1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1991, 1993, 1995 5 * The Regents of the University of California. 6 * Copyright (c) 2007-2009 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * Copyright (c) 2021-2022 Gleb Smirnoff <glebius@FreeBSD.org> 9 * All rights reserved. 10 * 11 * Portions of this software were developed by Robert N. M. Watson under 12 * contract to Juniper Networks, Inc. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 3. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 */ 38 39 #include <sys/cdefs.h> 40 #include "opt_ddb.h" 41 #include "opt_ipsec.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 #include "opt_ratelimit.h" 45 #include "opt_route.h" 46 #include "opt_rss.h" 47 48 #include <sys/param.h> 49 #include <sys/hash.h> 50 #include <sys/systm.h> 51 #include <sys/libkern.h> 52 #include <sys/lock.h> 53 #include <sys/malloc.h> 54 #include <sys/mbuf.h> 55 #include <sys/eventhandler.h> 56 #include <sys/domain.h> 57 #include <sys/proc.h> 58 #include <sys/protosw.h> 59 #include <sys/smp.h> 60 #include <sys/smr.h> 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/sockio.h> 64 #include <sys/priv.h> 65 #include <sys/proc.h> 66 #include <sys/refcount.h> 67 #include <sys/jail.h> 68 #include <sys/kernel.h> 69 #include <sys/sysctl.h> 70 71 #ifdef DDB 72 #include <ddb/ddb.h> 73 #endif 74 75 #include <vm/uma.h> 76 #include <vm/vm.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/if_private.h> 81 #include <net/if_types.h> 82 #include <net/if_llatbl.h> 83 #include <net/route.h> 84 #include <net/rss_config.h> 85 #include <net/vnet.h> 86 87 #if defined(INET) || defined(INET6) 88 #include <netinet/in.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/in_pcb_var.h> 91 #include <netinet/tcp.h> 92 #ifdef INET 93 #include <netinet/in_var.h> 94 #include <netinet/in_fib.h> 95 #endif 96 #include <netinet/ip_var.h> 97 #ifdef INET6 98 #include <netinet/ip6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_var.h> 101 #include <netinet6/ip6_var.h> 102 #endif /* INET6 */ 103 #include <net/route/nhop.h> 104 #endif 105 106 #include <netipsec/ipsec_support.h> 107 108 #include <security/mac/mac_framework.h> 109 110 #define INPCBLBGROUP_SIZMIN 8 111 #define INPCBLBGROUP_SIZMAX 256 112 113 #define INP_FREED 0x00000200 /* Went through in_pcbfree(). */ 114 #define INP_INLBGROUP 0x01000000 /* Inserted into inpcblbgroup. */ 115 116 /* 117 * These configure the range of local port addresses assigned to 118 * "unspecified" outgoing connections/packets/whatever. 119 */ 120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 126 127 /* 128 * Reserved ports accessible only to root. There are significant 129 * security considerations that must be accounted for when changing these, 130 * but the security benefits can be great. Please be careful. 131 */ 132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 133 VNET_DEFINE(int, ipport_reservedlow); 134 135 /* Enable random ephemeral port allocation by default. */ 136 VNET_DEFINE(int, ipport_randomized) = 1; 137 138 #ifdef INET 139 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 140 struct in_addr faddr, u_int fport_arg, 141 struct in_addr laddr, u_int lport_arg, 142 int lookupflags, uint8_t numa_domain); 143 144 #define RANGECHK(var, min, max) \ 145 if ((var) < (min)) { (var) = (min); } \ 146 else if ((var) > (max)) { (var) = (max); } 147 148 static int 149 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 150 { 151 int error; 152 153 error = sysctl_handle_int(oidp, arg1, arg2, req); 154 if (error == 0) { 155 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 156 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 157 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 158 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 159 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 160 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 161 } 162 return (error); 163 } 164 165 #undef RANGECHK 166 167 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, 168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 169 "IP Ports"); 170 171 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 172 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 173 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", 174 ""); 175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 176 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 177 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", 178 ""); 179 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 180 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 181 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", 182 ""); 183 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 184 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 185 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", 186 ""); 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 189 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", 190 ""); 191 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 192 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 193 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", 194 ""); 195 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 196 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 197 &VNET_NAME(ipport_reservedhigh), 0, ""); 198 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 199 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 201 CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 203 204 #ifdef RATELIMIT 205 counter_u64_t rate_limit_new; 206 counter_u64_t rate_limit_chg; 207 counter_u64_t rate_limit_active; 208 counter_u64_t rate_limit_alloc_fail; 209 counter_u64_t rate_limit_set_ok; 210 211 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 212 "IP Rate Limiting"); 213 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, 214 &rate_limit_active, "Active rate limited connections"); 215 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, 216 &rate_limit_alloc_fail, "Rate limited connection failures"); 217 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, 218 &rate_limit_set_ok, "Rate limited setting succeeded"); 219 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD, 220 &rate_limit_new, "Total Rate limit new attempts"); 221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD, 222 &rate_limit_chg, "Total Rate limited change attempts"); 223 #endif /* RATELIMIT */ 224 225 #endif /* INET */ 226 227 VNET_DEFINE(uint32_t, in_pcbhashseed); 228 static void 229 in_pcbhashseed_init(void) 230 { 231 232 V_in_pcbhashseed = arc4random(); 233 } 234 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, 235 in_pcbhashseed_init, 0); 236 237 static void in_pcbremhash(struct inpcb *); 238 239 /* 240 * in_pcb.c: manage the Protocol Control Blocks. 241 * 242 * NOTE: It is assumed that most of these functions will be called with 243 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 244 * functions often modify hash chains or addresses in pcbs. 245 */ 246 247 static struct inpcblbgroup * 248 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred, 249 u_char vflag, uint16_t port, const union in_dependaddr *addr, int size, 250 uint8_t numa_domain) 251 { 252 struct inpcblbgroup *grp; 253 size_t bytes; 254 255 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 256 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 257 if (grp == NULL) 258 return (NULL); 259 grp->il_cred = crhold(cred); 260 grp->il_vflag = vflag; 261 grp->il_lport = port; 262 grp->il_numa_domain = numa_domain; 263 grp->il_dependladdr = *addr; 264 grp->il_inpsiz = size; 265 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 266 return (grp); 267 } 268 269 static void 270 in_pcblbgroup_free_deferred(epoch_context_t ctx) 271 { 272 struct inpcblbgroup *grp; 273 274 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 275 crfree(grp->il_cred); 276 free(grp, M_PCB); 277 } 278 279 static void 280 in_pcblbgroup_free(struct inpcblbgroup *grp) 281 { 282 283 CK_LIST_REMOVE(grp, il_list); 284 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); 285 } 286 287 static struct inpcblbgroup * 288 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 289 struct inpcblbgroup *old_grp, int size) 290 { 291 struct inpcblbgroup *grp; 292 int i; 293 294 grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag, 295 old_grp->il_lport, &old_grp->il_dependladdr, size, 296 old_grp->il_numa_domain); 297 if (grp == NULL) 298 return (NULL); 299 300 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 301 ("invalid new local group size %d and old local group count %d", 302 grp->il_inpsiz, old_grp->il_inpcnt)); 303 304 for (i = 0; i < old_grp->il_inpcnt; ++i) 305 grp->il_inp[i] = old_grp->il_inp[i]; 306 grp->il_inpcnt = old_grp->il_inpcnt; 307 in_pcblbgroup_free(old_grp); 308 return (grp); 309 } 310 311 /* 312 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 313 * and shrink group if possible. 314 */ 315 static void 316 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 317 int i) 318 { 319 struct inpcblbgroup *grp, *new_grp; 320 321 grp = *grpp; 322 for (; i + 1 < grp->il_inpcnt; ++i) 323 grp->il_inp[i] = grp->il_inp[i + 1]; 324 grp->il_inpcnt--; 325 326 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 327 grp->il_inpcnt <= grp->il_inpsiz / 4) { 328 /* Shrink this group. */ 329 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 330 if (new_grp != NULL) 331 *grpp = new_grp; 332 } 333 } 334 335 /* 336 * Add PCB to load balance group for SO_REUSEPORT_LB option. 337 */ 338 static int 339 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain) 340 { 341 const static struct timeval interval = { 60, 0 }; 342 static struct timeval lastprint; 343 struct inpcbinfo *pcbinfo; 344 struct inpcblbgrouphead *hdr; 345 struct inpcblbgroup *grp; 346 uint32_t idx; 347 348 pcbinfo = inp->inp_pcbinfo; 349 350 INP_WLOCK_ASSERT(inp); 351 INP_HASH_WLOCK_ASSERT(pcbinfo); 352 353 #ifdef INET6 354 /* 355 * Don't allow IPv4 mapped INET6 wild socket. 