifdef INET 238 VNET_DEFINE_STATIC(int, connect_inaddr_wild) = 1; 239 #define V_connect_inaddr_wild VNET(connect_inaddr_wild) 240 SYSCTL_INT(_net_inet_ip, OID_AUTO, connect_inaddr_wild, 241 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(connect_inaddr_wild), 0, 242 "Allow connecting to INADDR_ANY or INADDR_BROADCAST for connect(2)"); 243 #endif 244 245 static void in_pcbremhash(struct inpcb *); 246 247 /* 248 * in_pcb.c: manage the Protocol Control Blocks. 249 * 250 * NOTE: It is assumed that most of these functions will be called with 251 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 252 * functions often modify hash chains or addresses in pcbs. 253 */ 254 255 static struct inpcblbgroup * 256 in_pcblbgroup_alloc(struct ucred *cred, u_char vflag, uint16_t port, 257 const union in_dependaddr *addr, int size, uint8_t numa_domain, int fib) 258 { 259 struct inpcblbgroup *grp; 260 size_t bytes; 261 262 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 263 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 264 if (grp == NULL) 265 return (NULL); 266 LIST_INIT(&grp->il_pending); 267 grp->il_cred = crhold(cred); 268 grp->il_vflag = vflag; 269 grp->il_lport = port; 270 grp->il_numa_domain = numa_domain; 271 grp->il_fibnum = fib; 272 grp->il_dependladdr = *addr; 273 grp->il_inpsiz = size; 274 return (grp); 275 } 276 277 static void 278 in_pcblbgroup_free_deferred(epoch_context_t ctx) 279 { 280 struct inpcblbgroup *grp; 281 282 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 283 crfree(grp->il_cred); 284 free(grp, M_PCB); 285 } 286 287 static void 288 in_pcblbgroup_free(struct inpcblbgroup *grp) 289 { 290 KASSERT(LIST_EMPTY(&grp->il_pending), 291 ("local group %p still has pending inps", grp)); 292 293 CK_LIST_REMOVE(grp, il_list); 294 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); 295 } 296 297 static struct inpcblbgroup * 298 in_pcblbgroup_find(struct inpcb *inp) 299 { 300 struct inpcbinfo *pcbinfo; 301 struct inpcblbgroup *grp; 302 struct inpcblbgrouphead *hdr; 303 304 INP_LOCK_ASSERT(inp); 305 306 pcbinfo = inp->inp_pcbinfo; 307 INP_HASH_LOCK_ASSERT(pcbinfo); 308 309 hdr = &pcbinfo->ipi_lbgrouphashbase[ 310 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 311 CK_LIST_FOREACH(grp, hdr, il_list) { 312 struct inpcb *inp1; 313 314 for (unsigned int i = 0; i < grp->il_inpcnt; i++) { 315 if (inp == grp->il_inp[i]) 316 goto found; 317 } 318 LIST_FOREACH(inp1, &grp->il_pending, inp_lbgroup_list) { 319 if (inp == inp1) 320 goto found; 321 } 322 } 323 found: 324 return (grp); 325 } 326 327 static void 328 in_pcblbgroup_insert(struct inpcblbgroup *grp, struct inpcb *inp) 329 { 330 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 331 ("invalid local group size %d and count %d", grp->il_inpsiz, 332 grp->il_inpcnt)); 333 INP_WLOCK_ASSERT(inp); 334 335 if (inp->inp_socket->so_proto->pr_listen != pr_listen_notsupp && 336 !SOLISTENING(inp->inp_socket)) { 337 /* 338 * If this is a TCP socket, it should not be visible to lbgroup 339 * lookups until listen() has been called. 340 */ 341 LIST_INSERT_HEAD(&grp->il_pending, inp, inp_lbgroup_list); 342 } else { 343 grp->il_inp[grp->il_inpcnt] = inp; 344 345 /* 346 * Synchronize with in_pcblookup_lbgroup(): make sure that we 347 * don't expose a null slot to the lookup path. 348 */ 349 atomic_store_rel_int(&grp->il_inpcnt, grp->il_inpcnt + 1); 350 } 351 352 inp->inp_flags |= INP_INLBGROUP; 353 } 354 355 static struct inpcblbgroup * 356 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 357 struct inpcblbgroup *old_grp, int size) 358 { 359 struct inpcblbgroup *grp; 360 int i; 361 362 grp = in_pcblbgroup_alloc(old_grp->il_cred, old_grp->il_vflag, 363 old_grp->il_lport, &old_grp->il_dependladdr, size, 364 old_grp->il_numa_domain, old_grp->il_fibnum); 365 if (grp == NULL) 366 return (NULL); 367 368 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 369 ("invalid new local group size %d and old local group count %d", 370 grp->il_inpsiz, old_grp->il_inpcnt)); 371 372 for (i = 0; i < old_grp->il_inpcnt; ++i) 373 grp->il_inp[i] = old_grp->il_inp[i]; 374 grp->il_inpcnt = old_grp->il_inpcnt; 375 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 376 LIST_SWAP(&old_grp->il_pending, &grp->il_pending, inpcb, 377 inp_lbgroup_list); 378 in_pcblbgroup_free(old_grp); 379 return (grp); 380 } 381 382 /* 383 * Add PCB to load balance group for SO_REUSEPORT_LB option. 384 */ 385 static int 386 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain) 387 { 388 const static struct timeval interval = { 60, 0 }; 389 static struct timeval lastprint; 390 struct inpcbinfo *pcbinfo; 391 struct inpcblbgrouphead *hdr; 392 struct inpcblbgroup *grp; 393 uint32_t idx; 394 int fib; 395 396 pcbinfo = inp->inp_pcbinfo; 397 398 INP_WLOCK_ASSERT(inp); 399 INP_HASH_WLOCK_ASSERT(pcbinfo); 400 401 fib = (inp->inp_flags & INP_BOUNDFIB) != 0 ? 402 inp->inp_inc.inc_fibnum : RT_ALL_FIBS; 403 404 #ifdef INET6 405 /* 406 * Don't allow IPv4 mapped INET6 wild socket. 407 */ 408 if ((inp->inp_vflag & INP_IPV4) && 409 inp->inp_laddr.s_addr == INADDR_ANY && 410 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 411 return (0); 412 } 413 #endif 414 415 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 416 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 417 CK_LIST_FOREACH(grp, hdr, il_list) { 418 if (grp->il_cred->cr_prison == inp->inp_cred->cr_prison && 419 grp->il_vflag == inp->inp_vflag && 420 grp->il_lport == inp->inp_lport && 421 grp->il_numa_domain == numa_domain && 422 grp->il_fibnum == fib && 423 memcmp(&grp->il_dependladdr, 424 &inp->inp_inc.inc_ie.ie_dependladdr, 425 sizeof(grp->il_dependladdr)) == 0) { 426 break; 427 } 428 } 429 if (grp == NULL) { 430 /* Create new load balance group. */ 431 grp = in_pcblbgroup_alloc(inp->inp_cred, inp->inp_vflag, 432 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 433 INPCBLBGROUP_SIZMIN, numa_domain, fib); 434 if (grp == NULL) 435 return (ENOBUFS); 436 in_pcblbgroup_insert(grp, inp); 437 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 438 } else if (grp->il_inpcnt == grp->il_inpsiz) { 439 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 440 if (ratecheck(&lastprint, &interval)) 441 printf("lb group port %d, limit reached\n", 442 ntohs(grp->il_lport)); 443 return (0); 444 } 445 446 /* Expand this local group. */ 447 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 448 if (grp == NULL) 449 return (ENOBUFS); 450 in_pcblbgroup_insert(grp, inp); 451 } else { 452 in_pcblbgroup_insert(grp, inp); 453 } 454 return (0); 455 } 456 457 /* 458 * Remove PCB from load balance group. 459 */ 460 static void 461 in_pcbremlbgrouphash(struct inpcb *inp) 462 { 463 struct inpcbinfo *pcbinfo; 464 struct inpcblbgrouphead *hdr; 465 struct inpcblbgroup *grp; 466 struct inpcb *inp1; 467 int i; 468 469 pcbinfo = inp->inp_pcbinfo; 470 471 INP_WLOCK_ASSERT(inp); 472 MPASS(inp->inp_flags & INP_INLBGROUP); 473 INP_HASH_WLOCK_ASSERT(pcbinfo); 474 475 hdr = &pcbinfo->ipi_lbgrouphashbase[ 476 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 477 CK_LIST_FOREACH(grp, hdr, il_list) { 478 for (i = 0; i < grp->il_inpcnt; ++i) { 479 if (grp->il_inp[i] != inp) 480 continue; 481 482 if (grp->il_inpcnt == 1 && 483 LIST_EMPTY(&grp->il_pending)) { 484 /* We are the last, free this local group. */ 485 in_pcblbgroup_free(grp); 486 } else { 487 grp->il_inp[i] = 488 grp->il_inp[grp->il_inpcnt - 1]; 489 490 /* 491 * Synchronize with in_pcblookup_lbgroup(). 492 */ 493 atomic_store_rel_int(&grp->il_inpcnt, 494 grp->il_inpcnt - 1); 495 } 496 inp->inp_flags &= ~INP_INLBGROUP; 497 return; 498 } 499 LIST_FOREACH(inp1, &grp->il_pending, inp_lbgroup_list) { 500 if (inp == inp1) { 501 LIST_REMOVE(inp, inp_lbgroup_list); 502 inp->inp_flags &= ~INP_INLBGROUP; 503 return; 504 } 505 } 506 } 507 __assert_unreachable(); 508 } 509 510 int 511 in_pcblbgroup_numa(struct inpcb *inp, int arg) 512 { 513 struct inpcbinfo *pcbinfo; 514 int error; 515 uint8_t numa_domain; 516 517 switch (arg) { 518 case TCP_REUSPORT_LB_NUMA_NODOM: 519 numa_domain = M_NODOM; 520 break; 521 case TCP_REUSPORT_LB_NUMA_CURDOM: 522 numa_domain = PCPU_GET(domain); 523 break; 524 default: 525 if (arg < 0 || arg >= vm_ndomains) 526 return (EINVAL); 527 numa_domain = arg; 528 } 529 530 pcbinfo = inp->inp_pcbinfo; 531 INP_WLOCK_ASSERT(inp); 532 INP_HASH_WLOCK(pcbinfo); 533 if (in_pcblbgroup_find(inp) != NULL) { 534 /* Remove it from the old group. */ 535 in_pcbremlbgrouphash(inp); 536 /* Add it to the new group based on numa domain. */ 537 in_pcbinslbgrouphash(inp, numa_domain); 538 error = 0; 539 } else { 540 error = ENOENT; 541 } 542 INP_HASH_WUNLOCK(pcbinfo); 543 return (error); 544 } 545 546 /* Make sure it is safe to use hashinit(9) on CK_LIST. */ 547 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb))); 548 549 /* 550 * Initialize an inpcbinfo - a per-VNET instance of connections db. 551 */ 552 void 553 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor, 554 u_int hash_nelements, u_int porthash_nelements) 555 { 556 557 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF); 558 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name, 559 NULL, MTX_DEF); 560 #ifdef VIMAGE 561 pcbinfo->ipi_vnet = curvnet; 562 #endif 563 CK_LIST_INIT(&pcbinfo->ipi_listhead); 564 pcbinfo->ipi_count = 0; 565 pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB, 566 &pcbinfo->ipi_hashmask); 567 pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB, 568 &pcbinfo->ipi_hashmask); 569 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 570 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 571 &pcbinfo->ipi_porthashmask); 572 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 573 &pcbinfo->ipi_lbgrouphashmask); 574 pcbinfo->ipi_zone = pcbstor->ips_zone; 575 pcbinfo->ipi_portzone = pcbstor->ips_portzone; 576 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone); 577 } 578 579 /* 580 * Destroy an inpcbinfo. 581 */ 582 void 583 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 584 { 585 586 KASSERT(pcbinfo->ipi_count == 0, 587 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 588 589 hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask); 590 hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask); 591 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 592 pcbinfo->ipi_porthashmask); 593 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 594 pcbinfo->ipi_lbgrouphashmask); 595 mtx_destroy(&pcbinfo->ipi_hash_lock); 596 mtx_destroy(&pcbinfo->ipi_lock); 597 } 598 599 /* 600 * Initialize a pcbstorage - per protocol zones to allocate inpcbs. 601 */ 602 static void inpcb_fini(void *, int); 603 void 604 in_pcbstorage_init(void *arg) 605 { 606 struct inpcbstorage *pcbstor = arg; 607 608 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name, 609 pcbstor->ips_size, NULL, NULL, pcbstor->ips_pcbinit, 610 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR); 611 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name, 612 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 613 uma_zone_set_smr(pcbstor->ips_portzone, 614 uma_zone_get_smr(pcbstor->ips_zone)); 615 } 616 617 /* 618 * Destroy a pcbstorage - used by unloadable protocols. 619 */ 620 void 621 in_pcbstorage_destroy(void *arg) 622 { 623 struct inpcbstorage *pcbstor = arg; 624 625 uma_zdestroy(pcbstor->ips_zone); 626 uma_zdestroy(pcbstor->ips_portzone); 627 } 628 629 /* 630 * Allocate a PCB and associate it with the socket. 631 * On success return with the PCB locked. 