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