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/proc.h> 62 #include <sys/protosw.h> 63 #include <sys/smp.h> 64 #include <sys/smr.h> 65 #include <sys/socket.h> 66 #include <sys/socketvar.h> 67 #include <sys/sockio.h> 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/refcount.h> 71 #include <sys/jail.h> 72 #include <sys/kernel.h> 73 #include <sys/sysctl.h> 74 75 #ifdef DDB 76 #include <ddb/ddb.h> 77 #endif 78 79 #include <vm/uma.h> 80 #include <vm/vm.h> 81 82 #include <net/if.h> 83 #include <net/if_var.h> 84 #include <net/if_private.h> 85 #include <net/if_types.h> 86 #include <net/if_llatbl.h> 87 #include <net/route.h> 88 #include <net/rss_config.h> 89 #include <net/vnet.h> 90 91 #if defined(INET) || defined(INET6) 92 #include <netinet/in.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_pcb_var.h> 95 #include <netinet/tcp.h> 96 #ifdef INET 97 #include <netinet/in_var.h> 98 #include <netinet/in_fib.h> 99 #endif 100 #include <netinet/ip_var.h> 101 #ifdef INET6 102 #include <netinet/ip6.h> 103 #include <netinet6/in6_pcb.h> 104 #include <netinet6/in6_var.h> 105 #include <netinet6/ip6_var.h> 106 #endif /* INET6 */ 107 #include <net/route/nhop.h> 108 #endif 109 110 #include <netipsec/ipsec_support.h> 111 112 #include <security/mac/mac_framework.h> 113 114 #define INPCBLBGROUP_SIZMIN 8 115 #define INPCBLBGROUP_SIZMAX 256 116 #define INP_FREED 0x00000200 /* See in_pcb.h. */ 117 118 /* 119 * These configure the range of local port addresses assigned to 120 * "unspecified" outgoing connections/packets/whatever. 121 */ 122 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 123 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 124 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 125 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 126 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 127 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 128 129 /* 130 * Reserved ports accessible only to root. There are significant 131 * security considerations that must be accounted for when changing these, 132 * but the security benefits can be great. Please be careful. 133 */ 134 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 135 VNET_DEFINE(int, ipport_reservedlow); 136 137 /* Enable random ephemeral port allocation by default. */ 138 VNET_DEFINE(int, ipport_randomized) = 1; 139 140 #ifdef INET 141 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 142 struct in_addr faddr, u_int fport_arg, 143 struct in_addr laddr, u_int lport_arg, 144 int lookupflags, uint8_t numa_domain); 145 146 #define RANGECHK(var, min, max) \ 147 if ((var) < (min)) { (var) = (min); } \ 148 else if ((var) > (max)) { (var) = (max); } 149 150 static int 151 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 152 { 153 int error; 154 155 error = sysctl_handle_int(oidp, arg1, arg2, req); 156 if (error == 0) { 157 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 158 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 159 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 160 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 161 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 162 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 163 } 164 return (error); 165 } 166 167 #undef RANGECHK 168 169 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, 170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 171 "IP Ports"); 172 173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 174 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 175 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", 176 ""); 177 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 178 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 179 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", 180 ""); 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 183 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", 184 ""); 185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 186 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 187 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", 188 ""); 189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 190 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 191 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", 192 ""); 193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", 196 ""); 197 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 198 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 199 &VNET_NAME(ipport_reservedhigh), 0, ""); 200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 201 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 202 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 203 CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 205 206 #ifdef RATELIMIT 207 counter_u64_t rate_limit_new; 208 counter_u64_t rate_limit_chg; 209 counter_u64_t rate_limit_active; 210 counter_u64_t rate_limit_alloc_fail; 211 counter_u64_t rate_limit_set_ok; 212 213 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 214 "IP Rate Limiting"); 215 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, 216 &rate_limit_active, "Active rate limited connections"); 217 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, 218 &rate_limit_alloc_fail, "Rate limited connection failures"); 219 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, 220 &rate_limit_set_ok, "Rate limited setting succeeded"); 221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD, 222 &rate_limit_new, "Total Rate limit new attempts"); 223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD, 224 &rate_limit_chg, "Total Rate limited change attempts"); 225 #endif /* RATELIMIT */ 226 227 #endif /* INET */ 228 229 VNET_DEFINE(uint32_t, in_pcbhashseed); 230 static void 231 in_pcbhashseed_init(void) 232 { 233 234 V_in_pcbhashseed = arc4random(); 235 } 236 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, 237 in_pcbhashseed_init, 0); 238 239 static void in_pcbremhash(struct inpcb *); 240 241 /* 242 * in_pcb.c: manage the Protocol Control Blocks. 243 * 244 * NOTE: It is assumed that most of these functions will be called with 245 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 246 * functions often modify hash chains or addresses in pcbs. 247 */ 248 249 static struct inpcblbgroup * 250 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred, 251 u_char vflag, uint16_t port, const union in_dependaddr *addr, int size, 252 uint8_t numa_domain) 253 { 254 struct inpcblbgroup *grp; 255 size_t bytes; 256 257 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 258 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 259 if (grp == NULL) 260 return (NULL); 261 grp->il_cred = crhold(cred); 262 grp->il_vflag = vflag; 263 grp->il_lport = port; 264 grp->il_numa_domain = numa_domain; 265 grp->il_dependladdr = *addr; 266 grp->il_inpsiz = size; 267 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 268 return (grp); 269 } 270 271 static void 272 in_pcblbgroup_free_deferred(epoch_context_t ctx) 273 { 274 struct inpcblbgroup *grp; 275 276 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 277 crfree(grp->il_cred); 278 free(grp, M_PCB); 279 } 280 281 static void 282 in_pcblbgroup_free(struct inpcblbgroup *grp) 283 { 284 285 CK_LIST_REMOVE(grp, il_list); 286 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); 287 } 288 289 static struct inpcblbgroup * 290 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 291 struct inpcblbgroup *old_grp, int size) 292 { 293 struct inpcblbgroup *grp; 294 int i; 295 296 grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag, 297 old_grp->il_lport, &old_grp->il_dependladdr, size, 298 old_grp->il_numa_domain); 299 if (grp == NULL) 300 return (NULL); 301 302 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 303 ("invalid new local group size %d and old local group count %d", 304 grp->il_inpsiz, old_grp->il_inpcnt)); 305 306 for (i = 0; i < old_grp->il_inpcnt; ++i) 307 grp->il_inp[i] = old_grp->il_inp[i]; 308 grp->il_inpcnt = old_grp->il_inpcnt; 309 in_pcblbgroup_free(old_grp); 310 return (grp); 311 } 312 313 /* 314 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 315 * and shrink group if possible. 316 */ 317 static void 318 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 319 int i) 320 { 321 struct inpcblbgroup *grp, *new_grp; 322 323 grp = *grpp; 324 for (; i + 1 < grp->il_inpcnt; ++i) 325 grp->il_inp[i] = grp->il_inp[i + 1]; 326 grp->il_inpcnt--; 327 328 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 329 grp->il_inpcnt <= grp->il_inpsiz / 4) { 330 /* Shrink this group. */ 331 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 332 if (new_grp != NULL) 333 *grpp = new_grp; 334 } 335 } 336 337 /* 338 * Add PCB to load balance group for SO_REUSEPORT_LB option. 339 */ 340 static int 341 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain) 342 { 343 const static struct timeval interval = { 60, 0 }; 344 static struct timeval lastprint; 345 struct inpcbinfo *pcbinfo; 346 struct inpcblbgrouphead *hdr; 347 struct inpcblbgroup *grp; 348 uint32_t idx; 349 350 pcbinfo = inp->inp_pcbinfo; 351 352 INP_WLOCK_ASSERT(inp); 353 INP_HASH_WLOCK_ASSERT(pcbinfo); 354 355 #ifdef INET6 356 /* 357 * Don't allow IPv4 mapped INET6 wild socket. 358 */ 359 if ((inp->inp_vflag & INP_IPV4) && 360 inp->inp_laddr.s_addr == INADDR_ANY && 361 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 362 return (0); 363 } 364 #endif 365 366 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 367 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 368 CK_LIST_FOREACH(grp, hdr, il_list) { 369 if (grp->il_cred->cr_prison == inp->inp_cred->cr_prison && 370 grp->il_vflag == inp->inp_vflag && 371 grp->il_lport == inp->inp_lport && 372 grp->il_numa_domain == numa_domain && 373 memcmp(&grp->il_dependladdr, 374 &inp->inp_inc.inc_ie.ie_dependladdr, 375 sizeof(grp->il_dependladdr)) == 0) { 376 break; 377 } 378 } 379 if (grp == NULL) { 380 /* Create new load balance group. */ 381 grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag, 382 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 383 INPCBLBGROUP_SIZMIN, numa_domain); 384 if (grp == NULL) 385 return (ENOBUFS); 386 } else if (grp->il_inpcnt == grp->il_inpsiz) { 387 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 388 if (ratecheck(&lastprint, &interval)) 389 printf("lb group port %d, limit reached\n", 390 ntohs(grp->il_lport)); 391 return (0); 392 } 393 394 /* Expand this local group. */ 395 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 396 if (grp == NULL) 397 return (ENOBUFS); 398 } 399 400 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 401 ("invalid local group size %d and count %d", grp->il_inpsiz, 402 grp->il_inpcnt)); 403 404 grp->il_inp[grp->il_inpcnt] = inp; 405 grp->il_inpcnt++; 406 return (0); 407 } 408 409 /* 410 * Remove PCB from load balance group. 411 */ 412 static void 413 in_pcbremlbgrouphash(struct inpcb *inp) 414 { 415 struct inpcbinfo *pcbinfo; 416 struct inpcblbgrouphead *hdr; 417 struct inpcblbgroup *grp; 418 int i; 419 420 pcbinfo = inp->inp_pcbinfo; 421 422 INP_WLOCK_ASSERT(inp); 423 INP_HASH_WLOCK_ASSERT(pcbinfo); 424 425 hdr = &pcbinfo->ipi_lbgrouphashbase[ 426 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 427 CK_LIST_FOREACH(grp, hdr, il_list) { 428 for (i = 0; i < grp->il_inpcnt; ++i) { 429 if (grp->il_inp[i] != inp) 430 continue; 431 432 if (grp->il_inpcnt == 1) { 433 /* We are the last, free this local group. */ 434 in_pcblbgroup_free(grp); 435 } else { 436 /* Pull up inpcbs, shrink group if possible. */ 437 in_pcblbgroup_reorder(hdr, &grp, i); 438 } 439 return; 440 } 441 } 442 } 443 444 int 445 in_pcblbgroup_numa(struct inpcb *inp, int arg) 446 { 447 struct inpcbinfo *pcbinfo; 448 struct inpcblbgrouphead *hdr; 449 struct inpcblbgroup *grp; 450 int err, i; 451 uint8_t numa_domain; 452 453 switch (arg) { 454 case TCP_REUSPORT_LB_NUMA_NODOM: 455 numa_domain = M_NODOM; 456 break; 457 case TCP_REUSPORT_LB_NUMA_CURDOM: 458 numa_domain = PCPU_GET(domain); 459 break; 460 default: 461 if (arg < 0 || arg >= vm_ndomains) 462 return (EINVAL); 463 numa_domain = arg; 464 } 465 466 err = 0; 467 pcbinfo = inp->inp_pcbinfo; 468 INP_WLOCK_ASSERT(inp); 469 INP_HASH_WLOCK(pcbinfo); 470 hdr = &pcbinfo->ipi_lbgrouphashbase[ 471 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 472 CK_LIST_FOREACH(grp, hdr, il_list) { 473 for (i = 0; i < grp->il_inpcnt; ++i) { 474 if (grp->il_inp[i] != inp) 475 continue; 476 477 if (grp->il_numa_domain == numa_domain) { 478 goto abort_with_hash_wlock; 479 } 480 481 /* Remove it from the old group. */ 482 in_pcbremlbgrouphash(inp); 483 484 /* Add it to the new group based on numa domain. */ 485 in_pcbinslbgrouphash(inp, numa_domain); 486 goto abort_with_hash_wlock; 487 } 488 } 489 err = ENOENT; 490 abort_with_hash_wlock: 491 INP_HASH_WUNLOCK(pcbinfo); 492 return (err); 493 } 494 495 /* Make sure it is safe to use hashinit(9) on CK_LIST. */ 496 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb))); 497 498 /* 499 * Initialize an inpcbinfo - a per-VNET instance of connections db. 500 */ 501 void 502 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor, 503 u_int hash_nelements, u_int porthash_nelements) 504 { 505 506 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF); 507 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name, 508 NULL, MTX_DEF); 509 #ifdef VIMAGE 510 pcbinfo->ipi_vnet = curvnet; 511 #endif 512 CK_LIST_INIT(&pcbinfo->ipi_listhead); 513 pcbinfo->ipi_count = 0; 514 pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB, 515 &pcbinfo->ipi_hashmask); 516 pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB, 517 &pcbinfo->ipi_hashmask); 518 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 519 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 520 &pcbinfo->ipi_porthashmask); 521 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 522 &pcbinfo->ipi_lbgrouphashmask); 523 pcbinfo->ipi_zone = pcbstor->ips_zone; 524 pcbinfo->ipi_portzone = pcbstor->ips_portzone; 525 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone); 526 } 527 528 /* 529 * Destroy an inpcbinfo. 