356 */ 357 if ((inp->inp_vflag & INP_IPV4) && 358 inp->inp_laddr.s_addr == INADDR_ANY && 359 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 360 return (0); 361 } 362 #endif 363 364 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 365 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 366 CK_LIST_FOREACH(grp, hdr, il_list) { 367 if (grp->il_cred->cr_prison == inp->inp_cred->cr_prison && 368 grp->il_vflag == inp->inp_vflag && 369 grp->il_lport == inp->inp_lport && 370 grp->il_numa_domain == numa_domain && 371 memcmp(&grp->il_dependladdr, 372 &inp->inp_inc.inc_ie.ie_dependladdr, 373 sizeof(grp->il_dependladdr)) == 0) { 374 break; 375 } 376 } 377 if (grp == NULL) { 378 /* Create new load balance group. */ 379 grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag, 380 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 381 INPCBLBGROUP_SIZMIN, numa_domain); 382 if (grp == NULL) 383 return (ENOBUFS); 384 } else if (grp->il_inpcnt == grp->il_inpsiz) { 385 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 386 if (ratecheck(&lastprint, &interval)) 387 printf("lb group port %d, limit reached\n", 388 ntohs(grp->il_lport)); 389 return (0); 390 } 391 392 /* Expand this local group. */ 393 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 394 if (grp == NULL) 395 return (ENOBUFS); 396 } 397 398 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 399 ("invalid local group size %d and count %d", grp->il_inpsiz, 400 grp->il_inpcnt)); 401 402 grp->il_inp[grp->il_inpcnt] = inp; 403 grp->il_inpcnt++; 404 inp->inp_flags |= INP_INLBGROUP; 405 return (0); 406 } 407 408 /* 409 * Remove PCB from load balance group. 410 */ 411 static void 412 in_pcbremlbgrouphash(struct inpcb *inp) 413 { 414 struct inpcbinfo *pcbinfo; 415 struct inpcblbgrouphead *hdr; 416 struct inpcblbgroup *grp; 417 int i; 418 419 pcbinfo = inp->inp_pcbinfo; 420 421 INP_WLOCK_ASSERT(inp); 422 MPASS(inp->inp_flags & INP_INLBGROUP); 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 inp->inp_flags &= ~INP_INLBGROUP; 440 return; 441 } 442 } 443 KASSERT(0, ("%s: did not find %p", __func__, inp)); 444 } 445 446 int 447 in_pcblbgroup_numa(struct inpcb *inp, int arg) 448 { 449 struct inpcbinfo *pcbinfo; 450 struct inpcblbgrouphead *hdr; 451 struct inpcblbgroup *grp; 452 int err, i; 453 uint8_t numa_domain; 454 455 switch (arg) { 456 case TCP_REUSPORT_LB_NUMA_NODOM: 457 numa_domain = M_NODOM; 458 break; 459 case TCP_REUSPORT_LB_NUMA_CURDOM: 460 numa_domain = PCPU_GET(domain); 461 break; 462 default: 463 if (arg < 0 || arg >= vm_ndomains) 464 return (EINVAL); 465 numa_domain = arg; 466 } 467 468 err = 0; 469 pcbinfo = inp->inp_pcbinfo; 470 INP_WLOCK_ASSERT(inp); 471 INP_HASH_WLOCK(pcbinfo); 472 hdr = &pcbinfo->ipi_lbgrouphashbase[ 473 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 474 CK_LIST_FOREACH(grp, hdr, il_list) { 475 for (i = 0; i < grp->il_inpcnt; ++i) { 476 if (grp->il_inp[i] != inp) 477 continue; 478 479 if (grp->il_numa_domain == numa_domain) { 480 goto abort_with_hash_wlock; 481 } 482 483 /* Remove it from the old group. */ 484 in_pcbremlbgrouphash(inp); 485 486 /* Add it to the new group based on numa domain. */ 487 in_pcbinslbgrouphash(inp, numa_domain); 488 goto abort_with_hash_wlock; 489 } 490 } 491 err = ENOENT; 492 abort_with_hash_wlock: 493 INP_HASH_WUNLOCK(pcbinfo); 494 return (err); 495 } 496 497 /* Make sure it is safe to use hashinit(9) on CK_LIST. */ 498 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb))); 499 500 /* 501 * Initialize an inpcbinfo - a per-VNET instance of connections db. 502 */ 503 void 504 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor, 505 u_int hash_nelements, u_int porthash_nelements) 506 { 507 508 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF); 509 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name, 510 NULL, MTX_DEF); 511 #ifdef VIMAGE 512 pcbinfo->ipi_vnet = curvnet; 513 #endif 514 CK_LIST_INIT(&pcbinfo->ipi_listhead); 515 pcbinfo->ipi_count = 0; 516 pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB, 517 &pcbinfo->ipi_hashmask); 518 pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB, 519 &pcbinfo->ipi_hashmask); 520 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 521 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 522 &pcbinfo->ipi_porthashmask); 523 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 524 &pcbinfo->ipi_lbgrouphashmask); 525 pcbinfo->ipi_zone = pcbstor->ips_zone; 526 pcbinfo->ipi_portzone = pcbstor->ips_portzone; 527 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone); 528 } 529 530 /* 531 * Destroy an inpcbinfo. 532 */ 533 void 534 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 535 { 536 537 KASSERT(pcbinfo->ipi_count == 0, 538 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 539 540 hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask); 541 hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask); 542 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 543 pcbinfo->ipi_porthashmask); 544 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 545 pcbinfo->ipi_lbgrouphashmask); 546 mtx_destroy(&pcbinfo->ipi_hash_lock); 547 mtx_destroy(&pcbinfo->ipi_lock); 548 } 549 550 /* 551 * Initialize a pcbstorage - per protocol zones to allocate inpcbs. 552 */ 553 static void inpcb_fini(void *, int); 554 void 555 in_pcbstorage_init(void *arg) 556 { 557 struct inpcbstorage *pcbstor = arg; 558 559 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name, 560 pcbstor->ips_size, NULL, NULL, pcbstor->ips_pcbinit, 561 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR); 562 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name, 563 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 564 uma_zone_set_smr(pcbstor->ips_portzone, 565 uma_zone_get_smr(pcbstor->ips_zone)); 566 } 567 568 /* 569 * Destroy a pcbstorage - used by unloadable protocols. 570 */ 571 void 572 in_pcbstorage_destroy(void *arg) 573 { 574 struct inpcbstorage *pcbstor = arg; 575 576 uma_zdestroy(pcbstor->ips_zone); 577 uma_zdestroy(pcbstor->ips_portzone); 578 } 579 580 /* 581 * Allocate a PCB and associate it with the socket. 582 * On success return with the PCB locked. 583 */ 584 int 585 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 586 { 587 struct inpcb *inp; 588 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 589 int error; 590 #endif 591 592 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT); 593 if (inp == NULL) 594 return (ENOBUFS); 595 bzero(&inp->inp_start_zero, inp_zero_size); 596 #ifdef NUMA 597 inp->inp_numa_domain = M_NODOM; 598 #endif 599 inp->inp_pcbinfo = pcbinfo; 600 inp->inp_socket = so; 601 inp->inp_cred = crhold(so->so_cred); 602 inp->inp_inc.inc_fibnum = so->so_fibnum; 603 #ifdef MAC 604 error = mac_inpcb_init(inp, M_NOWAIT); 605 if (error != 0) 606 goto out; 607 mac_inpcb_create(so, inp); 608 #endif 609 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 610 error = ipsec_init_pcbpolicy(inp); 611 if (error != 0) { 612 #ifdef MAC 613 mac_inpcb_destroy(inp); 614 #endif 615 goto out; 616 } 617 #endif /*IPSEC*/ 618 #ifdef INET6 619 if (INP_SOCKAF(so) == AF_INET6) { 620 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6; 621 if (V_ip6_v6only) 622 inp->inp_flags |= IN6P_IPV6_V6ONLY; 623 #ifdef INET 624 else 625 inp->inp_vflag |= INP_IPV4; 626 #endif 627 if (V_ip6_auto_flowlabel) 628 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 629 inp->in6p_hops = -1; /* use kernel default */ 630 } 631 #endif 632 #if defined(INET) && defined(INET6) 633 else 634 #endif 635 #ifdef INET 636 inp->inp_vflag |= INP_IPV4; 637 #endif 638 inp->inp_smr = SMR_SEQ_INVALID; 639 640 /* 641 * Routes in inpcb's can cache L2 as well; they are guaranteed 642 * to be cleaned up. 643 */ 644 inp->inp_route.ro_flags = RT_LLE_CACHE; 645 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */ 646 INP_WLOCK(inp); 647 INP_INFO_WLOCK(pcbinfo); 648 pcbinfo->ipi_count++; 649 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 650 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list); 651 INP_INFO_WUNLOCK(pcbinfo); 652 so->so_pcb = inp; 653 654 return (0); 655 656 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 657 out: 658 uma_zfree_smr(pcbinfo->ipi_zone, inp); 659 return (error); 660 #endif 661 } 662 663 #ifdef INET 664 int 665 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred) 666 { 667 int anonport, error; 668 669 KASSERT(sin == NULL || sin->sin_family == AF_INET, 670 ("%s: invalid address family for %p", __func__, sin)); 671 KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in), 672 ("%s: invalid address length for %p", __func__, sin)); 673 INP_WLOCK_ASSERT(inp); 674 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 675 676 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 677 return (EINVAL); 678 anonport = sin == NULL || sin->sin_port == 0; 679 error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr, 680 &inp->inp_lport, cred); 681 if (error) 682 return (error); 683 if (in_pcbinshash(inp) != 0) { 684 inp->inp_laddr.s_addr = INADDR_ANY; 685 inp->inp_lport = 0; 686 return (EAGAIN); 687 } 688 if (anonport) 689 inp->inp_flags |= INP_ANONPORT; 690 return (0); 691 } 692 #endif 693 694 #if defined(INET) || defined(INET6) 695 /* 696 * Assign a local port like in_pcb_lport(), but also used with connect() 697 * and a foreign address and port. If fsa is non-NULL, choose a local port 698 * that is unused with those, otherwise one that is completely unused. 699 * lsa can be NULL for IPv6. 700 */ 701 int 702 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 703 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 704 { 705 struct inpcbinfo *pcbinfo; 706 struct inpcb *tmpinp; 707 unsigned short *lastport; 708 int count, error; 709 u_short aux, first, last, lport; 710 #ifdef INET 711 struct in_addr laddr, faddr; 712 #endif 713 #ifdef INET6 714 struct in6_addr *laddr6, *faddr6; 715 #endif 716 717 pcbinfo = inp->inp_pcbinfo; 718 719 /* 720 * Because no actual state changes occur here, a global write lock on 721 * the pcbinfo isn't required. 