632 */ 633 int 634 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 635 { 636 struct inpcb *inp; 637 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 638 int error; 639 #endif 640 641 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT); 642 if (inp == NULL) 643 return (ENOBUFS); 644 bzero(&inp->inp_start_zero, inp_zero_size); 645 #ifdef NUMA 646 inp->inp_numa_domain = M_NODOM; 647 #endif 648 inp->inp_pcbinfo = pcbinfo; 649 inp->inp_socket = so; 650 inp->inp_cred = crhold(so->so_cred); 651 inp->inp_inc.inc_fibnum = so->so_fibnum; 652 #ifdef MAC 653 error = mac_inpcb_init(inp, M_NOWAIT); 654 if (error != 0) 655 goto out; 656 mac_inpcb_create(so, inp); 657 #endif 658 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 659 error = ipsec_init_pcbpolicy(inp); 660 if (error != 0) { 661 #ifdef MAC 662 mac_inpcb_destroy(inp); 663 #endif 664 goto out; 665 } 666 #endif /*IPSEC*/ 667 #ifdef INET6 668 if (INP_SOCKAF(so) == AF_INET6) { 669 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6; 670 if (V_ip6_v6only) 671 inp->inp_flags |= IN6P_IPV6_V6ONLY; 672 #ifdef INET 673 else 674 inp->inp_vflag |= INP_IPV4; 675 #endif 676 if (V_ip6_auto_flowlabel) 677 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 678 inp->in6p_hops = -1; /* use kernel default */ 679 } 680 #endif 681 #if defined(INET) && defined(INET6) 682 else 683 #endif 684 #ifdef INET 685 inp->inp_vflag |= INP_IPV4; 686 #endif 687 inp->inp_smr = SMR_SEQ_INVALID; 688 689 /* 690 * Routes in inpcb's can cache L2 as well; they are guaranteed 691 * to be cleaned up. 692 */ 693 inp->inp_route.ro_flags = RT_LLE_CACHE; 694 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */ 695 INP_WLOCK(inp); 696 INP_INFO_WLOCK(pcbinfo); 697 pcbinfo->ipi_count++; 698 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 699 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list); 700 INP_INFO_WUNLOCK(pcbinfo); 701 so->so_pcb = inp; 702 703 return (0); 704 705 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 706 out: 707 crfree(inp->inp_cred); 708 #ifdef INVARIANTS 709 inp->inp_cred = NULL; 710 #endif 711 uma_zfree_smr(pcbinfo->ipi_zone, inp); 712 return (error); 713 #endif 714 } 715 716 #ifdef INET 717 int 718 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, int flags, 719 struct ucred *cred) 720 { 721 int anonport, error; 722 723 KASSERT(sin == NULL || sin->sin_family == AF_INET, 724 ("%s: invalid address family for %p", __func__, sin)); 725 KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in), 726 ("%s: invalid address length for %p", __func__, sin)); 727 INP_WLOCK_ASSERT(inp); 728 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 729 730 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 731 return (EINVAL); 732 anonport = sin == NULL || sin->sin_port == 0; 733 error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr, 734 &inp->inp_lport, flags, cred); 735 if (error) 736 return (error); 737 if (in_pcbinshash(inp) != 0) { 738 inp->inp_laddr.s_addr = INADDR_ANY; 739 inp->inp_lport = 0; 740 inp->inp_flags &= ~INP_BOUNDFIB; 741 return (EAGAIN); 742 } 743 if (anonport) 744 inp->inp_flags |= INP_ANONPORT; 745 return (0); 746 } 747 #endif 748 749 #if defined(INET) || defined(INET6) 750 /* 751 * Assign a local port like in_pcb_lport(), but also used with connect() 752 * and a foreign address and port. If fsa is non-NULL, choose a local port 753 * that is unused with those, otherwise one that is completely unused. 754 * lsa can be NULL for IPv6. 755 */ 756 int 757 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 758 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 759 { 760 struct inpcbinfo *pcbinfo; 761 struct inpcb *tmpinp; 762 unsigned short *lastport; 763 int count, error; 764 u_short aux, first, last, lport; 765 #ifdef INET 766 struct in_addr laddr, faddr; 767 #endif 768 #ifdef INET6 769 struct in6_addr *laddr6, *faddr6; 770 #endif 771 772 pcbinfo = inp->inp_pcbinfo; 773 774 /* 775 * Because no actual state changes occur here, a global write lock on 776 * the pcbinfo isn't required. 777 */ 778 INP_LOCK_ASSERT(inp); 779 INP_HASH_LOCK_ASSERT(pcbinfo); 780 781 if (inp->inp_flags & INP_HIGHPORT) { 782 first = V_ipport_hifirstauto; /* sysctl */ 783 last = V_ipport_hilastauto; 784 lastport = &pcbinfo->ipi_lasthi; 785 } else if (inp->inp_flags & INP_LOWPORT) { 786 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 787 if (error) 788 return (error); 789 first = V_ipport_lowfirstauto; /* 1023 */ 790 last = V_ipport_lowlastauto; /* 600 */ 791 lastport = &pcbinfo->ipi_lastlow; 792 } else { 793 first = V_ipport_firstauto; /* sysctl */ 794 last = V_ipport_lastauto; 795 lastport = &pcbinfo->ipi_lastport; 796 } 797 798 /* 799 * Instead of having two loops further down counting up or down 800 * make sure that first is always <= last and go with only one 801 * code path implementing all logic. 802 */ 803 if (first > last) { 804 aux = first; 805 first = last; 806 last = aux; 807 } 808 809 #ifdef INET 810 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */ 811 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 812 if (lsa != NULL) 813 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 814 if (fsa != NULL) 815 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 816 } 817 #endif 818 #ifdef INET6 819 laddr6 = NULL; 820 if ((inp->inp_vflag & INP_IPV6) != 0) { 821 if (lsa != NULL) 822 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 823 if (fsa != NULL) 824 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 825 } 826 #endif 827 828 tmpinp = NULL; 829 lport = *lportp; 830 831 if (V_ipport_randomized) 832 *lastport = first + (arc4random() % (last - first)); 833 834 count = last - first; 835 836 do { 837 if (count-- < 0) /* completely used? */ 838 return (EADDRNOTAVAIL); 839 ++*lastport; 840 if (*lastport < first || *lastport > last) 841 *lastport = first; 842 lport = htons(*lastport); 843 844 if (fsa != NULL) { 845 #ifdef INET 846 if (lsa->sa_family == AF_INET) { 847 tmpinp = in_pcblookup_hash_locked(pcbinfo, 848 faddr, fport, laddr, lport, lookupflags, 849 M_NODOM, RT_ALL_FIBS); 850 } 851 #endif 852 #ifdef INET6 853 if (lsa->sa_family == AF_INET6) { 854 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 855 faddr6, fport, laddr6, lport, lookupflags, 856 M_NODOM, RT_ALL_FIBS); 857 } 858 #endif 859 } else { 860 #ifdef INET6 861 if ((inp->inp_vflag & INP_IPV6) != 0) { 862 tmpinp = in6_pcblookup_local(pcbinfo, 863 &inp->in6p_laddr, lport, RT_ALL_FIBS, 864 lookupflags, cred); 865 #ifdef INET 866 if (tmpinp == NULL && 867 (inp->inp_vflag & INP_IPV4)) 868 tmpinp = in_pcblookup_local(pcbinfo, 869 laddr, lport, RT_ALL_FIBS, 870 lookupflags, cred); 871 #endif 872 } 873 #endif 874 #if defined(INET) && defined(INET6) 875 else 876 #endif 877 #ifdef INET 878 tmpinp = in_pcblookup_local(pcbinfo, laddr, 879 lport, RT_ALL_FIBS, lookupflags, cred); 880 #endif 881 } 882 } while (tmpinp != NULL); 883 884 *lportp = lport; 885 886 return (0); 887 } 888 889 /* 890 * Select a local port (number) to use. 891 */ 892 int 893 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 894 struct ucred *cred, int lookupflags) 895 { 896 struct sockaddr_in laddr; 897 898 if (laddrp) { 899 bzero(&laddr, sizeof(laddr)); 900 laddr.sin_family = AF_INET; 901 laddr.sin_addr = *laddrp; 902 } 903 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 904 NULL, lportp, NULL, 0, cred, lookupflags)); 905 } 906 #endif /* INET || INET6 */ 907 908 #ifdef INET 909 /* 910 * Determine whether the inpcb can be bound to the specified address/port tuple. 911 */ 912 static int 913 in_pcbbind_avail(struct inpcb *inp, const struct in_addr laddr, 914 const u_short lport, const int fib, int sooptions, int lookupflags, 915 struct ucred *cred) 916 { 917 int reuseport, reuseport_lb; 918 919 INP_LOCK_ASSERT(inp); 920 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 921 922 reuseport = (sooptions & SO_REUSEPORT); 923 reuseport_lb = (sooptions & SO_REUSEPORT_LB); 924 925 if (IN_MULTICAST(ntohl(laddr.s_addr))) { 926 /* 927 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 928 * allow complete duplication of binding if 929 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 930 * and a multicast address is bound on both 931 * new and duplicated sockets. 932 */ 933 if ((sooptions & (SO_REUSEADDR | SO_REUSEPORT)) != 0) 934 reuseport = SO_REUSEADDR | SO_REUSEPORT; 935 /* 936 * XXX: How to deal with SO_REUSEPORT_LB here? 937 * Treat same as SO_REUSEPORT for now. 938 */ 939 if ((sooptions & (SO_REUSEADDR | SO_REUSEPORT_LB)) != 0) 940 reuseport_lb = SO_REUSEADDR | SO_REUSEPORT_LB; 941 } else if (!in_nullhost(laddr)) { 942 struct sockaddr_in sin; 943 944 memset(&sin, 0, sizeof(sin)); 945 sin.sin_family = AF_INET; 946 sin.sin_len = sizeof(sin); 947 sin.sin_addr = laddr; 948 949 /* 950 * Is the address a local IP address? 951 * If INP_BINDANY is set, then the socket may be bound 952 * to any endpoint address, local or not. 953 */ 954 if ((inp->inp_flags & INP_BINDANY) == 0 && 955 ifa_ifwithaddr_check((const struct sockaddr *)&sin) == 0) 956 return (EADDRNOTAVAIL); 957 } 958 959 if (lport != 0) { 960 struct inpcb *t; 961 962 if (ntohs(lport) <= V_ipport_reservedhigh && 963 ntohs(lport) >= V_ipport_reservedlow && 964 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 965 return (EACCES); 966 967 if (!IN_MULTICAST(ntohl(laddr.s_addr)) && 968 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 969 /* 970 * If a socket owned by a different user is already 971 * bound to this port, fail. In particular, SO_REUSE* 972 * can only be used to share a port among sockets owned 973 * by the same user. 974 * 975 * However, we can share a port with a connected socket 976 * which has a unique 4-tuple. 977 */ 978 t = in_pcblookup_local(inp->inp_pcbinfo, laddr, lport, 979 RT_ALL_FIBS, INPLOOKUP_WILDCARD, cred); 980 if (t != NULL && 981 (inp->inp_socket->so_type != SOCK_STREAM || 982 in_nullhost(t->inp_faddr)) && 983 (inp->inp_cred->cr_uid != t->inp_cred->cr_uid)) 984 return (EADDRINUSE); 985 } 986 t = in_pcblookup_local(inp->inp_pcbinfo, laddr, lport, fib, 987 lookupflags, cred); 988 if (t != NULL && ((reuseport | reuseport_lb) & 989 t->inp_socket->so_options) == 0) { 990 #ifdef INET6 991 if (!in_nullhost(laddr) || 992 !in_nullhost(t->inp_laddr) || 993 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 994 (t->inp_vflag & INP_IPV6PROTO) == 0) 995 #endif 996 return (EADDRINUSE); 997 } 998 } 999 return (0); 1000 } 1001 1002 /* 1003 * Set up a bind operation on a PCB, performing port allocation 1004 * as required, but do not actually modify the PCB. Callers can 1005 * either complete the bind by setting inp_laddr/inp_lport and 1006 * calling in_pcbinshash(), or they can just use the resulting 1007 * port and address to authorise the sending of a once-off packet. 1008 * 1009 * On error, the values of *laddrp and *lportp are not changed. 1010 */ 1011 int 1012 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp, 1013 u_short *lportp, int flags, struct ucred *cred) 1014 { 1015 struct socket *so = inp->inp_socket; 1016 struct in_addr laddr; 1017 u_short lport = 0; 1018 int error, fib, lookupflags, sooptions; 1019 1020 /* 1021 * No state changes, so read locks are sufficient here. 1022 */ 1023 INP_LOCK_ASSERT(inp); 1024 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1025 1026 laddr.s_addr = *laddrp; 1027 if (sin != NULL && laddr.s_addr != INADDR_ANY) 1028 return (EINVAL); 1029 1030 lookupflags = 0; 1031 sooptions = atomic_load_int(&so->so_options); 1032 if ((sooptions & (SO_REUSEADDR | SO_REUSEPORT | SO_REUSEPORT_LB)) == 0) 1033 lookupflags = INPLOOKUP_WILDCARD; 1034 if (sin == NULL) { 1035 if ((error = prison_local_ip4(cred, &laddr)) != 0) 1036 return (error); 1037 } else { 1038 KASSERT(sin->sin_family == AF_INET, 1039 ("%s: invalid family for address %p", __func__, sin)); 1040 KASSERT(sin->sin_len == sizeof(*sin), 1041 ("%s: invalid length for address %p", __func__, sin)); 1042 1043 error = prison_local_ip4(cred, &sin->sin_addr); 1044 if (error) 1045 return (error); 1046 if (sin->sin_port != *lportp) { 1047 /* Don't allow the port to change. */ 1048 if (*lportp != 0) 1049 return (EINVAL); 1050 lport = sin->sin_port; 1051 } 1052 laddr = sin->sin_addr; 1053 1054 fib = (flags & INPBIND_FIB) != 0 ? inp->inp_inc.inc_fibnum : 1055 RT_ALL_FIBS; 1056 1057 /* See if this address/port combo is available. */ 1058 error = in_pcbbind_avail(inp, laddr, lport, fib, sooptions, 1059 lookupflags, cred); 1060 if (error != 0) 1061 return (error); 1062 } 1063 if (*lportp != 0) 1064 lport = *lportp; 1065 if (lport == 0) { 1066 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 1067 if (error != 0) 1068 return (error); 1069 } 1070 *laddrp = laddr.s_addr; 1071 *lportp = lport; 1072 if ((flags & INPBIND_FIB) != 0) 1073 inp->inp_flags |= INP_BOUNDFIB; 1074 return (0); 1075 } 1076 1077 /* 1078 * Connect from a socket to a specified address. 