530 */ 531 void 532 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 533 { 534 535 KASSERT(pcbinfo->ipi_count == 0, 536 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 537 538 hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask); 539 hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask); 540 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 541 pcbinfo->ipi_porthashmask); 542 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 543 pcbinfo->ipi_lbgrouphashmask); 544 mtx_destroy(&pcbinfo->ipi_hash_lock); 545 mtx_destroy(&pcbinfo->ipi_lock); 546 } 547 548 /* 549 * Initialize a pcbstorage - per protocol zones to allocate inpcbs. 550 */ 551 static void inpcb_fini(void *, int); 552 void 553 in_pcbstorage_init(void *arg) 554 { 555 struct inpcbstorage *pcbstor = arg; 556 557 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name, 558 pcbstor->ips_size, NULL, NULL, pcbstor->ips_pcbinit, 559 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR); 560 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name, 561 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 562 uma_zone_set_smr(pcbstor->ips_portzone, 563 uma_zone_get_smr(pcbstor->ips_zone)); 564 } 565 566 /* 567 * Destroy a pcbstorage - used by unloadable protocols. 568 */ 569 void 570 in_pcbstorage_destroy(void *arg) 571 { 572 struct inpcbstorage *pcbstor = arg; 573 574 uma_zdestroy(pcbstor->ips_zone); 575 uma_zdestroy(pcbstor->ips_portzone); 576 } 577 578 /* 579 * Allocate a PCB and associate it with the socket. 580 * On success return with the PCB locked. 581 */ 582 int 583 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 584 { 585 struct inpcb *inp; 586 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 587 int error; 588 #endif 589 590 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT); 591 if (inp == NULL) 592 return (ENOBUFS); 593 bzero(&inp->inp_start_zero, inp_zero_size); 594 #ifdef NUMA 595 inp->inp_numa_domain = M_NODOM; 596 #endif 597 inp->inp_pcbinfo = pcbinfo; 598 inp->inp_socket = so; 599 inp->inp_cred = crhold(so->so_cred); 600 inp->inp_inc.inc_fibnum = so->so_fibnum; 601 #ifdef MAC 602 error = mac_inpcb_init(inp, M_NOWAIT); 603 if (error != 0) 604 goto out; 605 mac_inpcb_create(so, inp); 606 #endif 607 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 608 error = ipsec_init_pcbpolicy(inp); 609 if (error != 0) { 610 #ifdef MAC 611 mac_inpcb_destroy(inp); 612 #endif 613 goto out; 614 } 615 #endif /*IPSEC*/ 616 #ifdef INET6 617 if (INP_SOCKAF(so) == AF_INET6) { 618 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6; 619 if (V_ip6_v6only) 620 inp->inp_flags |= IN6P_IPV6_V6ONLY; 621 #ifdef INET 622 else 623 inp->inp_vflag |= INP_IPV4; 624 #endif 625 if (V_ip6_auto_flowlabel) 626 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 627 inp->in6p_hops = -1; /* use kernel default */ 628 } 629 #endif 630 #if defined(INET) && defined(INET6) 631 else 632 #endif 633 #ifdef INET 634 inp->inp_vflag |= INP_IPV4; 635 #endif 636 inp->inp_smr = SMR_SEQ_INVALID; 637 638 /* 639 * Routes in inpcb's can cache L2 as well; they are guaranteed 640 * to be cleaned up. 641 */ 642 inp->inp_route.ro_flags = RT_LLE_CACHE; 643 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */ 644 INP_WLOCK(inp); 645 INP_INFO_WLOCK(pcbinfo); 646 pcbinfo->ipi_count++; 647 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 648 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list); 649 INP_INFO_WUNLOCK(pcbinfo); 650 so->so_pcb = inp; 651 652 return (0); 653 654 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 655 out: 656 uma_zfree_smr(pcbinfo->ipi_zone, inp); 657 return (error); 658 #endif 659 } 660 661 #ifdef INET 662 int 663 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred) 664 { 665 int anonport, error; 666 667 KASSERT(sin == NULL || sin->sin_family == AF_INET, 668 ("%s: invalid address family for %p", __func__, sin)); 669 KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in), 670 ("%s: invalid address length for %p", __func__, sin)); 671 INP_WLOCK_ASSERT(inp); 672 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 673 674 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 675 return (EINVAL); 676 anonport = sin == NULL || sin->sin_port == 0; 677 error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr, 678 &inp->inp_lport, cred); 679 if (error) 680 return (error); 681 if (in_pcbinshash(inp) != 0) { 682 inp->inp_laddr.s_addr = INADDR_ANY; 683 inp->inp_lport = 0; 684 return (EAGAIN); 685 } 686 if (anonport) 687 inp->inp_flags |= INP_ANONPORT; 688 return (0); 689 } 690 #endif 691 692 #if defined(INET) || defined(INET6) 693 /* 694 * Assign a local port like in_pcb_lport(), but also used with connect() 695 * and a foreign address and port. If fsa is non-NULL, choose a local port 696 * that is unused with those, otherwise one that is completely unused. 697 * lsa can be NULL for IPv6. 698 */ 699 int 700 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 701 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 702 { 703 struct inpcbinfo *pcbinfo; 704 struct inpcb *tmpinp; 705 unsigned short *lastport; 706 int count, error; 707 u_short aux, first, last, lport; 708 #ifdef INET 709 struct in_addr laddr, faddr; 710 #endif 711 #ifdef INET6 712 struct in6_addr *laddr6, *faddr6; 713 #endif 714 715 pcbinfo = inp->inp_pcbinfo; 716 717 /* 718 * Because no actual state changes occur here, a global write lock on 719 * the pcbinfo isn't required. 720 */ 721 INP_LOCK_ASSERT(inp); 722 INP_HASH_LOCK_ASSERT(pcbinfo); 723 724 if (inp->inp_flags & INP_HIGHPORT) { 725 first = V_ipport_hifirstauto; /* sysctl */ 726 last = V_ipport_hilastauto; 727 lastport = &pcbinfo->ipi_lasthi; 728 } else if (inp->inp_flags & INP_LOWPORT) { 729 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 730 if (error) 731 return (error); 732 first = V_ipport_lowfirstauto; /* 1023 */ 733 last = V_ipport_lowlastauto; /* 600 */ 734 lastport = &pcbinfo->ipi_lastlow; 735 } else { 736 first = V_ipport_firstauto; /* sysctl */ 737 last = V_ipport_lastauto; 738 lastport = &pcbinfo->ipi_lastport; 739 } 740 741 /* 742 * Instead of having two loops further down counting up or down 743 * make sure that first is always <= last and go with only one 744 * code path implementing all logic. 745 */ 746 if (first > last) { 747 aux = first; 748 first = last; 749 last = aux; 750 } 751 752 #ifdef INET 753 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */ 754 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 755 if (lsa != NULL) 756 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 757 if (fsa != NULL) 758 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 759 } 760 #endif 761 #ifdef INET6 762 laddr6 = NULL; 763 if ((inp->inp_vflag & INP_IPV6) != 0) { 764 if (lsa != NULL) 765 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 766 if (fsa != NULL) 767 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 768 } 769 #endif 770 771 tmpinp = NULL; 772 lport = *lportp; 773 774 if (V_ipport_randomized) 775 *lastport = first + (arc4random() % (last - first)); 776 777 count = last - first; 778 779 do { 780 if (count-- < 0) /* completely used? */ 781 return (EADDRNOTAVAIL); 782 ++*lastport; 783 if (*lastport < first || *lastport > last) 784 *lastport = first; 785 lport = htons(*lastport); 786 787 if (fsa != NULL) { 788 #ifdef INET 789 if (lsa->sa_family == AF_INET) { 790 tmpinp = in_pcblookup_hash_locked(pcbinfo, 791 faddr, fport, laddr, lport, lookupflags, 792 M_NODOM); 793 } 794 #endif 795 #ifdef INET6 796 if (lsa->sa_family == AF_INET6) { 797 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 798 faddr6, fport, laddr6, lport, lookupflags, 799 M_NODOM); 800 } 801 #endif 802 } else { 803 #ifdef INET6 804 if ((inp->inp_vflag & INP_IPV6) != 0) { 805 tmpinp = in6_pcblookup_local(pcbinfo, 806 &inp->in6p_laddr, lport, lookupflags, cred); 807 #ifdef INET 808 if (tmpinp == NULL && 809 (inp->inp_vflag & INP_IPV4)) 810 tmpinp = in_pcblookup_local(pcbinfo, 811 laddr, lport, lookupflags, cred); 812 #endif 813 } 814 #endif 815 #if defined(INET) && defined(INET6) 816 else 817 #endif 818 #ifdef INET 819 tmpinp = in_pcblookup_local(pcbinfo, laddr, 820 lport, lookupflags, cred); 821 #endif 822 } 823 } while (tmpinp != NULL); 824 825 *lportp = lport; 826 827 return (0); 828 } 829 830 /* 831 * Select a local port (number) to use. 832 */ 833 int 834 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 835 struct ucred *cred, int lookupflags) 836 { 837 struct sockaddr_in laddr; 838 839 if (laddrp) { 840 bzero(&laddr, sizeof(laddr)); 841 laddr.sin_family = AF_INET; 842 laddr.sin_addr = *laddrp; 843 } 844 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 845 NULL, lportp, NULL, 0, cred, lookupflags)); 846 } 847 848 /* 849 * Return cached socket options. 850 */ 851 int 852 inp_so_options(const struct inpcb *inp) 853 { 854 int so_options; 855 856 so_options = 0; 857 858 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 859 so_options |= SO_REUSEPORT_LB; 860 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 861 so_options |= SO_REUSEPORT; 862 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 863 so_options |= SO_REUSEADDR; 864 return (so_options); 865 } 866 #endif /* INET || INET6 */ 867 868 #ifdef INET 869 /* 870 * Set up a bind operation on a PCB, performing port allocation 871 * as required, but do not actually modify the PCB. Callers can 872 * either complete the bind by setting inp_laddr/inp_lport and 873 * calling in_pcbinshash(), or they can just use the resulting 874 * port and address to authorise the sending of a once-off packet. 875 * 876 * On error, the values of *laddrp and *lportp are not changed. 877 */ 878 int 879 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp, 880 u_short *lportp, struct ucred *cred) 881 { 882 struct socket *so = inp->inp_socket; 883 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 884 struct in_addr laddr; 885 u_short lport = 0; 886 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 887 int error; 888 889 /* 890 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 891 * so that we don't have to add to the (already messy) code below. 892 */ 893 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 894 895 /* 896 * No state changes, so read locks are sufficient here. 897 */ 898 INP_LOCK_ASSERT(inp); 899 INP_HASH_LOCK_ASSERT(pcbinfo); 900 901 laddr.s_addr = *laddrp; 902 if (sin != NULL && laddr.s_addr != INADDR_ANY) 903 return (EINVAL); 904 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 905 lookupflags = INPLOOKUP_WILDCARD; 906 if (sin == NULL) { 907 if ((error = prison_local_ip4(cred, &laddr)) != 0) 908 return (error); 909 } else { 910 KASSERT(sin->sin_family == AF_INET, 911 ("%s: invalid family for address %p", __func__, sin)); 912 KASSERT(sin->sin_len == sizeof(*sin), 913 ("%s: invalid length for address %p", __func__, sin)); 914 915 error = prison_local_ip4(cred, &sin->sin_addr); 916 if (error) 917 return (error); 918 if (sin->sin_port != *lportp) { 919 /* Don't allow the port to change. */ 920 if (*lportp != 0) 921 return (EINVAL); 922 lport = sin->sin_port; 923 } 924 /* NB: lport is left as 0 if the port isn't being changed. */ 925 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 926 /* 927 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 928 * allow complete duplication of binding if 929 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 930 * and a multicast address is bound on both 931 * new and duplicated sockets. 