722 */ 723 INP_LOCK_ASSERT(inp); 724 INP_HASH_LOCK_ASSERT(pcbinfo); 725 726 if (inp->inp_flags & INP_HIGHPORT) { 727 first = V_ipport_hifirstauto; /* sysctl */ 728 last = V_ipport_hilastauto; 729 lastport = &pcbinfo->ipi_lasthi; 730 } else if (inp->inp_flags & INP_LOWPORT) { 731 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 732 if (error) 733 return (error); 734 first = V_ipport_lowfirstauto; /* 1023 */ 735 last = V_ipport_lowlastauto; /* 600 */ 736 lastport = &pcbinfo->ipi_lastlow; 737 } else { 738 first = V_ipport_firstauto; /* sysctl */ 739 last = V_ipport_lastauto; 740 lastport = &pcbinfo->ipi_lastport; 741 } 742 743 /* 744 * Instead of having two loops further down counting up or down 745 * make sure that first is always <= last and go with only one 746 * code path implementing all logic. 747 */ 748 if (first > last) { 749 aux = first; 750 first = last; 751 last = aux; 752 } 753 754 #ifdef INET 755 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */ 756 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 757 if (lsa != NULL) 758 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 759 if (fsa != NULL) 760 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 761 } 762 #endif 763 #ifdef INET6 764 laddr6 = NULL; 765 if ((inp->inp_vflag & INP_IPV6) != 0) { 766 if (lsa != NULL) 767 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 768 if (fsa != NULL) 769 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 770 } 771 #endif 772 773 tmpinp = NULL; 774 lport = *lportp; 775 776 if (V_ipport_randomized) 777 *lastport = first + (arc4random() % (last - first)); 778 779 count = last - first; 780 781 do { 782 if (count-- < 0) /* completely used? */ 783 return (EADDRNOTAVAIL); 784 ++*lastport; 785 if (*lastport < first || *lastport > last) 786 *lastport = first; 787 lport = htons(*lastport); 788 789 if (fsa != NULL) { 790 #ifdef INET 791 if (lsa->sa_family == AF_INET) { 792 tmpinp = in_pcblookup_hash_locked(pcbinfo, 793 faddr, fport, laddr, lport, lookupflags, 794 M_NODOM); 795 } 796 #endif 797 #ifdef INET6 798 if (lsa->sa_family == AF_INET6) { 799 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 800 faddr6, fport, laddr6, lport, lookupflags, 801 M_NODOM); 802 } 803 #endif 804 } else { 805 #ifdef INET6 806 if ((inp->inp_vflag & INP_IPV6) != 0) { 807 tmpinp = in6_pcblookup_local(pcbinfo, 808 &inp->in6p_laddr, lport, lookupflags, cred); 809 #ifdef INET 810 if (tmpinp == NULL && 811 (inp->inp_vflag & INP_IPV4)) 812 tmpinp = in_pcblookup_local(pcbinfo, 813 laddr, lport, lookupflags, cred); 814 #endif 815 } 816 #endif 817 #if defined(INET) && defined(INET6) 818 else 819 #endif 820 #ifdef INET 821 tmpinp = in_pcblookup_local(pcbinfo, laddr, 822 lport, lookupflags, cred); 823 #endif 824 } 825 } while (tmpinp != NULL); 826 827 *lportp = lport; 828 829 return (0); 830 } 831 832 /* 833 * Select a local port (number) to use. 834 */ 835 int 836 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 837 struct ucred *cred, int lookupflags) 838 { 839 struct sockaddr_in laddr; 840 841 if (laddrp) { 842 bzero(&laddr, sizeof(laddr)); 843 laddr.sin_family = AF_INET; 844 laddr.sin_addr = *laddrp; 845 } 846 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 847 NULL, lportp, NULL, 0, cred, lookupflags)); 848 } 849 #endif /* INET || INET6 */ 850 851 #ifdef INET 852 /* 853 * Set up a bind operation on a PCB, performing port allocation 854 * as required, but do not actually modify the PCB. Callers can 855 * either complete the bind by setting inp_laddr/inp_lport and 856 * calling in_pcbinshash(), or they can just use the resulting 857 * port and address to authorise the sending of a once-off packet. 858 * 859 * On error, the values of *laddrp and *lportp are not changed. 860 */ 861 int 862 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp, 863 u_short *lportp, struct ucred *cred) 864 { 865 struct socket *so = inp->inp_socket; 866 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 867 struct in_addr laddr; 868 u_short lport = 0; 869 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 870 int error; 871 872 /* 873 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 874 * so that we don't have to add to the (already messy) code below. 875 */ 876 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 877 878 /* 879 * No state changes, so read locks are sufficient here. 880 */ 881 INP_LOCK_ASSERT(inp); 882 INP_HASH_LOCK_ASSERT(pcbinfo); 883 884 laddr.s_addr = *laddrp; 885 if (sin != NULL && laddr.s_addr != INADDR_ANY) 886 return (EINVAL); 887 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 888 lookupflags = INPLOOKUP_WILDCARD; 889 if (sin == NULL) { 890 if ((error = prison_local_ip4(cred, &laddr)) != 0) 891 return (error); 892 } else { 893 KASSERT(sin->sin_family == AF_INET, 894 ("%s: invalid family for address %p", __func__, sin)); 895 KASSERT(sin->sin_len == sizeof(*sin), 896 ("%s: invalid length for address %p", __func__, sin)); 897 898 error = prison_local_ip4(cred, &sin->sin_addr); 899 if (error) 900 return (error); 901 if (sin->sin_port != *lportp) { 902 /* Don't allow the port to change. */ 903 if (*lportp != 0) 904 return (EINVAL); 905 lport = sin->sin_port; 906 } 907 /* NB: lport is left as 0 if the port isn't being changed. */ 908 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 909 /* 910 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 911 * allow complete duplication of binding if 912 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 913 * and a multicast address is bound on both 914 * new and duplicated sockets. 915 */ 916 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 917 reuseport = SO_REUSEADDR|SO_REUSEPORT; 918 /* 919 * XXX: How to deal with SO_REUSEPORT_LB here? 920 * Treat same as SO_REUSEPORT for now. 921 */ 922 if ((so->so_options & 923 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 924 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 925 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 926 sin->sin_port = 0; /* yech... */ 927 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 928 /* 929 * Is the address a local IP address? 930 * If INP_BINDANY is set, then the socket may be bound 931 * to any endpoint address, local or not. 932 */ 933 if ((inp->inp_flags & INP_BINDANY) == 0 && 934 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 935 return (EADDRNOTAVAIL); 936 } 937 laddr = sin->sin_addr; 938 if (lport) { 939 struct inpcb *t; 940 941 /* GROSS */ 942 if (ntohs(lport) <= V_ipport_reservedhigh && 943 ntohs(lport) >= V_ipport_reservedlow && 944 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 945 return (EACCES); 946 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 947 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 948 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 949 lport, INPLOOKUP_WILDCARD, cred); 950 /* 951 * XXX 952 * This entire block sorely needs a rewrite. 953 */ 954 if (t != NULL && 955 (so->so_type != SOCK_STREAM || 956 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 957 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 958 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 959 (t->inp_socket->so_options & SO_REUSEPORT) || 960 (t->inp_socket->so_options & SO_REUSEPORT_LB) == 0) && 961 (inp->inp_cred->cr_uid != 962 t->inp_cred->cr_uid)) 963 return (EADDRINUSE); 964 } 965 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 966 lport, lookupflags, cred); 967 if (t != NULL && (reuseport & t->inp_socket->so_options) == 0 && 968 (reuseport_lb & t->inp_socket->so_options) == 0) { 969 #ifdef INET6 970 if (ntohl(sin->sin_addr.s_addr) != 971 INADDR_ANY || 972 ntohl(t->inp_laddr.s_addr) != 973 INADDR_ANY || 974 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 975 (t->inp_vflag & INP_IPV6PROTO) == 0) 976 #endif 977 return (EADDRINUSE); 978 } 979 } 980 } 981 if (*lportp != 0) 982 lport = *lportp; 983 if (lport == 0) { 984 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 985 if (error != 0) 986 return (error); 987 } 988 *laddrp = laddr.s_addr; 989 *lportp = lport; 990 return (0); 991 } 992 993 /* 994 * Connect from a socket to a specified address. 995 * Both address and port must be specified in argument sin. 996 * If don't have a local address for this socket yet, 997 * then pick one. 998 */ 999 int 1000 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred, 1001 bool rehash __unused) 1002 { 1003 u_short lport, fport; 1004 in_addr_t laddr, faddr; 1005 int anonport, error; 1006 1007 INP_WLOCK_ASSERT(inp); 1008 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1009 KASSERT(in_nullhost(inp->inp_faddr), 1010 ("%s: inp is already connected", __func__)); 1011 1012 lport = inp->inp_lport; 1013 laddr = inp->inp_laddr.s_addr; 1014 anonport = (lport == 0); 1015 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport, 1016 cred); 1017 if (error) 1018 return (error); 1019 1020 inp->inp_faddr.s_addr = faddr; 1021 inp->inp_fport = fport; 1022 1023 /* Do the initial binding of the local address if required. */ 1024 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1025 inp->inp_lport = lport; 1026 inp->inp_laddr.s_addr = laddr; 1027 if (in_pcbinshash(inp) != 0) { 1028 inp->inp_laddr.s_addr = inp->inp_faddr.s_addr = 1029 INADDR_ANY; 1030 inp->inp_lport = inp->inp_fport = 0; 1031 return (EAGAIN); 1032 } 1033 } else { 1034 inp->inp_lport = lport; 1035 inp->inp_laddr.s_addr = laddr; 1036 if ((inp->inp_flags & INP_INHASHLIST) != 0) 1037 in_pcbrehash(inp); 1038 else 1039 in_pcbinshash(inp); 1040 } 1041 1042 if (anonport) 1043 inp->inp_flags |= INP_ANONPORT; 1044 return (0); 1045 } 1046 1047 /* 1048 * Do proper source address selection on an unbound socket in case 1049 * of connect. Take jails into account as well. 1050 */ 1051 int 1052 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1053 struct ucred *cred) 1054 { 1055 struct ifaddr *ifa; 1056 struct sockaddr *sa; 1057 struct sockaddr_in *sin, dst; 1058 struct nhop_object *nh; 1059 int error; 1060 1061 NET_EPOCH_ASSERT(); 1062 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1063 1064 /* 1065 * Bypass source address selection and use the primary jail IP 1066 * if requested. 1067 */ 1068 if (!prison_saddrsel_ip4(cred, laddr)) 1069 return (0); 1070 1071 error = 0; 1072 1073 nh = NULL; 1074 bzero(&dst, sizeof(dst)); 1075 sin = &dst; 1076 sin->sin_family = AF_INET; 1077 sin->sin_len = sizeof(struct sockaddr_in); 1078 sin->sin_addr.s_addr = faddr->s_addr; 1079 1080 /* 1081 * If route is known our src addr is taken from the i/f, 1082 * else punt. 1083 * 1084 * Find out route to destination. 