1079 * Both address and port must be specified in argument sin. 1080 * If don't have a local address for this socket yet, 1081 * then pick one. 1082 */ 1083 int 1084 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred) 1085 { 1086 u_short lport, fport; 1087 in_addr_t laddr, faddr; 1088 int anonport, error; 1089 1090 INP_WLOCK_ASSERT(inp); 1091 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1092 KASSERT(in_nullhost(inp->inp_faddr), 1093 ("%s: inp is already connected", __func__)); 1094 1095 lport = inp->inp_lport; 1096 laddr = inp->inp_laddr.s_addr; 1097 anonport = (lport == 0); 1098 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport, 1099 cred); 1100 if (error) 1101 return (error); 1102 1103 inp->inp_faddr.s_addr = faddr; 1104 inp->inp_fport = fport; 1105 1106 /* Do the initial binding of the local address if required. */ 1107 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1108 inp->inp_lport = lport; 1109 inp->inp_laddr.s_addr = laddr; 1110 if (in_pcbinshash(inp) != 0) { 1111 inp->inp_laddr.s_addr = inp->inp_faddr.s_addr = 1112 INADDR_ANY; 1113 inp->inp_lport = inp->inp_fport = 0; 1114 return (EAGAIN); 1115 } 1116 } else { 1117 inp->inp_lport = lport; 1118 inp->inp_laddr.s_addr = laddr; 1119 if ((inp->inp_flags & INP_INHASHLIST) != 0) 1120 in_pcbrehash(inp); 1121 else 1122 in_pcbinshash(inp); 1123 } 1124 1125 if (anonport) 1126 inp->inp_flags |= INP_ANONPORT; 1127 return (0); 1128 } 1129 1130 /* 1131 * Do proper source address selection on an unbound socket in case 1132 * of connect. Take jails into account as well. 1133 */ 1134 int 1135 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1136 struct ucred *cred) 1137 { 1138 struct ifaddr *ifa; 1139 struct sockaddr *sa; 1140 struct sockaddr_in *sin, dst; 1141 struct nhop_object *nh; 1142 int error; 1143 1144 NET_EPOCH_ASSERT(); 1145 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1146 1147 /* 1148 * Bypass source address selection and use the primary jail IP 1149 * if requested. 1150 */ 1151 if (!prison_saddrsel_ip4(cred, laddr)) 1152 return (0); 1153 1154 error = 0; 1155 1156 nh = NULL; 1157 bzero(&dst, sizeof(dst)); 1158 sin = &dst; 1159 sin->sin_family = AF_INET; 1160 sin->sin_len = sizeof(struct sockaddr_in); 1161 sin->sin_addr.s_addr = faddr->s_addr; 1162 1163 /* 1164 * If route is known our src addr is taken from the i/f, 1165 * else punt. 1166 * 1167 * Find out route to destination. 1168 */ 1169 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1170 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1171 0, NHR_NONE, 0); 1172 1173 /* 1174 * If we found a route, use the address corresponding to 1175 * the outgoing interface. 1176 * 1177 * Otherwise assume faddr is reachable on a directly connected 1178 * network and try to find a corresponding interface to take 1179 * the source address from. 1180 */ 1181 if (nh == NULL || nh->nh_ifp == NULL) { 1182 struct in_ifaddr *ia; 1183 struct ifnet *ifp; 1184 1185 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1186 inp->inp_socket->so_fibnum)); 1187 if (ia == NULL) { 1188 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1189 inp->inp_socket->so_fibnum)); 1190 } 1191 if (ia == NULL) { 1192 error = ENETUNREACH; 1193 goto done; 1194 } 1195 1196 if (!prison_flag(cred, PR_IP4)) { 1197 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1198 goto done; 1199 } 1200 1201 ifp = ia->ia_ifp; 1202 ia = NULL; 1203 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1204 sa = ifa->ifa_addr; 1205 if (sa->sa_family != AF_INET) 1206 continue; 1207 sin = (struct sockaddr_in *)sa; 1208 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1209 ia = (struct in_ifaddr *)ifa; 1210 break; 1211 } 1212 } 1213 if (ia != NULL) { 1214 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1215 goto done; 1216 } 1217 1218 /* 3. As a last resort return the 'default' jail address. */ 1219 error = prison_get_ip4(cred, laddr); 1220 goto done; 1221 } 1222 1223 /* 1224 * If the outgoing interface on the route found is not 1225 * a loopback interface, use the address from that interface. 1226 * In case of jails do those three steps: 1227 * 1. check if the interface address belongs to the jail. If so use it. 1228 * 2. check if we have any address on the outgoing interface 1229 * belonging to this jail. If so use it. 1230 * 3. as a last resort return the 'default' jail address. 1231 */ 1232 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1233 struct in_ifaddr *ia; 1234 struct ifnet *ifp; 1235 1236 /* If not jailed, use the default returned. */ 1237 if (!prison_flag(cred, PR_IP4)) { 1238 ia = (struct in_ifaddr *)nh->nh_ifa; 1239 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1240 goto done; 1241 } 1242 1243 /* Jailed. */ 1244 /* 1. Check if the iface address belongs to the jail. */ 1245 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1246 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1247 ia = (struct in_ifaddr *)nh->nh_ifa; 1248 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1249 goto done; 1250 } 1251 1252 /* 1253 * 2. Check if we have any address on the outgoing interface 1254 * belonging to this jail. 1255 */ 1256 ia = NULL; 1257 ifp = nh->nh_ifp; 1258 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1259 sa = ifa->ifa_addr; 1260 if (sa->sa_family != AF_INET) 1261 continue; 1262 sin = (struct sockaddr_in *)sa; 1263 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1264 ia = (struct in_ifaddr *)ifa; 1265 break; 1266 } 1267 } 1268 if (ia != NULL) { 1269 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1270 goto done; 1271 } 1272 1273 /* 3. As a last resort return the 'default' jail address. */ 1274 error = prison_get_ip4(cred, laddr); 1275 goto done; 1276 } 1277 1278 /* 1279 * The outgoing interface is marked with 'loopback net', so a route 1280 * to ourselves is here. 1281 * Try to find the interface of the destination address and then 1282 * take the address from there. That interface is not necessarily 1283 * a loopback interface. 1284 * In case of jails, check that it is an address of the jail 1285 * and if we cannot find, fall back to the 'default' jail address. 1286 */ 1287 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1288 struct in_ifaddr *ia; 1289 1290 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1291 inp->inp_socket->so_fibnum)); 1292 if (ia == NULL) 1293 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1294 inp->inp_socket->so_fibnum)); 1295 if (ia == NULL) 1296 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1297 1298 if (!prison_flag(cred, PR_IP4)) { 1299 if (ia == NULL) { 1300 error = ENETUNREACH; 1301 goto done; 1302 } 1303 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1304 goto done; 1305 } 1306 1307 /* Jailed. */ 1308 if (ia != NULL) { 1309 struct ifnet *ifp; 1310 1311 ifp = ia->ia_ifp; 1312 ia = NULL; 1313 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1314 sa = ifa->ifa_addr; 1315 if (sa->sa_family != AF_INET) 1316 continue; 1317 sin = (struct sockaddr_in *)sa; 1318 if (prison_check_ip4(cred, 1319 &sin->sin_addr) == 0) { 1320 ia = (struct in_ifaddr *)ifa; 1321 break; 1322 } 1323 } 1324 if (ia != NULL) { 1325 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1326 goto done; 1327 } 1328 } 1329 1330 /* 3. As a last resort return the 'default' jail address. */ 1331 error = prison_get_ip4(cred, laddr); 1332 goto done; 1333 } 1334 1335 done: 1336 if (error == 0 && laddr->s_addr == INADDR_ANY) 1337 return (EHOSTUNREACH); 1338 return (error); 1339 } 1340 1341 /* 1342 * Set up for a connect from a socket to the specified address. 1343 * On entry, *laddrp and *lportp should contain the current local 1344 * address and port for the PCB; these are updated to the values 1345 * that should be placed in inp_laddr and inp_lport to complete 1346 * the connect. 1347 * 1348 * On success, *faddrp and *fportp will be set to the remote address 1349 * and port. These are not updated in the error case. 1350 */ 1351 int 1352 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin, 1353 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1354 struct ucred *cred) 1355 { 1356 struct in_ifaddr *ia; 1357 struct in_addr laddr, faddr; 1358 u_short lport, fport; 1359 int error; 1360 1361 KASSERT(sin->sin_family == AF_INET, 1362 ("%s: invalid address family for %p", __func__, sin)); 1363 KASSERT(sin->sin_len == sizeof(*sin), 1364 ("%s: invalid address length for %p", __func__, sin)); 1365 1366 /* 1367 * Because a global state change doesn't actually occur here, a read 1368 * lock is sufficient. 1369 */ 1370 NET_EPOCH_ASSERT(); 1371 INP_LOCK_ASSERT(inp); 1372 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1373 1374 if (sin->sin_port == 0) 1375 return (EADDRNOTAVAIL); 1376 laddr.s_addr = *laddrp; 1377 lport = *lportp; 1378 faddr = sin->sin_addr; 1379 fport = sin->sin_port; 1380 #ifdef ROUTE_MPATH 1381 if (CALC_FLOWID_OUTBOUND) { 1382 uint32_t hash_val, hash_type; 1383 1384 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, 1385 inp->inp_socket->so_proto->pr_protocol, &hash_type); 1386 1387 inp->inp_flowid = hash_val; 1388 inp->inp_flowtype = hash_type; 1389 } 1390 #endif 1391 if (V_connect_inaddr_wild && !CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1392 /* 1393 * If the destination address is INADDR_ANY, 1394 * use the primary local address. 1395 * If the supplied address is INADDR_BROADCAST, 1396 * and the primary interface supports broadcast, 1397 * choose the broadcast address for that interface. 1398 */ 1399 if (faddr.s_addr == INADDR_ANY) { 1400 faddr = 1401 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1402 if ((error = prison_get_ip4(cred, &faddr)) != 0) 1403 return (error); 1404 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1405 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1406 IFF_BROADCAST) 1407 faddr = satosin(&CK_STAILQ_FIRST( 1408 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1409 } 1410 } else if (faddr.s_addr == INADDR_ANY) { 1411 return (ENETUNREACH); 1412 } 1413 if (laddr.s_addr == INADDR_ANY) { 1414 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1415 /* 1416 * If the destination address is multicast and an outgoing 1417 * interface has been set as a multicast option, prefer the 1418 * address of that interface as our source address. 1419 */ 1420 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1421 inp->inp_moptions != NULL) { 1422 struct ip_moptions *imo; 1423 struct ifnet *ifp; 1424 1425 imo = inp->inp_moptions; 1426 if (imo->imo_multicast_ifp != NULL) { 1427 ifp = imo->imo_multicast_ifp; 1428 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1429 if (ia->ia_ifp == ifp && 1430 prison_check_ip4(cred, 1431 &ia->ia_addr.sin_addr) == 0) 1432 break; 1433 } 1434 if (ia == NULL) 1435 error = EADDRNOTAVAIL; 1436 else { 1437 laddr = ia->ia_addr.sin_addr; 1438 error = 0; 1439 } 1440 } 1441 } 1442 if (error) 1443 return (error); 1444 } 1445 1446 if (lport != 0) { 1447 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, 1448 fport, laddr, lport, 0, M_NODOM, RT_ALL_FIBS) != NULL) 1449 return (EADDRINUSE); 1450 } else { 1451 struct sockaddr_in lsin, fsin; 1452 1453 bzero(&lsin, sizeof(lsin)); 1454 bzero(&fsin, sizeof(fsin)); 1455 lsin.sin_family = AF_INET; 1456 lsin.sin_addr = laddr; 1457 fsin.sin_family = AF_INET; 1458 fsin.sin_addr = faddr; 1459 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, 1460 &lport, (struct sockaddr *)& fsin, fport, cred, 1461 INPLOOKUP_WILDCARD); 1462 if (error) 1463 return (error); 1464 } 1465 *laddrp = laddr.s_addr; 1466 *lportp = lport; 1467 *faddrp = faddr.s_addr; 1468 *fportp = fport; 1469 return (0); 1470 } 1471 1472 void 1473 in_pcbdisconnect(struct inpcb *inp) 1474 { 1475 1476 INP_WLOCK_ASSERT(inp); 1477 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1478 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 1479 ("%s: inp %p was already disconnected", __func__, inp)); 1480 1481 in_pcbremhash_locked(inp); 1482 1483 /* See the comment in in_pcbinshash(). */ 1484 inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr); 1485 inp->inp_laddr.s_addr = INADDR_ANY; 1486 inp->inp_faddr.s_addr = INADDR_ANY; 1487 inp->inp_fport = 0; 1488 } 1489 #endif /* INET */ 1490 1491 void 1492 in_pcblisten(struct inpcb *inp) 1493 { 1494 struct inpcblbgroup *grp; 1495 1496 INP_WLOCK_ASSERT(inp); 1497 1498 if ((inp->inp_flags & INP_INLBGROUP) != 0) { 1499 struct inpcbinfo *pcbinfo; 1500 1501 pcbinfo = inp->inp_pcbinfo; 1502 INP_HASH_WLOCK(pcbinfo); 1503 grp = in_pcblbgroup_find(inp); 1504 LIST_REMOVE(inp, inp_lbgroup_list); 1505 in_pcblbgroup_insert(grp, inp); 1506 INP_HASH_WUNLOCK(pcbinfo); 1507 } 1508 } 1509 1510 /* 1511 * inpcb hash lookups are protected by SMR section. 