932 */ 933 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 934 reuseport = SO_REUSEADDR|SO_REUSEPORT; 935 /* 936 * XXX: How to deal with SO_REUSEPORT_LB here? 937 * Treat same as SO_REUSEPORT for now. 938 */ 939 if ((so->so_options & 940 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 941 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 942 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 943 sin->sin_port = 0; /* yech... */ 944 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 945 /* 946 * Is the address a local IP address? 947 * If INP_BINDANY is set, then the socket may be bound 948 * to any endpoint address, local or not. 949 */ 950 if ((inp->inp_flags & INP_BINDANY) == 0 && 951 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 952 return (EADDRNOTAVAIL); 953 } 954 laddr = sin->sin_addr; 955 if (lport) { 956 struct inpcb *t; 957 958 /* GROSS */ 959 if (ntohs(lport) <= V_ipport_reservedhigh && 960 ntohs(lport) >= V_ipport_reservedlow && 961 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 962 return (EACCES); 963 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 964 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 965 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 966 lport, INPLOOKUP_WILDCARD, cred); 967 /* 968 * XXX 969 * This entire block sorely needs a rewrite. 970 */ 971 if (t != NULL && 972 (so->so_type != SOCK_STREAM || 973 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 974 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 975 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 976 (t->inp_flags2 & INP_REUSEPORT) || 977 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 978 (inp->inp_cred->cr_uid != 979 t->inp_cred->cr_uid)) 980 return (EADDRINUSE); 981 } 982 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 983 lport, lookupflags, cred); 984 if (t != NULL && (reuseport & inp_so_options(t)) == 0 && 985 (reuseport_lb & inp_so_options(t)) == 0) { 986 #ifdef INET6 987 if (ntohl(sin->sin_addr.s_addr) != 988 INADDR_ANY || 989 ntohl(t->inp_laddr.s_addr) != 990 INADDR_ANY || 991 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 992 (t->inp_vflag & INP_IPV6PROTO) == 0) 993 #endif 994 return (EADDRINUSE); 995 } 996 } 997 } 998 if (*lportp != 0) 999 lport = *lportp; 1000 if (lport == 0) { 1001 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 1002 if (error != 0) 1003 return (error); 1004 } 1005 *laddrp = laddr.s_addr; 1006 *lportp = lport; 1007 return (0); 1008 } 1009 1010 /* 1011 * Connect from a socket to a specified address. 1012 * Both address and port must be specified in argument sin. 1013 * If don't have a local address for this socket yet, 1014 * then pick one. 1015 */ 1016 int 1017 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred, 1018 bool rehash __unused) 1019 { 1020 u_short lport, fport; 1021 in_addr_t laddr, faddr; 1022 int anonport, error; 1023 1024 INP_WLOCK_ASSERT(inp); 1025 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1026 KASSERT(in_nullhost(inp->inp_faddr), 1027 ("%s: inp is already connected", __func__)); 1028 1029 lport = inp->inp_lport; 1030 laddr = inp->inp_laddr.s_addr; 1031 anonport = (lport == 0); 1032 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport, 1033 cred); 1034 if (error) 1035 return (error); 1036 1037 inp->inp_faddr.s_addr = faddr; 1038 inp->inp_fport = fport; 1039 1040 /* Do the initial binding of the local address if required. */ 1041 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1042 inp->inp_lport = lport; 1043 inp->inp_laddr.s_addr = laddr; 1044 if (in_pcbinshash(inp) != 0) { 1045 inp->inp_laddr.s_addr = inp->inp_faddr.s_addr = 1046 INADDR_ANY; 1047 inp->inp_lport = inp->inp_fport = 0; 1048 return (EAGAIN); 1049 } 1050 } else { 1051 inp->inp_lport = lport; 1052 inp->inp_laddr.s_addr = laddr; 1053 if ((inp->inp_flags & INP_INHASHLIST) != 0) 1054 in_pcbrehash(inp); 1055 else 1056 in_pcbinshash(inp); 1057 } 1058 1059 if (anonport) 1060 inp->inp_flags |= INP_ANONPORT; 1061 return (0); 1062 } 1063 1064 /* 1065 * Do proper source address selection on an unbound socket in case 1066 * of connect. Take jails into account as well. 1067 */ 1068 int 1069 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1070 struct ucred *cred) 1071 { 1072 struct ifaddr *ifa; 1073 struct sockaddr *sa; 1074 struct sockaddr_in *sin, dst; 1075 struct nhop_object *nh; 1076 int error; 1077 1078 NET_EPOCH_ASSERT(); 1079 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1080 1081 /* 1082 * Bypass source address selection and use the primary jail IP 1083 * if requested. 1084 */ 1085 if (!prison_saddrsel_ip4(cred, laddr)) 1086 return (0); 1087 1088 error = 0; 1089 1090 nh = NULL; 1091 bzero(&dst, sizeof(dst)); 1092 sin = &dst; 1093 sin->sin_family = AF_INET; 1094 sin->sin_len = sizeof(struct sockaddr_in); 1095 sin->sin_addr.s_addr = faddr->s_addr; 1096 1097 /* 1098 * If route is known our src addr is taken from the i/f, 1099 * else punt. 1100 * 1101 * Find out route to destination. 1102 */ 1103 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1104 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1105 0, NHR_NONE, 0); 1106 1107 /* 1108 * If we found a route, use the address corresponding to 1109 * the outgoing interface. 1110 * 1111 * Otherwise assume faddr is reachable on a directly connected 1112 * network and try to find a corresponding interface to take 1113 * the source address from. 1114 */ 1115 if (nh == NULL || nh->nh_ifp == NULL) { 1116 struct in_ifaddr *ia; 1117 struct ifnet *ifp; 1118 1119 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1120 inp->inp_socket->so_fibnum)); 1121 if (ia == NULL) { 1122 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1123 inp->inp_socket->so_fibnum)); 1124 } 1125 if (ia == NULL) { 1126 error = ENETUNREACH; 1127 goto done; 1128 } 1129 1130 if (!prison_flag(cred, PR_IP4)) { 1131 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1132 goto done; 1133 } 1134 1135 ifp = ia->ia_ifp; 1136 ia = NULL; 1137 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1138 sa = ifa->ifa_addr; 1139 if (sa->sa_family != AF_INET) 1140 continue; 1141 sin = (struct sockaddr_in *)sa; 1142 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1143 ia = (struct in_ifaddr *)ifa; 1144 break; 1145 } 1146 } 1147 if (ia != NULL) { 1148 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1149 goto done; 1150 } 1151 1152 /* 3. As a last resort return the 'default' jail address. */ 1153 error = prison_get_ip4(cred, laddr); 1154 goto done; 1155 } 1156 1157 /* 1158 * If the outgoing interface on the route found is not 1159 * a loopback interface, use the address from that interface. 1160 * In case of jails do those three steps: 1161 * 1. check if the interface address belongs to the jail. If so use it. 1162 * 2. check if we have any address on the outgoing interface 1163 * belonging to this jail. If so use it. 1164 * 3. as a last resort return the 'default' jail address. 1165 */ 1166 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1167 struct in_ifaddr *ia; 1168 struct ifnet *ifp; 1169 1170 /* If not jailed, use the default returned. */ 1171 if (!prison_flag(cred, PR_IP4)) { 1172 ia = (struct in_ifaddr *)nh->nh_ifa; 1173 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1174 goto done; 1175 } 1176 1177 /* Jailed. */ 1178 /* 1. Check if the iface address belongs to the jail. */ 1179 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1180 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1181 ia = (struct in_ifaddr *)nh->nh_ifa; 1182 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1183 goto done; 1184 } 1185 1186 /* 1187 * 2. Check if we have any address on the outgoing interface 1188 * belonging to this jail. 1189 */ 1190 ia = NULL; 1191 ifp = nh->nh_ifp; 1192 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1193 sa = ifa->ifa_addr; 1194 if (sa->sa_family != AF_INET) 1195 continue; 1196 sin = (struct sockaddr_in *)sa; 1197 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1198 ia = (struct in_ifaddr *)ifa; 1199 break; 1200 } 1201 } 1202 if (ia != NULL) { 1203 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1204 goto done; 1205 } 1206 1207 /* 3. As a last resort return the 'default' jail address. */ 1208 error = prison_get_ip4(cred, laddr); 1209 goto done; 1210 } 1211 1212 /* 1213 * The outgoing interface is marked with 'loopback net', so a route 1214 * to ourselves is here. 1215 * Try to find the interface of the destination address and then 1216 * take the address from there. That interface is not necessarily 1217 * a loopback interface. 1218 * In case of jails, check that it is an address of the jail 1219 * and if we cannot find, fall back to the 'default' jail address. 1220 */ 1221 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1222 struct in_ifaddr *ia; 1223 1224 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1225 inp->inp_socket->so_fibnum)); 1226 if (ia == NULL) 1227 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1228 inp->inp_socket->so_fibnum)); 1229 if (ia == NULL) 1230 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1231 1232 if (!prison_flag(cred, PR_IP4)) { 1233 if (ia == NULL) { 1234 error = ENETUNREACH; 1235 goto done; 1236 } 1237 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1238 goto done; 1239 } 1240 1241 /* Jailed. */ 1242 if (ia != NULL) { 1243 struct ifnet *ifp; 1244 1245 ifp = ia->ia_ifp; 1246 ia = NULL; 1247 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1248 sa = ifa->ifa_addr; 1249 if (sa->sa_family != AF_INET) 1250 continue; 1251 sin = (struct sockaddr_in *)sa; 1252 if (prison_check_ip4(cred, 1253 &sin->sin_addr) == 0) { 1254 ia = (struct in_ifaddr *)ifa; 1255 break; 1256 } 1257 } 1258 if (ia != NULL) { 1259 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1260 goto done; 1261 } 1262 } 1263 1264 /* 3. As a last resort return the 'default' jail address. */ 1265 error = prison_get_ip4(cred, laddr); 1266 goto done; 1267 } 1268 1269 done: 1270 if (error == 0 && laddr->s_addr == INADDR_ANY) 1271 return (EHOSTUNREACH); 1272 return (error); 1273 } 1274 1275 /* 1276 * Set up for a connect from a socket to the specified address. 1277 * On entry, *laddrp and *lportp should contain the current local 1278 * address and port for the PCB; these are updated to the values 1279 * that should be placed in inp_laddr and inp_lport to complete 1280 * the connect. 1281 * 1282 * On success, *faddrp and *fportp will be set to the remote address 1283 * and port. These are not updated in the error case. 1284 */ 1285 int 1286 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin, 1287 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1288 struct ucred *cred) 1289 { 1290 struct in_ifaddr *ia; 1291 struct in_addr laddr, faddr; 1292 u_short lport, fport; 1293 int error; 1294 1295 KASSERT(sin->sin_family == AF_INET, 1296 ("%s: invalid address family for %p", __func__, sin)); 1297 KASSERT(sin->sin_len == sizeof(*sin), 1298 ("%s: invalid address length for %p", __func__, sin)); 1299 1300 /* 1301 * Because a global state change doesn't actually occur here, a read 1302 * lock is sufficient. 1303 */ 1304 NET_EPOCH_ASSERT(); 1305 INP_LOCK_ASSERT(inp); 1306 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1307 1308 if (sin->sin_port == 0) 1309 return (EADDRNOTAVAIL); 1310 laddr.s_addr = *laddrp; 1311 lport = *lportp; 1312 faddr = sin->sin_addr; 1313 fport = sin->sin_port; 1314 #ifdef ROUTE_MPATH 1315 if (CALC_FLOWID_OUTBOUND) { 1316 uint32_t hash_val, hash_type; 1317 1318 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, 1319 inp->inp_socket->so_proto->pr_protocol, &hash_type); 1320 1321 inp->inp_flowid = hash_val; 1322 inp->inp_flowtype = hash_type; 1323 } 1324 #endif 1325 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1326 /* 1327 * If the destination address is INADDR_ANY, 1328 * use the primary local address. 1329 * If the supplied address is INADDR_BROADCAST, 1330 * and the primary interface supports broadcast, 1331 * choose the broadcast address for that interface. 1332 */ 1333 if (faddr.s_addr == INADDR_ANY) { 1334 faddr = 1335 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1336 if ((error = prison_get_ip4(cred, &faddr)) != 0) 1337 return (error); 1338 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1339 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1340 IFF_BROADCAST) 1341 faddr = satosin(&CK_STAILQ_FIRST( 1342 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1343 } 1344 } 1345 if (laddr.s_addr == INADDR_ANY) { 1346 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1347 /* 1348 * If the destination address is multicast and an outgoing 1349 * interface has been set as a multicast option, prefer the 1350 * address of that interface as our source address. 