1085 */ 1086 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1087 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1088 0, NHR_NONE, 0); 1089 1090 /* 1091 * If we found a route, use the address corresponding to 1092 * the outgoing interface. 1093 * 1094 * Otherwise assume faddr is reachable on a directly connected 1095 * network and try to find a corresponding interface to take 1096 * the source address from. 1097 */ 1098 if (nh == NULL || nh->nh_ifp == NULL) { 1099 struct in_ifaddr *ia; 1100 struct ifnet *ifp; 1101 1102 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1103 inp->inp_socket->so_fibnum)); 1104 if (ia == NULL) { 1105 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1106 inp->inp_socket->so_fibnum)); 1107 } 1108 if (ia == NULL) { 1109 error = ENETUNREACH; 1110 goto done; 1111 } 1112 1113 if (!prison_flag(cred, PR_IP4)) { 1114 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1115 goto done; 1116 } 1117 1118 ifp = ia->ia_ifp; 1119 ia = NULL; 1120 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1121 sa = ifa->ifa_addr; 1122 if (sa->sa_family != AF_INET) 1123 continue; 1124 sin = (struct sockaddr_in *)sa; 1125 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1126 ia = (struct in_ifaddr *)ifa; 1127 break; 1128 } 1129 } 1130 if (ia != NULL) { 1131 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1132 goto done; 1133 } 1134 1135 /* 3. As a last resort return the 'default' jail address. */ 1136 error = prison_get_ip4(cred, laddr); 1137 goto done; 1138 } 1139 1140 /* 1141 * If the outgoing interface on the route found is not 1142 * a loopback interface, use the address from that interface. 1143 * In case of jails do those three steps: 1144 * 1. check if the interface address belongs to the jail. If so use it. 1145 * 2. check if we have any address on the outgoing interface 1146 * belonging to this jail. If so use it. 1147 * 3. as a last resort return the 'default' jail address. 1148 */ 1149 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1150 struct in_ifaddr *ia; 1151 struct ifnet *ifp; 1152 1153 /* If not jailed, use the default returned. */ 1154 if (!prison_flag(cred, PR_IP4)) { 1155 ia = (struct in_ifaddr *)nh->nh_ifa; 1156 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1157 goto done; 1158 } 1159 1160 /* Jailed. */ 1161 /* 1. Check if the iface address belongs to the jail. */ 1162 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1163 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1164 ia = (struct in_ifaddr *)nh->nh_ifa; 1165 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1166 goto done; 1167 } 1168 1169 /* 1170 * 2. Check if we have any address on the outgoing interface 1171 * belonging to this jail. 1172 */ 1173 ia = NULL; 1174 ifp = nh->nh_ifp; 1175 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1176 sa = ifa->ifa_addr; 1177 if (sa->sa_family != AF_INET) 1178 continue; 1179 sin = (struct sockaddr_in *)sa; 1180 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1181 ia = (struct in_ifaddr *)ifa; 1182 break; 1183 } 1184 } 1185 if (ia != NULL) { 1186 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1187 goto done; 1188 } 1189 1190 /* 3. As a last resort return the 'default' jail address. */ 1191 error = prison_get_ip4(cred, laddr); 1192 goto done; 1193 } 1194 1195 /* 1196 * The outgoing interface is marked with 'loopback net', so a route 1197 * to ourselves is here. 1198 * Try to find the interface of the destination address and then 1199 * take the address from there. That interface is not necessarily 1200 * a loopback interface. 1201 * In case of jails, check that it is an address of the jail 1202 * and if we cannot find, fall back to the 'default' jail address. 1203 */ 1204 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1205 struct in_ifaddr *ia; 1206 1207 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1208 inp->inp_socket->so_fibnum)); 1209 if (ia == NULL) 1210 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1211 inp->inp_socket->so_fibnum)); 1212 if (ia == NULL) 1213 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1214 1215 if (!prison_flag(cred, PR_IP4)) { 1216 if (ia == NULL) { 1217 error = ENETUNREACH; 1218 goto done; 1219 } 1220 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1221 goto done; 1222 } 1223 1224 /* Jailed. */ 1225 if (ia != NULL) { 1226 struct ifnet *ifp; 1227 1228 ifp = ia->ia_ifp; 1229 ia = NULL; 1230 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1231 sa = ifa->ifa_addr; 1232 if (sa->sa_family != AF_INET) 1233 continue; 1234 sin = (struct sockaddr_in *)sa; 1235 if (prison_check_ip4(cred, 1236 &sin->sin_addr) == 0) { 1237 ia = (struct in_ifaddr *)ifa; 1238 break; 1239 } 1240 } 1241 if (ia != NULL) { 1242 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1243 goto done; 1244 } 1245 } 1246 1247 /* 3. As a last resort return the 'default' jail address. */ 1248 error = prison_get_ip4(cred, laddr); 1249 goto done; 1250 } 1251 1252 done: 1253 if (error == 0 && laddr->s_addr == INADDR_ANY) 1254 return (EHOSTUNREACH); 1255 return (error); 1256 } 1257 1258 /* 1259 * Set up for a connect from a socket to the specified address. 1260 * On entry, *laddrp and *lportp should contain the current local 1261 * address and port for the PCB; these are updated to the values 1262 * that should be placed in inp_laddr and inp_lport to complete 1263 * the connect. 1264 * 1265 * On success, *faddrp and *fportp will be set to the remote address 1266 * and port. These are not updated in the error case. 1267 */ 1268 int 1269 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin, 1270 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1271 struct ucred *cred) 1272 { 1273 struct in_ifaddr *ia; 1274 struct in_addr laddr, faddr; 1275 u_short lport, fport; 1276 int error; 1277 1278 KASSERT(sin->sin_family == AF_INET, 1279 ("%s: invalid address family for %p", __func__, sin)); 1280 KASSERT(sin->sin_len == sizeof(*sin), 1281 ("%s: invalid address length for %p", __func__, sin)); 1282 1283 /* 1284 * Because a global state change doesn't actually occur here, a read 1285 * lock is sufficient. 1286 */ 1287 NET_EPOCH_ASSERT(); 1288 INP_LOCK_ASSERT(inp); 1289 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1290 1291 if (sin->sin_port == 0) 1292 return (EADDRNOTAVAIL); 1293 laddr.s_addr = *laddrp; 1294 lport = *lportp; 1295 faddr = sin->sin_addr; 1296 fport = sin->sin_port; 1297 #ifdef ROUTE_MPATH 1298 if (CALC_FLOWID_OUTBOUND) { 1299 uint32_t hash_val, hash_type; 1300 1301 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, 1302 inp->inp_socket->so_proto->pr_protocol, &hash_type); 1303 1304 inp->inp_flowid = hash_val; 1305 inp->inp_flowtype = hash_type; 1306 } 1307 #endif 1308 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1309 /* 1310 * If the destination address is INADDR_ANY, 1311 * use the primary local address. 1312 * If the supplied address is INADDR_BROADCAST, 1313 * and the primary interface supports broadcast, 1314 * choose the broadcast address for that interface. 1315 */ 1316 if (faddr.s_addr == INADDR_ANY) { 1317 faddr = 1318 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1319 if ((error = prison_get_ip4(cred, &faddr)) != 0) 1320 return (error); 1321 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1322 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1323 IFF_BROADCAST) 1324 faddr = satosin(&CK_STAILQ_FIRST( 1325 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1326 } 1327 } 1328 if (laddr.s_addr == INADDR_ANY) { 1329 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1330 /* 1331 * If the destination address is multicast and an outgoing 1332 * interface has been set as a multicast option, prefer the 1333 * address of that interface as our source address. 1334 */ 1335 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1336 inp->inp_moptions != NULL) { 1337 struct ip_moptions *imo; 1338 struct ifnet *ifp; 1339 1340 imo = inp->inp_moptions; 1341 if (imo->imo_multicast_ifp != NULL) { 1342 ifp = imo->imo_multicast_ifp; 1343 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1344 if (ia->ia_ifp == ifp && 1345 prison_check_ip4(cred, 1346 &ia->ia_addr.sin_addr) == 0) 1347 break; 1348 } 1349 if (ia == NULL) 1350 error = EADDRNOTAVAIL; 1351 else { 1352 laddr = ia->ia_addr.sin_addr; 1353 error = 0; 1354 } 1355 } 1356 } 1357 if (error) 1358 return (error); 1359 } 1360 1361 if (lport != 0) { 1362 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, 1363 fport, laddr, lport, 0, M_NODOM) != NULL) 1364 return (EADDRINUSE); 1365 } else { 1366 struct sockaddr_in lsin, fsin; 1367 1368 bzero(&lsin, sizeof(lsin)); 1369 bzero(&fsin, sizeof(fsin)); 1370 lsin.sin_family = AF_INET; 1371 lsin.sin_addr = laddr; 1372 fsin.sin_family = AF_INET; 1373 fsin.sin_addr = faddr; 1374 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, 1375 &lport, (struct sockaddr *)& fsin, fport, cred, 1376 INPLOOKUP_WILDCARD); 1377 if (error) 1378 return (error); 1379 } 1380 *laddrp = laddr.s_addr; 1381 *lportp = lport; 1382 *faddrp = faddr.s_addr; 1383 *fportp = fport; 1384 return (0); 1385 } 1386 1387 void 1388 in_pcbdisconnect(struct inpcb *inp) 1389 { 1390 1391 INP_WLOCK_ASSERT(inp); 1392 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1393 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 1394 ("%s: inp %p was already disconnected", __func__, inp)); 1395 1396 in_pcbremhash_locked(inp); 1397 1398 /* See the comment in in_pcbinshash(). */ 1399 inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr); 1400 inp->inp_laddr.s_addr = INADDR_ANY; 1401 inp->inp_faddr.s_addr = INADDR_ANY; 1402 inp->inp_fport = 0; 1403 } 1404 #endif /* INET */ 1405 1406 /* 1407 * inpcb hash lookups are protected by SMR section. 1408 * 1409 * Once desired pcb has been found, switching from SMR section to a pcb 1410 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK 1411 * here because SMR is a critical section. 