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 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 39 */ 40 41 #include <sys/cdefs.h> 42 __FBSDID("$FreeBSD$"); 43 44 #include "opt_ddb.h" 45 #include "opt_ipsec.h" 46 #include "opt_inet.h" 47 #include "opt_inet6.h" 48 #include "opt_ratelimit.h" 49 #include "opt_route.h" 50 #include "opt_rss.h" 51 52 #include <sys/param.h> 53 #include <sys/hash.h> 54 #include <sys/systm.h> 55 #include <sys/libkern.h> 56 #include <sys/lock.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.h> 59 #include <sys/eventhandler.h> 60 #include <sys/domain.h> 61 #include <sys/protosw.h> 62 #include <sys/smp.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/sockio.h> 66 #include <sys/priv.h> 67 #include <sys/proc.h> 68 #include <sys/refcount.h> 69 #include <sys/jail.h> 70 #include <sys/kernel.h> 71 #include <sys/sysctl.h> 72 73 #ifdef DDB 74 #include <ddb/ddb.h> 75 #endif 76 77 #include <vm/uma.h> 78 #include <vm/vm.h> 79 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/if_private.h> 83 #include <net/if_types.h> 84 #include <net/if_llatbl.h> 85 #include <net/route.h> 86 #include <net/rss_config.h> 87 #include <net/vnet.h> 88 89 #if defined(INET) || defined(INET6) 90 #include <netinet/in.h> 91 #include <netinet/in_pcb.h> 92 #include <netinet/in_pcb_var.h> 93 #include <netinet/tcp.h> 94 #ifdef INET 95 #include <netinet/in_var.h> 96 #include <netinet/in_fib.h> 97 #endif 98 #include <netinet/ip_var.h> 99 #ifdef INET6 100 #include <netinet/ip6.h> 101 #include <netinet6/in6_pcb.h> 102 #include <netinet6/in6_var.h> 103 #include <netinet6/ip6_var.h> 104 #endif /* INET6 */ 105 #include <net/route/nhop.h> 106 #endif 107 108 #include <netipsec/ipsec_support.h> 109 110 #include <security/mac/mac_framework.h> 111 112 #define INPCBLBGROUP_SIZMIN 8 113 #define INPCBLBGROUP_SIZMAX 256 114 #define INP_FREED 0x00000200 /* See in_pcb.h. */ 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"); 223 224 #endif /* RATELIMIT */ 225 226 #endif /* INET */ 227 228 VNET_DEFINE(uint32_t, in_pcbhashseed); 229 static void 230 in_pcbhashseed_init(void) 231 { 232 233 V_in_pcbhashseed = arc4random(); 234 } 235 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, 236 in_pcbhashseed_init, 0); 237 238 static void in_pcbremhash(struct inpcb *); 239 240 /* 241 * in_pcb.c: manage the Protocol Control Blocks. 242 * 243 * NOTE: It is assumed that most of these functions will be called with 244 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 245 * functions often modify hash chains or addresses in pcbs. 246 */ 247 248 static struct inpcblbgroup * 249 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred, 250 u_char vflag, uint16_t port, const union in_dependaddr *addr, int size, 251 uint8_t numa_domain) 252 { 253 struct inpcblbgroup *grp; 254 size_t bytes; 255 256 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 257 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 258 if (grp == NULL) 259 return (NULL); 260 grp->il_cred = crhold(cred); 261 grp->il_vflag = vflag; 262 grp->il_lport = port; 263 grp->il_numa_domain = numa_domain; 264 grp->il_dependladdr = *addr; 265 grp->il_inpsiz = size; 266 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 267 return (grp); 268 } 269 270 static void 271 in_pcblbgroup_free_deferred(epoch_context_t ctx) 272 { 273 struct inpcblbgroup *grp; 274 275 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 276 crfree(grp->il_cred); 277 free(grp, M_PCB); 278 } 279 280 static void 281 in_pcblbgroup_free(struct inpcblbgroup *grp) 282 { 283 284 CK_LIST_REMOVE(grp, il_list); 285 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); 286 } 287 288 static struct inpcblbgroup * 289 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 290 struct inpcblbgroup *old_grp, int size) 291 { 292 struct inpcblbgroup *grp; 293 int i; 294 295 grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag, 296 old_grp->il_lport, &old_grp->il_dependladdr, size, 297 old_grp->il_numa_domain); 298 if (grp == NULL) 299 return (NULL); 300 301 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 302 ("invalid new local group size %d and old local group count %d", 303 grp->il_inpsiz, old_grp->il_inpcnt)); 304 305 for (i = 0; i < old_grp->il_inpcnt; ++i) 306 grp->il_inp[i] = old_grp->il_inp[i]; 307 grp->il_inpcnt = old_grp->il_inpcnt; 308 in_pcblbgroup_free(old_grp); 309 return (grp); 310 } 311 312 /* 313 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 314 * and shrink group if possible. 315 */ 316 static void 317 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 318 int i) 319 { 320 struct inpcblbgroup *grp, *new_grp; 321 322 grp = *grpp; 323 for (; i + 1 < grp->il_inpcnt; ++i) 324 grp->il_inp[i] = grp->il_inp[i + 1]; 325 grp->il_inpcnt--; 326 327 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 328 grp->il_inpcnt <= grp->il_inpsiz / 4) { 329 /* Shrink this group. */ 330 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 331 if (new_grp != NULL) 332 *grpp = new_grp; 333 } 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(hdr, 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 } else if (grp->il_inpcnt == grp->il_inpsiz) { 386 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 387 if (ratecheck(&lastprint, &interval)) 388 printf("lb group port %d, limit reached\n", 389 ntohs(grp->il_lport)); 390 return (0); 391 } 392 393 /* Expand this local group. */ 394 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 395 if (grp == NULL) 396 return (ENOBUFS); 397 } 398 399 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 400 ("invalid local group size %d and count %d", grp->il_inpsiz, 401 grp->il_inpcnt)); 402 403 grp->il_inp[grp->il_inpcnt] = inp; 404 grp->il_inpcnt++; 405 return (0); 406 } 407 408 /* 409 * Remove PCB from load balance group. 410 */ 411 static void 412 in_pcbremlbgrouphash(struct inpcb *inp) 413 { 414 struct inpcbinfo *pcbinfo; 415 struct inpcblbgrouphead *hdr; 416 struct inpcblbgroup *grp; 417 int i; 418 419 pcbinfo = inp->inp_pcbinfo; 420 421 INP_WLOCK_ASSERT(inp); 422 INP_HASH_WLOCK_ASSERT(pcbinfo); 423 424 hdr = &pcbinfo->ipi_lbgrouphashbase[ 425 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 426 CK_LIST_FOREACH(grp, hdr, il_list) { 427 for (i = 0; i < grp->il_inpcnt; ++i) { 428 if (grp->il_inp[i] != inp) 429 continue; 430 431 if (grp->il_inpcnt == 1) { 432 /* We are the last, free this local group. */ 433 in_pcblbgroup_free(grp); 434 } else { 435 /* Pull up inpcbs, shrink group if possible. */ 436 in_pcblbgroup_reorder(hdr, &grp, i); 437 } 438 return; 439 } 440 } 441 } 442 443 int 444 in_pcblbgroup_numa(struct inpcb *inp, int arg) 445 { 446 struct inpcbinfo *pcbinfo; 447 struct inpcblbgrouphead *hdr; 448 struct inpcblbgroup *grp; 449 int err, i; 450 uint8_t numa_domain; 451 452 switch (arg) { 453 case TCP_REUSPORT_LB_NUMA_NODOM: 454 numa_domain = M_NODOM; 455 break; 456 case TCP_REUSPORT_LB_NUMA_CURDOM: 457 numa_domain = PCPU_GET(domain); 458 break; 459 default: 460 if (arg < 0 || arg >= vm_ndomains) 461 return (EINVAL); 462 numa_domain = arg; 463 } 464 465 err = 0; 466 pcbinfo = inp->inp_pcbinfo; 467 INP_WLOCK_ASSERT(inp); 468 INP_HASH_WLOCK(pcbinfo); 469 hdr = &pcbinfo->ipi_lbgrouphashbase[ 470 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 471 CK_LIST_FOREACH(grp, hdr, il_list) { 472 for (i = 0; i < grp->il_inpcnt; ++i) { 473 if (grp->il_inp[i] != inp) 474 continue; 475 476 if (grp->il_numa_domain == numa_domain) { 477 goto abort_with_hash_wlock; 478 } 479 480 /* Remove it from the old group. */ 481 in_pcbremlbgrouphash(inp); 482 483 /* Add it to the new group based on numa domain. */ 484 in_pcbinslbgrouphash(inp, numa_domain); 485 goto abort_with_hash_wlock; 486 } 487 } 488 err = ENOENT; 489 abort_with_hash_wlock: 490 INP_HASH_WUNLOCK(pcbinfo); 491 return (err); 492 } 493 494 /* Make sure it is safe to use hashinit(9) on CK_LIST. */ 495 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb))); 496 497 /* 498 * Initialize an inpcbinfo - a per-VNET instance of connections db. 499 */ 500 void 501 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor, 502 u_int hash_nelements, u_int porthash_nelements) 503 { 504 505 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF); 506 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name, 507 NULL, MTX_DEF); 508 #ifdef VIMAGE 509 pcbinfo->ipi_vnet = curvnet; 510 #endif 511 CK_LIST_INIT(&pcbinfo->ipi_listhead); 512 pcbinfo->ipi_count = 0; 513 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 514 &pcbinfo->ipi_hashmask); 515 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 516 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 517 &pcbinfo->ipi_porthashmask); 518 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 519 &pcbinfo->ipi_lbgrouphashmask); 520 pcbinfo->ipi_zone = pcbstor->ips_zone; 521 pcbinfo->ipi_portzone = pcbstor->ips_portzone; 522 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone); 523 } 524 525 /* 526 * Destroy an inpcbinfo. 527 */ 528 void 529 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 530 { 531 532 KASSERT(pcbinfo->ipi_count == 0, 533 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 534 535 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 536 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 537 pcbinfo->ipi_porthashmask); 538 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 539 pcbinfo->ipi_lbgrouphashmask); 540 mtx_destroy(&pcbinfo->ipi_hash_lock); 541 mtx_destroy(&pcbinfo->ipi_lock); 542 } 543 544 /* 545 * Initialize a pcbstorage - per protocol zones to allocate inpcbs. 546 */ 547 static void inpcb_dtor(void *, int, void *); 548 static void inpcb_fini(void *, int); 549 void 550 in_pcbstorage_init(void *arg) 551 { 552 struct inpcbstorage *pcbstor = arg; 553 554 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name, 555 pcbstor->ips_size, NULL, inpcb_dtor, pcbstor->ips_pcbinit, 556 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR); 557 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name, 558 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 559 uma_zone_set_smr(pcbstor->ips_portzone, 560 uma_zone_get_smr(pcbstor->ips_zone)); 561 } 562 563 /* 564 * Destroy a pcbstorage - used by unloadable protocols. 565 */ 566 void 567 in_pcbstorage_destroy(void *arg) 568 { 569 struct inpcbstorage *pcbstor = arg; 570 571 uma_zdestroy(pcbstor->ips_zone); 572 uma_zdestroy(pcbstor->ips_portzone); 573 } 574 575 /* 576 * Allocate a PCB and associate it with the socket. 577 * On success return with the PCB locked. 