1351 */ 1352 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1353 inp->inp_moptions != NULL) { 1354 struct ip_moptions *imo; 1355 struct ifnet *ifp; 1356 1357 imo = inp->inp_moptions; 1358 if (imo->imo_multicast_ifp != NULL) { 1359 ifp = imo->imo_multicast_ifp; 1360 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1361 if (ia->ia_ifp == ifp && 1362 prison_check_ip4(cred, 1363 &ia->ia_addr.sin_addr) == 0) 1364 break; 1365 } 1366 if (ia == NULL) 1367 error = EADDRNOTAVAIL; 1368 else { 1369 laddr = ia->ia_addr.sin_addr; 1370 error = 0; 1371 } 1372 } 1373 } 1374 if (error) 1375 return (error); 1376 } 1377 1378 if (lport != 0) { 1379 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, 1380 fport, laddr, lport, 0, M_NODOM) != NULL) 1381 return (EADDRINUSE); 1382 } else { 1383 struct sockaddr_in lsin, fsin; 1384 1385 bzero(&lsin, sizeof(lsin)); 1386 bzero(&fsin, sizeof(fsin)); 1387 lsin.sin_family = AF_INET; 1388 lsin.sin_addr = laddr; 1389 fsin.sin_family = AF_INET; 1390 fsin.sin_addr = faddr; 1391 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, 1392 &lport, (struct sockaddr *)& fsin, fport, cred, 1393 INPLOOKUP_WILDCARD); 1394 if (error) 1395 return (error); 1396 } 1397 *laddrp = laddr.s_addr; 1398 *lportp = lport; 1399 *faddrp = faddr.s_addr; 1400 *fportp = fport; 1401 return (0); 1402 } 1403 1404 void 1405 in_pcbdisconnect(struct inpcb *inp) 1406 { 1407 1408 INP_WLOCK_ASSERT(inp); 1409 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1410 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 1411 ("%s: inp %p was already disconnected", __func__, inp)); 1412 1413 in_pcbremhash_locked(inp); 1414 1415 /* See the comment in in_pcbinshash(). */ 1416 inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr); 1417 inp->inp_laddr.s_addr = INADDR_ANY; 1418 inp->inp_faddr.s_addr = INADDR_ANY; 1419 inp->inp_fport = 0; 1420 } 1421 #endif /* INET */ 1422 1423 /* 1424 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1425 * For most protocols, this will be invoked immediately prior to calling 1426 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1427 * socket, in which case in_pcbfree() is deferred. 1428 */ 1429 void 1430 in_pcbdetach(struct inpcb *inp) 1431 { 1432 1433 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1434 1435 #ifdef RATELIMIT 1436 if (inp->inp_snd_tag != NULL) 1437 in_pcbdetach_txrtlmt(inp); 1438 #endif 1439 inp->inp_socket->so_pcb = NULL; 1440 inp->inp_socket = NULL; 1441 } 1442 1443 /* 1444 * inpcb hash lookups are protected by SMR section. 1445 * 1446 * Once desired pcb has been found, switching from SMR section to a pcb 1447 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK 1448 * here because SMR is a critical section. 1449 * In 99%+ cases inp_smr_lock() would obtain the lock immediately. 1450 */ 1451 void 1452 inp_lock(struct inpcb *inp, const inp_lookup_t lock) 1453 { 1454 1455 lock == INPLOOKUP_RLOCKPCB ? 1456 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock); 1457 } 1458 1459 void 1460 inp_unlock(struct inpcb *inp, const inp_lookup_t lock) 1461 { 1462 1463 lock == INPLOOKUP_RLOCKPCB ? 1464 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock); 1465 } 1466 1467 int 1468 inp_trylock(struct inpcb *inp, const inp_lookup_t lock) 1469 { 1470 1471 return (lock == INPLOOKUP_RLOCKPCB ? 1472 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock)); 1473 } 1474 1475 static inline bool 1476 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags) 1477 { 1478 1479 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB); 1480 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr); 1481 1482 if (__predict_true(inp_trylock(inp, lock))) { 1483 if (__predict_false(inp->inp_flags & ignflags)) { 1484 smr_exit(inp->inp_pcbinfo->ipi_smr); 1485 inp_unlock(inp, lock); 1486 return (false); 1487 } 1488 smr_exit(inp->inp_pcbinfo->ipi_smr); 1489 return (true); 1490 } 1491 1492 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1493 smr_exit(inp->inp_pcbinfo->ipi_smr); 1494 inp_lock(inp, lock); 1495 if (__predict_false(in_pcbrele(inp, lock))) 1496 return (false); 1497 /* 1498 * inp acquired through refcount & lock for sure didn't went 1499 * through uma_zfree(). However, it may have already went 1500 * through in_pcbfree() and has another reference, that 1501 * prevented its release by our in_pcbrele(). 1502 */ 1503 if (__predict_false(inp->inp_flags & ignflags)) { 1504 inp_unlock(inp, lock); 1505 return (false); 1506 } 1507 return (true); 1508 } else { 1509 smr_exit(inp->inp_pcbinfo->ipi_smr); 1510 return (false); 1511 } 1512 } 1513 1514 bool 1515 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock) 1516 { 1517 1518 /* 1519 * in_pcblookup() family of functions ignore not only freed entries, 1520 * that may be found due to lockless access to the hash, but dropped 1521 * entries, too. 1522 */ 1523 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED)); 1524 } 1525 1526 /* 1527 * inp_next() - inpcb hash/list traversal iterator 1528 * 1529 * Requires initialized struct inpcb_iterator for context. 1530 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR(). 1531 * 1532 * - Iterator can have either write-lock or read-lock semantics, that can not 1533 * be changed later. 1534 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through 1535 * a single hash slot. Note: only rip_input() does the latter. 1536 * - Iterator may have optional bool matching function. The matching function 1537 * will be executed for each inpcb in the SMR context, so it can not acquire 1538 * locks and can safely access only immutable fields of inpcb. 1539 * 1540 * A fresh initialized iterator has NULL inpcb in its context and that 1541 * means that inp_next() call would return the very first inpcb on the list 1542 * locked with desired semantic. In all following calls the context pointer 1543 * shall hold the current inpcb pointer. The KPI user is not supposed to 1544 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL 1545 * and write NULL to its context. After end of traversal an iterator can be 1546 * reused. 1547 * 1548 * List traversals have the following features/constraints: 1549 * - New entries won't be seen, as they are always added to the head of a list. 1550 * - Removed entries won't stop traversal as long as they are not added to 1551 * a different list. This is violated by in_pcbrehash(). 1552 */ 1553 #define II_LIST_FIRST(ipi, hash) \ 1554 (((hash) == INP_ALL_LIST) ? \ 1555 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \ 1556 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)])) 1557 #define II_LIST_NEXT(inp, hash) \ 1558 (((hash) == INP_ALL_LIST) ? \ 1559 CK_LIST_NEXT((inp), inp_list) : \ 1560 CK_LIST_NEXT((inp), inp_hash_exact)) 1561 #define II_LOCK_ASSERT(inp, lock) \ 1562 rw_assert(&(inp)->inp_lock, \ 1563 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED ) 1564 struct inpcb * 1565 inp_next(struct inpcb_iterator *ii) 1566 { 1567 const struct inpcbinfo *ipi = ii->ipi; 1568 inp_match_t *match = ii->match; 1569 void *ctx = ii->ctx; 1570 inp_lookup_t lock = ii->lock; 1571 int hash = ii->hash; 1572 struct inpcb *inp; 1573 1574 if (ii->inp == NULL) { /* First call. */ 1575 smr_enter(ipi->ipi_smr); 1576 /* This is unrolled CK_LIST_FOREACH(). */ 1577 for (inp = II_LIST_FIRST(ipi, hash); 1578 inp != NULL; 1579 inp = II_LIST_NEXT(inp, hash)) { 1580 if (match != NULL && (match)(inp, ctx) == false) 1581 continue; 1582 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED))) 1583 break; 1584 else { 1585 smr_enter(ipi->ipi_smr); 1586 MPASS(inp != II_LIST_FIRST(ipi, hash)); 1587 inp = II_LIST_FIRST(ipi, hash); 1588 if (inp == NULL) 1589 break; 1590 } 1591 } 1592 1593 if (inp == NULL) 1594 smr_exit(ipi->ipi_smr); 1595 else 1596 ii->inp = inp; 1597 1598 return (inp); 1599 } 1600 1601 /* Not a first call. */ 1602 smr_enter(ipi->ipi_smr); 1603 restart: 1604 inp = ii->inp; 1605 II_LOCK_ASSERT(inp, lock); 1606 next: 1607 inp = II_LIST_NEXT(inp, hash); 1608 if (inp == NULL) { 1609 smr_exit(ipi->ipi_smr); 1610 goto found; 1611 } 1612 1613 if (match != NULL && (match)(inp, ctx) == false) 1614 goto next; 1615 1616 if (__predict_true(inp_trylock(inp, lock))) { 1617 if (__predict_false(inp->inp_flags & INP_FREED)) { 1618 /* 1619 * Entries are never inserted in middle of a list, thus 1620 * as long as we are in SMR, we can continue traversal. 1621 * Jump to 'restart' should yield in the same result, 1622 * but could produce unnecessary looping. Could this 1623 * looping be unbound? 1624 */ 1625 inp_unlock(inp, lock); 1626 goto next; 1627 } else { 1628 smr_exit(ipi->ipi_smr); 1629 goto found; 1630 } 1631 } 1632 1633 /* 1634 * Can't obtain lock immediately, thus going hard. Once we exit the 1635 * SMR section we can no longer jump to 'next', and our only stable 1636 * anchoring point is ii->inp, which we keep locked for this case, so 1637 * we jump to 'restart'. 1638 */ 1639 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1640 smr_exit(ipi->ipi_smr); 1641 inp_lock(inp, lock); 1642 if (__predict_false(in_pcbrele(inp, lock))) { 1643 smr_enter(ipi->ipi_smr); 1644 goto restart; 1645 } 1646 /* 1647 * See comment in inp_smr_lock(). 1648 */ 1649 if (__predict_false(inp->inp_flags & INP_FREED)) { 1650 inp_unlock(inp, lock); 1651 smr_enter(ipi->ipi_smr); 1652 goto restart; 1653 } 1654 } else 1655 goto next; 1656 1657 found: 1658 inp_unlock(ii->inp, lock); 1659 ii->inp = inp; 1660 1661 return (ii->inp); 1662 } 1663 1664 /* 1665 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1666 * stability of an inpcb pointer despite the inpcb lock being released or 1667 * SMR section exited. 1668 * 1669 * To free a reference later in_pcbrele_(r|w)locked() must be performed. 1670 */ 1671 void 1672 in_pcbref(struct inpcb *inp) 1673 { 1674 u_int old __diagused; 1675 1676 old = refcount_acquire(&inp->inp_refcount); 1677 KASSERT(old > 0, ("%s: refcount 0", __func__)); 1678 } 1679 1680 /* 1681 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially 1682 * freeing the pcb, if the reference was very last. 1683 */ 1684 bool 1685 in_pcbrele_rlocked(struct inpcb *inp) 1686 { 1687 1688 INP_RLOCK_ASSERT(inp); 1689 1690 if (!refcount_release(&inp->inp_refcount)) 1691 return (false); 1692 1693 MPASS(inp->inp_flags & INP_FREED); 1694 MPASS(inp->inp_socket == NULL); 1695 MPASS(inp->inp_in_hpts == 0); 1696 crfree(inp->inp_cred); 1697 #ifdef INVARIANTS 1698 inp->inp_cred = NULL; 1699 #endif 1700 INP_RUNLOCK(inp); 1701 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1702 return (true); 1703 } 1704 1705 bool 1706 in_pcbrele_wlocked(struct inpcb *inp) 1707 { 1708 1709 INP_WLOCK_ASSERT(inp); 1710 1711 if (!refcount_release(&inp->inp_refcount)) 1712 return (false); 1713 1714 MPASS(inp->inp_flags & INP_FREED); 1715 MPASS(inp->inp_socket == NULL); 1716 MPASS(inp->inp_in_hpts == 0); 1717 crfree(inp->inp_cred); 1718 #ifdef INVARIANTS 1719 inp->inp_cred = NULL; 1720 #endif 1721 INP_WUNLOCK(inp); 1722 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1723 return (true); 1724 } 1725 1726 bool 1727 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock) 1728 { 1729 1730 return (lock == INPLOOKUP_RLOCKPCB ? 