1512 * 1513 * Once desired pcb has been found, switching from SMR section to a pcb 1514 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK 1515 * here because SMR is a critical section. 1516 * In 99%+ cases inp_smr_lock() would obtain the lock immediately. 1517 */ 1518 void 1519 inp_lock(struct inpcb *inp, const inp_lookup_t lock) 1520 { 1521 1522 lock == INPLOOKUP_RLOCKPCB ? 1523 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock); 1524 } 1525 1526 void 1527 inp_unlock(struct inpcb *inp, const inp_lookup_t lock) 1528 { 1529 1530 lock == INPLOOKUP_RLOCKPCB ? 1531 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock); 1532 } 1533 1534 int 1535 inp_trylock(struct inpcb *inp, const inp_lookup_t lock) 1536 { 1537 1538 return (lock == INPLOOKUP_RLOCKPCB ? 1539 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock)); 1540 } 1541 1542 static inline bool 1543 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags) 1544 { 1545 1546 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB); 1547 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr); 1548 1549 if (__predict_true(inp_trylock(inp, lock))) { 1550 if (__predict_false(inp->inp_flags & ignflags)) { 1551 smr_exit(inp->inp_pcbinfo->ipi_smr); 1552 inp_unlock(inp, lock); 1553 return (false); 1554 } 1555 smr_exit(inp->inp_pcbinfo->ipi_smr); 1556 return (true); 1557 } 1558 1559 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1560 smr_exit(inp->inp_pcbinfo->ipi_smr); 1561 inp_lock(inp, lock); 1562 if (__predict_false(in_pcbrele(inp, lock))) 1563 return (false); 1564 /* 1565 * inp acquired through refcount & lock for sure didn't went 1566 * through uma_zfree(). However, it may have already went 1567 * through in_pcbfree() and has another reference, that 1568 * prevented its release by our in_pcbrele(). 1569 */ 1570 if (__predict_false(inp->inp_flags & ignflags)) { 1571 inp_unlock(inp, lock); 1572 return (false); 1573 } 1574 return (true); 1575 } else { 1576 smr_exit(inp->inp_pcbinfo->ipi_smr); 1577 return (false); 1578 } 1579 } 1580 1581 bool 1582 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock) 1583 { 1584 1585 /* 1586 * in_pcblookup() family of functions ignore not only freed entries, 1587 * that may be found due to lockless access to the hash, but dropped 1588 * entries, too. 1589 */ 1590 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED)); 1591 } 1592 1593 /* 1594 * inp_next() - inpcb hash/list traversal iterator 1595 * 1596 * Requires initialized struct inpcb_iterator for context. 1597 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR(). 1598 * 1599 * - Iterator can have either write-lock or read-lock semantics, that can not 1600 * be changed later. 1601 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through 1602 * a single hash slot. Note: only rip_input() does the latter. 1603 * - Iterator may have optional bool matching function. The matching function 1604 * will be executed for each inpcb in the SMR context, so it can not acquire 1605 * locks and can safely access only immutable fields of inpcb. 1606 * 1607 * A fresh initialized iterator has NULL inpcb in its context and that 1608 * means that inp_next() call would return the very first inpcb on the list 1609 * locked with desired semantic. In all following calls the context pointer 1610 * shall hold the current inpcb pointer. The KPI user is not supposed to 1611 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL 1612 * and write NULL to its context. After end of traversal an iterator can be 1613 * reused. 1614 * 1615 * List traversals have the following features/constraints: 1616 * - New entries won't be seen, as they are always added to the head of a list. 1617 * - Removed entries won't stop traversal as long as they are not added to 1618 * a different list. This is violated by in_pcbrehash(). 1619 */ 1620 #define II_LIST_FIRST(ipi, hash) \ 1621 (((hash) == INP_ALL_LIST) ? \ 1622 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \ 1623 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)])) 1624 #define II_LIST_NEXT(inp, hash) \ 1625 (((hash) == INP_ALL_LIST) ? \ 1626 CK_LIST_NEXT((inp), inp_list) : \ 1627 CK_LIST_NEXT((inp), inp_hash_exact)) 1628 #define II_LOCK_ASSERT(inp, lock) \ 1629 rw_assert(&(inp)->inp_lock, \ 1630 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED ) 1631 struct inpcb * 1632 inp_next(struct inpcb_iterator *ii) 1633 { 1634 const struct inpcbinfo *ipi = ii->ipi; 1635 inp_match_t *match = ii->match; 1636 void *ctx = ii->ctx; 1637 inp_lookup_t lock = ii->lock; 1638 int hash = ii->hash; 1639 struct inpcb *inp; 1640 1641 if (ii->inp == NULL) { /* First call. */ 1642 smr_enter(ipi->ipi_smr); 1643 /* This is unrolled CK_LIST_FOREACH(). */ 1644 for (inp = II_LIST_FIRST(ipi, hash); 1645 inp != NULL; 1646 inp = II_LIST_NEXT(inp, hash)) { 1647 if (match != NULL && (match)(inp, ctx) == false) 1648 continue; 1649 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED))) 1650 break; 1651 else { 1652 smr_enter(ipi->ipi_smr); 1653 MPASS(inp != II_LIST_FIRST(ipi, hash)); 1654 inp = II_LIST_FIRST(ipi, hash); 1655 if (inp == NULL) 1656 break; 1657 } 1658 } 1659 1660 if (inp == NULL) 1661 smr_exit(ipi->ipi_smr); 1662 else 1663 ii->inp = inp; 1664 1665 return (inp); 1666 } 1667 1668 /* Not a first call. */ 1669 smr_enter(ipi->ipi_smr); 1670 restart: 1671 inp = ii->inp; 1672 II_LOCK_ASSERT(inp, lock); 1673 next: 1674 inp = II_LIST_NEXT(inp, hash); 1675 if (inp == NULL) { 1676 smr_exit(ipi->ipi_smr); 1677 goto found; 1678 } 1679 1680 if (match != NULL && (match)(inp, ctx) == false) 1681 goto next; 1682 1683 if (__predict_true(inp_trylock(inp, lock))) { 1684 if (__predict_false(inp->inp_flags & INP_FREED)) { 1685 /* 1686 * Entries are never inserted in middle of a list, thus 1687 * as long as we are in SMR, we can continue traversal. 1688 * Jump to 'restart' should yield in the same result, 1689 * but could produce unnecessary looping. Could this 1690 * looping be unbound? 1691 */ 1692 inp_unlock(inp, lock); 1693 goto next; 1694 } else { 1695 smr_exit(ipi->ipi_smr); 1696 goto found; 1697 } 1698 } 1699 1700 /* 1701 * Can't obtain lock immediately, thus going hard. Once we exit the 1702 * SMR section we can no longer jump to 'next', and our only stable 1703 * anchoring point is ii->inp, which we keep locked for this case, so 1704 * we jump to 'restart'. 1705 */ 1706 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1707 smr_exit(ipi->ipi_smr); 1708 inp_lock(inp, lock); 1709 if (__predict_false(in_pcbrele(inp, lock))) { 1710 smr_enter(ipi->ipi_smr); 1711 goto restart; 1712 } 1713 /* 1714 * See comment in inp_smr_lock(). 1715 */ 1716 if (__predict_false(inp->inp_flags & INP_FREED)) { 1717 inp_unlock(inp, lock); 1718 smr_enter(ipi->ipi_smr); 1719 goto restart; 1720 } 1721 } else 1722 goto next; 1723 1724 found: 1725 inp_unlock(ii->inp, lock); 1726 ii->inp = inp; 1727 1728 return (ii->inp); 1729 } 1730 1731 /* 1732 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1733 * stability of an inpcb pointer despite the inpcb lock being released or 1734 * SMR section exited. 1735 * 1736 * To free a reference later in_pcbrele_(r|w)locked() must be performed. 1737 */ 1738 void 1739 in_pcbref(struct inpcb *inp) 1740 { 1741 u_int old __diagused; 1742 1743 old = refcount_acquire(&inp->inp_refcount); 1744 KASSERT(old > 0, ("%s: refcount 0", __func__)); 1745 } 1746 1747 /* 1748 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially 1749 * freeing the pcb, if the reference was very last. 1750 */ 1751 bool 1752 in_pcbrele_rlocked(struct inpcb *inp) 1753 { 1754 1755 INP_RLOCK_ASSERT(inp); 1756 1757 if (!refcount_release(&inp->inp_refcount)) 1758 return (false); 1759 1760 MPASS(inp->inp_flags & INP_FREED); 1761 MPASS(inp->inp_socket == NULL); 1762 crfree(inp->inp_cred); 1763 #ifdef INVARIANTS 1764 inp->inp_cred = NULL; 1765 #endif 1766 INP_RUNLOCK(inp); 1767 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1768 return (true); 1769 } 1770 1771 bool 1772 in_pcbrele_wlocked(struct inpcb *inp) 1773 { 1774 1775 INP_WLOCK_ASSERT(inp); 1776 1777 if (!refcount_release(&inp->inp_refcount)) 1778 return (false); 1779 1780 MPASS(inp->inp_flags & INP_FREED); 1781 MPASS(inp->inp_socket == NULL); 1782 crfree(inp->inp_cred); 1783 #ifdef INVARIANTS 1784 inp->inp_cred = NULL; 1785 #endif 1786 INP_WUNLOCK(inp); 1787 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1788 return (true); 1789 } 1790 1791 bool 1792 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock) 1793 { 1794 1795 return (lock == INPLOOKUP_RLOCKPCB ? 1796 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp)); 1797 } 1798 1799 /* 1800 * Unconditionally schedule an inpcb to be freed by decrementing its 1801 * reference count, which should occur only after the inpcb has been detached 1802 * from its socket. If another thread holds a temporary reference (acquired 1803 * using in_pcbref()) then the free is deferred until that reference is 1804 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked. 1805 * Almost all work, including removal from global lists, is done in this 1806 * context, where the pcbinfo lock is held. 1807 */ 1808 void 1809 in_pcbfree(struct inpcb *inp) 1810 { 1811 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1812 #ifdef INET 1813 struct ip_moptions *imo; 1814 #endif 1815 #ifdef INET6 1816 struct ip6_moptions *im6o; 1817 #endif 1818 1819 INP_WLOCK_ASSERT(inp); 1820 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1821 KASSERT((inp->inp_flags & INP_FREED) == 0, 1822 ("%s: called twice for pcb %p", __func__, inp)); 1823 1824 /* 1825 * in_pcblookup_local() and in6_pcblookup_local() may return an inpcb 1826 * from the hash without acquiring inpcb lock, they rely on the hash 1827 * lock, thus in_pcbremhash() should be the first action. 1828 */ 1829 if (inp->inp_flags & INP_INHASHLIST) 1830 in_pcbremhash(inp); 1831 INP_INFO_WLOCK(pcbinfo); 1832 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1833 pcbinfo->ipi_count--; 1834 CK_LIST_REMOVE(inp, inp_list); 1835 INP_INFO_WUNLOCK(pcbinfo); 1836 1837 #ifdef RATELIMIT 1838 if (inp->inp_snd_tag != NULL) 1839 in_pcbdetach_txrtlmt(inp); 1840 #endif 1841 inp->inp_flags |= INP_FREED; 1842 inp->inp_socket->so_pcb = NULL; 1843 inp->inp_socket = NULL; 1844 1845 RO_INVALIDATE_CACHE(&inp->inp_route); 1846 #ifdef MAC 1847 mac_inpcb_destroy(inp); 1848 #endif 1849 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1850 if (inp->inp_sp != NULL) 1851 ipsec_delete_pcbpolicy(inp); 1852 #endif 1853 #ifdef INET 1854 if (inp->inp_options) 1855 (void)m_free(inp->inp_options); 1856 DEBUG_POISON_POINTER(inp->inp_options); 1857 imo = inp->inp_moptions; 1858 DEBUG_POISON_POINTER(inp->inp_moptions); 1859 #endif 1860 #ifdef INET6 1861 if (inp->inp_vflag & INP_IPV6PROTO) { 1862 ip6_freepcbopts(inp->in6p_outputopts); 1863 DEBUG_POISON_POINTER(inp->in6p_outputopts); 1864 im6o = inp->in6p_moptions; 1865 DEBUG_POISON_POINTER(inp->in6p_moptions); 1866 } else 1867 im6o = NULL; 1868 #endif 1869 1870 if (__predict_false(in_pcbrele_wlocked(inp) == false)) { 1871 INP_WUNLOCK(inp); 1872 } 1873 #ifdef INET6 1874 ip6_freemoptions(im6o); 1875 #endif 1876 #ifdef INET 1877 inp_freemoptions(imo); 1878 #endif 1879 } 1880 1881 /* 1882 * Different protocols initialize their inpcbs differently - giving 1883 * different name to the lock. But they all are disposed the same. 1884 */ 1885 static void 1886 inpcb_fini(void *mem, int size) 1887 { 1888 struct inpcb *inp = mem; 1889 1890 INP_LOCK_DESTROY(inp); 1891 } 1892 1893 /* 1894 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1895 * port reservation, and preventing it from being returned by inpcb lookups. 