578 */ 579 int 580 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 581 { 582 struct inpcb *inp; 583 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 584 int error; 585 #endif 586 587 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT); 588 if (inp == NULL) 589 return (ENOBUFS); 590 bzero(&inp->inp_start_zero, inp_zero_size); 591 #ifdef NUMA 592 inp->inp_numa_domain = M_NODOM; 593 #endif 594 inp->inp_pcbinfo = pcbinfo; 595 inp->inp_socket = so; 596 inp->inp_cred = crhold(so->so_cred); 597 inp->inp_inc.inc_fibnum = so->so_fibnum; 598 #ifdef MAC 599 error = mac_inpcb_init(inp, M_NOWAIT); 600 if (error != 0) 601 goto out; 602 mac_inpcb_create(so, inp); 603 #endif 604 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 605 error = ipsec_init_pcbpolicy(inp); 606 if (error != 0) { 607 #ifdef MAC 608 mac_inpcb_destroy(inp); 609 #endif 610 goto out; 611 } 612 #endif /*IPSEC*/ 613 #ifdef INET6 614 if (INP_SOCKAF(so) == AF_INET6) { 615 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6; 616 if (V_ip6_v6only) 617 inp->inp_flags |= IN6P_IPV6_V6ONLY; 618 #ifdef INET 619 else 620 inp->inp_vflag |= INP_IPV4; 621 #endif 622 if (V_ip6_auto_flowlabel) 623 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 624 inp->in6p_hops = -1; /* use kernel default */ 625 } 626 #endif 627 #if defined(INET) && defined(INET6) 628 else 629 #endif 630 #ifdef INET 631 inp->inp_vflag |= INP_IPV4; 632 #endif 633 /* 634 * Routes in inpcb's can cache L2 as well; they are guaranteed 635 * to be cleaned up. 636 */ 637 inp->inp_route.ro_flags = RT_LLE_CACHE; 638 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */ 639 INP_WLOCK(inp); 640 INP_INFO_WLOCK(pcbinfo); 641 pcbinfo->ipi_count++; 642 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 643 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list); 644 INP_INFO_WUNLOCK(pcbinfo); 645 so->so_pcb = inp; 646 647 return (0); 648 649 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 650 out: 651 uma_zfree_smr(pcbinfo->ipi_zone, inp); 652 return (error); 653 #endif 654 } 655 656 #ifdef INET 657 int 658 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 659 { 660 int anonport, error; 661 662 KASSERT(nam == NULL || nam->sa_family == AF_INET, 663 ("%s: invalid address family for %p", __func__, nam)); 664 KASSERT(nam == NULL || nam->sa_len == sizeof(struct sockaddr_in), 665 ("%s: invalid address length for %p", __func__, nam)); 666 INP_WLOCK_ASSERT(inp); 667 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 668 669 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 670 return (EINVAL); 671 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 672 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 673 &inp->inp_lport, cred); 674 if (error) 675 return (error); 676 if (in_pcbinshash(inp) != 0) { 677 inp->inp_laddr.s_addr = INADDR_ANY; 678 inp->inp_lport = 0; 679 return (EAGAIN); 680 } 681 if (anonport) 682 inp->inp_flags |= INP_ANONPORT; 683 return (0); 684 } 685 #endif 686 687 #if defined(INET) || defined(INET6) 688 /* 689 * Assign a local port like in_pcb_lport(), but also used with connect() 690 * and a foreign address and port. If fsa is non-NULL, choose a local port 691 * that is unused with those, otherwise one that is completely unused. 692 * lsa can be NULL for IPv6. 693 */ 694 int 695 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 696 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 697 { 698 struct inpcbinfo *pcbinfo; 699 struct inpcb *tmpinp; 700 unsigned short *lastport; 701 int count, error; 702 u_short aux, first, last, lport; 703 #ifdef INET 704 struct in_addr laddr, faddr; 705 #endif 706 #ifdef INET6 707 struct in6_addr *laddr6, *faddr6; 708 #endif 709 710 pcbinfo = inp->inp_pcbinfo; 711 712 /* 713 * Because no actual state changes occur here, a global write lock on 714 * the pcbinfo isn't required. 715 */ 716 INP_LOCK_ASSERT(inp); 717 INP_HASH_LOCK_ASSERT(pcbinfo); 718 719 if (inp->inp_flags & INP_HIGHPORT) { 720 first = V_ipport_hifirstauto; /* sysctl */ 721 last = V_ipport_hilastauto; 722 lastport = &pcbinfo->ipi_lasthi; 723 } else if (inp->inp_flags & INP_LOWPORT) { 724 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 725 if (error) 726 return (error); 727 first = V_ipport_lowfirstauto; /* 1023 */ 728 last = V_ipport_lowlastauto; /* 600 */ 729 lastport = &pcbinfo->ipi_lastlow; 730 } else { 731 first = V_ipport_firstauto; /* sysctl */ 732 last = V_ipport_lastauto; 733 lastport = &pcbinfo->ipi_lastport; 734 } 735 736 /* 737 * Instead of having two loops further down counting up or down 738 * make sure that first is always <= last and go with only one 739 * code path implementing all logic. 740 */ 741 if (first > last) { 742 aux = first; 743 first = last; 744 last = aux; 745 } 746 747 #ifdef INET 748 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */ 749 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 750 if (lsa != NULL) 751 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 752 if (fsa != NULL) 753 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 754 } 755 #endif 756 #ifdef INET6 757 laddr6 = NULL; 758 if ((inp->inp_vflag & INP_IPV6) != 0) { 759 if (lsa != NULL) 760 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 761 if (fsa != NULL) 762 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 763 } 764 #endif 765 766 tmpinp = NULL; 767 lport = *lportp; 768 769 if (V_ipport_randomized) 770 *lastport = first + (arc4random() % (last - first)); 771 772 count = last - first; 773 774 do { 775 if (count-- < 0) /* completely used? */ 776 return (EADDRNOTAVAIL); 777 ++*lastport; 778 if (*lastport < first || *lastport > last) 779 *lastport = first; 780 lport = htons(*lastport); 781 782 if (fsa != NULL) { 783 #ifdef INET 784 if (lsa->sa_family == AF_INET) { 785 tmpinp = in_pcblookup_hash_locked(pcbinfo, 786 faddr, fport, laddr, lport, lookupflags, 787 M_NODOM); 788 } 789 #endif 790 #ifdef INET6 791 if (lsa->sa_family == AF_INET6) { 792 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 793 faddr6, fport, laddr6, lport, lookupflags, 794 M_NODOM); 795 } 796 #endif 797 } else { 798 #ifdef INET6 799 if ((inp->inp_vflag & INP_IPV6) != 0) { 800 tmpinp = in6_pcblookup_local(pcbinfo, 801 &inp->in6p_laddr, lport, lookupflags, cred); 802 #ifdef INET 803 if (tmpinp == NULL && 804 (inp->inp_vflag & INP_IPV4)) 805 tmpinp = in_pcblookup_local(pcbinfo, 806 laddr, lport, lookupflags, cred); 807 #endif 808 } 809 #endif 810 #if defined(INET) && defined(INET6) 811 else 812 #endif 813 #ifdef INET 814 tmpinp = in_pcblookup_local(pcbinfo, laddr, 815 lport, lookupflags, cred); 816 #endif 817 } 818 } while (tmpinp != NULL); 819 820 *lportp = lport; 821 822 return (0); 823 } 824 825 /* 826 * Select a local port (number) to use. 827 */ 828 int 829 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 830 struct ucred *cred, int lookupflags) 831 { 832 struct sockaddr_in laddr; 833 834 if (laddrp) { 835 bzero(&laddr, sizeof(laddr)); 836 laddr.sin_family = AF_INET; 837 laddr.sin_addr = *laddrp; 838 } 839 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 840 NULL, lportp, NULL, 0, cred, lookupflags)); 841 } 842 843 /* 844 * Return cached socket options. 845 */ 846 int 847 inp_so_options(const struct inpcb *inp) 848 { 849 int so_options; 850 851 so_options = 0; 852 853 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 854 so_options |= SO_REUSEPORT_LB; 855 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 856 so_options |= SO_REUSEPORT; 857 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 858 so_options |= SO_REUSEADDR; 859 return (so_options); 860 } 861 #endif /* INET || INET6 */ 862 863 /* 864 * Check if a new BINDMULTI socket is allowed to be created. 865 * 866 * ni points to the new inp. 867 * oi points to the existing inp. 868 * 869 * This checks whether the existing inp also has BINDMULTI and 870 * whether the credentials match. 871 */ 872 int 873 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 874 { 875 /* Check permissions match */ 876 if ((ni->inp_flags2 & INP_BINDMULTI) && 877 (ni->inp_cred->cr_uid != 878 oi->inp_cred->cr_uid)) 879 return (0); 880 881 /* Check the existing inp has BINDMULTI set */ 882 if ((ni->inp_flags2 & INP_BINDMULTI) && 883 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 884 return (0); 885 886 /* 887 * We're okay - either INP_BINDMULTI isn't set on ni, or 888 * it is and it matches the checks. 889 */ 890 return (1); 891 } 892 893 #ifdef INET 894 /* 895 * Set up a bind operation on a PCB, performing port allocation 896 * as required, but do not actually modify the PCB. Callers can 897 * either complete the bind by setting inp_laddr/inp_lport and 898 * calling in_pcbinshash(), or they can just use the resulting 899 * port and address to authorise the sending of a once-off packet. 900 * 901 * On error, the values of *laddrp and *lportp are not changed. 902 */ 903 int 904 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 905 u_short *lportp, struct ucred *cred) 906 { 907 struct socket *so = inp->inp_socket; 908 struct sockaddr_in *sin; 909 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 910 struct in_addr laddr; 911 u_short lport = 0; 912 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 913 int error; 914 915 /* 916 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 917 * so that we don't have to add to the (already messy) code below. 918 */ 919 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 920 921 /* 922 * No state changes, so read locks are sufficient here. 923 */ 924 INP_LOCK_ASSERT(inp); 925 INP_HASH_LOCK_ASSERT(pcbinfo); 926 927 laddr.s_addr = *laddrp; 928 if (nam != NULL && laddr.s_addr != INADDR_ANY) 929 return (EINVAL); 930 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 931 lookupflags = INPLOOKUP_WILDCARD; 932 if (nam == NULL) { 933 if ((error = prison_local_ip4(cred, &laddr)) != 0) 934 return (error); 935 } else { 936 sin = (struct sockaddr_in *)nam; 937 KASSERT(sin->sin_family == AF_INET, 938 ("%s: invalid family for address %p", __func__, sin)); 939 KASSERT(sin->sin_len == sizeof(*sin), 940 ("%s: invalid length for address %p", __func__, sin)); 941 942 error = prison_local_ip4(cred, &sin->sin_addr); 943 if (error) 944 return (error); 945 if (sin->sin_port != *lportp) { 946 /* Don't allow the port to change. */ 947 if (*lportp != 0) 948 return (EINVAL); 949 lport = sin->sin_port; 950 } 951 /* NB: lport is left as 0 if the port isn't being changed. */ 952 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 953 /* 954 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 955 * allow complete duplication of binding if 956 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 957 * and a multicast address is bound on both 958 * new and duplicated sockets. 959 */ 960 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 961 reuseport = SO_REUSEADDR|SO_REUSEPORT; 962 /* 963 * XXX: How to deal with SO_REUSEPORT_LB here? 964 * Treat same as SO_REUSEPORT for now. 965 */ 966 if ((so->so_options & 967 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 968 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 969 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 970 sin->sin_port = 0; /* yech... */ 971 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 972 /* 973 * Is the address a local IP address? 