1731 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp)); 1732 } 1733 1734 /* 1735 * Unconditionally schedule an inpcb to be freed by decrementing its 1736 * reference count, which should occur only after the inpcb has been detached 1737 * from its socket. If another thread holds a temporary reference (acquired 1738 * using in_pcbref()) then the free is deferred until that reference is 1739 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked. 1740 * Almost all work, including removal from global lists, is done in this 1741 * context, where the pcbinfo lock is held. 1742 */ 1743 void 1744 in_pcbfree(struct inpcb *inp) 1745 { 1746 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1747 #ifdef INET 1748 struct ip_moptions *imo; 1749 #endif 1750 #ifdef INET6 1751 struct ip6_moptions *im6o; 1752 #endif 1753 1754 INP_WLOCK_ASSERT(inp); 1755 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1756 KASSERT((inp->inp_flags & INP_FREED) == 0, 1757 ("%s: called twice for pcb %p", __func__, inp)); 1758 1759 inp->inp_flags |= INP_FREED; 1760 INP_INFO_WLOCK(pcbinfo); 1761 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1762 pcbinfo->ipi_count--; 1763 CK_LIST_REMOVE(inp, inp_list); 1764 INP_INFO_WUNLOCK(pcbinfo); 1765 1766 if (inp->inp_flags & INP_INHASHLIST) 1767 in_pcbremhash(inp); 1768 1769 RO_INVALIDATE_CACHE(&inp->inp_route); 1770 #ifdef MAC 1771 mac_inpcb_destroy(inp); 1772 #endif 1773 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1774 if (inp->inp_sp != NULL) 1775 ipsec_delete_pcbpolicy(inp); 1776 #endif 1777 #ifdef INET 1778 if (inp->inp_options) 1779 (void)m_free(inp->inp_options); 1780 imo = inp->inp_moptions; 1781 #endif 1782 #ifdef INET6 1783 if (inp->inp_vflag & INP_IPV6PROTO) { 1784 ip6_freepcbopts(inp->in6p_outputopts); 1785 im6o = inp->in6p_moptions; 1786 } else 1787 im6o = NULL; 1788 #endif 1789 1790 if (__predict_false(in_pcbrele_wlocked(inp) == false)) { 1791 INP_WUNLOCK(inp); 1792 } 1793 #ifdef INET6 1794 ip6_freemoptions(im6o); 1795 #endif 1796 #ifdef INET 1797 inp_freemoptions(imo); 1798 #endif 1799 } 1800 1801 /* 1802 * Different protocols initialize their inpcbs differently - giving 1803 * different name to the lock. But they all are disposed the same. 1804 */ 1805 static void 1806 inpcb_fini(void *mem, int size) 1807 { 1808 struct inpcb *inp = mem; 1809 1810 INP_LOCK_DESTROY(inp); 1811 } 1812 1813 /* 1814 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1815 * port reservation, and preventing it from being returned by inpcb lookups. 1816 * 1817 * It is used by TCP to mark an inpcb as unused and avoid future packet 1818 * delivery or event notification when a socket remains open but TCP has 1819 * closed. This might occur as a result of a shutdown()-initiated TCP close 1820 * or a RST on the wire, and allows the port binding to be reused while still 1821 * maintaining the invariant that so_pcb always points to a valid inpcb until 1822 * in_pcbdetach(). 1823 * 1824 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1825 * in_pcbnotifyall() and in_pcbpurgeif0()? 1826 */ 1827 void 1828 in_pcbdrop(struct inpcb *inp) 1829 { 1830 1831 INP_WLOCK_ASSERT(inp); 1832 #ifdef INVARIANTS 1833 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1834 MPASS(inp->inp_refcount > 1); 1835 #endif 1836 1837 inp->inp_flags |= INP_DROPPED; 1838 if (inp->inp_flags & INP_INHASHLIST) 1839 in_pcbremhash(inp); 1840 } 1841 1842 #ifdef INET 1843 /* 1844 * Common routines to return the socket addresses associated with inpcbs. 1845 */ 1846 struct sockaddr * 1847 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1848 { 1849 struct sockaddr_in *sin; 1850 1851 sin = malloc(sizeof *sin, M_SONAME, 1852 M_WAITOK | M_ZERO); 1853 sin->sin_family = AF_INET; 1854 sin->sin_len = sizeof(*sin); 1855 sin->sin_addr = *addr_p; 1856 sin->sin_port = port; 1857 1858 return (struct sockaddr *)sin; 1859 } 1860 1861 int 1862 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1863 { 1864 struct inpcb *inp; 1865 struct in_addr addr; 1866 in_port_t port; 1867 1868 inp = sotoinpcb(so); 1869 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1870 1871 INP_RLOCK(inp); 1872 port = inp->inp_lport; 1873 addr = inp->inp_laddr; 1874 INP_RUNLOCK(inp); 1875 1876 *nam = in_sockaddr(port, &addr); 1877 return 0; 1878 } 1879 1880 int 1881 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1882 { 1883 struct inpcb *inp; 1884 struct in_addr addr; 1885 in_port_t port; 1886 1887 inp = sotoinpcb(so); 1888 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1889 1890 INP_RLOCK(inp); 1891 port = inp->inp_fport; 1892 addr = inp->inp_faddr; 1893 INP_RUNLOCK(inp); 1894 1895 *nam = in_sockaddr(port, &addr); 1896 return 0; 1897 } 1898 1899 void 1900 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1901 struct inpcb *(*notify)(struct inpcb *, int)) 1902 { 1903 struct inpcb *inp, *inp_temp; 1904 1905 INP_INFO_WLOCK(pcbinfo); 1906 CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) { 1907 INP_WLOCK(inp); 1908 #ifdef INET6 1909 if ((inp->inp_vflag & INP_IPV4) == 0) { 1910 INP_WUNLOCK(inp); 1911 continue; 1912 } 1913 #endif 1914 if (inp->inp_faddr.s_addr != faddr.s_addr || 1915 inp->inp_socket == NULL) { 1916 INP_WUNLOCK(inp); 1917 continue; 1918 } 1919 if ((*notify)(inp, errno)) 1920 INP_WUNLOCK(inp); 1921 } 1922 INP_INFO_WUNLOCK(pcbinfo); 1923 } 1924 1925 static bool 1926 inp_v4_multi_match(const struct inpcb *inp, void *v __unused) 1927 { 1928 1929 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL) 1930 return (true); 1931 else 1932 return (false); 1933 } 1934 1935 void 1936 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1937 { 1938 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB, 1939 inp_v4_multi_match, NULL); 1940 struct inpcb *inp; 1941 struct in_multi *inm; 1942 struct in_mfilter *imf; 1943 struct ip_moptions *imo; 1944 1945 IN_MULTI_LOCK_ASSERT(); 1946 1947 while ((inp = inp_next(&inpi)) != NULL) { 1948 INP_WLOCK_ASSERT(inp); 1949 1950 imo = inp->inp_moptions; 1951 /* 1952 * Unselect the outgoing interface if it is being 1953 * detached. 1954 */ 1955 if (imo->imo_multicast_ifp == ifp) 1956 imo->imo_multicast_ifp = NULL; 1957 1958 /* 1959 * Drop multicast group membership if we joined 1960 * through the interface being detached. 1961 * 1962 * XXX This can all be deferred to an epoch_call 1963 */ 1964 restart: 1965 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 1966 if ((inm = imf->imf_inm) == NULL) 1967 continue; 1968 if (inm->inm_ifp != ifp) 1969 continue; 1970 ip_mfilter_remove(&imo->imo_head, imf); 1971 in_leavegroup_locked(inm, NULL); 1972 ip_mfilter_free(imf); 1973 goto restart; 1974 } 1975 } 1976 } 1977 1978 /* 1979 * Lookup a PCB based on the local address and port. Caller must hold the 1980 * hash lock. No inpcb locks or references are acquired. 1981 */ 1982 #define INP_LOOKUP_MAPPED_PCB_COST 3 1983 struct inpcb * 1984 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1985 u_short lport, int lookupflags, struct ucred *cred) 1986 { 1987 struct inpcb *inp; 1988 #ifdef INET6 1989 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1990 #else 1991 int matchwild = 3; 1992 #endif 1993 int wildcard; 1994 1995 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1996 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1997 INP_HASH_LOCK_ASSERT(pcbinfo); 1998 1999 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 2000 struct inpcbhead *head; 2001 /* 2002 * Look for an unconnected (wildcard foreign addr) PCB that 2003 * matches the local address and port we're looking for. 2004 */ 2005 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2006 pcbinfo->ipi_hashmask)]; 2007 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2008 #ifdef INET6 2009 /* XXX inp locking */ 2010 if ((inp->inp_vflag & INP_IPV4) == 0) 2011 continue; 2012 #endif 2013 if (inp->inp_faddr.s_addr == INADDR_ANY && 2014 inp->inp_laddr.s_addr == laddr.s_addr && 2015 inp->inp_lport == lport) { 2016 /* 2017 * Found? 2018 */ 2019 if (prison_equal_ip4(cred->cr_prison, 2020 inp->inp_cred->cr_prison)) 2021 return (inp); 2022 } 2023 } 2024 /* 2025 * Not found. 2026 */ 2027 return (NULL); 2028 } else { 2029 struct inpcbporthead *porthash; 2030 struct inpcbport *phd; 2031 struct inpcb *match = NULL; 2032 /* 2033 * Best fit PCB lookup. 2034 * 2035 * First see if this local port is in use by looking on the 2036 * port hash list. 2037 */ 2038 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 2039 pcbinfo->ipi_porthashmask)]; 2040 CK_LIST_FOREACH(phd, porthash, phd_hash) { 2041 if (phd->phd_port == lport) 2042 break; 2043 } 2044 if (phd != NULL) { 2045 /* 2046 * Port is in use by one or more PCBs. Look for best 2047 * fit. 2048 */ 2049 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 2050 wildcard = 0; 2051 if (!prison_equal_ip4(inp->inp_cred->cr_prison, 2052 cred->cr_prison)) 2053 continue; 2054 #ifdef INET6 2055 /* XXX inp locking */ 2056 if ((inp->inp_vflag & INP_IPV4) == 0) 2057 continue; 2058 /* 2059 * We never select the PCB that has 2060 * INP_IPV6 flag and is bound to :: if 2061 * we have another PCB which is bound 2062 * to 0.0.0.0. If a PCB has the 2063 * INP_IPV6 flag, then we set its cost 2064 * higher than IPv4 only PCBs. 2065 * 2066 * Note that the case only happens 2067 * when a socket is bound to ::, under 2068 * the condition that the use of the 2069 * mapped address is allowed. 2070 */ 2071 if ((inp->inp_vflag & INP_IPV6) != 0) 2072 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 2073 #endif 2074 if (inp->inp_faddr.s_addr != INADDR_ANY) 2075 wildcard++; 2076 if (inp->inp_laddr.s_addr != INADDR_ANY) { 2077 if (laddr.s_addr == INADDR_ANY) 2078 wildcard++; 2079 else if (inp->inp_laddr.s_addr != laddr.s_addr) 2080 continue; 2081 } else { 2082 if (laddr.s_addr != INADDR_ANY) 2083 wildcard++; 2084 } 2085 if (wildcard < matchwild) { 2086 match = inp; 2087 matchwild = wildcard; 2088 if (matchwild == 0) 2089 break; 2090 } 2091 } 2092 } 2093 return (match); 2094 } 2095 } 2096 #undef INP_LOOKUP_MAPPED_PCB_COST 2097 2098 static bool 2099 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain) 2100 { 2101 return (domain == M_NODOM || domain == grp->il_numa_domain); 2102 } 2103 2104 static struct inpcb * 2105 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 2106 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr, 2107 uint16_t lport, int domain) 2108 { 2109 const struct inpcblbgrouphead *hdr; 2110 struct inpcblbgroup *grp; 2111 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild; 2112 2113 INP_HASH_LOCK_ASSERT(pcbinfo); 2114 2115 hdr = &pcbinfo->ipi_lbgrouphashbase[ 2116 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 2117 2118 /* 2119 * Search for an LB group match based on the following criteria: 2120 * - prefer jailed groups to non-jailed groups 2121 * - prefer exact source address matches to wildcard matches 2122 * - prefer groups bound to the specified NUMA domain 2123 */ 2124 jail_exact = jail_wild = local_exact = local_wild = NULL; 2125 CK_LIST_FOREACH(grp, hdr, il_list) { 2126 bool injail; 2127 2128 #ifdef INET6 2129 if (!(grp->il_vflag & INP_IPV4)) 2130 continue; 2131 #endif 2132 if (grp->il_lport != lport) 2133 continue; 2134 2135 injail = prison_flag(grp->il_cred, PR_IP4) != 0; 2136 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison, 2137 laddr) != 0) 2138 continue; 2139 2140 if (grp->il_laddr.s_addr == laddr->s_addr) { 2141 if (injail) { 2142 jail_exact = grp; 2143 if (in_pcblookup_lb_numa_match(grp, domain)) 2144 /* This is a perfect match. */ 2145 goto out; 2146 } else if (local_exact == NULL || 2147 in_pcblookup_lb_numa_match(grp, domain)) { 2148 local_exact = grp; 2149 } 2150 } else if (grp->il_laddr.s_addr == INADDR_ANY) { 2151 if (injail) { 2152 if (jail_wild == NULL || 2153 in_pcblookup_lb_numa_match(grp, domain)) 2154 jail_wild = grp; 2155 } else if (local_wild == NULL || 2156 in_pcblookup_lb_numa_match(grp, domain)) { 2157 local_wild = grp; 2158 } 2159 } 2160 } 2161 2162 if (jail_exact != NULL) 2163 grp = jail_exact; 2164 else if (jail_wild != NULL) 2165 grp = jail_wild; 2166 else if (local_exact != NULL) 2167 grp = local_exact; 2168 else 2169 grp = local_wild; 2170 if (grp == NULL) 2171 return (NULL); 2172 out: 2173 return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) % 2174 grp->il_inpcnt]); 2175 } 2176 2177 static bool 2178 in_pcblookup_exact_match(const struct inpcb *inp, struct in_addr faddr, 2179 u_short fport, struct in_addr laddr, u_short lport) 2180 { 2181 #ifdef INET6 2182 /* XXX inp locking */ 2183 if ((inp->inp_vflag & INP_IPV4) == 0) 2184 return (false); 2185 #endif 2186 if (inp->inp_faddr.s_addr == faddr.s_addr && 2187 inp->inp_laddr.s_addr == laddr.s_addr && 2188 inp->inp_fport == fport && 2189 inp->inp_lport == lport) 2190 return (true); 2191 return (false); 2192 } 2193 2194 static struct inpcb * 2195 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2196 u_short fport, struct in_addr laddr, u_short lport) 2197 { 2198 struct inpcbhead *head; 2199 struct inpcb *inp; 2200 2201 INP_HASH_LOCK_ASSERT(pcbinfo); 2202 2203 head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport, 2204 pcbinfo->ipi_hashmask)]; 2205 CK_LIST_FOREACH(inp, head, inp_hash_exact) { 2206 if (in_pcblookup_exact_match(inp, faddr, fport, laddr, lport)) 2207 return (inp); 2208 } 2209 return (NULL); 2210 } 2211 2212 typedef enum { 2213 INPLOOKUP_MATCH_NONE = 0, 2214 INPLOOKUP_MATCH_WILD = 1, 2215 INPLOOKUP_MATCH_LADDR = 2, 2216 } inp_lookup_match_t; 2217 2218 static inp_lookup_match_t 2219 in_pcblookup_wild_match(const struct inpcb *inp, struct in_addr laddr, 2220 u_short lport) 2221 { 2222 #ifdef INET6 2223 /* XXX inp locking */ 2224 if ((inp->inp_vflag & INP_IPV4) == 0) 2225 return (INPLOOKUP_MATCH_NONE); 2226 #endif 2227 if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) 2228 return (INPLOOKUP_MATCH_NONE); 2229 if (inp->inp_laddr.s_addr == INADDR_ANY) 2230 return (INPLOOKUP_MATCH_WILD); 2231 if (inp->inp_laddr.s_addr == laddr.s_addr) 2232 return (INPLOOKUP_MATCH_LADDR); 2233 return (INPLOOKUP_MATCH_NONE); 2234 } 2235 2236 #define INP_LOOKUP_AGAIN ((struct inpcb *)(uintptr_t)-1) 2237 2238 static struct inpcb * 2239 in_pcblookup_hash_wild_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2240 u_short fport, struct in_addr laddr, u_short lport, 2241 const inp_lookup_t lockflags) 2242 { 2243 struct inpcbhead *head; 2244 struct inpcb *inp; 2245 2246 KASSERT(SMR_ENTERED(pcbinfo->ipi_smr), 2247 ("%s: not in SMR read section", __func__)); 2248 2249 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2250 pcbinfo->ipi_hashmask)]; 2251 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2252 inp_lookup_match_t match; 2253 2254 match = in_pcblookup_wild_match(inp, laddr, lport); 2255 if (match == INPLOOKUP_MATCH_NONE) 2256 continue; 2257 2258 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2259 if (__predict_true(in_pcblookup_wild_match(inp, laddr, 2260 lport) != INPLOOKUP_MATCH_NONE)) 2261 return (inp); 2262 inp_unlock(inp, lockflags); 2263 } 2264 2265 /* 2266 * The matching socket disappeared out from under us. Fall back 2267 * to a serialized lookup. 