1412 * In 99%+ cases inp_smr_lock() would obtain the lock immediately. 1413 */ 1414 void 1415 inp_lock(struct inpcb *inp, const inp_lookup_t lock) 1416 { 1417 1418 lock == INPLOOKUP_RLOCKPCB ? 1419 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock); 1420 } 1421 1422 void 1423 inp_unlock(struct inpcb *inp, const inp_lookup_t lock) 1424 { 1425 1426 lock == INPLOOKUP_RLOCKPCB ? 1427 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock); 1428 } 1429 1430 int 1431 inp_trylock(struct inpcb *inp, const inp_lookup_t lock) 1432 { 1433 1434 return (lock == INPLOOKUP_RLOCKPCB ? 1435 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock)); 1436 } 1437 1438 static inline bool 1439 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags) 1440 { 1441 1442 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB); 1443 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr); 1444 1445 if (__predict_true(inp_trylock(inp, lock))) { 1446 if (__predict_false(inp->inp_flags & ignflags)) { 1447 smr_exit(inp->inp_pcbinfo->ipi_smr); 1448 inp_unlock(inp, lock); 1449 return (false); 1450 } 1451 smr_exit(inp->inp_pcbinfo->ipi_smr); 1452 return (true); 1453 } 1454 1455 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1456 smr_exit(inp->inp_pcbinfo->ipi_smr); 1457 inp_lock(inp, lock); 1458 if (__predict_false(in_pcbrele(inp, lock))) 1459 return (false); 1460 /* 1461 * inp acquired through refcount & lock for sure didn't went 1462 * through uma_zfree(). However, it may have already went 1463 * through in_pcbfree() and has another reference, that 1464 * prevented its release by our in_pcbrele(). 1465 */ 1466 if (__predict_false(inp->inp_flags & ignflags)) { 1467 inp_unlock(inp, lock); 1468 return (false); 1469 } 1470 return (true); 1471 } else { 1472 smr_exit(inp->inp_pcbinfo->ipi_smr); 1473 return (false); 1474 } 1475 } 1476 1477 bool 1478 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock) 1479 { 1480 1481 /* 1482 * in_pcblookup() family of functions ignore not only freed entries, 1483 * that may be found due to lockless access to the hash, but dropped 1484 * entries, too. 1485 */ 1486 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED)); 1487 } 1488 1489 /* 1490 * inp_next() - inpcb hash/list traversal iterator 1491 * 1492 * Requires initialized struct inpcb_iterator for context. 1493 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR(). 1494 * 1495 * - Iterator can have either write-lock or read-lock semantics, that can not 1496 * be changed later. 1497 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through 1498 * a single hash slot. Note: only rip_input() does the latter. 1499 * - Iterator may have optional bool matching function. The matching function 1500 * will be executed for each inpcb in the SMR context, so it can not acquire 1501 * locks and can safely access only immutable fields of inpcb. 1502 * 1503 * A fresh initialized iterator has NULL inpcb in its context and that 1504 * means that inp_next() call would return the very first inpcb on the list 1505 * locked with desired semantic. In all following calls the context pointer 1506 * shall hold the current inpcb pointer. The KPI user is not supposed to 1507 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL 1508 * and write NULL to its context. After end of traversal an iterator can be 1509 * reused. 1510 * 1511 * List traversals have the following features/constraints: 1512 * - New entries won't be seen, as they are always added to the head of a list. 1513 * - Removed entries won't stop traversal as long as they are not added to 1514 * a different list. This is violated by in_pcbrehash(). 1515 */ 1516 #define II_LIST_FIRST(ipi, hash) \ 1517 (((hash) == INP_ALL_LIST) ? \ 1518 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \ 1519 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)])) 1520 #define II_LIST_NEXT(inp, hash) \ 1521 (((hash) == INP_ALL_LIST) ? \ 1522 CK_LIST_NEXT((inp), inp_list) : \ 1523 CK_LIST_NEXT((inp), inp_hash_exact)) 1524 #define II_LOCK_ASSERT(inp, lock) \ 1525 rw_assert(&(inp)->inp_lock, \ 1526 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED ) 1527 struct inpcb * 1528 inp_next(struct inpcb_iterator *ii) 1529 { 1530 const struct inpcbinfo *ipi = ii->ipi; 1531 inp_match_t *match = ii->match; 1532 void *ctx = ii->ctx; 1533 inp_lookup_t lock = ii->lock; 1534 int hash = ii->hash; 1535 struct inpcb *inp; 1536 1537 if (ii->inp == NULL) { /* First call. */ 1538 smr_enter(ipi->ipi_smr); 1539 /* This is unrolled CK_LIST_FOREACH(). */ 1540 for (inp = II_LIST_FIRST(ipi, hash); 1541 inp != NULL; 1542 inp = II_LIST_NEXT(inp, hash)) { 1543 if (match != NULL && (match)(inp, ctx) == false) 1544 continue; 1545 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED))) 1546 break; 1547 else { 1548 smr_enter(ipi->ipi_smr); 1549 MPASS(inp != II_LIST_FIRST(ipi, hash)); 1550 inp = II_LIST_FIRST(ipi, hash); 1551 if (inp == NULL) 1552 break; 1553 } 1554 } 1555 1556 if (inp == NULL) 1557 smr_exit(ipi->ipi_smr); 1558 else 1559 ii->inp = inp; 1560 1561 return (inp); 1562 } 1563 1564 /* Not a first call. */ 1565 smr_enter(ipi->ipi_smr); 1566 restart: 1567 inp = ii->inp; 1568 II_LOCK_ASSERT(inp, lock); 1569 next: 1570 inp = II_LIST_NEXT(inp, hash); 1571 if (inp == NULL) { 1572 smr_exit(ipi->ipi_smr); 1573 goto found; 1574 } 1575 1576 if (match != NULL && (match)(inp, ctx) == false) 1577 goto next; 1578 1579 if (__predict_true(inp_trylock(inp, lock))) { 1580 if (__predict_false(inp->inp_flags & INP_FREED)) { 1581 /* 1582 * Entries are never inserted in middle of a list, thus 1583 * as long as we are in SMR, we can continue traversal. 1584 * Jump to 'restart' should yield in the same result, 1585 * but could produce unnecessary looping. Could this 1586 * looping be unbound? 1587 */ 1588 inp_unlock(inp, lock); 1589 goto next; 1590 } else { 1591 smr_exit(ipi->ipi_smr); 1592 goto found; 1593 } 1594 } 1595 1596 /* 1597 * Can't obtain lock immediately, thus going hard. Once we exit the 1598 * SMR section we can no longer jump to 'next', and our only stable 1599 * anchoring point is ii->inp, which we keep locked for this case, so 1600 * we jump to 'restart'. 1601 */ 1602 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1603 smr_exit(ipi->ipi_smr); 1604 inp_lock(inp, lock); 1605 if (__predict_false(in_pcbrele(inp, lock))) { 1606 smr_enter(ipi->ipi_smr); 1607 goto restart; 1608 } 1609 /* 1610 * See comment in inp_smr_lock(). 1611 */ 1612 if (__predict_false(inp->inp_flags & INP_FREED)) { 1613 inp_unlock(inp, lock); 1614 smr_enter(ipi->ipi_smr); 1615 goto restart; 1616 } 1617 } else 1618 goto next; 1619 1620 found: 1621 inp_unlock(ii->inp, lock); 1622 ii->inp = inp; 1623 1624 return (ii->inp); 1625 } 1626 1627 /* 1628 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1629 * stability of an inpcb pointer despite the inpcb lock being released or 1630 * SMR section exited. 1631 * 1632 * To free a reference later in_pcbrele_(r|w)locked() must be performed. 1633 */ 1634 void 1635 in_pcbref(struct inpcb *inp) 1636 { 1637 u_int old __diagused; 1638 1639 old = refcount_acquire(&inp->inp_refcount); 1640 KASSERT(old > 0, ("%s: refcount 0", __func__)); 1641 } 1642 1643 /* 1644 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially 1645 * freeing the pcb, if the reference was very last. 1646 */ 1647 bool 1648 in_pcbrele_rlocked(struct inpcb *inp) 1649 { 1650 1651 INP_RLOCK_ASSERT(inp); 1652 1653 if (!refcount_release(&inp->inp_refcount)) 1654 return (false); 1655 1656 MPASS(inp->inp_flags & INP_FREED); 1657 MPASS(inp->inp_socket == NULL); 1658 crfree(inp->inp_cred); 1659 #ifdef INVARIANTS 1660 inp->inp_cred = NULL; 1661 #endif 1662 INP_RUNLOCK(inp); 1663 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1664 return (true); 1665 } 1666 1667 bool 1668 in_pcbrele_wlocked(struct inpcb *inp) 1669 { 1670 1671 INP_WLOCK_ASSERT(inp); 1672 1673 if (!refcount_release(&inp->inp_refcount)) 1674 return (false); 1675 1676 MPASS(inp->inp_flags & INP_FREED); 1677 MPASS(inp->inp_socket == NULL); 1678 crfree(inp->inp_cred); 1679 #ifdef INVARIANTS 1680 inp->inp_cred = NULL; 1681 #endif 1682 INP_WUNLOCK(inp); 1683 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1684 return (true); 1685 } 1686 1687 bool 1688 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock) 1689 { 1690 1691 return (lock == INPLOOKUP_RLOCKPCB ? 1692 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp)); 1693 } 1694 1695 /* 1696 * Unconditionally schedule an inpcb to be freed by decrementing its 1697 * reference count, which should occur only after the inpcb has been detached 1698 * from its socket. If another thread holds a temporary reference (acquired 1699 * using in_pcbref()) then the free is deferred until that reference is 1700 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked. 1701 * Almost all work, including removal from global lists, is done in this 1702 * context, where the pcbinfo lock is held. 1703 */ 1704 void 1705 in_pcbfree(struct inpcb *inp) 1706 { 1707 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1708 #ifdef INET 1709 struct ip_moptions *imo; 1710 #endif 1711 #ifdef INET6 1712 struct ip6_moptions *im6o; 1713 #endif 1714 1715 INP_WLOCK_ASSERT(inp); 1716 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1717 KASSERT((inp->inp_flags & INP_FREED) == 0, 1718 ("%s: called twice for pcb %p", __func__, inp)); 1719 1720 /* 1721 * in_pcblookup_local() and in6_pcblookup_local() may return an inpcb 1722 * from the hash without acquiring inpcb lock, they rely on the hash 1723 * lock, thus in_pcbremhash() should be the first action. 