974 * If INP_BINDANY is set, then the socket may be bound 975 * to any endpoint address, local or not. 976 */ 977 if ((inp->inp_flags & INP_BINDANY) == 0 && 978 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 979 return (EADDRNOTAVAIL); 980 } 981 laddr = sin->sin_addr; 982 if (lport) { 983 struct inpcb *t; 984 985 /* GROSS */ 986 if (ntohs(lport) <= V_ipport_reservedhigh && 987 ntohs(lport) >= V_ipport_reservedlow && 988 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 989 return (EACCES); 990 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 991 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 992 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 993 lport, INPLOOKUP_WILDCARD, cred); 994 /* 995 * XXX 996 * This entire block sorely needs a rewrite. 997 */ 998 if (t && 999 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 1000 (so->so_type != SOCK_STREAM || 1001 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 1002 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 1003 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 1004 (t->inp_flags2 & INP_REUSEPORT) || 1005 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 1006 (inp->inp_cred->cr_uid != 1007 t->inp_cred->cr_uid)) 1008 return (EADDRINUSE); 1009 1010 /* 1011 * If the socket is a BINDMULTI socket, then 1012 * the credentials need to match and the 1013 * original socket also has to have been bound 1014 * with BINDMULTI. 1015 */ 1016 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 1017 return (EADDRINUSE); 1018 } 1019 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 1020 lport, lookupflags, cred); 1021 if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 1022 (reuseport & inp_so_options(t)) == 0 && 1023 (reuseport_lb & inp_so_options(t)) == 0) { 1024 #ifdef INET6 1025 if (ntohl(sin->sin_addr.s_addr) != 1026 INADDR_ANY || 1027 ntohl(t->inp_laddr.s_addr) != 1028 INADDR_ANY || 1029 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 1030 (t->inp_vflag & INP_IPV6PROTO) == 0) 1031 #endif 1032 return (EADDRINUSE); 1033 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 1034 return (EADDRINUSE); 1035 } 1036 } 1037 } 1038 if (*lportp != 0) 1039 lport = *lportp; 1040 if (lport == 0) { 1041 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 1042 if (error != 0) 1043 return (error); 1044 } 1045 *laddrp = laddr.s_addr; 1046 *lportp = lport; 1047 return (0); 1048 } 1049 1050 /* 1051 * Connect from a socket to a specified address. 1052 * Both address and port must be specified in argument sin. 1053 * If don't have a local address for this socket yet, 1054 * then pick one. 1055 */ 1056 int 1057 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred, 1058 bool rehash) 1059 { 1060 u_short lport, fport; 1061 in_addr_t laddr, faddr; 1062 int anonport, error; 1063 1064 INP_WLOCK_ASSERT(inp); 1065 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1066 1067 lport = inp->inp_lport; 1068 laddr = inp->inp_laddr.s_addr; 1069 anonport = (lport == 0); 1070 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport, 1071 cred); 1072 if (error) 1073 return (error); 1074 1075 /* Do the initial binding of the local address if required. */ 1076 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1077 KASSERT(rehash == true, 1078 ("Rehashing required for unbound inps")); 1079 inp->inp_lport = lport; 1080 inp->inp_laddr.s_addr = laddr; 1081 if (in_pcbinshash(inp) != 0) { 1082 inp->inp_laddr.s_addr = INADDR_ANY; 1083 inp->inp_lport = 0; 1084 return (EAGAIN); 1085 } 1086 } 1087 1088 /* Commit the remaining changes. */ 1089 inp->inp_lport = lport; 1090 inp->inp_laddr.s_addr = laddr; 1091 inp->inp_faddr.s_addr = faddr; 1092 inp->inp_fport = fport; 1093 if (rehash) { 1094 in_pcbrehash(inp); 1095 } else { 1096 in_pcbinshash(inp); 1097 } 1098 1099 if (anonport) 1100 inp->inp_flags |= INP_ANONPORT; 1101 return (0); 1102 } 1103 1104 /* 1105 * Do proper source address selection on an unbound socket in case 1106 * of connect. Take jails into account as well. 1107 */ 1108 int 1109 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1110 struct ucred *cred) 1111 { 1112 struct ifaddr *ifa; 1113 struct sockaddr *sa; 1114 struct sockaddr_in *sin, dst; 1115 struct nhop_object *nh; 1116 int error; 1117 1118 NET_EPOCH_ASSERT(); 1119 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1120 1121 /* 1122 * Bypass source address selection and use the primary jail IP 1123 * if requested. 1124 */ 1125 if (!prison_saddrsel_ip4(cred, laddr)) 1126 return (0); 1127 1128 error = 0; 1129 1130 nh = NULL; 1131 bzero(&dst, sizeof(dst)); 1132 sin = &dst; 1133 sin->sin_family = AF_INET; 1134 sin->sin_len = sizeof(struct sockaddr_in); 1135 sin->sin_addr.s_addr = faddr->s_addr; 1136 1137 /* 1138 * If route is known our src addr is taken from the i/f, 1139 * else punt. 1140 * 1141 * Find out route to destination. 1142 */ 1143 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1144 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1145 0, NHR_NONE, 0); 1146 1147 /* 1148 * If we found a route, use the address corresponding to 1149 * the outgoing interface. 1150 * 1151 * Otherwise assume faddr is reachable on a directly connected 1152 * network and try to find a corresponding interface to take 1153 * the source address from. 1154 */ 1155 if (nh == NULL || nh->nh_ifp == NULL) { 1156 struct in_ifaddr *ia; 1157 struct ifnet *ifp; 1158 1159 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1160 inp->inp_socket->so_fibnum)); 1161 if (ia == NULL) { 1162 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1163 inp->inp_socket->so_fibnum)); 1164 } 1165 if (ia == NULL) { 1166 error = ENETUNREACH; 1167 goto done; 1168 } 1169 1170 if (!prison_flag(cred, PR_IP4)) { 1171 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1172 goto done; 1173 } 1174 1175 ifp = ia->ia_ifp; 1176 ia = NULL; 1177 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1178 sa = ifa->ifa_addr; 1179 if (sa->sa_family != AF_INET) 1180 continue; 1181 sin = (struct sockaddr_in *)sa; 1182 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1183 ia = (struct in_ifaddr *)ifa; 1184 break; 1185 } 1186 } 1187 if (ia != NULL) { 1188 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1189 goto done; 1190 } 1191 1192 /* 3. As a last resort return the 'default' jail address. */ 1193 error = prison_get_ip4(cred, laddr); 1194 goto done; 1195 } 1196 1197 /* 1198 * If the outgoing interface on the route found is not 1199 * a loopback interface, use the address from that interface. 1200 * In case of jails do those three steps: 1201 * 1. check if the interface address belongs to the jail. If so use it. 1202 * 2. check if we have any address on the outgoing interface 1203 * belonging to this jail. If so use it. 1204 * 3. as a last resort return the 'default' jail address. 1205 */ 1206 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1207 struct in_ifaddr *ia; 1208 struct ifnet *ifp; 1209 1210 /* If not jailed, use the default returned. */ 1211 if (!prison_flag(cred, PR_IP4)) { 1212 ia = (struct in_ifaddr *)nh->nh_ifa; 1213 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1214 goto done; 1215 } 1216 1217 /* Jailed. */ 1218 /* 1. Check if the iface address belongs to the jail. */ 1219 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1220 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1221 ia = (struct in_ifaddr *)nh->nh_ifa; 1222 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1223 goto done; 1224 } 1225 1226 /* 1227 * 2. Check if we have any address on the outgoing interface 1228 * belonging to this jail. 1229 */ 1230 ia = NULL; 1231 ifp = nh->nh_ifp; 1232 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1233 sa = ifa->ifa_addr; 1234 if (sa->sa_family != AF_INET) 1235 continue; 1236 sin = (struct sockaddr_in *)sa; 1237 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1238 ia = (struct in_ifaddr *)ifa; 1239 break; 1240 } 1241 } 1242 if (ia != NULL) { 1243 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1244 goto done; 1245 } 1246 1247 /* 3. As a last resort return the 'default' jail address. */ 1248 error = prison_get_ip4(cred, laddr); 1249 goto done; 1250 } 1251 1252 /* 1253 * The outgoing interface is marked with 'loopback net', so a route 1254 * to ourselves is here. 1255 * Try to find the interface of the destination address and then 1256 * take the address from there. That interface is not necessarily 1257 * a loopback interface. 1258 * In case of jails, check that it is an address of the jail 1259 * and if we cannot find, fall back to the 'default' jail address. 1260 */ 1261 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1262 struct in_ifaddr *ia; 1263 1264 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1265 inp->inp_socket->so_fibnum)); 1266 if (ia == NULL) 1267 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1268 inp->inp_socket->so_fibnum)); 1269 if (ia == NULL) 1270 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1271 1272 if (!prison_flag(cred, PR_IP4)) { 1273 if (ia == NULL) { 1274 error = ENETUNREACH; 1275 goto done; 1276 } 1277 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1278 goto done; 1279 } 1280 1281 /* Jailed. */ 1282 if (ia != NULL) { 1283 struct ifnet *ifp; 1284 1285 ifp = ia->ia_ifp; 1286 ia = NULL; 1287 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1288 sa = ifa->ifa_addr; 1289 if (sa->sa_family != AF_INET) 1290 continue; 1291 sin = (struct sockaddr_in *)sa; 1292 if (prison_check_ip4(cred, 1293 &sin->sin_addr) == 0) { 1294 ia = (struct in_ifaddr *)ifa; 1295 break; 1296 } 1297 } 1298 if (ia != NULL) { 1299 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1300 goto done; 1301 } 1302 } 1303 1304 /* 3. As a last resort return the 'default' jail address. */ 1305 error = prison_get_ip4(cred, laddr); 1306 goto done; 1307 } 1308 1309 done: 1310 return (error); 1311 } 1312 1313 /* 1314 * Set up for a connect from a socket to the specified address. 1315 * On entry, *laddrp and *lportp should contain the current local 1316 * address and port for the PCB; these are updated to the values 1317 * that should be placed in inp_laddr and inp_lport to complete 1318 * the connect. 1319 * 1320 * On success, *faddrp and *fportp will be set to the remote address 1321 * and port. These are not updated in the error case. 1322 */ 1323 int 1324 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin, 1325 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1326 struct ucred *cred) 1327 { 1328 struct in_ifaddr *ia; 1329 struct in_addr laddr, faddr; 1330 u_short lport, fport; 1331 int error; 1332 1333 KASSERT(sin->sin_family == AF_INET, 1334 ("%s: invalid address family for %p", __func__, sin)); 1335 KASSERT(sin->sin_len == sizeof(*sin), 1336 ("%s: invalid address length for %p", __func__, sin)); 1337 1338 /* 1339 * Because a global state change doesn't actually occur here, a read 1340 * lock is sufficient. 1341 */ 1342 NET_EPOCH_ASSERT(); 1343 INP_LOCK_ASSERT(inp); 1344 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1345 1346 if (sin->sin_port == 0) 1347 return (EADDRNOTAVAIL); 1348 laddr.s_addr = *laddrp; 1349 lport = *lportp; 1350 faddr = sin->sin_addr; 1351 fport = sin->sin_port; 1352 #ifdef ROUTE_MPATH 1353 if (CALC_FLOWID_OUTBOUND) { 1354 uint32_t hash_val, hash_type; 1355 1356 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, 1357 inp->inp_socket->so_proto->pr_protocol, &hash_type); 1358 1359 inp->inp_flowid = hash_val; 1360 inp->inp_flowtype = hash_type; 1361 } 1362 #endif 1363 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1364 /* 1365 * If the destination address is INADDR_ANY, 1366 * use the primary local address. 