1896 * 1897 * It is used by TCP to mark an inpcb as unused and avoid future packet 1898 * delivery or event notification when a socket remains open but TCP has 1899 * closed. This might occur as a result of a shutdown()-initiated TCP close 1900 * or a RST on the wire, and allows the port binding to be reused while still 1901 * maintaining the invariant that so_pcb always points to a valid inpcb until 1902 * in_pcbdetach(). 1903 * 1904 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1905 * in_pcbpurgeif0()? 1906 */ 1907 void 1908 in_pcbdrop(struct inpcb *inp) 1909 { 1910 1911 INP_WLOCK_ASSERT(inp); 1912 1913 inp->inp_flags |= INP_DROPPED; 1914 if (inp->inp_flags & INP_INHASHLIST) 1915 in_pcbremhash(inp); 1916 } 1917 1918 #ifdef INET 1919 /* 1920 * Common routines to return the socket addresses associated with inpcbs. 1921 */ 1922 int 1923 in_getsockaddr(struct socket *so, struct sockaddr *sa) 1924 { 1925 struct inpcb *inp; 1926 1927 inp = sotoinpcb(so); 1928 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1929 1930 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 1931 .sin_len = sizeof(struct sockaddr_in), 1932 .sin_family = AF_INET, 1933 .sin_port = inp->inp_lport, 1934 .sin_addr = inp->inp_laddr, 1935 }; 1936 1937 return (0); 1938 } 1939 1940 int 1941 in_getpeeraddr(struct socket *so, struct sockaddr *sa) 1942 { 1943 struct inpcb *inp; 1944 1945 inp = sotoinpcb(so); 1946 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1947 1948 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 1949 .sin_len = sizeof(struct sockaddr_in), 1950 .sin_family = AF_INET, 1951 .sin_port = inp->inp_fport, 1952 .sin_addr = inp->inp_faddr, 1953 }; 1954 1955 return (0); 1956 } 1957 1958 static bool 1959 inp_v4_multi_match(const struct inpcb *inp, void *v __unused) 1960 { 1961 1962 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL) 1963 return (true); 1964 else 1965 return (false); 1966 } 1967 1968 void 1969 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1970 { 1971 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB, 1972 inp_v4_multi_match, NULL); 1973 struct inpcb *inp; 1974 struct in_multi *inm; 1975 struct in_mfilter *imf; 1976 struct ip_moptions *imo; 1977 1978 IN_MULTI_LOCK_ASSERT(); 1979 1980 while ((inp = inp_next(&inpi)) != NULL) { 1981 INP_WLOCK_ASSERT(inp); 1982 1983 imo = inp->inp_moptions; 1984 /* 1985 * Unselect the outgoing interface if it is being 1986 * detached. 1987 */ 1988 if (imo->imo_multicast_ifp == ifp) 1989 imo->imo_multicast_ifp = NULL; 1990 1991 /* 1992 * Drop multicast group membership if we joined 1993 * through the interface being detached. 1994 * 1995 * XXX This can all be deferred to an epoch_call 1996 */ 1997 restart: 1998 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 1999 if ((inm = imf->imf_inm) == NULL) 2000 continue; 2001 if (inm->inm_ifp != ifp) 2002 continue; 2003 ip_mfilter_remove(&imo->imo_head, imf); 2004 in_leavegroup_locked(inm, NULL); 2005 ip_mfilter_free(imf); 2006 goto restart; 2007 } 2008 } 2009 } 2010 2011 /* 2012 * Lookup a PCB based on the local address and port. Caller must hold the 2013 * hash lock. No inpcb locks or references are acquired. 2014 */ 2015 #define INP_LOOKUP_MAPPED_PCB_COST 3 2016 struct inpcb * 2017 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 2018 u_short lport, int fib, int lookupflags, struct ucred *cred) 2019 { 2020 struct inpcb *inp; 2021 #ifdef INET6 2022 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 2023 #else 2024 int matchwild = 3; 2025 #endif 2026 int wildcard; 2027 2028 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2029 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2030 KASSERT(fib == RT_ALL_FIBS || (fib >= 0 && fib < V_rt_numfibs), 2031 ("%s: invalid fib %d", __func__, fib)); 2032 2033 INP_HASH_LOCK_ASSERT(pcbinfo); 2034 2035 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 2036 struct inpcbhead *head; 2037 /* 2038 * Look for an unconnected (wildcard foreign addr) PCB that 2039 * matches the local address and port we're looking for. 2040 */ 2041 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2042 pcbinfo->ipi_hashmask)]; 2043 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2044 #ifdef INET6 2045 /* XXX inp locking */ 2046 if ((inp->inp_vflag & INP_IPV4) == 0) 2047 continue; 2048 #endif 2049 if (inp->inp_faddr.s_addr == INADDR_ANY && 2050 inp->inp_laddr.s_addr == laddr.s_addr && 2051 inp->inp_lport == lport && (fib == RT_ALL_FIBS || 2052 inp->inp_inc.inc_fibnum == fib)) { 2053 /* 2054 * Found? 2055 */ 2056 if (prison_equal_ip4(cred->cr_prison, 2057 inp->inp_cred->cr_prison)) 2058 return (inp); 2059 } 2060 } 2061 /* 2062 * Not found. 2063 */ 2064 return (NULL); 2065 } else { 2066 struct inpcbporthead *porthash; 2067 struct inpcbport *phd; 2068 struct inpcb *match = NULL; 2069 /* 2070 * Best fit PCB lookup. 2071 * 2072 * First see if this local port is in use by looking on the 2073 * port hash list. 2074 */ 2075 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 2076 pcbinfo->ipi_porthashmask)]; 2077 CK_LIST_FOREACH(phd, porthash, phd_hash) { 2078 if (phd->phd_port == lport) 2079 break; 2080 } 2081 if (phd != NULL) { 2082 /* 2083 * Port is in use by one or more PCBs. Look for best 2084 * fit. 2085 */ 2086 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 2087 wildcard = 0; 2088 if (!prison_equal_ip4(inp->inp_cred->cr_prison, 2089 cred->cr_prison)) 2090 continue; 2091 if (fib != RT_ALL_FIBS && 2092 inp->inp_inc.inc_fibnum != fib) 2093 continue; 2094 #ifdef INET6 2095 /* XXX inp locking */ 2096 if ((inp->inp_vflag & INP_IPV4) == 0) 2097 continue; 2098 /* 2099 * We never select the PCB that has 2100 * INP_IPV6 flag and is bound to :: if 2101 * we have another PCB which is bound 2102 * to 0.0.0.0. If a PCB has the 2103 * INP_IPV6 flag, then we set its cost 2104 * higher than IPv4 only PCBs. 2105 * 2106 * Note that the case only happens 2107 * when a socket is bound to ::, under 2108 * the condition that the use of the 2109 * mapped address is allowed. 2110 */ 2111 if ((inp->inp_vflag & INP_IPV6) != 0) 2112 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 2113 #endif 2114 if (inp->inp_faddr.s_addr != INADDR_ANY) 2115 wildcard++; 2116 if (inp->inp_laddr.s_addr != INADDR_ANY) { 2117 if (laddr.s_addr == INADDR_ANY) 2118 wildcard++; 2119 else if (inp->inp_laddr.s_addr != laddr.s_addr) 2120 continue; 2121 } else { 2122 if (laddr.s_addr != INADDR_ANY) 2123 wildcard++; 2124 } 2125 if (wildcard < matchwild) { 2126 match = inp; 2127 matchwild = wildcard; 2128 if (matchwild == 0) 2129 break; 2130 } 2131 } 2132 } 2133 return (match); 2134 } 2135 } 2136 #undef INP_LOOKUP_MAPPED_PCB_COST 2137 2138 static bool 2139 in_pcblookup_lb_match(const struct inpcblbgroup *grp, int domain, int fib) 2140 { 2141 return ((domain == M_NODOM || domain == grp->il_numa_domain) && 2142 (fib == RT_ALL_FIBS || fib == grp->il_fibnum)); 2143 } 2144 2145 static struct inpcb * 2146 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 2147 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr, 2148 uint16_t lport, int domain, int fib) 2149 { 2150 const struct inpcblbgrouphead *hdr; 2151 struct inpcblbgroup *grp; 2152 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild; 2153 struct inpcb *inp; 2154 u_int count; 2155 2156 INP_HASH_LOCK_ASSERT(pcbinfo); 2157 NET_EPOCH_ASSERT(); 2158 2159 hdr = &pcbinfo->ipi_lbgrouphashbase[ 2160 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 2161 2162 /* 2163 * Search for an LB group match based on the following criteria: 2164 * - prefer jailed groups to non-jailed groups 2165 * - prefer exact source address matches to wildcard matches 2166 * - prefer groups bound to the specified NUMA domain 2167 */ 2168 jail_exact = jail_wild = local_exact = local_wild = NULL; 2169 CK_LIST_FOREACH(grp, hdr, il_list) { 2170 bool injail; 2171 2172 #ifdef INET6 2173 if (!(grp->il_vflag & INP_IPV4)) 2174 continue; 2175 #endif 2176 if (grp->il_lport != lport) 2177 continue; 2178 2179 injail = prison_flag(grp->il_cred, PR_IP4) != 0; 2180 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison, 2181 laddr) != 0) 2182 continue; 2183 2184 if (grp->il_laddr.s_addr == laddr->s_addr) { 2185 if (injail) { 2186 jail_exact = grp; 2187 if (in_pcblookup_lb_match(grp, domain, fib)) 2188 /* This is a perfect match. */ 2189 goto out; 2190 } else if (local_exact == NULL || 2191 in_pcblookup_lb_match(grp, domain, fib)) { 2192 local_exact = grp; 2193 } 2194 } else if (grp->il_laddr.s_addr == INADDR_ANY) { 2195 if (injail) { 2196 if (jail_wild == NULL || 2197 in_pcblookup_lb_match(grp, domain, fib)) 2198 jail_wild = grp; 2199 } else if (local_wild == NULL || 2200 in_pcblookup_lb_match(grp, domain, fib)) { 2201 local_wild = grp; 2202 } 2203 } 2204 } 2205 2206 if (jail_exact != NULL) 2207 grp = jail_exact; 2208 else if (jail_wild != NULL) 2209 grp = jail_wild; 2210 else if (local_exact != NULL) 2211 grp = local_exact; 2212 else 2213 grp = local_wild; 2214 if (grp == NULL) 2215 return (NULL); 2216 2217 out: 2218 /* 2219 * Synchronize with in_pcblbgroup_insert(). 2220 */ 2221 count = atomic_load_acq_int(&grp->il_inpcnt); 2222 if (count == 0) 2223 return (NULL); 2224 inp = grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) % count]; 2225 KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); 2226 return (inp); 2227 } 2228 2229 static bool 2230 in_pcblookup_exact_match(const struct inpcb *inp, struct in_addr faddr, 2231 u_short fport, struct in_addr laddr, u_short lport) 2232 { 2233 #ifdef INET6 2234 /* XXX inp locking */ 2235 if ((inp->inp_vflag & INP_IPV4) == 0) 2236 return (false); 2237 #endif 2238 if (inp->inp_faddr.s_addr == faddr.s_addr && 2239 inp->inp_laddr.s_addr == laddr.s_addr && 2240 inp->inp_fport == fport && 2241 inp->inp_lport == lport) 2242 return (true); 2243 return (false); 2244 } 2245 2246 static struct inpcb * 2247 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2248 u_short fport, struct in_addr laddr, u_short lport) 2249 { 2250 struct inpcbhead *head; 2251 struct inpcb *inp; 2252 2253 INP_HASH_LOCK_ASSERT(pcbinfo); 2254 2255 head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport, 2256 pcbinfo->ipi_hashmask)]; 2257 CK_LIST_FOREACH(inp, head, inp_hash_exact) { 2258 if (in_pcblookup_exact_match(inp, faddr, fport, laddr, lport)) 2259 return (inp); 2260 } 2261 return (NULL); 2262 } 2263 2264 typedef enum { 2265 INPLOOKUP_MATCH_NONE = 0, 2266 INPLOOKUP_MATCH_WILD = 1, 2267 INPLOOKUP_MATCH_LADDR = 2, 2268 } inp_lookup_match_t; 2269 2270 static inp_lookup_match_t 2271 in_pcblookup_wild_match(const struct inpcb *inp, struct in_addr laddr, 2272 u_short lport, int fib) 2273 { 2274 #ifdef INET6 2275 /* XXX inp locking */ 2276 if ((inp->inp_vflag & INP_IPV4) == 0) 2277 return (INPLOOKUP_MATCH_NONE); 2278 #endif 2279 if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) 2280 return (INPLOOKUP_MATCH_NONE); 2281 if (fib != RT_ALL_FIBS && inp->inp_inc.inc_fibnum != fib) 2282 return (INPLOOKUP_MATCH_NONE); 2283 if (inp->inp_laddr.s_addr == INADDR_ANY) 2284 return (INPLOOKUP_MATCH_WILD); 2285 if (inp->inp_laddr.s_addr == laddr.s_addr) 2286 return (INPLOOKUP_MATCH_LADDR); 2287 return (INPLOOKUP_MATCH_NONE); 2288 } 2289 2290 #define INP_LOOKUP_AGAIN ((struct inpcb *)(uintptr_t)-1) 2291 2292 static struct inpcb * 2293 in_pcblookup_hash_wild_smr(struct inpcbinfo *pcbinfo, struct in_addr laddr, 2294 u_short lport, int fib, const inp_lookup_t lockflags) 2295 { 2296 struct inpcbhead *head; 2297 struct inpcb *inp; 2298 2299 KASSERT(SMR_ENTERED(pcbinfo->ipi_smr), 2300 ("%s: not in SMR read section", __func__)); 2301 2302 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2303 pcbinfo->ipi_hashmask)]; 2304 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2305 inp_lookup_match_t match; 2306 2307 match = in_pcblookup_wild_match(inp, laddr, lport, fib); 2308 if (match == INPLOOKUP_MATCH_NONE) 2309 continue; 2310 2311 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2312 match = in_pcblookup_wild_match(inp, laddr, lport, fib); 2313 if (match != INPLOOKUP_MATCH_NONE && 2314 prison_check_ip4_locked(inp->inp_cred->cr_prison, 2315 &laddr) == 0) 2316 return (inp); 2317 inp_unlock(inp, lockflags); 2318 } 2319 2320 /* 2321 * The matching socket disappeared out from under us. Fall back 2322 * to a serialized lookup. 