1724 */ 1725 if (inp->inp_flags & INP_INHASHLIST) 1726 in_pcbremhash(inp); 1727 INP_INFO_WLOCK(pcbinfo); 1728 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1729 pcbinfo->ipi_count--; 1730 CK_LIST_REMOVE(inp, inp_list); 1731 INP_INFO_WUNLOCK(pcbinfo); 1732 1733 #ifdef RATELIMIT 1734 if (inp->inp_snd_tag != NULL) 1735 in_pcbdetach_txrtlmt(inp); 1736 #endif 1737 inp->inp_flags |= INP_FREED; 1738 inp->inp_socket->so_pcb = NULL; 1739 inp->inp_socket = NULL; 1740 1741 RO_INVALIDATE_CACHE(&inp->inp_route); 1742 #ifdef MAC 1743 mac_inpcb_destroy(inp); 1744 #endif 1745 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1746 if (inp->inp_sp != NULL) 1747 ipsec_delete_pcbpolicy(inp); 1748 #endif 1749 #ifdef INET 1750 if (inp->inp_options) 1751 (void)m_free(inp->inp_options); 1752 DEBUG_POISON_POINTER(inp->inp_options); 1753 imo = inp->inp_moptions; 1754 DEBUG_POISON_POINTER(inp->inp_moptions); 1755 #endif 1756 #ifdef INET6 1757 if (inp->inp_vflag & INP_IPV6PROTO) { 1758 ip6_freepcbopts(inp->in6p_outputopts); 1759 DEBUG_POISON_POINTER(inp->in6p_outputopts); 1760 im6o = inp->in6p_moptions; 1761 DEBUG_POISON_POINTER(inp->in6p_moptions); 1762 } else 1763 im6o = NULL; 1764 #endif 1765 1766 if (__predict_false(in_pcbrele_wlocked(inp) == false)) { 1767 INP_WUNLOCK(inp); 1768 } 1769 #ifdef INET6 1770 ip6_freemoptions(im6o); 1771 #endif 1772 #ifdef INET 1773 inp_freemoptions(imo); 1774 #endif 1775 } 1776 1777 /* 1778 * Different protocols initialize their inpcbs differently - giving 1779 * different name to the lock. But they all are disposed the same. 1780 */ 1781 static void 1782 inpcb_fini(void *mem, int size) 1783 { 1784 struct inpcb *inp = mem; 1785 1786 INP_LOCK_DESTROY(inp); 1787 } 1788 1789 /* 1790 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1791 * port reservation, and preventing it from being returned by inpcb lookups. 1792 * 1793 * It is used by TCP to mark an inpcb as unused and avoid future packet 1794 * delivery or event notification when a socket remains open but TCP has 1795 * closed. This might occur as a result of a shutdown()-initiated TCP close 1796 * or a RST on the wire, and allows the port binding to be reused while still 1797 * maintaining the invariant that so_pcb always points to a valid inpcb until 1798 * in_pcbdetach(). 1799 * 1800 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1801 * in_pcbpurgeif0()? 1802 */ 1803 void 1804 in_pcbdrop(struct inpcb *inp) 1805 { 1806 1807 INP_WLOCK_ASSERT(inp); 1808 #ifdef INVARIANTS 1809 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1810 MPASS(inp->inp_refcount > 1); 1811 #endif 1812 1813 inp->inp_flags |= INP_DROPPED; 1814 if (inp->inp_flags & INP_INHASHLIST) 1815 in_pcbremhash(inp); 1816 } 1817 1818 #ifdef INET 1819 /* 1820 * Common routines to return the socket addresses associated with inpcbs. 1821 */ 1822 int 1823 in_getsockaddr(struct socket *so, struct sockaddr *sa) 1824 { 1825 struct inpcb *inp; 1826 1827 inp = sotoinpcb(so); 1828 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1829 1830 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 1831 .sin_len = sizeof(struct sockaddr_in), 1832 .sin_family = AF_INET, 1833 .sin_port = inp->inp_lport, 1834 .sin_addr = inp->inp_laddr, 1835 }; 1836 1837 return (0); 1838 } 1839 1840 int 1841 in_getpeeraddr(struct socket *so, struct sockaddr *sa) 1842 { 1843 struct inpcb *inp; 1844 1845 inp = sotoinpcb(so); 1846 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1847 1848 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 1849 .sin_len = sizeof(struct sockaddr_in), 1850 .sin_family = AF_INET, 1851 .sin_port = inp->inp_fport, 1852 .sin_addr = inp->inp_faddr, 1853 }; 1854 1855 return (0); 1856 } 1857 1858 static bool 1859 inp_v4_multi_match(const struct inpcb *inp, void *v __unused) 1860 { 1861 1862 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL) 1863 return (true); 1864 else 1865 return (false); 1866 } 1867 1868 void 1869 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1870 { 1871 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB, 1872 inp_v4_multi_match, NULL); 1873 struct inpcb *inp; 1874 struct in_multi *inm; 1875 struct in_mfilter *imf; 1876 struct ip_moptions *imo; 1877 1878 IN_MULTI_LOCK_ASSERT(); 1879 1880 while ((inp = inp_next(&inpi)) != NULL) { 1881 INP_WLOCK_ASSERT(inp); 1882 1883 imo = inp->inp_moptions; 1884 /* 1885 * Unselect the outgoing interface if it is being 1886 * detached. 1887 */ 1888 if (imo->imo_multicast_ifp == ifp) 1889 imo->imo_multicast_ifp = NULL; 1890 1891 /* 1892 * Drop multicast group membership if we joined 1893 * through the interface being detached. 1894 * 1895 * XXX This can all be deferred to an epoch_call 1896 */ 1897 restart: 1898 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 1899 if ((inm = imf->imf_inm) == NULL) 1900 continue; 1901 if (inm->inm_ifp != ifp) 1902 continue; 1903 ip_mfilter_remove(&imo->imo_head, imf); 1904 in_leavegroup_locked(inm, NULL); 1905 ip_mfilter_free(imf); 1906 goto restart; 1907 } 1908 } 1909 } 1910 1911 /* 1912 * Lookup a PCB based on the local address and port. Caller must hold the 1913 * hash lock. No inpcb locks or references are acquired. 1914 */ 1915 #define INP_LOOKUP_MAPPED_PCB_COST 3 1916 struct inpcb * 1917 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1918 u_short lport, int lookupflags, struct ucred *cred) 1919 { 1920 struct inpcb *inp; 1921 #ifdef INET6 1922 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1923 #else 1924 int matchwild = 3; 1925 #endif 1926 int wildcard; 1927 1928 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1929 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1930 INP_HASH_LOCK_ASSERT(pcbinfo); 1931 1932 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1933 struct inpcbhead *head; 1934 /* 1935 * Look for an unconnected (wildcard foreign addr) PCB that 1936 * matches the local address and port we're looking for. 1937 */ 1938 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 1939 pcbinfo->ipi_hashmask)]; 1940 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 1941 #ifdef INET6 1942 /* XXX inp locking */ 1943 if ((inp->inp_vflag & INP_IPV4) == 0) 1944 continue; 1945 #endif 1946 if (inp->inp_faddr.s_addr == INADDR_ANY && 1947 inp->inp_laddr.s_addr == laddr.s_addr && 1948 inp->inp_lport == lport) { 1949 /* 1950 * Found? 1951 */ 1952 if (prison_equal_ip4(cred->cr_prison, 1953 inp->inp_cred->cr_prison)) 1954 return (inp); 1955 } 1956 } 1957 /* 1958 * Not found. 1959 */ 1960 return (NULL); 1961 } else { 1962 struct inpcbporthead *porthash; 1963 struct inpcbport *phd; 1964 struct inpcb *match = NULL; 1965 /* 1966 * Best fit PCB lookup. 1967 * 1968 * First see if this local port is in use by looking on the 1969 * port hash list. 1970 */ 1971 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1972 pcbinfo->ipi_porthashmask)]; 1973 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1974 if (phd->phd_port == lport) 1975 break; 1976 } 1977 if (phd != NULL) { 1978 /* 1979 * Port is in use by one or more PCBs. Look for best 1980 * fit. 1981 */ 1982 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1983 wildcard = 0; 1984 if (!prison_equal_ip4(inp->inp_cred->cr_prison, 1985 cred->cr_prison)) 1986 continue; 1987 #ifdef INET6 1988 /* XXX inp locking */ 1989 if ((inp->inp_vflag & INP_IPV4) == 0) 1990 continue; 1991 /* 1992 * We never select the PCB that has 1993 * INP_IPV6 flag and is bound to :: if 1994 * we have another PCB which is bound 1995 * to 0.0.0.0. If a PCB has the 1996 * INP_IPV6 flag, then we set its cost 1997 * higher than IPv4 only PCBs. 1998 * 1999 * Note that the case only happens 2000 * when a socket is bound to ::, under 2001 * the condition that the use of the 2002 * mapped address is allowed. 2003 */ 2004 if ((inp->inp_vflag & INP_IPV6) != 0) 2005 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 2006 #endif 2007 if (inp->inp_faddr.s_addr != INADDR_ANY) 2008 wildcard++; 2009 if (inp->inp_laddr.s_addr != INADDR_ANY) { 2010 if (laddr.s_addr == INADDR_ANY) 2011 wildcard++; 2012 else if (inp->inp_laddr.s_addr != laddr.s_addr) 2013 continue; 2014 } else { 2015 if (laddr.s_addr != INADDR_ANY) 2016 wildcard++; 2017 } 2018 if (wildcard < matchwild) { 2019 match = inp; 2020 matchwild = wildcard; 2021 if (matchwild == 0) 2022 break; 2023 } 2024 } 2025 } 2026 return (match); 2027 } 2028 } 2029 #undef INP_LOOKUP_MAPPED_PCB_COST 2030 2031 static bool 2032 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain) 2033 { 2034 return (domain == M_NODOM || domain == grp->il_numa_domain); 2035 } 2036 2037 static struct inpcb * 2038 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 2039 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr, 2040 uint16_t lport, int domain) 2041 { 2042 const struct inpcblbgrouphead *hdr; 2043 struct inpcblbgroup *grp; 2044 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild; 2045 2046 INP_HASH_LOCK_ASSERT(pcbinfo); 2047 2048 hdr = &pcbinfo->ipi_lbgrouphashbase[ 2049 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 2050 2051 /* 2052 * Search for an LB group match based on the following criteria: 2053 * - prefer jailed groups to non-jailed groups 2054 * - prefer exact source address matches to wildcard matches 2055 * - prefer groups bound to the specified NUMA domain 2056 */ 2057 jail_exact = jail_wild = local_exact = local_wild = NULL; 2058 CK_LIST_FOREACH(grp, hdr, il_list) { 2059 bool injail; 2060 2061 #ifdef INET6 2062 if (!