1367 * If the supplied address is INADDR_BROADCAST, 1368 * and the primary interface supports broadcast, 1369 * choose the broadcast address for that interface. 1370 */ 1371 if (faddr.s_addr == INADDR_ANY) { 1372 faddr = 1373 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1374 if ((error = prison_get_ip4(cred, &faddr)) != 0) 1375 return (error); 1376 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1377 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1378 IFF_BROADCAST) 1379 faddr = satosin(&CK_STAILQ_FIRST( 1380 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1381 } 1382 } 1383 if (laddr.s_addr == INADDR_ANY) { 1384 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1385 /* 1386 * If the destination address is multicast and an outgoing 1387 * interface has been set as a multicast option, prefer the 1388 * address of that interface as our source address. 1389 */ 1390 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1391 inp->inp_moptions != NULL) { 1392 struct ip_moptions *imo; 1393 struct ifnet *ifp; 1394 1395 imo = inp->inp_moptions; 1396 if (imo->imo_multicast_ifp != NULL) { 1397 ifp = imo->imo_multicast_ifp; 1398 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1399 if (ia->ia_ifp == ifp && 1400 prison_check_ip4(cred, 1401 &ia->ia_addr.sin_addr) == 0) 1402 break; 1403 } 1404 if (ia == NULL) 1405 error = EADDRNOTAVAIL; 1406 else { 1407 laddr = ia->ia_addr.sin_addr; 1408 error = 0; 1409 } 1410 } 1411 } 1412 if (error) 1413 return (error); 1414 } 1415 1416 if (lport != 0) { 1417 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, 1418 fport, laddr, lport, 0, M_NODOM) != NULL) 1419 return (EADDRINUSE); 1420 } else { 1421 struct sockaddr_in lsin, fsin; 1422 1423 bzero(&lsin, sizeof(lsin)); 1424 bzero(&fsin, sizeof(fsin)); 1425 lsin.sin_family = AF_INET; 1426 lsin.sin_addr = laddr; 1427 fsin.sin_family = AF_INET; 1428 fsin.sin_addr = faddr; 1429 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, 1430 &lport, (struct sockaddr *)& fsin, fport, cred, 1431 INPLOOKUP_WILDCARD); 1432 if (error) 1433 return (error); 1434 } 1435 *laddrp = laddr.s_addr; 1436 *lportp = lport; 1437 *faddrp = faddr.s_addr; 1438 *fportp = fport; 1439 return (0); 1440 } 1441 1442 void 1443 in_pcbdisconnect(struct inpcb *inp) 1444 { 1445 1446 INP_WLOCK_ASSERT(inp); 1447 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1448 1449 inp->inp_laddr.s_addr = INADDR_ANY; 1450 inp->inp_faddr.s_addr = INADDR_ANY; 1451 inp->inp_fport = 0; 1452 in_pcbrehash(inp); 1453 } 1454 #endif /* INET */ 1455 1456 /* 1457 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1458 * For most protocols, this will be invoked immediately prior to calling 1459 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1460 * socket, in which case in_pcbfree() is deferred. 1461 */ 1462 void 1463 in_pcbdetach(struct inpcb *inp) 1464 { 1465 1466 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1467 1468 #ifdef RATELIMIT 1469 if (inp->inp_snd_tag != NULL) 1470 in_pcbdetach_txrtlmt(inp); 1471 #endif 1472 inp->inp_socket->so_pcb = NULL; 1473 inp->inp_socket = NULL; 1474 } 1475 1476 /* 1477 * inpcb hash lookups are protected by SMR section. 1478 * 1479 * Once desired pcb has been found, switching from SMR section to a pcb 1480 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK 1481 * here because SMR is a critical section. 1482 * In 99%+ cases inp_smr_lock() would obtain the lock immediately. 1483 */ 1484 static inline void 1485 inp_lock(struct inpcb *inp, const inp_lookup_t lock) 1486 { 1487 1488 lock == INPLOOKUP_RLOCKPCB ? 1489 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock); 1490 } 1491 1492 static inline void 1493 inp_unlock(struct inpcb *inp, const inp_lookup_t lock) 1494 { 1495 1496 lock == INPLOOKUP_RLOCKPCB ? 1497 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock); 1498 } 1499 1500 static inline int 1501 inp_trylock(struct inpcb *inp, const inp_lookup_t lock) 1502 { 1503 1504 return (lock == INPLOOKUP_RLOCKPCB ? 1505 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock)); 1506 } 1507 1508 static inline bool 1509 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock) 1510 { 1511 1512 return (lock == INPLOOKUP_RLOCKPCB ? 1513 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp)); 1514 } 1515 1516 static inline bool 1517 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags) 1518 { 1519 1520 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB); 1521 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr); 1522 1523 if (__predict_true(inp_trylock(inp, lock))) { 1524 if (__predict_false(inp->inp_flags & ignflags)) { 1525 smr_exit(inp->inp_pcbinfo->ipi_smr); 1526 inp_unlock(inp, lock); 1527 return (false); 1528 } 1529 smr_exit(inp->inp_pcbinfo->ipi_smr); 1530 return (true); 1531 } 1532 1533 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1534 smr_exit(inp->inp_pcbinfo->ipi_smr); 1535 inp_lock(inp, lock); 1536 if (__predict_false(in_pcbrele(inp, lock))) 1537 return (false); 1538 /* 1539 * inp acquired through refcount & lock for sure didn't went 1540 * through uma_zfree(). However, it may have already went 1541 * through in_pcbfree() and has another reference, that 1542 * prevented its release by our in_pcbrele(). 1543 */ 1544 if (__predict_false(inp->inp_flags & ignflags)) { 1545 inp_unlock(inp, lock); 1546 return (false); 1547 } 1548 return (true); 1549 } else { 1550 smr_exit(inp->inp_pcbinfo->ipi_smr); 1551 return (false); 1552 } 1553 } 1554 1555 bool 1556 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock) 1557 { 1558 1559 /* 1560 * in_pcblookup() family of functions ignore not only freed entries, 1561 * that may be found due to lockless access to the hash, but dropped 1562 * entries, too. 1563 */ 1564 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED)); 1565 } 1566 1567 /* 1568 * inp_next() - inpcb hash/list traversal iterator 1569 * 1570 * Requires initialized struct inpcb_iterator for context. 1571 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR(). 1572 * 1573 * - Iterator can have either write-lock or read-lock semantics, that can not 1574 * be changed later. 1575 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through 1576 * a single hash slot. Note: only rip_input() does the latter. 1577 * - Iterator may have optional bool matching function. The matching function 1578 * will be executed for each inpcb in the SMR context, so it can not acquire 1579 * locks and can safely access only immutable fields of inpcb. 1580 * 1581 * A fresh initialized iterator has NULL inpcb in its context and that 1582 * means that inp_next() call would return the very first inpcb on the list 1583 * locked with desired semantic. In all following calls the context pointer 1584 * shall hold the current inpcb pointer. The KPI user is not supposed to 1585 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL 1586 * and write NULL to its context. After end of traversal an iterator can be 1587 * reused. 1588 * 1589 * List traversals have the following features/constraints: 1590 * - New entries won't be seen, as they are always added to the head of a list. 1591 * - Removed entries won't stop traversal as long as they are not added to 1592 * a different list. This is violated by in_pcbrehash(). 1593 */ 1594 #define II_LIST_FIRST(ipi, hash) \ 1595 (((hash) == INP_ALL_LIST) ? \ 1596 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \ 1597 CK_LIST_FIRST(&(ipi)->ipi_hashbase[(hash)])) 1598 #define II_LIST_NEXT(inp, hash) \ 1599 (((hash) == INP_ALL_LIST) ? \ 1600 CK_LIST_NEXT((inp), inp_list) : \ 1601 CK_LIST_NEXT((inp), inp_hash)) 1602 #define II_LOCK_ASSERT(inp, lock) \ 1603 rw_assert(&(inp)->inp_lock, \ 1604 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED ) 1605 struct inpcb * 1606 inp_next(struct inpcb_iterator *ii) 1607 { 1608 const struct inpcbinfo *ipi = ii->ipi; 1609 inp_match_t *match = ii->match; 1610 void *ctx = ii->ctx; 1611 inp_lookup_t lock = ii->lock; 1612 int hash = ii->hash; 1613 struct inpcb *inp; 1614 1615 if (ii->inp == NULL) { /* First call. */ 1616 smr_enter(ipi->ipi_smr); 1617 /* This is unrolled CK_LIST_FOREACH(). */ 1618 for (inp = II_LIST_FIRST(ipi, hash); 1619 inp != NULL; 1620 inp = II_LIST_NEXT(inp, hash)) { 1621 if (match != NULL && (match)(inp, ctx) == false) 1622 continue; 1623 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED))) 1624 break; 1625 else { 1626 smr_enter(ipi->ipi_smr); 1627 MPASS(inp != II_LIST_FIRST(ipi, hash)); 1628 inp = II_LIST_FIRST(ipi, hash); 1629 if (inp == NULL) 1630 break; 1631 } 1632 } 1633 1634 if (inp == NULL) 1635 smr_exit(ipi->ipi_smr); 1636 else 1637 ii->inp = inp; 1638 1639 return (inp); 1640 } 1641 1642 /* Not a first call. */ 1643 smr_enter(ipi->ipi_smr); 1644 restart: 1645 inp = ii->inp; 1646 II_LOCK_ASSERT(inp, lock); 1647 next: 1648 inp = II_LIST_NEXT(inp, hash); 1649 if (inp == NULL) { 1650 smr_exit(ipi->ipi_smr); 1651 goto found; 1652 } 1653 1654 if (match != NULL && (match)(inp, ctx) == false) 1655 goto next; 1656 1657 if (__predict_true(inp_trylock(inp, lock))) { 1658 if (__predict_false(inp->inp_flags & INP_FREED)) { 1659 /* 1660 * Entries are never inserted in middle of a list, thus 1661 * as long as we are in SMR, we can continue traversal. 1662 * Jump to 'restart' should yield in the same result, 1663 * but could produce unnecessary looping. Could this 1664 * looping be unbound? 1665 */ 1666 inp_unlock(inp, lock); 1667 goto next; 1668 } else { 1669 smr_exit(ipi->ipi_smr); 1670 goto found; 1671 } 1672 } 1673 1674 /* 1675 * Can't obtain lock immediately, thus going hard. Once we exit the 1676 * SMR section we can no longer jump to 'next', and our only stable 1677 * anchoring point is ii->inp, which we keep locked for this case, so 1678 * we jump to 'restart'. 1679 */ 1680 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1681 smr_exit(ipi->ipi_smr); 1682 inp_lock(inp, lock); 1683 if (__predict_false(in_pcbrele(inp, lock))) { 1684 smr_enter(ipi->ipi_smr); 1685 goto restart; 1686 } 1687 /* 1688 * See comment in inp_smr_lock(). 1689 */ 1690 if (__predict_false(inp->inp_flags & INP_FREED)) { 1691 inp_unlock(inp, lock); 1692 smr_enter(ipi->ipi_smr); 1693 goto restart; 1694 } 1695 } else 1696 goto next; 1697 1698 found: 1699 inp_unlock(ii->inp, lock); 1700 ii->inp = inp; 1701 1702 return (ii->inp); 1703 } 1704 1705 /* 1706 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1707 * stability of an inpcb pointer despite the inpcb lock being released or 1708 * SMR section exited. 1709 * 1710 * To free a reference later in_pcbrele_(r|w)locked() must be performed. 