2268 */ 2269 return (INP_LOOKUP_AGAIN); 2270 } 2271 return (NULL); 2272 } 2273 2274 static struct inpcb * 2275 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2276 u_short fport, struct in_addr laddr, u_short lport) 2277 { 2278 struct inpcbhead *head; 2279 struct inpcb *inp, *local_wild, *local_exact, *jail_wild; 2280 #ifdef INET6 2281 struct inpcb *local_wild_mapped; 2282 #endif 2283 2284 INP_HASH_LOCK_ASSERT(pcbinfo); 2285 2286 /* 2287 * Order of socket selection - we always prefer jails. 2288 * 1. jailed, non-wild. 2289 * 2. jailed, wild. 2290 * 3. non-jailed, non-wild. 2291 * 4. non-jailed, wild. 2292 */ 2293 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2294 pcbinfo->ipi_hashmask)]; 2295 local_wild = local_exact = jail_wild = NULL; 2296 #ifdef INET6 2297 local_wild_mapped = NULL; 2298 #endif 2299 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2300 inp_lookup_match_t match; 2301 bool injail; 2302 2303 match = in_pcblookup_wild_match(inp, laddr, lport); 2304 if (match == INPLOOKUP_MATCH_NONE) 2305 continue; 2306 2307 injail = prison_flag(inp->inp_cred, PR_IP4) != 0; 2308 if (injail) { 2309 if (prison_check_ip4_locked(inp->inp_cred->cr_prison, 2310 &laddr) != 0) 2311 continue; 2312 } else { 2313 if (local_exact != NULL) 2314 continue; 2315 } 2316 2317 if (match == INPLOOKUP_MATCH_LADDR) { 2318 if (injail) 2319 return (inp); 2320 local_exact = inp; 2321 } else { 2322 #ifdef INET6 2323 /* XXX inp locking, NULL check */ 2324 if (inp->inp_vflag & INP_IPV6PROTO) 2325 local_wild_mapped = inp; 2326 else 2327 #endif 2328 if (injail) 2329 jail_wild = inp; 2330 else 2331 local_wild = inp; 2332 } 2333 } 2334 if (jail_wild != NULL) 2335 return (jail_wild); 2336 if (local_exact != NULL) 2337 return (local_exact); 2338 if (local_wild != NULL) 2339 return (local_wild); 2340 #ifdef INET6 2341 if (local_wild_mapped != NULL) 2342 return (local_wild_mapped); 2343 #endif 2344 return (NULL); 2345 } 2346 2347 /* 2348 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2349 * that the caller has either locked the hash list, which usually happens 2350 * for bind(2) operations, or is in SMR section, which happens when sorting 2351 * out incoming packets. 2352 */ 2353 static struct inpcb * 2354 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2355 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2356 uint8_t numa_domain) 2357 { 2358 struct inpcb *inp; 2359 const u_short fport = fport_arg, lport = lport_arg; 2360 2361 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0, 2362 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2363 KASSERT(faddr.s_addr != INADDR_ANY, 2364 ("%s: invalid foreign address", __func__)); 2365 KASSERT(laddr.s_addr != INADDR_ANY, 2366 ("%s: invalid local address", __func__)); 2367 INP_HASH_WLOCK_ASSERT(pcbinfo); 2368 2369 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2370 if (inp != NULL) 2371 return (inp); 2372 2373 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2374 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2375 &laddr, lport, numa_domain); 2376 if (inp == NULL) { 2377 inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr, 2378 fport, laddr, lport); 2379 } 2380 } 2381 2382 return (inp); 2383 } 2384 2385 static struct inpcb * 2386 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2387 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2388 uint8_t numa_domain) 2389 { 2390 struct inpcb *inp; 2391 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2392 2393 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2394 ("%s: LOCKPCB not set", __func__)); 2395 2396 INP_HASH_WLOCK(pcbinfo); 2397 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2398 lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain); 2399 if (inp != NULL && !inp_trylock(inp, lockflags)) { 2400 in_pcbref(inp); 2401 INP_HASH_WUNLOCK(pcbinfo); 2402 inp_lock(inp, lockflags); 2403 if (in_pcbrele(inp, lockflags)) 2404 /* XXX-MJ or retry until we get a negative match? */ 2405 inp = NULL; 2406 } else { 2407 INP_HASH_WUNLOCK(pcbinfo); 2408 } 2409 return (inp); 2410 } 2411 2412 static struct inpcb * 2413 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2414 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2415 uint8_t numa_domain) 2416 { 2417 struct inpcb *inp; 2418 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2419 const u_short fport = fport_arg, lport = lport_arg; 2420 2421 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2422 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2423 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2424 ("%s: LOCKPCB not set", __func__)); 2425 2426 smr_enter(pcbinfo->ipi_smr); 2427 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2428 if (inp != NULL) { 2429 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2430 /* 2431 * Revalidate the 4-tuple, the socket could have been 2432 * disconnected. 2433 */ 2434 if (__predict_true(in_pcblookup_exact_match(inp, 2435 faddr, fport, laddr, lport))) 2436 return (inp); 2437 inp_unlock(inp, lockflags); 2438 } 2439 2440 /* 2441 * We failed to lock the inpcb, or its connection state changed 2442 * out from under us. Fall back to a precise search. 2443 */ 2444 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2445 lookupflags, numa_domain)); 2446 } 2447 2448 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2449 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2450 &laddr, lport, numa_domain); 2451 if (inp != NULL) { 2452 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2453 if (__predict_true(in_pcblookup_wild_match(inp, 2454 laddr, lport) != INPLOOKUP_MATCH_NONE)) 2455 return (inp); 2456 inp_unlock(inp, lockflags); 2457 } 2458 inp = INP_LOOKUP_AGAIN; 2459 } else { 2460 inp = in_pcblookup_hash_wild_smr(pcbinfo, faddr, fport, 2461 laddr, lport, lockflags); 2462 } 2463 if (inp == INP_LOOKUP_AGAIN) { 2464 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, 2465 lport, lookupflags, numa_domain)); 2466 } 2467 } 2468 2469 if (inp == NULL) 2470 smr_exit(pcbinfo->ipi_smr); 2471 2472 return (inp); 2473 } 2474 2475 /* 2476 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2477 * from which a pre-calculated hash value may be extracted. 2478 */ 2479 struct inpcb * 2480 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2481 struct in_addr laddr, u_int lport, int lookupflags, 2482 struct ifnet *ifp __unused) 2483 { 2484 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2485 lookupflags, M_NODOM)); 2486 } 2487 2488 struct inpcb * 2489 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2490 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2491 struct ifnet *ifp __unused, struct mbuf *m) 2492 { 2493 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2494 lookupflags, m->m_pkthdr.numa_domain)); 2495 } 2496 #endif /* INET */ 2497 2498 static bool 2499 in_pcbjailed(const struct inpcb *inp, unsigned int flag) 2500 { 2501 return (prison_flag(inp->inp_cred, flag) != 0); 2502 } 2503 2504 /* 2505 * Insert the PCB into a hash chain using ordering rules which ensure that 2506 * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first. 2507 * 2508 * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs 2509 * with exact local addresses ahead of wildcard PCBs. 2510 */ 2511 static void 2512 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2513 { 2514 struct inpcb *last; 2515 bool bound, injail; 2516 2517 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2518 2519 last = NULL; 2520 bound = inp->inp_laddr.s_addr != INADDR_ANY; 2521 injail = in_pcbjailed(inp, PR_IP4); 2522 if (!injail) { 2523 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2524 if (in_pcbjailed(inp, PR_IP4)) 2525 break; 2526 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2527 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2528 return; 2529 } 2530 } 2531 } else if (!CK_LIST_EMPTY(pcbhash) && 2532 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) { 2533 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2534 return; 2535 } 2536 if (!bound) { 2537 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2538 if (last->inp_laddr.s_addr == INADDR_ANY) 2539 break; 2540 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2541 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2542 return; 2543 } 2544 } 2545 } 2546 if (last == NULL) 2547 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2548 else 2549 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2550 } 2551 2552 #ifdef INET6 2553 /* 2554 * See the comment above _in_pcbinshash_wild(). 2555 */ 2556 static void 2557 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2558 { 2559 struct inpcb *last; 2560 bool bound, injail; 2561 2562 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2563 2564 last = NULL; 2565 bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr); 2566 injail = in_pcbjailed(inp, PR_IP6); 2567 if (!injail) { 2568 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2569 if (in_pcbjailed(last, PR_IP6)) 2570 break; 2571 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2572 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2573 return; 2574 } 2575 } 2576 } else if (!CK_LIST_EMPTY(pcbhash) && 2577 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) { 2578 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2579 return; 2580 } 2581 if (!bound) { 2582 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2583 if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr)) 2584 break; 2585 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2586 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2587 return; 2588 } 2589 } 2590 } 2591 if (last == NULL) 2592 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2593 else 2594 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2595 } 2596 #endif 2597 2598 /* 2599 * Insert PCB onto various hash lists. 2600 */ 2601 int 2602 in_pcbinshash(struct inpcb *inp) 2603 { 2604 struct inpcbhead *pcbhash; 2605 struct inpcbporthead *pcbporthash; 2606 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2607 struct inpcbport *phd; 2608 uint32_t hash; 2609 bool connected; 2610 2611 INP_WLOCK_ASSERT(inp); 2612 INP_HASH_WLOCK_ASSERT(pcbinfo); 2613 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2614 ("in_pcbinshash: INP_INHASHLIST")); 2615 2616 #ifdef INET6 2617 if (inp->inp_vflag & INP_IPV6) { 2618 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2619 inp->inp_fport, pcbinfo->ipi_hashmask); 2620 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2621 } else 2622 #endif 2623 { 2624 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2625 inp->inp_fport, pcbinfo->ipi_hashmask); 2626 connected = !in_nullhost(inp->inp_faddr); 2627 } 2628 2629 if (connected) 2630 pcbhash = &pcbinfo->ipi_hash_exact[hash]; 2631 else 2632 pcbhash = &pcbinfo->ipi_hash_wild[hash]; 2633 2634 pcbporthash = &pcbinfo->ipi_porthashbase[ 2635 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2636 2637 /* 2638 * Add entry to load balance group. 2639 * Only do this if SO_REUSEPORT_LB is set. 2640 */ 2641 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) { 2642 int error = in_pcbinslbgrouphash(inp, M_NODOM); 2643 if (error != 0) 2644 return (error); 2645 } 2646 2647 /* 2648 * Go through port list and look for a head for this lport. 