(grp->il_vflag & INP_IPV4)) 2063 continue; 2064 #endif 2065 if (grp->il_lport != lport) 2066 continue; 2067 2068 injail = prison_flag(grp->il_cred, PR_IP4) != 0; 2069 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison, 2070 laddr) != 0) 2071 continue; 2072 2073 if (grp->il_laddr.s_addr == laddr->s_addr) { 2074 if (injail) { 2075 jail_exact = grp; 2076 if (in_pcblookup_lb_numa_match(grp, domain)) 2077 /* This is a perfect match. */ 2078 goto out; 2079 } else if (local_exact == NULL || 2080 in_pcblookup_lb_numa_match(grp, domain)) { 2081 local_exact = grp; 2082 } 2083 } else if (grp->il_laddr.s_addr == INADDR_ANY) { 2084 if (injail) { 2085 if (jail_wild == NULL || 2086 in_pcblookup_lb_numa_match(grp, domain)) 2087 jail_wild = grp; 2088 } else if (local_wild == NULL || 2089 in_pcblookup_lb_numa_match(grp, domain)) { 2090 local_wild = grp; 2091 } 2092 } 2093 } 2094 2095 if (jail_exact != NULL) 2096 grp = jail_exact; 2097 else if (jail_wild != NULL) 2098 grp = jail_wild; 2099 else if (local_exact != NULL) 2100 grp = local_exact; 2101 else 2102 grp = local_wild; 2103 if (grp == NULL) 2104 return (NULL); 2105 out: 2106 return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) % 2107 grp->il_inpcnt]); 2108 } 2109 2110 static bool 2111 in_pcblookup_exact_match(const struct inpcb *inp, struct in_addr faddr, 2112 u_short fport, struct in_addr laddr, u_short lport) 2113 { 2114 #ifdef INET6 2115 /* XXX inp locking */ 2116 if ((inp->inp_vflag & INP_IPV4) == 0) 2117 return (false); 2118 #endif 2119 if (inp->inp_faddr.s_addr == faddr.s_addr && 2120 inp->inp_laddr.s_addr == laddr.s_addr && 2121 inp->inp_fport == fport && 2122 inp->inp_lport == lport) 2123 return (true); 2124 return (false); 2125 } 2126 2127 static struct inpcb * 2128 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2129 u_short fport, struct in_addr laddr, u_short lport) 2130 { 2131 struct inpcbhead *head; 2132 struct inpcb *inp; 2133 2134 INP_HASH_LOCK_ASSERT(pcbinfo); 2135 2136 head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport, 2137 pcbinfo->ipi_hashmask)]; 2138 CK_LIST_FOREACH(inp, head, inp_hash_exact) { 2139 if (in_pcblookup_exact_match(inp, faddr, fport, laddr, lport)) 2140 return (inp); 2141 } 2142 return (NULL); 2143 } 2144 2145 typedef enum { 2146 INPLOOKUP_MATCH_NONE = 0, 2147 INPLOOKUP_MATCH_WILD = 1, 2148 INPLOOKUP_MATCH_LADDR = 2, 2149 } inp_lookup_match_t; 2150 2151 static inp_lookup_match_t 2152 in_pcblookup_wild_match(const struct inpcb *inp, struct in_addr laddr, 2153 u_short lport) 2154 { 2155 #ifdef INET6 2156 /* XXX inp locking */ 2157 if ((inp->inp_vflag & INP_IPV4) == 0) 2158 return (INPLOOKUP_MATCH_NONE); 2159 #endif 2160 if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) 2161 return (INPLOOKUP_MATCH_NONE); 2162 if (inp->inp_laddr.s_addr == INADDR_ANY) 2163 return (INPLOOKUP_MATCH_WILD); 2164 if (inp->inp_laddr.s_addr == laddr.s_addr) 2165 return (INPLOOKUP_MATCH_LADDR); 2166 return (INPLOOKUP_MATCH_NONE); 2167 } 2168 2169 #define INP_LOOKUP_AGAIN ((struct inpcb *)(uintptr_t)-1) 2170 2171 static struct inpcb * 2172 in_pcblookup_hash_wild_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2173 u_short fport, struct in_addr laddr, u_short lport, 2174 const inp_lookup_t lockflags) 2175 { 2176 struct inpcbhead *head; 2177 struct inpcb *inp; 2178 2179 KASSERT(SMR_ENTERED(pcbinfo->ipi_smr), 2180 ("%s: not in SMR read section", __func__)); 2181 2182 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2183 pcbinfo->ipi_hashmask)]; 2184 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2185 inp_lookup_match_t match; 2186 2187 match = in_pcblookup_wild_match(inp, laddr, lport); 2188 if (match == INPLOOKUP_MATCH_NONE) 2189 continue; 2190 2191 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2192 match = in_pcblookup_wild_match(inp, laddr, lport); 2193 if (match != INPLOOKUP_MATCH_NONE && 2194 prison_check_ip4_locked(inp->inp_cred->cr_prison, 2195 &laddr) == 0) 2196 return (inp); 2197 inp_unlock(inp, lockflags); 2198 } 2199 2200 /* 2201 * The matching socket disappeared out from under us. Fall back 2202 * to a serialized lookup. 2203 */ 2204 return (INP_LOOKUP_AGAIN); 2205 } 2206 return (NULL); 2207 } 2208 2209 static struct inpcb * 2210 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2211 u_short fport, struct in_addr laddr, u_short lport) 2212 { 2213 struct inpcbhead *head; 2214 struct inpcb *inp, *local_wild, *local_exact, *jail_wild; 2215 #ifdef INET6 2216 struct inpcb *local_wild_mapped; 2217 #endif 2218 2219 INP_HASH_LOCK_ASSERT(pcbinfo); 2220 2221 /* 2222 * Order of socket selection - we always prefer jails. 2223 * 1. jailed, non-wild. 2224 * 2. jailed, wild. 2225 * 3. non-jailed, non-wild. 2226 * 4. non-jailed, wild. 2227 */ 2228 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2229 pcbinfo->ipi_hashmask)]; 2230 local_wild = local_exact = jail_wild = NULL; 2231 #ifdef INET6 2232 local_wild_mapped = NULL; 2233 #endif 2234 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2235 inp_lookup_match_t match; 2236 bool injail; 2237 2238 match = in_pcblookup_wild_match(inp, laddr, lport); 2239 if (match == INPLOOKUP_MATCH_NONE) 2240 continue; 2241 2242 injail = prison_flag(inp->inp_cred, PR_IP4) != 0; 2243 if (injail) { 2244 if (prison_check_ip4_locked(inp->inp_cred->cr_prison, 2245 &laddr) != 0) 2246 continue; 2247 } else { 2248 if (local_exact != NULL) 2249 continue; 2250 } 2251 2252 if (match == INPLOOKUP_MATCH_LADDR) { 2253 if (injail) 2254 return (inp); 2255 local_exact = inp; 2256 } else { 2257 #ifdef INET6 2258 /* XXX inp locking, NULL check */ 2259 if (inp->inp_vflag & INP_IPV6PROTO) 2260 local_wild_mapped = inp; 2261 else 2262 #endif 2263 if (injail) 2264 jail_wild = inp; 2265 else 2266 local_wild = inp; 2267 } 2268 } 2269 if (jail_wild != NULL) 2270 return (jail_wild); 2271 if (local_exact != NULL) 2272 return (local_exact); 2273 if (local_wild != NULL) 2274 return (local_wild); 2275 #ifdef INET6 2276 if (local_wild_mapped != NULL) 2277 return (local_wild_mapped); 2278 #endif 2279 return (NULL); 2280 } 2281 2282 /* 2283 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2284 * that the caller has either locked the hash list, which usually happens 2285 * for bind(2) operations, or is in SMR section, which happens when sorting 2286 * out incoming packets. 2287 */ 2288 static struct inpcb * 2289 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2290 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2291 uint8_t numa_domain) 2292 { 2293 struct inpcb *inp; 2294 const u_short fport = fport_arg, lport = lport_arg; 2295 2296 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0, 2297 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2298 KASSERT(faddr.s_addr != INADDR_ANY, 2299 ("%s: invalid foreign address", __func__)); 2300 KASSERT(laddr.s_addr != INADDR_ANY, 2301 ("%s: invalid local address", __func__)); 2302 INP_HASH_WLOCK_ASSERT(pcbinfo); 2303 2304 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2305 if (inp != NULL) 2306 return (inp); 2307 2308 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2309 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2310 &laddr, lport, numa_domain); 2311 if (inp == NULL) { 2312 inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr, 2313 fport, laddr, lport); 2314 } 2315 } 2316 2317 return (inp); 2318 } 2319 2320 static struct inpcb * 2321 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2322 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2323 uint8_t numa_domain) 2324 { 2325 struct inpcb *inp; 2326 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2327 2328 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2329 ("%s: LOCKPCB not set", __func__)); 2330 2331 INP_HASH_WLOCK(pcbinfo); 2332 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2333 lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain); 2334 if (inp != NULL && !inp_trylock(inp, lockflags)) { 2335 in_pcbref(inp); 2336 INP_HASH_WUNLOCK(pcbinfo); 2337 inp_lock(inp, lockflags); 2338 if (in_pcbrele(inp, lockflags)) 2339 /* XXX-MJ or retry until we get a negative match? */ 2340 inp = NULL; 2341 } else { 2342 INP_HASH_WUNLOCK(pcbinfo); 2343 } 2344 return (inp); 2345 } 2346 2347 static struct inpcb * 2348 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2349 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2350 uint8_t numa_domain) 2351 { 2352 struct inpcb *inp; 2353 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2354 const u_short fport = fport_arg, lport = lport_arg; 2355 2356 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2357 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2358 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2359 ("%s: LOCKPCB not set", __func__)); 2360 2361 smr_enter(pcbinfo->ipi_smr); 2362 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2363 if (inp != NULL) { 2364 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2365 /* 2366 * Revalidate the 4-tuple, the socket could have been 2367 * disconnected. 2368 */ 2369 if (__predict_true(in_pcblookup_exact_match(inp, 2370 faddr, fport, laddr, lport))) 2371 return (inp); 2372 inp_unlock(inp, lockflags); 2373 } 2374 2375 /* 2376 * We failed to lock the inpcb, or its connection state changed 2377 * out from under us. Fall back to a precise search. 2378 */ 2379 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2380 lookupflags, numa_domain)); 2381 } 2382 2383 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2384 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2385 &laddr, lport, numa_domain); 2386 if (inp != NULL) { 2387 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2388 if (__predict_true(in_pcblookup_wild_match(inp, 2389 laddr, lport) != INPLOOKUP_MATCH_NONE)) 2390 return (inp); 2391 inp_unlock(inp, lockflags); 2392 } 2393 inp = INP_LOOKUP_AGAIN; 2394 } else { 2395 inp = in_pcblookup_hash_wild_smr(pcbinfo, faddr, fport, 2396 laddr, lport, lockflags); 2397 } 2398 if (inp == INP_LOOKUP_AGAIN) { 2399 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, 2400 lport, lookupflags, numa_domain)); 2401 } 2402 } 2403 2404 if (inp == NULL) 2405 smr_exit(pcbinfo->ipi_smr); 2406 2407 return (inp); 2408 } 2409 2410 /* 2411 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2412 * from which a pre-calculated hash value may be extracted. 2413 */ 2414 struct inpcb * 2415 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2416 struct in_addr laddr, u_int lport, int lookupflags, 2417 struct ifnet *ifp __unused) 2418 { 2419 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2420 lookupflags, M_NODOM)); 2421 } 2422 2423 struct inpcb * 2424 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2425 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2426 struct ifnet *ifp __unused, struct mbuf *m) 2427 { 2428 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2429 lookupflags, m->m_pkthdr.numa_domain)); 2430 } 2431 #endif /* INET */ 2432 2433 static bool 2434 in_pcbjailed(const struct inpcb *inp, unsigned int flag) 2435 { 2436 return (prison_flag(inp->inp_cred, flag) != 0); 2437 } 2438 2439 /* 2440 * Insert the PCB into a hash chain using ordering rules which ensure that 2441 * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first. 2442 * 2443 * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs 2444 * with exact local addresses ahead of wildcard PCBs. Unbound v4-mapped v6 PCBs 2445 * always appear last no matter whether they are jailed. 