1711 */ 1712 void 1713 in_pcbref(struct inpcb *inp) 1714 { 1715 u_int old __diagused; 1716 1717 old = refcount_acquire(&inp->inp_refcount); 1718 KASSERT(old > 0, ("%s: refcount 0", __func__)); 1719 } 1720 1721 /* 1722 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially 1723 * freeing the pcb, if the reference was very last. 1724 */ 1725 bool 1726 in_pcbrele_rlocked(struct inpcb *inp) 1727 { 1728 1729 INP_RLOCK_ASSERT(inp); 1730 1731 if (refcount_release(&inp->inp_refcount) == 0) 1732 return (false); 1733 1734 MPASS(inp->inp_flags & INP_FREED); 1735 MPASS(inp->inp_socket == NULL); 1736 MPASS(inp->inp_in_hpts == 0); 1737 INP_RUNLOCK(inp); 1738 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1739 return (true); 1740 } 1741 1742 bool 1743 in_pcbrele_wlocked(struct inpcb *inp) 1744 { 1745 1746 INP_WLOCK_ASSERT(inp); 1747 1748 if (refcount_release(&inp->inp_refcount) == 0) 1749 return (false); 1750 1751 MPASS(inp->inp_flags & INP_FREED); 1752 MPASS(inp->inp_socket == NULL); 1753 MPASS(inp->inp_in_hpts == 0); 1754 INP_WUNLOCK(inp); 1755 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1756 return (true); 1757 } 1758 1759 /* 1760 * Unconditionally schedule an inpcb to be freed by decrementing its 1761 * reference count, which should occur only after the inpcb has been detached 1762 * from its socket. If another thread holds a temporary reference (acquired 1763 * using in_pcbref()) then the free is deferred until that reference is 1764 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked. 1765 * Almost all work, including removal from global lists, is done in this 1766 * context, where the pcbinfo lock is held. 1767 */ 1768 void 1769 in_pcbfree(struct inpcb *inp) 1770 { 1771 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1772 #ifdef INET 1773 struct ip_moptions *imo; 1774 #endif 1775 #ifdef INET6 1776 struct ip6_moptions *im6o; 1777 #endif 1778 1779 INP_WLOCK_ASSERT(inp); 1780 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1781 KASSERT((inp->inp_flags & INP_FREED) == 0, 1782 ("%s: called twice for pcb %p", __func__, inp)); 1783 1784 inp->inp_flags |= INP_FREED; 1785 INP_INFO_WLOCK(pcbinfo); 1786 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1787 pcbinfo->ipi_count--; 1788 CK_LIST_REMOVE(inp, inp_list); 1789 INP_INFO_WUNLOCK(pcbinfo); 1790 1791 if (inp->inp_flags & INP_INHASHLIST) 1792 in_pcbremhash(inp); 1793 1794 RO_INVALIDATE_CACHE(&inp->inp_route); 1795 #ifdef MAC 1796 mac_inpcb_destroy(inp); 1797 #endif 1798 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1799 if (inp->inp_sp != NULL) 1800 ipsec_delete_pcbpolicy(inp); 1801 #endif 1802 #ifdef INET 1803 if (inp->inp_options) 1804 (void)m_free(inp->inp_options); 1805 imo = inp->inp_moptions; 1806 #endif 1807 #ifdef INET6 1808 if (inp->inp_vflag & INP_IPV6PROTO) { 1809 ip6_freepcbopts(inp->in6p_outputopts); 1810 im6o = inp->in6p_moptions; 1811 } else 1812 im6o = NULL; 1813 #endif 1814 1815 if (__predict_false(in_pcbrele_wlocked(inp) == false)) { 1816 INP_WUNLOCK(inp); 1817 } 1818 #ifdef INET6 1819 ip6_freemoptions(im6o); 1820 #endif 1821 #ifdef INET 1822 inp_freemoptions(imo); 1823 #endif 1824 /* Destruction is finalized in inpcb_dtor(). */ 1825 } 1826 1827 static void 1828 inpcb_dtor(void *mem, int size, void *arg) 1829 { 1830 struct inpcb *inp = mem; 1831 1832 crfree(inp->inp_cred); 1833 #ifdef INVARIANTS 1834 inp->inp_cred = NULL; 1835 #endif 1836 } 1837 1838 /* 1839 * Different protocols initialize their inpcbs differently - giving 1840 * different name to the lock. But they all are disposed the same. 1841 */ 1842 static void 1843 inpcb_fini(void *mem, int size) 1844 { 1845 struct inpcb *inp = mem; 1846 1847 INP_LOCK_DESTROY(inp); 1848 } 1849 1850 /* 1851 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1852 * port reservation, and preventing it from being returned by inpcb lookups. 1853 * 1854 * It is used by TCP to mark an inpcb as unused and avoid future packet 1855 * delivery or event notification when a socket remains open but TCP has 1856 * closed. This might occur as a result of a shutdown()-initiated TCP close 1857 * or a RST on the wire, and allows the port binding to be reused while still 1858 * maintaining the invariant that so_pcb always points to a valid inpcb until 1859 * in_pcbdetach(). 1860 * 1861 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1862 * in_pcbnotifyall() and in_pcbpurgeif0()? 1863 */ 1864 void 1865 in_pcbdrop(struct inpcb *inp) 1866 { 1867 1868 INP_WLOCK_ASSERT(inp); 1869 #ifdef INVARIANTS 1870 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1871 MPASS(inp->inp_refcount > 1); 1872 #endif 1873 1874 inp->inp_flags |= INP_DROPPED; 1875 if (inp->inp_flags & INP_INHASHLIST) 1876 in_pcbremhash(inp); 1877 } 1878 1879 #ifdef INET 1880 /* 1881 * Common routines to return the socket addresses associated with inpcbs. 1882 */ 1883 struct sockaddr * 1884 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1885 { 1886 struct sockaddr_in *sin; 1887 1888 sin = malloc(sizeof *sin, M_SONAME, 1889 M_WAITOK | M_ZERO); 1890 sin->sin_family = AF_INET; 1891 sin->sin_len = sizeof(*sin); 1892 sin->sin_addr = *addr_p; 1893 sin->sin_port = port; 1894 1895 return (struct sockaddr *)sin; 1896 } 1897 1898 int 1899 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1900 { 1901 struct inpcb *inp; 1902 struct in_addr addr; 1903 in_port_t port; 1904 1905 inp = sotoinpcb(so); 1906 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1907 1908 INP_RLOCK(inp); 1909 port = inp->inp_lport; 1910 addr = inp->inp_laddr; 1911 INP_RUNLOCK(inp); 1912 1913 *nam = in_sockaddr(port, &addr); 1914 return 0; 1915 } 1916 1917 int 1918 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1919 { 1920 struct inpcb *inp; 1921 struct in_addr addr; 1922 in_port_t port; 1923 1924 inp = sotoinpcb(so); 1925 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1926 1927 INP_RLOCK(inp); 1928 port = inp->inp_fport; 1929 addr = inp->inp_faddr; 1930 INP_RUNLOCK(inp); 1931 1932 *nam = in_sockaddr(port, &addr); 1933 return 0; 1934 } 1935 1936 void 1937 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1938 struct inpcb *(*notify)(struct inpcb *, int)) 1939 { 1940 struct inpcb *inp, *inp_temp; 1941 1942 INP_INFO_WLOCK(pcbinfo); 1943 CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) { 1944 INP_WLOCK(inp); 1945 #ifdef INET6 1946 if ((inp->inp_vflag & INP_IPV4) == 0) { 1947 INP_WUNLOCK(inp); 1948 continue; 1949 } 1950 #endif 1951 if (inp->inp_faddr.s_addr != faddr.s_addr || 1952 inp->inp_socket == NULL) { 1953 INP_WUNLOCK(inp); 1954 continue; 1955 } 1956 if ((*notify)(inp, errno)) 1957 INP_WUNLOCK(inp); 1958 } 1959 INP_INFO_WUNLOCK(pcbinfo); 1960 } 1961 1962 static bool 1963 inp_v4_multi_match(const struct inpcb *inp, void *v __unused) 1964 { 1965 1966 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL) 1967 return (true); 1968 else 1969 return (false); 1970 } 1971 1972 void 1973 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1974 { 1975 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB, 1976 inp_v4_multi_match, NULL); 1977 struct inpcb *inp; 1978 struct in_multi *inm; 1979 struct in_mfilter *imf; 1980 struct ip_moptions *imo; 1981 1982 IN_MULTI_LOCK_ASSERT(); 1983 1984 while ((inp = inp_next(&inpi)) != NULL) { 1985 INP_WLOCK_ASSERT(inp); 1986 1987 imo = inp->inp_moptions; 1988 /* 1989 * Unselect the outgoing interface if it is being 1990 * detached. 1991 */ 1992 if (imo->imo_multicast_ifp == ifp) 1993 imo->imo_multicast_ifp = NULL; 1994 1995 /* 1996 * Drop multicast group membership if we joined 1997 * through the interface being detached. 1998 * 1999 * XXX This can all be deferred to an epoch_call 2000 */ 2001 restart: 2002 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 2003 if ((inm = imf->imf_inm) == NULL) 2004 continue; 2005 if (inm->inm_ifp != ifp) 2006 continue; 2007 ip_mfilter_remove(&imo->imo_head, imf); 2008 in_leavegroup_locked(inm, NULL); 2009 ip_mfilter_free(imf); 2010 goto restart; 2011 } 2012 } 2013 } 2014 2015 /* 2016 * Lookup a PCB based on the local address and port. Caller must hold the 2017 * hash lock. No inpcb locks or references are acquired. 2018 */ 2019 #define INP_LOOKUP_MAPPED_PCB_COST 3 2020 struct inpcb * 2021 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 2022 u_short lport, int lookupflags, struct ucred *cred) 2023 { 2024 struct inpcb *inp; 2025 #ifdef INET6 2026 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 2027 #else 2028 int matchwild = 3; 2029 #endif 2030 int wildcard; 2031 2032 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2033 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2034 INP_HASH_LOCK_ASSERT(pcbinfo); 2035 2036 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 2037 struct inpcbhead *head; 2038 /* 2039 * Look for an unconnected (wildcard foreign addr) PCB that 2040 * matches the local address and port we're looking for. 2041 */ 2042 head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport, 2043 pcbinfo->ipi_hashmask)]; 2044 CK_LIST_FOREACH(inp, head, inp_hash) { 2045 #ifdef INET6 2046 /* XXX inp locking */ 2047 if ((inp->inp_vflag & INP_IPV4) == 0) 2048 continue; 2049 #endif 2050 if (inp->inp_faddr.s_addr == INADDR_ANY && 2051 inp->inp_laddr.s_addr == laddr.s_addr && 2052 inp->inp_lport == lport) { 2053 /* 2054 * Found? 2055 */ 2056 if (prison_equal_ip4(cred->cr_prison, 2057 inp->inp_cred->cr_prison)) 2058 return (inp); 2059 } 2060 } 2061 /* 2062 * Not found. 2063 */ 2064 return (NULL); 2065 } else { 2066 struct inpcbporthead *porthash; 2067 struct inpcbport *phd; 2068 struct inpcb *match = NULL; 2069 /* 2070 * Best fit PCB lookup. 2071 * 2072 * First see if this local port is in use by looking on the 2073 * port hash list. 2074 */ 2075 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 2076 pcbinfo->ipi_porthashmask)]; 2077 CK_LIST_FOREACH(phd, porthash, phd_hash) { 2078 if (phd->phd_port == lport) 2079 break; 2080 } 2081 if (phd != NULL) { 2082 /* 2083 * Port is in use by one or more PCBs. Look for best 2084 * fit. 2085 */ 2086 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 2087 wildcard = 0; 2088 if (!prison_equal_ip4(inp->inp_cred->cr_prison, 2089 cred->cr_prison)) 2090 continue; 2091 #ifdef INET6 2092 /* XXX inp locking */ 2093 if ((inp->inp_vflag & INP_IPV4) == 0) 2094 continue; 2095 /* 2096 * We never select the PCB that has 2097 * INP_IPV6 flag and is bound to :: if 2098 * we have another PCB which is bound 2099 * to 0.0.0.0. If a PCB has the 2100 * INP_IPV6 flag, then we set its cost 2101 * higher than IPv4 only PCBs. 2102 * 2103 * Note that the case only happens 2104 * when a socket is bound to ::, under 2105 * the condition that the use of the 2106 * mapped address is allowed. 