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