2323 */ 2324 return (INP_LOOKUP_AGAIN); 2325 } 2326 return (NULL); 2327 } 2328 2329 static struct inpcb * 2330 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr laddr, 2331 u_short lport, int fib) 2332 { 2333 struct inpcbhead *head; 2334 struct inpcb *inp, *local_wild, *local_exact, *jail_wild; 2335 #ifdef INET6 2336 struct inpcb *local_wild_mapped; 2337 #endif 2338 2339 INP_HASH_LOCK_ASSERT(pcbinfo); 2340 2341 /* 2342 * Order of socket selection - we always prefer jails. 2343 * 1. jailed, non-wild. 2344 * 2. jailed, wild. 2345 * 3. non-jailed, non-wild. 2346 * 4. non-jailed, wild. 2347 */ 2348 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2349 pcbinfo->ipi_hashmask)]; 2350 local_wild = local_exact = jail_wild = NULL; 2351 #ifdef INET6 2352 local_wild_mapped = NULL; 2353 #endif 2354 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2355 inp_lookup_match_t match; 2356 bool injail; 2357 2358 match = in_pcblookup_wild_match(inp, laddr, lport, fib); 2359 if (match == INPLOOKUP_MATCH_NONE) 2360 continue; 2361 2362 injail = prison_flag(inp->inp_cred, PR_IP4) != 0; 2363 if (injail) { 2364 if (prison_check_ip4_locked(inp->inp_cred->cr_prison, 2365 &laddr) != 0) 2366 continue; 2367 } else { 2368 if (local_exact != NULL) 2369 continue; 2370 } 2371 2372 if (match == INPLOOKUP_MATCH_LADDR) { 2373 if (injail) 2374 return (inp); 2375 local_exact = inp; 2376 } else { 2377 #ifdef INET6 2378 /* XXX inp locking, NULL check */ 2379 if (inp->inp_vflag & INP_IPV6PROTO) 2380 local_wild_mapped = inp; 2381 else 2382 #endif 2383 if (injail) 2384 jail_wild = inp; 2385 else 2386 local_wild = inp; 2387 } 2388 } 2389 if (jail_wild != NULL) 2390 return (jail_wild); 2391 if (local_exact != NULL) 2392 return (local_exact); 2393 if (local_wild != NULL) 2394 return (local_wild); 2395 #ifdef INET6 2396 if (local_wild_mapped != NULL) 2397 return (local_wild_mapped); 2398 #endif 2399 return (NULL); 2400 } 2401 2402 /* 2403 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2404 * that the caller has either locked the hash list, which usually happens 2405 * for bind(2) operations, or is in SMR section, which happens when sorting 2406 * out incoming packets. 2407 */ 2408 static struct inpcb * 2409 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2410 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2411 uint8_t numa_domain, int fib) 2412 { 2413 struct inpcb *inp; 2414 const u_short fport = fport_arg, lport = lport_arg; 2415 2416 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD | INPLOOKUP_FIB)) == 0, 2417 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2418 KASSERT(faddr.s_addr != INADDR_ANY, 2419 ("%s: invalid foreign address", __func__)); 2420 KASSERT(laddr.s_addr != INADDR_ANY, 2421 ("%s: invalid local address", __func__)); 2422 INP_HASH_WLOCK_ASSERT(pcbinfo); 2423 2424 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2425 if (inp != NULL) 2426 return (inp); 2427 2428 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2429 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2430 &laddr, lport, numa_domain, fib); 2431 if (inp == NULL) { 2432 inp = in_pcblookup_hash_wild_locked(pcbinfo, laddr, 2433 lport, fib); 2434 } 2435 } 2436 2437 return (inp); 2438 } 2439 2440 static struct inpcb * 2441 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2442 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2443 uint8_t numa_domain, int fib) 2444 { 2445 struct inpcb *inp; 2446 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2447 2448 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2449 ("%s: LOCKPCB not set", __func__)); 2450 2451 INP_HASH_WLOCK(pcbinfo); 2452 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2453 lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain, fib); 2454 if (inp != NULL && !inp_trylock(inp, lockflags)) { 2455 in_pcbref(inp); 2456 INP_HASH_WUNLOCK(pcbinfo); 2457 inp_lock(inp, lockflags); 2458 if (in_pcbrele(inp, lockflags)) 2459 /* XXX-MJ or retry until we get a negative match? */ 2460 inp = NULL; 2461 } else { 2462 INP_HASH_WUNLOCK(pcbinfo); 2463 } 2464 return (inp); 2465 } 2466 2467 static struct inpcb * 2468 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2469 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2470 uint8_t numa_domain, int fib) 2471 { 2472 struct inpcb *inp; 2473 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2474 const u_short fport = fport_arg, lport = lport_arg; 2475 2476 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2477 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2478 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2479 ("%s: LOCKPCB not set", __func__)); 2480 2481 smr_enter(pcbinfo->ipi_smr); 2482 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2483 if (inp != NULL) { 2484 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2485 /* 2486 * Revalidate the 4-tuple, the socket could have been 2487 * disconnected. 2488 */ 2489 if (__predict_true(in_pcblookup_exact_match(inp, 2490 faddr, fport, laddr, lport))) 2491 return (inp); 2492 inp_unlock(inp, lockflags); 2493 } 2494 2495 /* 2496 * We failed to lock the inpcb, or its connection state changed 2497 * out from under us. Fall back to a precise search. 2498 */ 2499 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2500 lookupflags, numa_domain, fib)); 2501 } 2502 2503 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2504 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2505 &laddr, lport, numa_domain, fib); 2506 if (inp != NULL) { 2507 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2508 if (__predict_true(in_pcblookup_wild_match(inp, 2509 laddr, lport, fib) != INPLOOKUP_MATCH_NONE)) 2510 return (inp); 2511 inp_unlock(inp, lockflags); 2512 } 2513 inp = INP_LOOKUP_AGAIN; 2514 } else { 2515 inp = in_pcblookup_hash_wild_smr(pcbinfo, laddr, lport, 2516 fib, lockflags); 2517 } 2518 if (inp == INP_LOOKUP_AGAIN) { 2519 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, 2520 lport, lookupflags, numa_domain, fib)); 2521 } 2522 } 2523 2524 if (inp == NULL) 2525 smr_exit(pcbinfo->ipi_smr); 2526 2527 return (inp); 2528 } 2529 2530 /* 2531 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2532 * from which a pre-calculated hash value may be extracted. 2533 */ 2534 struct inpcb * 2535 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2536 struct in_addr laddr, u_int lport, int lookupflags, 2537 struct ifnet *ifp) 2538 { 2539 int fib; 2540 2541 fib = (lookupflags & INPLOOKUP_FIB) ? if_getfib(ifp) : RT_ALL_FIBS; 2542 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2543 lookupflags, M_NODOM, fib)); 2544 } 2545 2546 struct inpcb * 2547 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2548 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2549 struct ifnet *ifp __unused, struct mbuf *m) 2550 { 2551 int fib; 2552 2553 M_ASSERTPKTHDR(m); 2554 fib = (lookupflags & INPLOOKUP_FIB) ? M_GETFIB(m) : RT_ALL_FIBS; 2555 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2556 lookupflags, m->m_pkthdr.numa_domain, fib)); 2557 } 2558 #endif /* INET */ 2559 2560 static bool 2561 in_pcbjailed(const struct inpcb *inp, unsigned int flag) 2562 { 2563 return (prison_flag(inp->inp_cred, flag) != 0); 2564 } 2565 2566 /* 2567 * Insert the PCB into a hash chain using ordering rules which ensure that 2568 * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first. 2569 * 2570 * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs 2571 * with exact local addresses ahead of wildcard PCBs. Unbound v4-mapped v6 PCBs 2572 * always appear last no matter whether they are jailed. 2573 */ 2574 static void 2575 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2576 { 2577 struct inpcb *last; 2578 bool bound, injail; 2579 2580 INP_LOCK_ASSERT(inp); 2581 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2582 2583 last = NULL; 2584 bound = inp->inp_laddr.s_addr != INADDR_ANY; 2585 if (!bound && (inp->inp_vflag & INP_IPV6PROTO) != 0) { 2586 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2587 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2588 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2589 return; 2590 } 2591 } 2592 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2593 return; 2594 } 2595 2596 injail = in_pcbjailed(inp, PR_IP4); 2597 if (!injail) { 2598 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2599 if (!in_pcbjailed(last, PR_IP4)) 2600 break; 2601 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2602 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2603 return; 2604 } 2605 } 2606 } else if (!CK_LIST_EMPTY(pcbhash) && 2607 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) { 2608 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2609 return; 2610 } 2611 if (!bound) { 2612 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2613 if (last->inp_laddr.s_addr == INADDR_ANY) 2614 break; 2615 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2616 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2617 return; 2618 } 2619 } 2620 } 2621 if (last == NULL) 2622 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2623 else 2624 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2625 } 2626 2627 #ifdef INET6 2628 /* 2629 * See the comment above _in_pcbinshash_wild(). 2630 */ 2631 static void 2632 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2633 { 2634 struct inpcb *last; 2635 bool bound, injail; 2636 2637 INP_LOCK_ASSERT(inp); 2638 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2639 2640 last = NULL; 2641 bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr); 2642 injail = in_pcbjailed(inp, PR_IP6); 2643 if (!injail) { 2644 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2645 if (!in_pcbjailed(last, PR_IP6)) 2646 break; 2647 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2648 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2649 return; 2650 } 2651 } 2652 } else if (!CK_LIST_EMPTY(pcbhash) && 2653 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) { 2654 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2655 return; 2656 } 2657 if (!bound) { 2658 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2659 if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr)) 2660 break; 2661 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2662 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2663 return; 2664 } 2665 } 2666 } 2667 if (last == NULL) 2668 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2669 else 2670 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2671 } 2672 #endif 2673 2674 /* 2675 * Insert PCB onto various hash lists. 2676 */ 2677 int 2678 in_pcbinshash(struct inpcb *inp) 2679 { 2680 struct inpcbhead *pcbhash; 2681 struct inpcbporthead *pcbporthash; 2682 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2683 struct inpcbport *phd; 2684 uint32_t hash; 2685 bool connected; 2686 2687 INP_WLOCK_ASSERT(inp); 2688 INP_HASH_WLOCK_ASSERT(pcbinfo); 2689 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2690 ("in_pcbinshash: INP_INHASHLIST")); 2691 2692 #ifdef INET6 2693 if (inp->inp_vflag & INP_IPV6) { 2694 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2695 inp->inp_fport, pcbinfo->ipi_hashmask); 2696 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2697 } else 2698 #endif 2699 { 2700 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2701 inp->inp_fport, pcbinfo->ipi_hashmask); 2702 connected = !in_nullhost(inp->inp_faddr); 2703 } 2704 2705 if (connected) 2706 pcbhash = &pcbinfo->ipi_hash_exact[hash]; 2707 else 2708 pcbhash = &pcbinfo->ipi_hash_wild[hash]; 2709 2710 pcbporthash = &pcbinfo->ipi_porthashbase[ 2711 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2712 2713 /* 2714 * Add entry to load balance group. 2715 * Only do this if SO_REUSEPORT_LB is set. 2716 */ 2717 if ((inp->inp_socket->so_options & SO_REUSEPORT_LB) != 0) { 2718 int error = in_pcbinslbgrouphash(inp, M_NODOM); 2719 if (error != 0) 2720 return (error); 2721 } 2722 2723 /* 2724 * Go through port list and look for a head for this lport. 