2107 */ 2108 if ((inp->inp_vflag & INP_IPV6) != 0) 2109 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 2110 #endif 2111 if (inp->inp_faddr.s_addr != INADDR_ANY) 2112 wildcard++; 2113 if (inp->inp_laddr.s_addr != INADDR_ANY) { 2114 if (laddr.s_addr == INADDR_ANY) 2115 wildcard++; 2116 else if (inp->inp_laddr.s_addr != laddr.s_addr) 2117 continue; 2118 } else { 2119 if (laddr.s_addr != INADDR_ANY) 2120 wildcard++; 2121 } 2122 if (wildcard < matchwild) { 2123 match = inp; 2124 matchwild = wildcard; 2125 if (matchwild == 0) 2126 break; 2127 } 2128 } 2129 } 2130 return (match); 2131 } 2132 } 2133 #undef INP_LOOKUP_MAPPED_PCB_COST 2134 2135 static bool 2136 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain) 2137 { 2138 return (domain == M_NODOM || domain == grp->il_numa_domain); 2139 } 2140 2141 static struct inpcb * 2142 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 2143 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr, 2144 uint16_t lport, int domain) 2145 { 2146 const struct inpcblbgrouphead *hdr; 2147 struct inpcblbgroup *grp; 2148 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild; 2149 2150 INP_HASH_LOCK_ASSERT(pcbinfo); 2151 2152 hdr = &pcbinfo->ipi_lbgrouphashbase[ 2153 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 2154 2155 /* 2156 * Search for an LB group match based on the following criteria: 2157 * - prefer jailed groups to non-jailed groups 2158 * - prefer exact source address matches to wildcard matches 2159 * - prefer groups bound to the specified NUMA domain 2160 */ 2161 jail_exact = jail_wild = local_exact = local_wild = NULL; 2162 CK_LIST_FOREACH(grp, hdr, il_list) { 2163 bool injail; 2164 2165 #ifdef INET6 2166 if (!(grp->il_vflag & INP_IPV4)) 2167 continue; 2168 #endif 2169 if (grp->il_lport != lport) 2170 continue; 2171 2172 injail = prison_flag(grp->il_cred, PR_IP4) != 0; 2173 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison, 2174 laddr) != 0) 2175 continue; 2176 2177 if (grp->il_laddr.s_addr == laddr->s_addr) { 2178 if (injail) { 2179 jail_exact = grp; 2180 if (in_pcblookup_lb_numa_match(grp, domain)) 2181 /* This is a perfect match. */ 2182 goto out; 2183 } else if (local_exact == NULL || 2184 in_pcblookup_lb_numa_match(grp, domain)) { 2185 local_exact = grp; 2186 } 2187 } else if (grp->il_laddr.s_addr == INADDR_ANY) { 2188 if (injail) { 2189 if (jail_wild == NULL || 2190 in_pcblookup_lb_numa_match(grp, domain)) 2191 jail_wild = grp; 2192 } else if (local_wild == NULL || 2193 in_pcblookup_lb_numa_match(grp, domain)) { 2194 local_wild = grp; 2195 } 2196 } 2197 } 2198 2199 if (jail_exact != NULL) 2200 grp = jail_exact; 2201 else if (jail_wild != NULL) 2202 grp = jail_wild; 2203 else if (local_exact != NULL) 2204 grp = local_exact; 2205 else 2206 grp = local_wild; 2207 if (grp == NULL) 2208 return (NULL); 2209 out: 2210 return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) % 2211 grp->il_inpcnt]); 2212 } 2213 2214 static struct inpcb * 2215 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2216 u_short fport, struct in_addr laddr, u_short lport) 2217 { 2218 struct inpcbhead *head; 2219 struct inpcb *inp, *match; 2220 2221 INP_HASH_LOCK_ASSERT(pcbinfo); 2222 2223 match = NULL; 2224 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&faddr, lport, fport, 2225 pcbinfo->ipi_hashmask)]; 2226 CK_LIST_FOREACH(inp, head, inp_hash) { 2227 #ifdef INET6 2228 /* XXX inp locking */ 2229 if ((inp->inp_vflag & INP_IPV4) == 0) 2230 continue; 2231 #endif 2232 if (inp->inp_faddr.s_addr == faddr.s_addr && 2233 inp->inp_laddr.s_addr == laddr.s_addr && 2234 inp->inp_fport == fport && 2235 inp->inp_lport == lport) 2236 return (inp); 2237 } 2238 return (match); 2239 } 2240 2241 static struct inpcb * 2242 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2243 u_short fport, struct in_addr laddr, u_short lport) 2244 { 2245 struct inpcbhead *head; 2246 struct inpcb *inp, *local_wild, *local_exact, *jail_wild; 2247 #ifdef INET6 2248 struct inpcb *local_wild_mapped; 2249 #endif 2250 2251 INP_HASH_LOCK_ASSERT(pcbinfo); 2252 2253 /* 2254 * Order of socket selection - we always prefer jails. 2255 * 1. jailed, non-wild. 2256 * 2. jailed, wild. 2257 * 3. non-jailed, non-wild. 2258 * 4. non-jailed, wild. 2259 */ 2260 head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport, 2261 pcbinfo->ipi_hashmask)]; 2262 local_wild = local_exact = jail_wild = NULL; 2263 #ifdef INET6 2264 local_wild_mapped = NULL; 2265 #endif 2266 CK_LIST_FOREACH(inp, head, inp_hash) { 2267 bool injail; 2268 2269 #ifdef INET6 2270 /* XXX inp locking */ 2271 if ((inp->inp_vflag & INP_IPV4) == 0) 2272 continue; 2273 #endif 2274 if (inp->inp_faddr.s_addr != INADDR_ANY || 2275 inp->inp_lport != lport) 2276 continue; 2277 2278 injail = prison_flag(inp->inp_cred, PR_IP4) != 0; 2279 if (injail) { 2280 if (prison_check_ip4_locked(inp->inp_cred->cr_prison, 2281 &laddr) != 0) 2282 continue; 2283 } else { 2284 if (local_exact != NULL) 2285 continue; 2286 } 2287 2288 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2289 if (injail) 2290 return (inp); 2291 local_exact = inp; 2292 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2293 #ifdef INET6 2294 /* XXX inp locking, NULL check */ 2295 if (inp->inp_vflag & INP_IPV6PROTO) 2296 local_wild_mapped = inp; 2297 else 2298 #endif 2299 if (injail) 2300 jail_wild = inp; 2301 else 2302 local_wild = inp; 2303 } 2304 } 2305 if (jail_wild != NULL) 2306 return (jail_wild); 2307 if (local_exact != NULL) 2308 return (local_exact); 2309 if (local_wild != NULL) 2310 return (local_wild); 2311 #ifdef INET6 2312 if (local_wild_mapped != NULL) 2313 return (local_wild_mapped); 2314 #endif 2315 return (NULL); 2316 } 2317 2318 /* 2319 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2320 * that the caller has either locked the hash list, which usually happens 2321 * for bind(2) operations, or is in SMR section, which happens when sorting 2322 * out incoming packets. 2323 */ 2324 static struct inpcb * 2325 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2326 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2327 uint8_t numa_domain) 2328 { 2329 struct inpcb *inp; 2330 const u_short fport = fport_arg, lport = lport_arg; 2331 2332 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0, 2333 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2334 KASSERT(faddr.s_addr != INADDR_ANY, 2335 ("%s: invalid foreign address", __func__)); 2336 KASSERT(laddr.s_addr != INADDR_ANY, 2337 ("%s: invalid local address", __func__)); 2338 INP_HASH_LOCK_ASSERT(pcbinfo); 2339 2340 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2341 if (inp != NULL) 2342 return (inp); 2343 2344 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2345 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, &laddr, 2346 lport, numa_domain); 2347 if (inp == NULL) { 2348 inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr, 2349 fport, laddr, lport); 2350 } 2351 } 2352 2353 return (inp); 2354 } 2355 2356 static struct inpcb * 2357 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2358 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2359 uint8_t numa_domain) 2360 { 2361 struct inpcb *inp; 2362 2363 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2364 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2365 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2366 ("%s: LOCKPCB not set", __func__)); 2367 2368 smr_enter(pcbinfo->ipi_smr); 2369 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2370 lookupflags & INPLOOKUP_WILDCARD, numa_domain); 2371 if (inp != NULL) { 2372 if (__predict_false(inp_smr_lock(inp, 2373 (lookupflags & INPLOOKUP_LOCKMASK)) == false)) 2374 inp = NULL; 2375 } else 2376 smr_exit(pcbinfo->ipi_smr); 2377 2378 return (inp); 2379 } 2380 2381 /* 2382 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2383 * from which a pre-calculated hash value may be extracted. 2384 */ 2385 struct inpcb * 2386 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2387 struct in_addr laddr, u_int lport, int lookupflags, 2388 struct ifnet *ifp __unused) 2389 { 2390 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2391 lookupflags, M_NODOM)); 2392 } 2393 2394 struct inpcb * 2395 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2396 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2397 struct ifnet *ifp __unused, struct mbuf *m) 2398 { 2399 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2400 lookupflags, m->m_pkthdr.numa_domain)); 2401 } 2402 #endif /* INET */ 2403 2404 /* 2405 * Insert PCB onto various hash lists. 2406 */ 2407 int 2408 in_pcbinshash(struct inpcb *inp) 2409 { 2410 struct inpcbhead *pcbhash; 2411 struct inpcbporthead *pcbporthash; 2412 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2413 struct inpcbport *phd; 2414 2415 INP_WLOCK_ASSERT(inp); 2416 INP_HASH_WLOCK_ASSERT(pcbinfo); 2417 2418 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2419 ("in_pcbinshash: INP_INHASHLIST")); 2420 2421 #ifdef INET6 2422 if (inp->inp_vflag & INP_IPV6) 2423 pcbhash = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr, 2424 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2425 else 2426 #endif 2427 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr, 2428 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2429 2430 pcbporthash = &pcbinfo->ipi_porthashbase[ 2431 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2432 2433 /* 2434 * Add entry to load balance group. 2435 * Only do this if SO_REUSEPORT_LB is set. 2436 */ 2437 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) { 2438 int error = in_pcbinslbgrouphash(inp, M_NODOM); 2439 if (error != 0) 2440 return (error); 2441 } 2442 2443 /* 2444 * Go through port list and look for a head for this lport. 2445 */ 2446 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2447 if (phd->phd_port == inp->inp_lport) 2448 break; 2449 } 2450 2451 /* 2452 * If none exists, malloc one and tack it on. 2453 */ 2454 if (phd == NULL) { 2455 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT); 2456 if (phd == NULL) { 2457 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 2458 in_pcbremlbgrouphash(inp); 2459 return (ENOMEM); 2460 } 2461 phd->phd_port = inp->inp_lport; 2462 CK_LIST_INIT(&phd->phd_pcblist); 2463 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2464 } 2465 inp->inp_phd = phd; 2466 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2467 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2468 inp->inp_flags |= INP_INHASHLIST; 2469 2470 return (0); 2471 } 2472 2473 static void 2474 in_pcbremhash(struct inpcb *inp) 2475 { 2476 struct inpcbport *phd = inp->inp_phd; 2477 2478 INP_WLOCK_ASSERT(inp); 2479 MPASS(inp->inp_flags & INP_INHASHLIST); 2480 2481 INP_HASH_WLOCK(inp->inp_pcbinfo); 2482 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 2483 in_pcbremlbgrouphash(inp); 2484 CK_LIST_REMOVE(inp, inp_hash); 2485 CK_LIST_REMOVE(inp, inp_portlist); 2486 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2487 CK_LIST_REMOVE(phd, phd_hash); 2488 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd); 2489 } 2490 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 2491 inp->inp_flags &= ~INP_INHASHLIST; 2492 } 2493 2494 /* 2495 * Move PCB to the proper hash bucket when { faddr, fport } have been 2496 * changed. NOTE: This does not handle the case of the lport changing (the 2497 * hashed port list would have to be updated as well), so the lport must 2498 * not change after in_pcbinshash() has been called. 2499 * 2500 * XXXGL: a race between this function and SMR-protected hash iterator 2501 * will lead to iterator traversing a possibly wrong hash list. However, 2502 * this race should have been here since change from rwlock to epoch. 