2649 */ 2650 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2651 if (phd->phd_port == inp->inp_lport) 2652 break; 2653 } 2654 2655 /* 2656 * If none exists, malloc one and tack it on. 2657 */ 2658 if (phd == NULL) { 2659 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT); 2660 if (phd == NULL) { 2661 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 2662 in_pcbremlbgrouphash(inp); 2663 return (ENOMEM); 2664 } 2665 phd->phd_port = inp->inp_lport; 2666 CK_LIST_INIT(&phd->phd_pcblist); 2667 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2668 } 2669 inp->inp_phd = phd; 2670 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2671 2672 /* 2673 * The PCB may have been disconnected in the past. Before we can safely 2674 * make it visible in the hash table, we must wait for all readers which 2675 * may be traversing this PCB to finish. 2676 */ 2677 if (inp->inp_smr != SMR_SEQ_INVALID) { 2678 smr_wait(pcbinfo->ipi_smr, inp->inp_smr); 2679 inp->inp_smr = SMR_SEQ_INVALID; 2680 } 2681 2682 if (connected) 2683 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact); 2684 else { 2685 #ifdef INET6 2686 if ((inp->inp_vflag & INP_IPV6) != 0) 2687 _in6_pcbinshash_wild(pcbhash, inp); 2688 else 2689 #endif 2690 _in_pcbinshash_wild(pcbhash, inp); 2691 } 2692 inp->inp_flags |= INP_INHASHLIST; 2693 2694 return (0); 2695 } 2696 2697 void 2698 in_pcbremhash_locked(struct inpcb *inp) 2699 { 2700 struct inpcbport *phd = inp->inp_phd; 2701 2702 INP_WLOCK_ASSERT(inp); 2703 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2704 MPASS(inp->inp_flags & INP_INHASHLIST); 2705 2706 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 2707 in_pcbremlbgrouphash(inp); 2708 #ifdef INET6 2709 if (inp->inp_vflag & INP_IPV6) { 2710 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) 2711 CK_LIST_REMOVE(inp, inp_hash_wild); 2712 else 2713 CK_LIST_REMOVE(inp, inp_hash_exact); 2714 } else 2715 #endif 2716 { 2717 if (in_nullhost(inp->inp_faddr)) 2718 CK_LIST_REMOVE(inp, inp_hash_wild); 2719 else 2720 CK_LIST_REMOVE(inp, inp_hash_exact); 2721 } 2722 CK_LIST_REMOVE(inp, inp_portlist); 2723 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2724 CK_LIST_REMOVE(phd, phd_hash); 2725 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd); 2726 } 2727 inp->inp_flags &= ~INP_INHASHLIST; 2728 } 2729 2730 static void 2731 in_pcbremhash(struct inpcb *inp) 2732 { 2733 INP_HASH_WLOCK(inp->inp_pcbinfo); 2734 in_pcbremhash_locked(inp); 2735 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 2736 } 2737 2738 /* 2739 * Move PCB to the proper hash bucket when { faddr, fport } have been 2740 * changed. NOTE: This does not handle the case of the lport changing (the 2741 * hashed port list would have to be updated as well), so the lport must 2742 * not change after in_pcbinshash() has been called. 2743 */ 2744 void 2745 in_pcbrehash(struct inpcb *inp) 2746 { 2747 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2748 struct inpcbhead *head; 2749 uint32_t hash; 2750 bool connected; 2751 2752 INP_WLOCK_ASSERT(inp); 2753 INP_HASH_WLOCK_ASSERT(pcbinfo); 2754 KASSERT(inp->inp_flags & INP_INHASHLIST, 2755 ("%s: !INP_INHASHLIST", __func__)); 2756 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 2757 ("%s: inp was disconnected", __func__)); 2758 2759 #ifdef INET6 2760 if (inp->inp_vflag & INP_IPV6) { 2761 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2762 inp->inp_fport, pcbinfo->ipi_hashmask); 2763 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2764 } else 2765 #endif 2766 { 2767 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2768 inp->inp_fport, pcbinfo->ipi_hashmask); 2769 connected = !in_nullhost(inp->inp_faddr); 2770 } 2771 2772 /* 2773 * When rehashing, the caller must ensure that either the new or the old 2774 * foreign address was unspecified. 2775 */ 2776 if (connected) 2777 CK_LIST_REMOVE(inp, inp_hash_wild); 2778 else 2779 CK_LIST_REMOVE(inp, inp_hash_exact); 2780 2781 if (connected) { 2782 head = &pcbinfo->ipi_hash_exact[hash]; 2783 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact); 2784 } else { 2785 head = &pcbinfo->ipi_hash_wild[hash]; 2786 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild); 2787 } 2788 } 2789 2790 /* 2791 * Check for alternatives when higher level complains 2792 * about service problems. For now, invalidate cached 2793 * routing information. If the route was created dynamically 2794 * (by a redirect), time to try a default gateway again. 2795 */ 2796 void 2797 in_losing(struct inpcb *inp) 2798 { 2799 2800 RO_INVALIDATE_CACHE(&inp->inp_route); 2801 return; 2802 } 2803 2804 /* 2805 * A set label operation has occurred at the socket layer, propagate the 2806 * label change into the in_pcb for the socket. 2807 */ 2808 void 2809 in_pcbsosetlabel(struct socket *so) 2810 { 2811 #ifdef MAC 2812 struct inpcb *inp; 2813 2814 inp = sotoinpcb(so); 2815 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2816 2817 INP_WLOCK(inp); 2818 SOCK_LOCK(so); 2819 mac_inpcb_sosetlabel(so, inp); 2820 SOCK_UNLOCK(so); 2821 INP_WUNLOCK(inp); 2822 #endif 2823 } 2824 2825 void 2826 inp_wlock(struct inpcb *inp) 2827 { 2828 2829 INP_WLOCK(inp); 2830 } 2831 2832 void 2833 inp_wunlock(struct inpcb *inp) 2834 { 2835 2836 INP_WUNLOCK(inp); 2837 } 2838 2839 void 2840 inp_rlock(struct inpcb *inp) 2841 { 2842 2843 INP_RLOCK(inp); 2844 } 2845 2846 void 2847 inp_runlock(struct inpcb *inp) 2848 { 2849 2850 INP_RUNLOCK(inp); 2851 } 2852 2853 #ifdef INVARIANT_SUPPORT 2854 void 2855 inp_lock_assert(struct inpcb *inp) 2856 { 2857 2858 INP_WLOCK_ASSERT(inp); 2859 } 2860 2861 void 2862 inp_unlock_assert(struct inpcb *inp) 2863 { 2864 2865 INP_UNLOCK_ASSERT(inp); 2866 } 2867 #endif 2868 2869 void 2870 inp_apply_all(struct inpcbinfo *pcbinfo, 2871 void (*func)(struct inpcb *, void *), void *arg) 2872 { 2873 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2874 INPLOOKUP_WLOCKPCB); 2875 struct inpcb *inp; 2876 2877 while ((inp = inp_next(&inpi)) != NULL) 2878 func(inp, arg); 2879 } 2880 2881 struct socket * 2882 inp_inpcbtosocket(struct inpcb *inp) 2883 { 2884 2885 INP_WLOCK_ASSERT(inp); 2886 return (inp->inp_socket); 2887 } 2888 2889 struct tcpcb * 2890 inp_inpcbtotcpcb(struct inpcb *inp) 2891 { 2892 2893 INP_WLOCK_ASSERT(inp); 2894 return ((struct tcpcb *)inp->inp_ppcb); 2895 } 2896 2897 int 2898 inp_ip_tos_get(const struct inpcb *inp) 2899 { 2900 2901 return (inp->inp_ip_tos); 2902 } 2903 2904 void 2905 inp_ip_tos_set(struct inpcb *inp, int val) 2906 { 2907 2908 inp->inp_ip_tos = val; 2909 } 2910 2911 void 2912 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2913 uint32_t *faddr, uint16_t *fp) 2914 { 2915 2916 INP_LOCK_ASSERT(inp); 2917 *laddr = inp->inp_laddr.s_addr; 2918 *faddr = inp->inp_faddr.s_addr; 2919 *lp = inp->inp_lport; 2920 *fp = inp->inp_fport; 2921 } 2922 2923 struct inpcb * 2924 so_sotoinpcb(struct socket *so) 2925 { 2926 2927 return (sotoinpcb(so)); 2928 } 2929 2930 /* 2931 * Create an external-format (``xinpcb'') structure using the information in 2932 * the kernel-format in_pcb structure pointed to by inp. This is done to 2933 * reduce the spew of irrelevant information over this interface, to isolate 2934 * user code from changes in the kernel structure, and potentially to provide 2935 * information-hiding if we decide that some of this information should be 2936 * hidden from users. 2937 */ 2938 void 2939 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2940 { 2941 2942 bzero(xi, sizeof(*xi)); 2943 xi->xi_len = sizeof(struct xinpcb); 2944 if (inp->inp_socket) 2945 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2946 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2947 xi->inp_gencnt = inp->inp_gencnt; 2948 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2949 xi->inp_flow = inp->inp_flow; 2950 xi->inp_flowid = inp->inp_flowid; 2951 xi->inp_flowtype = inp->inp_flowtype; 2952 xi->inp_flags = inp->inp_flags; 2953 xi->inp_flags2 = inp->inp_flags2; 2954 xi->in6p_cksum = inp->in6p_cksum; 2955 xi->in6p_hops = inp->in6p_hops; 2956 xi->inp_ip_tos = inp->inp_ip_tos; 2957 xi->inp_vflag = inp->inp_vflag; 2958 xi->inp_ip_ttl = inp->inp_ip_ttl; 2959 xi->inp_ip_p = inp->inp_ip_p; 2960 xi->inp_ip_minttl = inp->inp_ip_minttl; 2961 } 2962 2963 int 2964 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo, 2965 int (*ctloutput_set)(struct inpcb *, struct sockopt *)) 2966 { 2967 struct sockopt sopt; 2968 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2969 INPLOOKUP_WLOCKPCB); 2970 struct inpcb *inp; 2971 struct sockopt_parameters *params; 2972 struct socket *so; 2973 int error; 2974 char buf[1024]; 2975 2976 if (req->oldptr != NULL || req->oldlen != 0) 2977 return (EINVAL); 2978 if (req->newptr == NULL) 2979 return (EPERM); 2980 if (req->newlen > sizeof(buf)) 2981 return (ENOMEM); 2982 error = SYSCTL_IN(req, buf, req->newlen); 2983 if (error != 0) 2984 return (error); 2985 if (req->newlen < sizeof(struct sockopt_parameters)) 2986 return (EINVAL); 2987 params = (struct sockopt_parameters *)buf; 2988 sopt.sopt_level = params->sop_level; 2989 sopt.sopt_name = params->sop_optname; 2990 sopt.sopt_dir = SOPT_SET; 2991 sopt.sopt_val = params->sop_optval; 2992 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters); 2993 sopt.sopt_td = NULL; 2994 #ifdef INET6 2995 if (params->sop_inc.inc_flags & INC_ISIPV6) { 2996 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr)) 2997 params->sop_inc.inc6_laddr.s6_addr16[1] = 2998 htons(params->sop_inc.inc6_zoneid & 0xffff); 2999 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr)) 3000 params->sop_inc.inc6_faddr.s6_addr16[1] = 3001 htons(params->sop_inc.inc6_zoneid & 0xffff); 3002 } 3003 #endif 3004 if (params->sop_inc.inc_lport != htons(0)) { 3005 if (params->sop_inc.inc_fport == htons(0)) 3006 inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport, 3007 pcbinfo->ipi_hashmask); 3008 else 3009 #ifdef INET6 3010 if (params->sop_inc.inc_flags & INC_ISIPV6) 3011 inpi.hash = INP6_PCBHASH( 3012 ¶ms->sop_inc.inc6_faddr, 3013 params->sop_inc.inc_lport, 3014 params->sop_inc.inc_fport, 3015 pcbinfo->ipi_hashmask); 3016 else 3017 #endif 3018 inpi.hash = INP_PCBHASH( 3019 ¶ms->sop_inc.inc_faddr, 3020 params->sop_inc.inc_lport, 3021 params->sop_inc.inc_fport, 3022 pcbinfo->ipi_hashmask); 3023 } 3024 while ((inp = inp_next(&inpi)) != NULL) 3025 if (inp->inp_gencnt == params->sop_id) { 3026 if (inp->inp_flags & INP_DROPPED) { 3027 INP_WUNLOCK(inp); 3028 return (ECONNRESET); 3029 } 3030 so = inp->inp_socket; 3031 KASSERT(so != NULL, ("inp_socket == NULL")); 3032 soref(so); 3033 error = (*ctloutput_set)(inp, &sopt); 3034 sorele(so); 3035 break; 3036 } 3037 if (inp == NULL) 3038 error = ESRCH; 3039 return (error); 3040 } 3041 3042 #ifdef DDB 3043 static void 3044 db_print_indent(int indent) 3045 { 3046 int i; 3047 3048 for (i = 0; i < indent; i++) 3049 db_printf(" "); 3050 } 3051 3052 static void 3053 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 3054 { 3055 char faddr_str[48], laddr_str[48]; 3056 3057 db_print_indent(indent); 3058 db_printf("%s at %p\n", name, inc); 3059 3060 indent += 2; 3061 3062 #ifdef INET6 3063 if (inc->inc_flags & INC_ISIPV6) { 3064 /* IPv6. */ 3065 ip6_sprintf(laddr_str, &inc->inc6_laddr); 3066 ip6_sprintf(faddr_str, &inc->inc6_faddr); 3067 } else 3068 #endif 3069 { 3070 /* IPv4. */ 3071 inet_ntoa_r(inc->inc_laddr, laddr_str); 3072 inet_ntoa_r(inc->inc_faddr, faddr_str); 3073 } 3074 db_print_indent(indent); 3075 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3076 ntohs(inc->inc_lport)); 3077 db_print_indent(indent); 3078 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3079 ntohs(inc->inc_fport)); 3080 } 3081 3082 static void 3083 db_print_inpflags(int inp_flags) 3084 { 3085 int comma; 3086 3087 comma = 0; 3088 if (inp_flags & INP_RECVOPTS) { 3089 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3090 comma = 1; 3091 } 3092 if (inp_flags & INP_RECVRETOPTS) { 3093 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3094 comma = 1; 3095 } 3096 if (inp_flags & INP_RECVDSTADDR) { 3097 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3098 comma = 1; 3099 } 3100 if (inp_flags & INP_ORIGDSTADDR) { 3101 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3102 comma = 1; 3103 } 3104 if (inp_flags & INP_HDRINCL) { 3105 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3106 comma = 1; 3107 } 3108 if (inp_flags & INP_HIGHPORT) { 3109 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3110 comma = 1; 3111 } 3112 if (inp_flags & INP_LOWPORT) { 3113 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3114 comma = 1; 3115 } 3116 if (inp_flags & INP_ANONPORT) { 3117 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3118 comma = 1; 3119 } 3120 if (inp_flags & INP_RECVIF) { 3121 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3122 