2725 */ 2726 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2727 if (phd->phd_port == inp->inp_lport) 2728 break; 2729 } 2730 2731 /* 2732 * If none exists, malloc one and tack it on. 2733 */ 2734 if (phd == NULL) { 2735 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT); 2736 if (phd == NULL) { 2737 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2738 in_pcbremlbgrouphash(inp); 2739 return (ENOMEM); 2740 } 2741 phd->phd_port = inp->inp_lport; 2742 CK_LIST_INIT(&phd->phd_pcblist); 2743 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2744 } 2745 inp->inp_phd = phd; 2746 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2747 2748 /* 2749 * The PCB may have been disconnected in the past. Before we can safely 2750 * make it visible in the hash table, we must wait for all readers which 2751 * may be traversing this PCB to finish. 2752 */ 2753 if (inp->inp_smr != SMR_SEQ_INVALID) { 2754 smr_wait(pcbinfo->ipi_smr, inp->inp_smr); 2755 inp->inp_smr = SMR_SEQ_INVALID; 2756 } 2757 2758 if (connected) 2759 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact); 2760 else { 2761 #ifdef INET6 2762 if ((inp->inp_vflag & INP_IPV6) != 0) 2763 _in6_pcbinshash_wild(pcbhash, inp); 2764 else 2765 #endif 2766 _in_pcbinshash_wild(pcbhash, inp); 2767 } 2768 inp->inp_flags |= INP_INHASHLIST; 2769 2770 return (0); 2771 } 2772 2773 void 2774 in_pcbremhash_locked(struct inpcb *inp) 2775 { 2776 struct inpcbport *phd = inp->inp_phd; 2777 2778 INP_WLOCK_ASSERT(inp); 2779 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2780 MPASS(inp->inp_flags & INP_INHASHLIST); 2781 2782 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2783 in_pcbremlbgrouphash(inp); 2784 #ifdef INET6 2785 if (inp->inp_vflag & INP_IPV6) { 2786 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) 2787 CK_LIST_REMOVE(inp, inp_hash_wild); 2788 else 2789 CK_LIST_REMOVE(inp, inp_hash_exact); 2790 } else 2791 #endif 2792 { 2793 if (in_nullhost(inp->inp_faddr)) 2794 CK_LIST_REMOVE(inp, inp_hash_wild); 2795 else 2796 CK_LIST_REMOVE(inp, inp_hash_exact); 2797 } 2798 CK_LIST_REMOVE(inp, inp_portlist); 2799 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2800 CK_LIST_REMOVE(phd, phd_hash); 2801 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd); 2802 } 2803 inp->inp_flags &= ~INP_INHASHLIST; 2804 } 2805 2806 static void 2807 in_pcbremhash(struct inpcb *inp) 2808 { 2809 INP_HASH_WLOCK(inp->inp_pcbinfo); 2810 in_pcbremhash_locked(inp); 2811 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 2812 } 2813 2814 /* 2815 * Move PCB to the proper hash bucket when { faddr, fport } have been 2816 * changed. NOTE: This does not handle the case of the lport changing (the 2817 * hashed port list would have to be updated as well), so the lport must 2818 * not change after in_pcbinshash() has been called. 2819 */ 2820 void 2821 in_pcbrehash(struct inpcb *inp) 2822 { 2823 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2824 struct inpcbhead *head; 2825 uint32_t hash; 2826 bool connected; 2827 2828 INP_WLOCK_ASSERT(inp); 2829 INP_HASH_WLOCK_ASSERT(pcbinfo); 2830 KASSERT(inp->inp_flags & INP_INHASHLIST, 2831 ("%s: !INP_INHASHLIST", __func__)); 2832 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 2833 ("%s: inp was disconnected", __func__)); 2834 2835 #ifdef INET6 2836 if (inp->inp_vflag & INP_IPV6) { 2837 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2838 inp->inp_fport, pcbinfo->ipi_hashmask); 2839 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2840 } else 2841 #endif 2842 { 2843 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2844 inp->inp_fport, pcbinfo->ipi_hashmask); 2845 connected = !in_nullhost(inp->inp_faddr); 2846 } 2847 2848 /* 2849 * When rehashing, the caller must ensure that either the new or the old 2850 * foreign address was unspecified. 2851 */ 2852 if (connected) 2853 CK_LIST_REMOVE(inp, inp_hash_wild); 2854 else 2855 CK_LIST_REMOVE(inp, inp_hash_exact); 2856 2857 if (connected) { 2858 head = &pcbinfo->ipi_hash_exact[hash]; 2859 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact); 2860 } else { 2861 head = &pcbinfo->ipi_hash_wild[hash]; 2862 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild); 2863 } 2864 } 2865 2866 /* 2867 * Check for alternatives when higher level complains 2868 * about service problems. For now, invalidate cached 2869 * routing information. If the route was created dynamically 2870 * (by a redirect), time to try a default gateway again. 2871 */ 2872 void 2873 in_losing(struct inpcb *inp) 2874 { 2875 2876 RO_INVALIDATE_CACHE(&inp->inp_route); 2877 return; 2878 } 2879 2880 /* 2881 * A set label operation has occurred at the socket layer, propagate the 2882 * label change into the in_pcb for the socket. 2883 */ 2884 void 2885 in_pcbsosetlabel(struct socket *so) 2886 { 2887 #ifdef MAC 2888 struct inpcb *inp; 2889 2890 inp = sotoinpcb(so); 2891 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2892 2893 INP_WLOCK(inp); 2894 SOCK_LOCK(so); 2895 mac_inpcb_sosetlabel(so, inp); 2896 SOCK_UNLOCK(so); 2897 INP_WUNLOCK(inp); 2898 #endif 2899 } 2900 2901 void 2902 inp_wlock(struct inpcb *inp) 2903 { 2904 2905 INP_WLOCK(inp); 2906 } 2907 2908 void 2909 inp_wunlock(struct inpcb *inp) 2910 { 2911 2912 INP_WUNLOCK(inp); 2913 } 2914 2915 void 2916 inp_rlock(struct inpcb *inp) 2917 { 2918 2919 INP_RLOCK(inp); 2920 } 2921 2922 void 2923 inp_runlock(struct inpcb *inp) 2924 { 2925 2926 INP_RUNLOCK(inp); 2927 } 2928 2929 #ifdef INVARIANT_SUPPORT 2930 void 2931 inp_lock_assert(struct inpcb *inp) 2932 { 2933 2934 INP_WLOCK_ASSERT(inp); 2935 } 2936 2937 void 2938 inp_unlock_assert(struct inpcb *inp) 2939 { 2940 2941 INP_UNLOCK_ASSERT(inp); 2942 } 2943 #endif 2944 2945 void 2946 inp_apply_all(struct inpcbinfo *pcbinfo, 2947 void (*func)(struct inpcb *, void *), void *arg) 2948 { 2949 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2950 INPLOOKUP_WLOCKPCB); 2951 struct inpcb *inp; 2952 2953 while ((inp = inp_next(&inpi)) != NULL) 2954 func(inp, arg); 2955 } 2956 2957 struct socket * 2958 inp_inpcbtosocket(struct inpcb *inp) 2959 { 2960 2961 INP_WLOCK_ASSERT(inp); 2962 return (inp->inp_socket); 2963 } 2964 2965 void 2966 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2967 uint32_t *faddr, uint16_t *fp) 2968 { 2969 2970 INP_LOCK_ASSERT(inp); 2971 *laddr = inp->inp_laddr.s_addr; 2972 *faddr = inp->inp_faddr.s_addr; 2973 *lp = inp->inp_lport; 2974 *fp = inp->inp_fport; 2975 } 2976 2977 /* 2978 * Create an external-format (``xinpcb'') structure using the information in 2979 * the kernel-format in_pcb structure pointed to by inp. This is done to 2980 * reduce the spew of irrelevant information over this interface, to isolate 2981 * user code from changes in the kernel structure, and potentially to provide 2982 * information-hiding if we decide that some of this information should be 2983 * hidden from users. 2984 */ 2985 void 2986 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2987 { 2988 2989 bzero(xi, sizeof(*xi)); 2990 xi->xi_len = sizeof(struct xinpcb); 2991 if (inp->inp_socket) 2992 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2993 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2994 xi->inp_gencnt = inp->inp_gencnt; 2995 xi->inp_flow = inp->inp_flow; 2996 xi->inp_flowid = inp->inp_flowid; 2997 xi->inp_flowtype = inp->inp_flowtype; 2998 xi->inp_flags = inp->inp_flags; 2999 xi->inp_flags2 = inp->inp_flags2; 3000 xi->in6p_cksum = inp->in6p_cksum; 3001 xi->in6p_hops = inp->in6p_hops; 3002 xi->inp_ip_tos = inp->inp_ip_tos; 3003 xi->inp_vflag = inp->inp_vflag; 3004 xi->inp_ip_ttl = inp->inp_ip_ttl; 3005 xi->inp_ip_p = inp->inp_ip_p; 3006 xi->inp_ip_minttl = inp->inp_ip_minttl; 3007 } 3008 3009 int 3010 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo, 3011 int (*ctloutput_set)(struct inpcb *, struct sockopt *)) 3012 { 3013 struct sockopt sopt; 3014 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 3015 INPLOOKUP_WLOCKPCB); 3016 struct inpcb *inp; 3017 struct sockopt_parameters *params; 3018 struct socket *so; 3019 int error; 3020 char buf[1024]; 3021 3022 if (req->oldptr != NULL || req->oldlen != 0) 3023 return (EINVAL); 3024 if (req->newptr == NULL) 3025 return (EPERM); 3026 if (req->newlen > sizeof(buf)) 3027 return (ENOMEM); 3028 error = SYSCTL_IN(req, buf, req->newlen); 3029 if (error != 0) 3030 return (error); 3031 if (req->newlen < sizeof(struct sockopt_parameters)) 3032 return (EINVAL); 3033 params = (struct sockopt_parameters *)buf; 3034 sopt.sopt_level = params->sop_level; 3035 sopt.sopt_name = params->sop_optname; 3036 sopt.sopt_dir = SOPT_SET; 3037 sopt.sopt_val = params->sop_optval; 3038 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters); 3039 sopt.sopt_td = NULL; 3040 #ifdef INET6 3041 if (params->sop_inc.inc_flags & INC_ISIPV6) { 3042 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr)) 3043 params->sop_inc.inc6_laddr.s6_addr16[1] = 3044 htons(params->sop_inc.inc6_zoneid & 0xffff); 3045 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr)) 3046 params->sop_inc.inc6_faddr.s6_addr16[1] = 3047 htons(params->sop_inc.inc6_zoneid & 0xffff); 3048 } 3049 #endif 3050 if (params->sop_inc.inc_lport != htons(0) && 3051 params->sop_inc.inc_fport != htons(0)) { 3052 #ifdef INET6 3053 if (params->sop_inc.inc_flags & INC_ISIPV6) 3054 inpi.hash = INP6_PCBHASH( 3055 ¶ms->sop_inc.inc6_faddr, 3056 params->sop_inc.inc_lport, 3057 params->sop_inc.inc_fport, 3058 pcbinfo->ipi_hashmask); 3059 else 3060 #endif 3061 inpi.hash = INP_PCBHASH( 3062 ¶ms->sop_inc.inc_faddr, 3063 params->sop_inc.inc_lport, 3064 params->sop_inc.inc_fport, 3065 pcbinfo->ipi_hashmask); 3066 } 3067 while ((inp = inp_next(&inpi)) != NULL) 3068 if (inp->inp_gencnt == params->sop_id) { 3069 if (inp->inp_flags & INP_DROPPED) { 3070 INP_WUNLOCK(inp); 3071 return (ECONNRESET); 3072 } 3073 so = inp->inp_socket; 3074 KASSERT(so != NULL, ("inp_socket == NULL")); 3075 soref(so); 3076 if (params->sop_level == SOL_SOCKET) { 3077 INP_WUNLOCK(inp); 3078 error = sosetopt(so, &sopt); 3079 } else 3080 error = (*ctloutput_set)(inp, &sopt); 3081 sorele(so); 3082 break; 3083 } 3084 if (inp == NULL) 3085 error = ESRCH; 3086 return (error); 3087 } 3088 3089 #ifdef DDB 3090 static void 3091 db_print_indent(int indent) 3092 { 3093 int i; 3094 3095 for (i = 0; i < indent; i++) 3096 db_printf(" "); 3097 } 3098 3099 static void 3100 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 3101 { 3102 char faddr_str[48], laddr_str[48]; 3103 3104 db_print_indent(indent); 3105 db_printf("%s at %p\n", name, inc); 3106 3107 indent += 2; 3108 3109 #ifdef INET6 3110 if (inc->inc_flags & INC_ISIPV6) { 3111 /* IPv6. */ 3112 ip6_sprintf(laddr_str, &inc->inc6_laddr); 3113 ip6_sprintf(faddr_str, &inc->inc6_faddr); 3114 } else 3115 #endif 3116 { 3117 /* IPv4. */ 3118 inet_ntoa_r(inc->inc_laddr, laddr_str); 3119 inet_ntoa_r(inc->inc_faddr, faddr_str); 3120 } 3121 db_print_indent(indent); 3122 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3123 ntohs(inc->inc_lport)); 3124 db_print_indent(indent); 3125 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3126 ntohs(inc->inc_fport)); 3127 } 3128 3129 static void 3130 db_print_inpflags(int inp_flags) 3131 { 3132 int comma; 3133 3134 comma = 0; 3135 if (inp_flags & INP_RECVOPTS) { 3136 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3137 comma = 1; 3138 } 3139 if (inp_flags & INP_RECVRETOPTS) { 3140 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3141 comma = 1; 3142 } 3143 if (inp_flags & INP_RECVDSTADDR) { 3144 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3145 comma = 1; 3146 } 3147 if (inp_flags & INP_ORIGDSTADDR) { 3148 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3149 comma = 1; 3150 } 3151 if (inp_flags & INP_HDRINCL) { 3152 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3153 comma = 1; 3154 } 3155 if (inp_flags & INP_HIGHPORT) { 3156 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3157 comma = 1; 3158 } 3159 if (inp_flags & INP_LOWPORT) { 3160 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3161 comma = 1; 3162 } 3163 if (inp_flags & INP_ANONPORT) { 3164 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3165 comma = 1; 3166 } 3167 if (inp_flags & INP_RECVIF) { 3168 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3169 comma = 1; 3170 } 3171 if (inp_flags & INP_MTUDISC) { 3172 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3173 comma = 1; 3174 } 3175 if (inp_flags & INP_RECVTTL) { 3176 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3177 comma = 1; 3178 } 3179 if (inp_flags & INP_DONTFRAG) { 