2503 */ 2504 void 2505 in_pcbrehash(struct inpcb *inp) 2506 { 2507 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2508 struct inpcbhead *head; 2509 2510 INP_WLOCK_ASSERT(inp); 2511 INP_HASH_WLOCK_ASSERT(pcbinfo); 2512 2513 KASSERT(inp->inp_flags & INP_INHASHLIST, 2514 ("in_pcbrehash: !INP_INHASHLIST")); 2515 2516 #ifdef INET6 2517 if (inp->inp_vflag & INP_IPV6) 2518 head = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr, 2519 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2520 else 2521 #endif 2522 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr, 2523 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2524 2525 CK_LIST_REMOVE(inp, inp_hash); 2526 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2527 } 2528 2529 /* 2530 * Check for alternatives when higher level complains 2531 * about service problems. For now, invalidate cached 2532 * routing information. If the route was created dynamically 2533 * (by a redirect), time to try a default gateway again. 2534 */ 2535 void 2536 in_losing(struct inpcb *inp) 2537 { 2538 2539 RO_INVALIDATE_CACHE(&inp->inp_route); 2540 return; 2541 } 2542 2543 /* 2544 * A set label operation has occurred at the socket layer, propagate the 2545 * label change into the in_pcb for the socket. 2546 */ 2547 void 2548 in_pcbsosetlabel(struct socket *so) 2549 { 2550 #ifdef MAC 2551 struct inpcb *inp; 2552 2553 inp = sotoinpcb(so); 2554 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2555 2556 INP_WLOCK(inp); 2557 SOCK_LOCK(so); 2558 mac_inpcb_sosetlabel(so, inp); 2559 SOCK_UNLOCK(so); 2560 INP_WUNLOCK(inp); 2561 #endif 2562 } 2563 2564 void 2565 inp_wlock(struct inpcb *inp) 2566 { 2567 2568 INP_WLOCK(inp); 2569 } 2570 2571 void 2572 inp_wunlock(struct inpcb *inp) 2573 { 2574 2575 INP_WUNLOCK(inp); 2576 } 2577 2578 void 2579 inp_rlock(struct inpcb *inp) 2580 { 2581 2582 INP_RLOCK(inp); 2583 } 2584 2585 void 2586 inp_runlock(struct inpcb *inp) 2587 { 2588 2589 INP_RUNLOCK(inp); 2590 } 2591 2592 #ifdef INVARIANT_SUPPORT 2593 void 2594 inp_lock_assert(struct inpcb *inp) 2595 { 2596 2597 INP_WLOCK_ASSERT(inp); 2598 } 2599 2600 void 2601 inp_unlock_assert(struct inpcb *inp) 2602 { 2603 2604 INP_UNLOCK_ASSERT(inp); 2605 } 2606 #endif 2607 2608 void 2609 inp_apply_all(struct inpcbinfo *pcbinfo, 2610 void (*func)(struct inpcb *, void *), void *arg) 2611 { 2612 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2613 INPLOOKUP_WLOCKPCB); 2614 struct inpcb *inp; 2615 2616 while ((inp = inp_next(&inpi)) != NULL) 2617 func(inp, arg); 2618 } 2619 2620 struct socket * 2621 inp_inpcbtosocket(struct inpcb *inp) 2622 { 2623 2624 INP_WLOCK_ASSERT(inp); 2625 return (inp->inp_socket); 2626 } 2627 2628 struct tcpcb * 2629 inp_inpcbtotcpcb(struct inpcb *inp) 2630 { 2631 2632 INP_WLOCK_ASSERT(inp); 2633 return ((struct tcpcb *)inp->inp_ppcb); 2634 } 2635 2636 int 2637 inp_ip_tos_get(const struct inpcb *inp) 2638 { 2639 2640 return (inp->inp_ip_tos); 2641 } 2642 2643 void 2644 inp_ip_tos_set(struct inpcb *inp, int val) 2645 { 2646 2647 inp->inp_ip_tos = val; 2648 } 2649 2650 void 2651 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2652 uint32_t *faddr, uint16_t *fp) 2653 { 2654 2655 INP_LOCK_ASSERT(inp); 2656 *laddr = inp->inp_laddr.s_addr; 2657 *faddr = inp->inp_faddr.s_addr; 2658 *lp = inp->inp_lport; 2659 *fp = inp->inp_fport; 2660 } 2661 2662 struct inpcb * 2663 so_sotoinpcb(struct socket *so) 2664 { 2665 2666 return (sotoinpcb(so)); 2667 } 2668 2669 /* 2670 * Create an external-format (``xinpcb'') structure using the information in 2671 * the kernel-format in_pcb structure pointed to by inp. This is done to 2672 * reduce the spew of irrelevant information over this interface, to isolate 2673 * user code from changes in the kernel structure, and potentially to provide 2674 * information-hiding if we decide that some of this information should be 2675 * hidden from users. 2676 */ 2677 void 2678 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2679 { 2680 2681 bzero(xi, sizeof(*xi)); 2682 xi->xi_len = sizeof(struct xinpcb); 2683 if (inp->inp_socket) 2684 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2685 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2686 xi->inp_gencnt = inp->inp_gencnt; 2687 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2688 xi->inp_flow = inp->inp_flow; 2689 xi->inp_flowid = inp->inp_flowid; 2690 xi->inp_flowtype = inp->inp_flowtype; 2691 xi->inp_flags = inp->inp_flags; 2692 xi->inp_flags2 = inp->inp_flags2; 2693 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2694 xi->in6p_cksum = inp->in6p_cksum; 2695 xi->in6p_hops = inp->in6p_hops; 2696 xi->inp_ip_tos = inp->inp_ip_tos; 2697 xi->inp_vflag = inp->inp_vflag; 2698 xi->inp_ip_ttl = inp->inp_ip_ttl; 2699 xi->inp_ip_p = inp->inp_ip_p; 2700 xi->inp_ip_minttl = inp->inp_ip_minttl; 2701 } 2702 2703 int 2704 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo, 2705 int (*ctloutput_set)(struct inpcb *, struct sockopt *)) 2706 { 2707 struct sockopt sopt; 2708 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2709 INPLOOKUP_WLOCKPCB); 2710 struct inpcb *inp; 2711 struct sockopt_parameters *params; 2712 struct socket *so; 2713 int error; 2714 char buf[1024]; 2715 2716 if (req->oldptr != NULL || req->oldlen != 0) 2717 return (EINVAL); 2718 if (req->newptr == NULL) 2719 return (EPERM); 2720 if (req->newlen > sizeof(buf)) 2721 return (ENOMEM); 2722 error = SYSCTL_IN(req, buf, req->newlen); 2723 if (error != 0) 2724 return (error); 2725 if (req->newlen < sizeof(struct sockopt_parameters)) 2726 return (EINVAL); 2727 params = (struct sockopt_parameters *)buf; 2728 sopt.sopt_level = params->sop_level; 2729 sopt.sopt_name = params->sop_optname; 2730 sopt.sopt_dir = SOPT_SET; 2731 sopt.sopt_val = params->sop_optval; 2732 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters); 2733 sopt.sopt_td = NULL; 2734 #ifdef INET6 2735 if (params->sop_inc.inc_flags & INC_ISIPV6) { 2736 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr)) 2737 params->sop_inc.inc6_laddr.s6_addr16[1] = 2738 htons(params->sop_inc.inc6_zoneid & 0xffff); 2739 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr)) 2740 params->sop_inc.inc6_faddr.s6_addr16[1] = 2741 htons(params->sop_inc.inc6_zoneid & 0xffff); 2742 } 2743 #endif 2744 if (params->sop_inc.inc_lport != htons(0)) { 2745 if (params->sop_inc.inc_fport == htons(0)) 2746 inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport, 2747 pcbinfo->ipi_hashmask); 2748 else 2749 #ifdef INET6 2750 if (params->sop_inc.inc_flags & INC_ISIPV6) 2751 inpi.hash = INP6_PCBHASH( 2752 ¶ms->sop_inc.inc6_faddr, 2753 params->sop_inc.inc_lport, 2754 params->sop_inc.inc_fport, 2755 pcbinfo->ipi_hashmask); 2756 else 2757 #endif 2758 inpi.hash = INP_PCBHASH( 2759 ¶ms->sop_inc.inc_faddr, 2760 params->sop_inc.inc_lport, 2761 params->sop_inc.inc_fport, 2762 pcbinfo->ipi_hashmask); 2763 } 2764 while ((inp = inp_next(&inpi)) != NULL) 2765 if (inp->inp_gencnt == params->sop_id) { 2766 if (inp->inp_flags & INP_DROPPED) { 2767 INP_WUNLOCK(inp); 2768 return (ECONNRESET); 2769 } 2770 so = inp->inp_socket; 2771 KASSERT(so != NULL, ("inp_socket == NULL")); 2772 soref(so); 2773 error = (*ctloutput_set)(inp, &sopt); 2774 sorele(so); 2775 break; 2776 } 2777 if (inp == NULL) 2778 error = ESRCH; 2779 return (error); 2780 } 2781 2782 #ifdef DDB 2783 static void 2784 db_print_indent(int indent) 2785 { 2786 int i; 2787 2788 for (i = 0; i < indent; i++) 2789 db_printf(" "); 2790 } 2791 2792 static void 2793 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2794 { 2795 char faddr_str[48], laddr_str[48]; 2796 2797 db_print_indent(indent); 2798 db_printf("%s at %p\n", name, inc); 2799 2800 indent += 2; 2801 2802 #ifdef INET6 2803 if (inc->inc_flags & INC_ISIPV6) { 2804 /* IPv6. */ 2805 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2806 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2807 } else 2808 #endif 2809 { 2810 /* IPv4. */ 2811 inet_ntoa_r(inc->inc_laddr, laddr_str); 2812 inet_ntoa_r(inc->inc_faddr, faddr_str); 2813 } 2814 db_print_indent(indent); 2815 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2816 ntohs(inc->inc_lport)); 2817 db_print_indent(indent); 2818 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2819 ntohs(inc->inc_fport)); 2820 } 2821 2822 static void 2823 db_print_inpflags(int inp_flags) 2824 { 2825 int comma; 2826 2827 comma = 0; 2828 if (inp_flags & INP_RECVOPTS) { 2829 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2830 comma = 1; 2831 } 2832 if (inp_flags & INP_RECVRETOPTS) { 2833 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2834 comma = 1; 2835 } 2836 if (inp_flags & INP_RECVDSTADDR) { 2837 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2838 comma = 1; 2839 } 2840 if (inp_flags & INP_ORIGDSTADDR) { 2841 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2842 comma = 1; 2843 } 2844 if (inp_flags & INP_HDRINCL) { 2845 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2846 comma = 1; 2847 } 2848 if (inp_flags & INP_HIGHPORT) { 2849 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2850 comma = 1; 2851 } 2852 if (inp_flags & INP_LOWPORT) { 2853 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 2854 comma = 1; 2855 } 2856 if (inp_flags & INP_ANONPORT) { 2857 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 2858 comma = 1; 2859 } 2860 if (inp_flags & INP_RECVIF) { 2861 db_printf("%sINP_RECVIF", comma ? ", " : ""); 2862 comma = 1; 2863 } 2864 if (inp_flags & INP_MTUDISC) { 2865 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 2866 comma = 1; 2867 } 2868 if (inp_flags & INP_RECVTTL) { 2869 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 2870 comma = 1; 2871 } 2872 if (inp_flags & INP_DONTFRAG) { 2873 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 2874 comma = 1; 2875 } 2876 if (inp_flags & INP_RECVTOS) { 2877 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 2878 comma = 1; 2879 } 2880 if (inp_flags & IN6P_IPV6_V6ONLY) { 2881 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 2882 comma = 1; 2883 } 2884 if (inp_flags & IN6P_PKTINFO) { 2885 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 2886 comma = 1; 2887 } 2888 if (inp_flags & IN6P_HOPLIMIT) { 2889 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 2890 comma = 1; 2891 } 2892 if (inp_flags & IN6P_HOPOPTS) { 2893 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 2894 comma = 1; 2895 } 2896 if (inp_flags & IN6P_DSTOPTS) { 2897 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 2898 comma = 1; 2899 } 2900 if (inp_flags & IN6P_RTHDR) { 2901 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 2902 comma = 1; 2903 } 2904 if (inp_flags & IN6P_RTHDRDSTOPTS) { 2905 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 2906 comma = 1; 2907 } 2908 if (inp_flags & IN6P_TCLASS) { 2909 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 2910 comma = 1; 2911 } 2912 if (inp_flags & IN6P_AUTOFLOWLABEL) { 2913 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 2914 comma = 1; 2915 } 2916 