comma = 1; 3123 } 3124 if (inp_flags & INP_MTUDISC) { 3125 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3126 comma = 1; 3127 } 3128 if (inp_flags & INP_RECVTTL) { 3129 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3130 comma = 1; 3131 } 3132 if (inp_flags & INP_DONTFRAG) { 3133 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3134 comma = 1; 3135 } 3136 if (inp_flags & INP_RECVTOS) { 3137 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3138 comma = 1; 3139 } 3140 if (inp_flags & IN6P_IPV6_V6ONLY) { 3141 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3142 comma = 1; 3143 } 3144 if (inp_flags & IN6P_PKTINFO) { 3145 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3146 comma = 1; 3147 } 3148 if (inp_flags & IN6P_HOPLIMIT) { 3149 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3150 comma = 1; 3151 } 3152 if (inp_flags & IN6P_HOPOPTS) { 3153 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3154 comma = 1; 3155 } 3156 if (inp_flags & IN6P_DSTOPTS) { 3157 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3158 comma = 1; 3159 } 3160 if (inp_flags & IN6P_RTHDR) { 3161 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3162 comma = 1; 3163 } 3164 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3165 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3166 comma = 1; 3167 } 3168 if (inp_flags & IN6P_TCLASS) { 3169 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3170 comma = 1; 3171 } 3172 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3173 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3174 comma = 1; 3175 } 3176 if (inp_flags & INP_ONESBCAST) { 3177 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3178 comma = 1; 3179 } 3180 if (inp_flags & INP_DROPPED) { 3181 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3182 comma = 1; 3183 } 3184 if (inp_flags & INP_SOCKREF) { 3185 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3186 comma = 1; 3187 } 3188 if (inp_flags & IN6P_RFC2292) { 3189 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3190 comma = 1; 3191 } 3192 if (inp_flags & IN6P_MTU) { 3193 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3194 comma = 1; 3195 } 3196 } 3197 3198 static void 3199 db_print_inpvflag(u_char inp_vflag) 3200 { 3201 int comma; 3202 3203 comma = 0; 3204 if (inp_vflag & INP_IPV4) { 3205 db_printf("%sINP_IPV4", comma ? ", " : ""); 3206 comma = 1; 3207 } 3208 if (inp_vflag & INP_IPV6) { 3209 db_printf("%sINP_IPV6", comma ? ", " : ""); 3210 comma = 1; 3211 } 3212 if (inp_vflag & INP_IPV6PROTO) { 3213 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3214 comma = 1; 3215 } 3216 } 3217 3218 static void 3219 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3220 { 3221 3222 db_print_indent(indent); 3223 db_printf("%s at %p\n", name, inp); 3224 3225 indent += 2; 3226 3227 db_print_indent(indent); 3228 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3229 3230 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3231 3232 db_print_indent(indent); 3233 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3234 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3235 3236 db_print_indent(indent); 3237 db_printf("inp_label: %p inp_flags: 0x%x (", 3238 inp->inp_label, inp->inp_flags); 3239 db_print_inpflags(inp->inp_flags); 3240 db_printf(")\n"); 3241 3242 db_print_indent(indent); 3243 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3244 inp->inp_vflag); 3245 db_print_inpvflag(inp->inp_vflag); 3246 db_printf(")\n"); 3247 3248 db_print_indent(indent); 3249 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3250 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3251 3252 db_print_indent(indent); 3253 #ifdef INET6 3254 if (inp->inp_vflag & INP_IPV6) { 3255 db_printf("in6p_options: %p in6p_outputopts: %p " 3256 "in6p_moptions: %p\n", inp->in6p_options, 3257 inp->in6p_outputopts, inp->in6p_moptions); 3258 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3259 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3260 inp->in6p_hops); 3261 } else 3262 #endif 3263 { 3264 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3265 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3266 inp->inp_options, inp->inp_moptions); 3267 } 3268 3269 db_print_indent(indent); 3270 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3271 (uintmax_t)inp->inp_gencnt); 3272 } 3273 3274 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3275 { 3276 struct inpcb *inp; 3277 3278 if (!have_addr) { 3279 db_printf("usage: show inpcb <addr>\n"); 3280 return; 3281 } 3282 inp = (struct inpcb *)addr; 3283 3284 db_print_inpcb(inp, "inpcb", 0); 3285 } 3286 #endif /* DDB */ 3287 3288 #ifdef RATELIMIT 3289 /* 3290 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3291 * if any. 3292 */ 3293 int 3294 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3295 { 3296 union if_snd_tag_modify_params params = { 3297 .rate_limit.max_rate = max_pacing_rate, 3298 .rate_limit.flags = M_NOWAIT, 3299 }; 3300 struct m_snd_tag *mst; 3301 int error; 3302 3303 mst = inp->inp_snd_tag; 3304 if (mst == NULL) 3305 return (EINVAL); 3306 3307 if (mst->sw->snd_tag_modify == NULL) { 3308 error = EOPNOTSUPP; 3309 } else { 3310 error = mst->sw->snd_tag_modify(mst, ¶ms); 3311 } 3312 return (error); 3313 } 3314 3315 /* 3316 * Query existing TX rate limit based on the existing 3317 * "inp->inp_snd_tag", if any. 3318 */ 3319 int 3320 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3321 { 3322 union if_snd_tag_query_params params = { }; 3323 struct m_snd_tag *mst; 3324 int error; 3325 3326 mst = inp->inp_snd_tag; 3327 if (mst == NULL) 3328 return (EINVAL); 3329 3330 if (mst->sw->snd_tag_query == NULL) { 3331 error = EOPNOTSUPP; 3332 } else { 3333 error = mst->sw->snd_tag_query(mst, ¶ms); 3334 if (error == 0 && p_max_pacing_rate != NULL) 3335 *p_max_pacing_rate = params.rate_limit.max_rate; 3336 } 3337 return (error); 3338 } 3339 3340 /* 3341 * Query existing TX queue level based on the existing 3342 * "inp->inp_snd_tag", if any. 3343 */ 3344 int 3345 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3346 { 3347 union if_snd_tag_query_params params = { }; 3348 struct m_snd_tag *mst; 3349 int error; 3350 3351 mst = inp->inp_snd_tag; 3352 if (mst == NULL) 3353 return (EINVAL); 3354 3355 if (mst->sw->snd_tag_query == NULL) 3356 return (EOPNOTSUPP); 3357 3358 error = mst->sw->snd_tag_query(mst, ¶ms); 3359 if (error == 0 && p_txqueue_level != NULL) 3360 *p_txqueue_level = params.rate_limit.queue_level; 3361 return (error); 3362 } 3363 3364 /* 3365 * Allocate a new TX rate limit send tag from the network interface 3366 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3367 */ 3368 int 3369 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3370 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3371 3372 { 3373 union if_snd_tag_alloc_params params = { 3374 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3375 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3376 .rate_limit.hdr.flowid = flowid, 3377 .rate_limit.hdr.flowtype = flowtype, 3378 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3379 .rate_limit.max_rate = max_pacing_rate, 3380 .rate_limit.flags = M_NOWAIT, 3381 }; 3382 int error; 3383 3384 INP_WLOCK_ASSERT(inp); 3385 3386 /* 3387 * If there is already a send tag, or the INP is being torn 3388 * down, allocating a new send tag is not allowed. Else send 3389 * tags may leak. 3390 */ 3391 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0) 3392 return (EINVAL); 3393 3394 error = m_snd_tag_alloc(ifp, ¶ms, st); 3395 #ifdef INET 3396 if (error == 0) { 3397 counter_u64_add(rate_limit_set_ok, 1); 3398 counter_u64_add(rate_limit_active, 1); 3399 } else if (error != EOPNOTSUPP) 3400 counter_u64_add(rate_limit_alloc_fail, 1); 3401 #endif 3402 return (error); 3403 } 3404 3405 void 3406 in_pcbdetach_tag(struct m_snd_tag *mst) 3407 { 3408 3409 m_snd_tag_rele(mst); 3410 #ifdef INET 3411 counter_u64_add(rate_limit_active, -1); 3412 #endif 3413 } 3414 3415 /* 3416 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3417 * if any: 3418 */ 3419 void 3420 in_pcbdetach_txrtlmt(struct inpcb *inp) 3421 { 3422 struct m_snd_tag *mst; 3423 3424 INP_WLOCK_ASSERT(inp); 3425 3426 mst = inp->inp_snd_tag; 3427 inp->inp_snd_tag = NULL; 3428 3429 if (mst == NULL) 3430 return; 3431 3432 m_snd_tag_rele(mst); 3433 #ifdef INET 3434 counter_u64_add(rate_limit_active, -1); 3435 #endif 3436 } 3437 3438 int 3439 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3440 { 3441 int error; 3442 3443 /* 3444 * If the existing send tag is for the wrong interface due to 3445 * a route change, first drop the existing tag. Set the 3446 * CHANGED flag so that we will keep trying to allocate a new 3447 * tag if we fail to allocate one this time. 3448 */ 3449 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3450 in_pcbdetach_txrtlmt(inp); 3451 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3452 } 3453 3454 /* 3455 * NOTE: When attaching to a network interface a reference is 3456 * made to ensure the network interface doesn't go away until 3457 * all ratelimit connections are gone. The network interface 3458 * pointers compared below represent valid network interfaces, 3459 * except when comparing towards NULL. 3460 */ 3461 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3462 error = 0; 3463 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3464 if (inp->inp_snd_tag != NULL) 3465 in_pcbdetach_txrtlmt(inp); 3466 error = 0; 3467 } else if (inp->inp_snd_tag == NULL) { 3468 /* 3469 * In order to utilize packet pacing with RSS, we need 3470 * to wait until there is a valid RSS hash before we 3471 * can proceed: 3472 */ 3473 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3474 error = EAGAIN; 3475 } else { 3476 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3477 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3478 } 3479 } else { 3480 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3481 } 3482 if (error == 0 || error == EOPNOTSUPP) 3483 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3484 3485 return (error); 3486 } 3487 3488 /* 3489 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3490 * is set in the fast path and will attach/detach/modify the TX rate 3491 * limit send tag based on the socket's so_max_pacing_rate value. 3492 */ 3493 void 3494 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3495 { 3496 struct socket *socket; 3497 uint32_t max_pacing_rate; 3498 bool did_upgrade; 3499 3500 if (inp == NULL) 3501 return; 3502 3503 socket = inp->inp_socket; 3504 if (socket == NULL) 3505 return; 3506 3507 if (!INP_WLOCKED(inp)) { 3508 /* 3509 * NOTE: If the write locking fails, we need to bail 3510 * out and use the non-ratelimited ring for the 3511 * transmit until there is a new chance to get the 3512 * write lock. 3513 */ 3514 if (!INP_TRY_UPGRADE(inp)) 3515 return; 3516 did_upgrade = 1; 3517 } else { 3518 did_upgrade = 0; 3519 } 3520 3521 /* 3522 * NOTE: The so_max_pacing_rate value is read unlocked, 3523 * because atomic updates are not required since the variable 3524 * is checked at every mbuf we send. It is assumed that the 3525 * variable read itself will be atomic. 3526 */ 3527 max_pacing_rate = socket->so_max_pacing_rate; 3528 3529 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3530 3531 if (did_upgrade) 3532 INP_DOWNGRADE(inp); 3533 } 3534 3535 /* 3536 * Track route changes for TX rate limiting. 3537 */ 3538 void 3539 in_pcboutput_eagain(struct inpcb *inp) 3540 { 3541 bool did_upgrade; 3542 3543 if (inp == NULL) 3544 return; 3545 3546 if (inp->inp_snd_tag == NULL) 3547 return; 3548 3549 if (!INP_WLOCKED(inp)) { 3550 /* 3551 * NOTE: If the write locking fails, we need to bail 3552 * out and use the non-ratelimited ring for the 3553 * transmit until there is a new chance to get the 3554 * write lock. 3555 */ 3556 if (!INP_TRY_UPGRADE(inp)) 3557 return; 3558 did_upgrade = 1; 3559 } else { 3560 did_upgrade = 0; 3561 } 3562 3563 /* detach rate limiting */ 3564 in_pcbdetach_txrtlmt(inp); 3565 3566 /* make sure new mbuf send tag allocation is made */ 3567 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3568 3569 if (did_upgrade) 3570 INP_DOWNGRADE(inp); 3571 } 3572 3573 #ifdef INET 3574 static void 3575 rl_init(void *st) 3576 { 3577 rate_limit_new = counter_u64_alloc(M_WAITOK); 3578 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3579 rate_limit_active = counter_u64_alloc(M_WAITOK); 3580 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3581 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3582 } 3583 3584 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3585 #endif 3586 #endif /* RATELIMIT */ 3587