3180 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3181 comma = 1; 3182 } 3183 if (inp_flags & INP_RECVTOS) { 3184 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3185 comma = 1; 3186 } 3187 if (inp_flags & IN6P_IPV6_V6ONLY) { 3188 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3189 comma = 1; 3190 } 3191 if (inp_flags & IN6P_PKTINFO) { 3192 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3193 comma = 1; 3194 } 3195 if (inp_flags & IN6P_HOPLIMIT) { 3196 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3197 comma = 1; 3198 } 3199 if (inp_flags & IN6P_HOPOPTS) { 3200 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3201 comma = 1; 3202 } 3203 if (inp_flags & IN6P_DSTOPTS) { 3204 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3205 comma = 1; 3206 } 3207 if (inp_flags & IN6P_RTHDR) { 3208 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3209 comma = 1; 3210 } 3211 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3212 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3213 comma = 1; 3214 } 3215 if (inp_flags & IN6P_TCLASS) { 3216 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3217 comma = 1; 3218 } 3219 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3220 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3221 comma = 1; 3222 } 3223 if (inp_flags & INP_ONESBCAST) { 3224 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3225 comma = 1; 3226 } 3227 if (inp_flags & INP_DROPPED) { 3228 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3229 comma = 1; 3230 } 3231 if (inp_flags & INP_SOCKREF) { 3232 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3233 comma = 1; 3234 } 3235 if (inp_flags & IN6P_RFC2292) { 3236 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3237 comma = 1; 3238 } 3239 if (inp_flags & IN6P_MTU) { 3240 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3241 comma = 1; 3242 } 3243 } 3244 3245 static void 3246 db_print_inpvflag(u_char inp_vflag) 3247 { 3248 int comma; 3249 3250 comma = 0; 3251 if (inp_vflag & INP_IPV4) { 3252 db_printf("%sINP_IPV4", comma ? ", " : ""); 3253 comma = 1; 3254 } 3255 if (inp_vflag & INP_IPV6) { 3256 db_printf("%sINP_IPV6", comma ? ", " : ""); 3257 comma = 1; 3258 } 3259 if (inp_vflag & INP_IPV6PROTO) { 3260 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3261 comma = 1; 3262 } 3263 } 3264 3265 static void 3266 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3267 { 3268 3269 db_print_indent(indent); 3270 db_printf("%s at %p\n", name, inp); 3271 3272 indent += 2; 3273 3274 db_print_indent(indent); 3275 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3276 3277 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3278 3279 db_print_indent(indent); 3280 db_printf("inp_label: %p inp_flags: 0x%x (", 3281 inp->inp_label, inp->inp_flags); 3282 db_print_inpflags(inp->inp_flags); 3283 db_printf(")\n"); 3284 3285 db_print_indent(indent); 3286 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3287 inp->inp_vflag); 3288 db_print_inpvflag(inp->inp_vflag); 3289 db_printf(")\n"); 3290 3291 db_print_indent(indent); 3292 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3293 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3294 3295 db_print_indent(indent); 3296 #ifdef INET6 3297 if (inp->inp_vflag & INP_IPV6) { 3298 db_printf("in6p_options: %p in6p_outputopts: %p " 3299 "in6p_moptions: %p\n", inp->in6p_options, 3300 inp->in6p_outputopts, inp->in6p_moptions); 3301 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3302 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3303 inp->in6p_hops); 3304 } else 3305 #endif 3306 { 3307 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3308 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3309 inp->inp_options, inp->inp_moptions); 3310 } 3311 3312 db_print_indent(indent); 3313 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3314 (uintmax_t)inp->inp_gencnt); 3315 } 3316 3317 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3318 { 3319 struct inpcb *inp; 3320 3321 if (!have_addr) { 3322 db_printf("usage: show inpcb <addr>\n"); 3323 return; 3324 } 3325 inp = (struct inpcb *)addr; 3326 3327 db_print_inpcb(inp, "inpcb", 0); 3328 } 3329 #endif /* DDB */ 3330 3331 #ifdef RATELIMIT 3332 /* 3333 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3334 * if any. 3335 */ 3336 int 3337 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3338 { 3339 union if_snd_tag_modify_params params = { 3340 .rate_limit.max_rate = max_pacing_rate, 3341 .rate_limit.flags = M_NOWAIT, 3342 }; 3343 struct m_snd_tag *mst; 3344 int error; 3345 3346 mst = inp->inp_snd_tag; 3347 if (mst == NULL) 3348 return (EINVAL); 3349 3350 if (mst->sw->snd_tag_modify == NULL) { 3351 error = EOPNOTSUPP; 3352 } else { 3353 error = mst->sw->snd_tag_modify(mst, ¶ms); 3354 } 3355 return (error); 3356 } 3357 3358 /* 3359 * Query existing TX rate limit based on the existing 3360 * "inp->inp_snd_tag", if any. 3361 */ 3362 int 3363 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3364 { 3365 union if_snd_tag_query_params params = { }; 3366 struct m_snd_tag *mst; 3367 int error; 3368 3369 mst = inp->inp_snd_tag; 3370 if (mst == NULL) 3371 return (EINVAL); 3372 3373 if (mst->sw->snd_tag_query == NULL) { 3374 error = EOPNOTSUPP; 3375 } else { 3376 error = mst->sw->snd_tag_query(mst, ¶ms); 3377 if (error == 0 && p_max_pacing_rate != NULL) 3378 *p_max_pacing_rate = params.rate_limit.max_rate; 3379 } 3380 return (error); 3381 } 3382 3383 /* 3384 * Query existing TX queue level based on the existing 3385 * "inp->inp_snd_tag", if any. 3386 */ 3387 int 3388 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3389 { 3390 union if_snd_tag_query_params params = { }; 3391 struct m_snd_tag *mst; 3392 int error; 3393 3394 mst = inp->inp_snd_tag; 3395 if (mst == NULL) 3396 return (EINVAL); 3397 3398 if (mst->sw->snd_tag_query == NULL) 3399 return (EOPNOTSUPP); 3400 3401 error = mst->sw->snd_tag_query(mst, ¶ms); 3402 if (error == 0 && p_txqueue_level != NULL) 3403 *p_txqueue_level = params.rate_limit.queue_level; 3404 return (error); 3405 } 3406 3407 /* 3408 * Allocate a new TX rate limit send tag from the network interface 3409 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3410 */ 3411 int 3412 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3413 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3414 3415 { 3416 union if_snd_tag_alloc_params params = { 3417 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3418 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3419 .rate_limit.hdr.flowid = flowid, 3420 .rate_limit.hdr.flowtype = flowtype, 3421 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3422 .rate_limit.max_rate = max_pacing_rate, 3423 .rate_limit.flags = M_NOWAIT, 3424 }; 3425 int error; 3426 3427 INP_WLOCK_ASSERT(inp); 3428 3429 /* 3430 * If there is already a send tag, or the INP is being torn 3431 * down, allocating a new send tag is not allowed. Else send 3432 * tags may leak. 3433 */ 3434 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0) 3435 return (EINVAL); 3436 3437 error = m_snd_tag_alloc(ifp, ¶ms, st); 3438 #ifdef INET 3439 if (error == 0) { 3440 counter_u64_add(rate_limit_set_ok, 1); 3441 counter_u64_add(rate_limit_active, 1); 3442 } else if (error != EOPNOTSUPP) 3443 counter_u64_add(rate_limit_alloc_fail, 1); 3444 #endif 3445 return (error); 3446 } 3447 3448 void 3449 in_pcbdetach_tag(struct m_snd_tag *mst) 3450 { 3451 3452 m_snd_tag_rele(mst); 3453 #ifdef INET 3454 counter_u64_add(rate_limit_active, -1); 3455 #endif 3456 } 3457 3458 /* 3459 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3460 * if any: 3461 */ 3462 void 3463 in_pcbdetach_txrtlmt(struct inpcb *inp) 3464 { 3465 struct m_snd_tag *mst; 3466 3467 INP_WLOCK_ASSERT(inp); 3468 3469 mst = inp->inp_snd_tag; 3470 inp->inp_snd_tag = NULL; 3471 3472 if (mst == NULL) 3473 return; 3474 3475 m_snd_tag_rele(mst); 3476 #ifdef INET 3477 counter_u64_add(rate_limit_active, -1); 3478 #endif 3479 } 3480 3481 int 3482 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3483 { 3484 int error; 3485 3486 /* 3487 * If the existing send tag is for the wrong interface due to 3488 * a route change, first drop the existing tag. Set the 3489 * CHANGED flag so that we will keep trying to allocate a new 3490 * tag if we fail to allocate one this time. 3491 */ 3492 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3493 in_pcbdetach_txrtlmt(inp); 3494 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3495 } 3496 3497 /* 3498 * NOTE: When attaching to a network interface a reference is 3499 * made to ensure the network interface doesn't go away until 3500 * all ratelimit connections are gone. The network interface 3501 * pointers compared below represent valid network interfaces, 3502 * except when comparing towards NULL. 3503 */ 3504 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3505 error = 0; 3506 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3507 if (inp->inp_snd_tag != NULL) 3508 in_pcbdetach_txrtlmt(inp); 3509 error = 0; 3510 } else if (inp->inp_snd_tag == NULL) { 3511 /* 3512 * In order to utilize packet pacing with RSS, we need 3513 * to wait until there is a valid RSS hash before we 3514 * can proceed: 3515 */ 3516 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3517 error = EAGAIN; 3518 } else { 3519 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3520 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3521 } 3522 } else { 3523 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3524 } 3525 if (error == 0 || error == EOPNOTSUPP) 3526 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3527 3528 return (error); 3529 } 3530 3531 /* 3532 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3533 * is set in the fast path and will attach/detach/modify the TX rate 3534 * limit send tag based on the socket's so_max_pacing_rate value. 3535 */ 3536 void 3537 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3538 { 3539 struct socket *socket; 3540 uint32_t max_pacing_rate; 3541 bool did_upgrade; 3542 3543 if (inp == NULL) 3544 return; 3545 3546 socket = inp->inp_socket; 3547 if (socket == NULL) 3548 return; 3549 3550 if (!INP_WLOCKED(inp)) { 3551 /* 3552 * NOTE: If the write locking fails, we need to bail 3553 * out and use the non-ratelimited ring for the 3554 * transmit until there is a new chance to get the 3555 * write lock. 3556 */ 3557 if (!INP_TRY_UPGRADE(inp)) 3558 return; 3559 did_upgrade = 1; 3560 } else { 3561 did_upgrade = 0; 3562 } 3563 3564 /* 3565 * NOTE: The so_max_pacing_rate value is read unlocked, 3566 * because atomic updates are not required since the variable 3567 * is checked at every mbuf we send. It is assumed that the 3568 * variable read itself will be atomic. 3569 */ 3570 max_pacing_rate = socket->so_max_pacing_rate; 3571 3572 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3573 3574 if (did_upgrade) 3575 INP_DOWNGRADE(inp); 3576 } 3577 3578 /* 3579 * Track route changes for TX rate limiting. 3580 */ 3581 void 3582 in_pcboutput_eagain(struct inpcb *inp) 3583 { 3584 bool did_upgrade; 3585 3586 if (inp == NULL) 3587 return; 3588 3589 if (inp->inp_snd_tag == NULL) 3590 return; 3591 3592 if (!INP_WLOCKED(inp)) { 3593 /* 3594 * NOTE: If the write locking fails, we need to bail 3595 * out and use the non-ratelimited ring for the 3596 * transmit until there is a new chance to get the 3597 * write lock. 3598 */ 3599 if (!INP_TRY_UPGRADE(inp)) 3600 return; 3601 did_upgrade = 1; 3602 } else { 3603 did_upgrade = 0; 3604 } 3605 3606 /* detach rate limiting */ 3607 in_pcbdetach_txrtlmt(inp); 3608 3609 /* make sure new mbuf send tag allocation is made */ 3610 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3611 3612 if (did_upgrade) 3613 INP_DOWNGRADE(inp); 3614 } 3615 3616 #ifdef INET 3617 static void 3618 rl_init(void *st) 3619 { 3620 rate_limit_new = counter_u64_alloc(M_WAITOK); 3621 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3622 rate_limit_active = counter_u64_alloc(M_WAITOK); 3623 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3624 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3625 } 3626 3627 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3628 #endif 3629 #endif /* RATELIMIT */ 3630