if (inp_flags & INP_ONESBCAST) { 2917 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 2918 comma = 1; 2919 } 2920 if (inp_flags & INP_DROPPED) { 2921 db_printf("%sINP_DROPPED", comma ? ", " : ""); 2922 comma = 1; 2923 } 2924 if (inp_flags & INP_SOCKREF) { 2925 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 2926 comma = 1; 2927 } 2928 if (inp_flags & IN6P_RFC2292) { 2929 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 2930 comma = 1; 2931 } 2932 if (inp_flags & IN6P_MTU) { 2933 db_printf("IN6P_MTU%s", comma ? ", " : ""); 2934 comma = 1; 2935 } 2936 } 2937 2938 static void 2939 db_print_inpvflag(u_char inp_vflag) 2940 { 2941 int comma; 2942 2943 comma = 0; 2944 if (inp_vflag & INP_IPV4) { 2945 db_printf("%sINP_IPV4", comma ? ", " : ""); 2946 comma = 1; 2947 } 2948 if (inp_vflag & INP_IPV6) { 2949 db_printf("%sINP_IPV6", comma ? ", " : ""); 2950 comma = 1; 2951 } 2952 if (inp_vflag & INP_IPV6PROTO) { 2953 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 2954 comma = 1; 2955 } 2956 } 2957 2958 static void 2959 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 2960 { 2961 2962 db_print_indent(indent); 2963 db_printf("%s at %p\n", name, inp); 2964 2965 indent += 2; 2966 2967 db_print_indent(indent); 2968 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 2969 2970 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 2971 2972 db_print_indent(indent); 2973 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 2974 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 2975 2976 db_print_indent(indent); 2977 db_printf("inp_label: %p inp_flags: 0x%x (", 2978 inp->inp_label, inp->inp_flags); 2979 db_print_inpflags(inp->inp_flags); 2980 db_printf(")\n"); 2981 2982 db_print_indent(indent); 2983 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 2984 inp->inp_vflag); 2985 db_print_inpvflag(inp->inp_vflag); 2986 db_printf(")\n"); 2987 2988 db_print_indent(indent); 2989 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 2990 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 2991 2992 db_print_indent(indent); 2993 #ifdef INET6 2994 if (inp->inp_vflag & INP_IPV6) { 2995 db_printf("in6p_options: %p in6p_outputopts: %p " 2996 "in6p_moptions: %p\n", inp->in6p_options, 2997 inp->in6p_outputopts, inp->in6p_moptions); 2998 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 2999 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3000 inp->in6p_hops); 3001 } else 3002 #endif 3003 { 3004 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3005 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3006 inp->inp_options, inp->inp_moptions); 3007 } 3008 3009 db_print_indent(indent); 3010 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3011 (uintmax_t)inp->inp_gencnt); 3012 } 3013 3014 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3015 { 3016 struct inpcb *inp; 3017 3018 if (!have_addr) { 3019 db_printf("usage: show inpcb <addr>\n"); 3020 return; 3021 } 3022 inp = (struct inpcb *)addr; 3023 3024 db_print_inpcb(inp, "inpcb", 0); 3025 } 3026 #endif /* DDB */ 3027 3028 #ifdef RATELIMIT 3029 /* 3030 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3031 * if any. 3032 */ 3033 int 3034 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3035 { 3036 union if_snd_tag_modify_params params = { 3037 .rate_limit.max_rate = max_pacing_rate, 3038 .rate_limit.flags = M_NOWAIT, 3039 }; 3040 struct m_snd_tag *mst; 3041 int error; 3042 3043 mst = inp->inp_snd_tag; 3044 if (mst == NULL) 3045 return (EINVAL); 3046 3047 if (mst->sw->snd_tag_modify == NULL) { 3048 error = EOPNOTSUPP; 3049 } else { 3050 error = mst->sw->snd_tag_modify(mst, ¶ms); 3051 } 3052 return (error); 3053 } 3054 3055 /* 3056 * Query existing TX rate limit based on the existing 3057 * "inp->inp_snd_tag", if any. 3058 */ 3059 int 3060 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3061 { 3062 union if_snd_tag_query_params params = { }; 3063 struct m_snd_tag *mst; 3064 int error; 3065 3066 mst = inp->inp_snd_tag; 3067 if (mst == NULL) 3068 return (EINVAL); 3069 3070 if (mst->sw->snd_tag_query == NULL) { 3071 error = EOPNOTSUPP; 3072 } else { 3073 error = mst->sw->snd_tag_query(mst, ¶ms); 3074 if (error == 0 && p_max_pacing_rate != NULL) 3075 *p_max_pacing_rate = params.rate_limit.max_rate; 3076 } 3077 return (error); 3078 } 3079 3080 /* 3081 * Query existing TX queue level based on the existing 3082 * "inp->inp_snd_tag", if any. 3083 */ 3084 int 3085 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3086 { 3087 union if_snd_tag_query_params params = { }; 3088 struct m_snd_tag *mst; 3089 int error; 3090 3091 mst = inp->inp_snd_tag; 3092 if (mst == NULL) 3093 return (EINVAL); 3094 3095 if (mst->sw->snd_tag_query == NULL) 3096 return (EOPNOTSUPP); 3097 3098 error = mst->sw->snd_tag_query(mst, ¶ms); 3099 if (error == 0 && p_txqueue_level != NULL) 3100 *p_txqueue_level = params.rate_limit.queue_level; 3101 return (error); 3102 } 3103 3104 /* 3105 * Allocate a new TX rate limit send tag from the network interface 3106 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3107 */ 3108 int 3109 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3110 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3111 3112 { 3113 union if_snd_tag_alloc_params params = { 3114 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3115 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3116 .rate_limit.hdr.flowid = flowid, 3117 .rate_limit.hdr.flowtype = flowtype, 3118 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3119 .rate_limit.max_rate = max_pacing_rate, 3120 .rate_limit.flags = M_NOWAIT, 3121 }; 3122 int error; 3123 3124 INP_WLOCK_ASSERT(inp); 3125 3126 /* 3127 * If there is already a send tag, or the INP is being torn 3128 * down, allocating a new send tag is not allowed. Else send 3129 * tags may leak. 3130 */ 3131 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0) 3132 return (EINVAL); 3133 3134 error = m_snd_tag_alloc(ifp, ¶ms, st); 3135 #ifdef INET 3136 if (error == 0) { 3137 counter_u64_add(rate_limit_set_ok, 1); 3138 counter_u64_add(rate_limit_active, 1); 3139 } else if (error != EOPNOTSUPP) 3140 counter_u64_add(rate_limit_alloc_fail, 1); 3141 #endif 3142 return (error); 3143 } 3144 3145 void 3146 in_pcbdetach_tag(struct m_snd_tag *mst) 3147 { 3148 3149 m_snd_tag_rele(mst); 3150 #ifdef INET 3151 counter_u64_add(rate_limit_active, -1); 3152 #endif 3153 } 3154 3155 /* 3156 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3157 * if any: 3158 */ 3159 void 3160 in_pcbdetach_txrtlmt(struct inpcb *inp) 3161 { 3162 struct m_snd_tag *mst; 3163 3164 INP_WLOCK_ASSERT(inp); 3165 3166 mst = inp->inp_snd_tag; 3167 inp->inp_snd_tag = NULL; 3168 3169 if (mst == NULL) 3170 return; 3171 3172 m_snd_tag_rele(mst); 3173 #ifdef INET 3174 counter_u64_add(rate_limit_active, -1); 3175 #endif 3176 } 3177 3178 int 3179 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3180 { 3181 int error; 3182 3183 /* 3184 * If the existing send tag is for the wrong interface due to 3185 * a route change, first drop the existing tag. Set the 3186 * CHANGED flag so that we will keep trying to allocate a new 3187 * tag if we fail to allocate one this time. 3188 */ 3189 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3190 in_pcbdetach_txrtlmt(inp); 3191 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3192 } 3193 3194 /* 3195 * NOTE: When attaching to a network interface a reference is 3196 * made to ensure the network interface doesn't go away until 3197 * all ratelimit connections are gone. The network interface 3198 * pointers compared below represent valid network interfaces, 3199 * except when comparing towards NULL. 3200 */ 3201 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3202 error = 0; 3203 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3204 if (inp->inp_snd_tag != NULL) 3205 in_pcbdetach_txrtlmt(inp); 3206 error = 0; 3207 } else if (inp->inp_snd_tag == NULL) { 3208 /* 3209 * In order to utilize packet pacing with RSS, we need 3210 * to wait until there is a valid RSS hash before we 3211 * can proceed: 3212 */ 3213 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3214 error = EAGAIN; 3215 } else { 3216 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3217 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3218 } 3219 } else { 3220 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3221 } 3222 if (error == 0 || error == EOPNOTSUPP) 3223 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3224 3225 return (error); 3226 } 3227 3228 /* 3229 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3230 * is set in the fast path and will attach/detach/modify the TX rate 3231 * limit send tag based on the socket's so_max_pacing_rate value. 3232 */ 3233 void 3234 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3235 { 3236 struct socket *socket; 3237 uint32_t max_pacing_rate; 3238 bool did_upgrade; 3239 3240 if (inp == NULL) 3241 return; 3242 3243 socket = inp->inp_socket; 3244 if (socket == NULL) 3245 return; 3246 3247 if (!INP_WLOCKED(inp)) { 3248 /* 3249 * NOTE: If the write locking fails, we need to bail 3250 * out and use the non-ratelimited ring for the 3251 * transmit until there is a new chance to get the 3252 * write lock. 3253 */ 3254 if (!INP_TRY_UPGRADE(inp)) 3255 return; 3256 did_upgrade = 1; 3257 } else { 3258 did_upgrade = 0; 3259 } 3260 3261 /* 3262 * NOTE: The so_max_pacing_rate value is read unlocked, 3263 * because atomic updates are not required since the variable 3264 * is checked at every mbuf we send. It is assumed that the 3265 * variable read itself will be atomic. 3266 */ 3267 max_pacing_rate = socket->so_max_pacing_rate; 3268 3269 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3270 3271 if (did_upgrade) 3272 INP_DOWNGRADE(inp); 3273 } 3274 3275 /* 3276 * Track route changes for TX rate limiting. 3277 */ 3278 void 3279 in_pcboutput_eagain(struct inpcb *inp) 3280 { 3281 bool did_upgrade; 3282 3283 if (inp == NULL) 3284 return; 3285 3286 if (inp->inp_snd_tag == NULL) 3287 return; 3288 3289 if (!INP_WLOCKED(inp)) { 3290 /* 3291 * NOTE: If the write locking fails, we need to bail 3292 * out and use the non-ratelimited ring for the 3293 * transmit until there is a new chance to get the 3294 * write lock. 3295 */ 3296 if (!INP_TRY_UPGRADE(inp)) 3297 return; 3298 did_upgrade = 1; 3299 } else { 3300 did_upgrade = 0; 3301 } 3302 3303 /* detach rate limiting */ 3304 in_pcbdetach_txrtlmt(inp); 3305 3306 /* make sure new mbuf send tag allocation is made */ 3307 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3308 3309 if (did_upgrade) 3310 INP_DOWNGRADE(inp); 3311 } 3312 3313 #ifdef INET 3314 static void 3315 rl_init(void *st) 3316 { 3317 rate_limit_new = counter_u64_alloc(M_WAITOK); 3318 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3319 rate_limit_active = counter_u64_alloc(M_WAITOK); 3320 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3321 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3322 } 3323 3324 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3325 #endif 3326 #endif /* RATELIMIT */ 3327