2446 */ 2447 static void 2448 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2449 { 2450 struct inpcb *last; 2451 bool bound, injail; 2452 2453 INP_LOCK_ASSERT(inp); 2454 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2455 2456 last = NULL; 2457 bound = inp->inp_laddr.s_addr != INADDR_ANY; 2458 if (!bound && (inp->inp_vflag & INP_IPV6PROTO) != 0) { 2459 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2460 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2461 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2462 return; 2463 } 2464 } 2465 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2466 return; 2467 } 2468 2469 injail = in_pcbjailed(inp, PR_IP4); 2470 if (!injail) { 2471 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2472 if (!in_pcbjailed(last, PR_IP4)) 2473 break; 2474 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2475 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2476 return; 2477 } 2478 } 2479 } else if (!CK_LIST_EMPTY(pcbhash) && 2480 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) { 2481 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2482 return; 2483 } 2484 if (!bound) { 2485 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2486 if (last->inp_laddr.s_addr == INADDR_ANY) 2487 break; 2488 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2489 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2490 return; 2491 } 2492 } 2493 } 2494 if (last == NULL) 2495 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2496 else 2497 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2498 } 2499 2500 #ifdef INET6 2501 /* 2502 * See the comment above _in_pcbinshash_wild(). 2503 */ 2504 static void 2505 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2506 { 2507 struct inpcb *last; 2508 bool bound, injail; 2509 2510 INP_LOCK_ASSERT(inp); 2511 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2512 2513 last = NULL; 2514 bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr); 2515 injail = in_pcbjailed(inp, PR_IP6); 2516 if (!injail) { 2517 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2518 if (!in_pcbjailed(last, PR_IP6)) 2519 break; 2520 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2521 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2522 return; 2523 } 2524 } 2525 } else if (!CK_LIST_EMPTY(pcbhash) && 2526 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) { 2527 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2528 return; 2529 } 2530 if (!bound) { 2531 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2532 if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr)) 2533 break; 2534 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2535 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2536 return; 2537 } 2538 } 2539 } 2540 if (last == NULL) 2541 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2542 else 2543 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2544 } 2545 #endif 2546 2547 /* 2548 * Insert PCB onto various hash lists. 2549 */ 2550 int 2551 in_pcbinshash(struct inpcb *inp) 2552 { 2553 struct inpcbhead *pcbhash; 2554 struct inpcbporthead *pcbporthash; 2555 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2556 struct inpcbport *phd; 2557 uint32_t hash; 2558 bool connected; 2559 2560 INP_WLOCK_ASSERT(inp); 2561 INP_HASH_WLOCK_ASSERT(pcbinfo); 2562 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2563 ("in_pcbinshash: INP_INHASHLIST")); 2564 2565 #ifdef INET6 2566 if (inp->inp_vflag & INP_IPV6) { 2567 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2568 inp->inp_fport, pcbinfo->ipi_hashmask); 2569 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2570 } else 2571 #endif 2572 { 2573 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2574 inp->inp_fport, pcbinfo->ipi_hashmask); 2575 connected = !in_nullhost(inp->inp_faddr); 2576 } 2577 2578 if (connected) 2579 pcbhash = &pcbinfo->ipi_hash_exact[hash]; 2580 else 2581 pcbhash = &pcbinfo->ipi_hash_wild[hash]; 2582 2583 pcbporthash = &pcbinfo->ipi_porthashbase[ 2584 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2585 2586 /* 2587 * Add entry to load balance group. 2588 * Only do this if SO_REUSEPORT_LB is set. 2589 */ 2590 if ((inp->inp_socket->so_options & SO_REUSEPORT_LB) != 0) { 2591 int error = in_pcbinslbgrouphash(inp, M_NODOM); 2592 if (error != 0) 2593 return (error); 2594 } 2595 2596 /* 2597 * Go through port list and look for a head for this lport. 2598 */ 2599 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2600 if (phd->phd_port == inp->inp_lport) 2601 break; 2602 } 2603 2604 /* 2605 * If none exists, malloc one and tack it on. 2606 */ 2607 if (phd == NULL) { 2608 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT); 2609 if (phd == NULL) { 2610 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2611 in_pcbremlbgrouphash(inp); 2612 return (ENOMEM); 2613 } 2614 phd->phd_port = inp->inp_lport; 2615 CK_LIST_INIT(&phd->phd_pcblist); 2616 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2617 } 2618 inp->inp_phd = phd; 2619 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2620 2621 /* 2622 * The PCB may have been disconnected in the past. Before we can safely 2623 * make it visible in the hash table, we must wait for all readers which 2624 * may be traversing this PCB to finish. 2625 */ 2626 if (inp->inp_smr != SMR_SEQ_INVALID) { 2627 smr_wait(pcbinfo->ipi_smr, inp->inp_smr); 2628 inp->inp_smr = SMR_SEQ_INVALID; 2629 } 2630 2631 if (connected) 2632 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact); 2633 else { 2634 #ifdef INET6 2635 if ((inp->inp_vflag & INP_IPV6) != 0) 2636 _in6_pcbinshash_wild(pcbhash, inp); 2637 else 2638 #endif 2639 _in_pcbinshash_wild(pcbhash, inp); 2640 } 2641 inp->inp_flags |= INP_INHASHLIST; 2642 2643 return (0); 2644 } 2645 2646 void 2647 in_pcbremhash_locked(struct inpcb *inp) 2648 { 2649 struct inpcbport *phd = inp->inp_phd; 2650 2651 INP_WLOCK_ASSERT(inp); 2652 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2653 MPASS(inp->inp_flags & INP_INHASHLIST); 2654 2655 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2656 in_pcbremlbgrouphash(inp); 2657 #ifdef INET6 2658 if (inp->inp_vflag & INP_IPV6) { 2659 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) 2660 CK_LIST_REMOVE(inp, inp_hash_wild); 2661 else 2662 CK_LIST_REMOVE(inp, inp_hash_exact); 2663 } else 2664 #endif 2665 { 2666 if (in_nullhost(inp->inp_faddr)) 2667 CK_LIST_REMOVE(inp, inp_hash_wild); 2668 else 2669 CK_LIST_REMOVE(inp, inp_hash_exact); 2670 } 2671 CK_LIST_REMOVE(inp, inp_portlist); 2672 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2673 CK_LIST_REMOVE(phd, phd_hash); 2674 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd); 2675 } 2676 inp->inp_flags &= ~INP_INHASHLIST; 2677 } 2678 2679 static void 2680 in_pcbremhash(struct inpcb *inp) 2681 { 2682 INP_HASH_WLOCK(inp->inp_pcbinfo); 2683 in_pcbremhash_locked(inp); 2684 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 2685 } 2686 2687 /* 2688 * Move PCB to the proper hash bucket when { faddr, fport } have been 2689 * changed. NOTE: This does not handle the case of the lport changing (the 2690 * hashed port list would have to be updated as well), so the lport must 2691 * not change after in_pcbinshash() has been called. 2692 */ 2693 void 2694 in_pcbrehash(struct inpcb *inp) 2695 { 2696 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2697 struct inpcbhead *head; 2698 uint32_t hash; 2699 bool connected; 2700 2701 INP_WLOCK_ASSERT(inp); 2702 INP_HASH_WLOCK_ASSERT(pcbinfo); 2703 KASSERT(inp->inp_flags & INP_INHASHLIST, 2704 ("%s: !INP_INHASHLIST", __func__)); 2705 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 2706 ("%s: inp was disconnected", __func__)); 2707 2708 #ifdef INET6 2709 if (inp->inp_vflag & INP_IPV6) { 2710 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2711 inp->inp_fport, pcbinfo->ipi_hashmask); 2712 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2713 } else 2714 #endif 2715 { 2716 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2717 inp->inp_fport, pcbinfo->ipi_hashmask); 2718 connected = !in_nullhost(inp->inp_faddr); 2719 } 2720 2721 /* 2722 * When rehashing, the caller must ensure that either the new or the old 2723 * foreign address was unspecified. 2724 */ 2725 if (connected) 2726 CK_LIST_REMOVE(inp, inp_hash_wild); 2727 else 2728 CK_LIST_REMOVE(inp, inp_hash_exact); 2729 2730 if (connected) { 2731 head = &pcbinfo->ipi_hash_exact[hash]; 2732 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact); 2733 } else { 2734 head = &pcbinfo->ipi_hash_wild[hash]; 2735 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild); 2736 } 2737 } 2738 2739 /* 2740 * Check for alternatives when higher level complains 2741 * about service problems. For now, invalidate cached 2742 * routing information. If the route was created dynamically 2743 * (by a redirect), time to try a default gateway again. 2744 */ 2745 void 2746 in_losing(struct inpcb *inp) 2747 { 2748 2749 RO_INVALIDATE_CACHE(&inp->inp_route); 2750 return; 2751 } 2752 2753 /* 2754 * A set label operation has occurred at the socket layer, propagate the 2755 * label change into the in_pcb for the socket. 2756 */ 2757 void 2758 in_pcbsosetlabel(struct socket *so) 2759 { 2760 #ifdef MAC 2761 struct inpcb *inp; 2762 2763 inp = sotoinpcb(so); 2764 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2765 2766 INP_WLOCK(inp); 2767 SOCK_LOCK(so); 2768 mac_inpcb_sosetlabel(so, inp); 2769 SOCK_UNLOCK(so); 2770 INP_WUNLOCK(inp); 2771 #endif 2772 } 2773 2774 void 2775 inp_wlock(struct inpcb *inp) 2776 { 2777 2778 INP_WLOCK(inp); 2779 } 2780 2781 void 2782 inp_wunlock(struct inpcb *inp) 2783 { 2784 2785 INP_WUNLOCK(inp); 2786 } 2787 2788 void 2789 inp_rlock(struct inpcb *inp) 2790 { 2791 2792 INP_RLOCK(inp); 2793 } 2794 2795 void 2796 inp_runlock(struct inpcb *inp) 2797 { 2798 2799 INP_RUNLOCK(inp); 2800 } 2801 2802 #ifdef INVARIANT_SUPPORT 2803 void 2804 inp_lock_assert(struct inpcb *inp) 2805 { 2806 2807 INP_WLOCK_ASSERT(inp); 2808 } 2809 2810 void 2811 inp_unlock_assert(struct inpcb *inp) 2812 { 2813 2814 INP_UNLOCK_ASSERT(inp); 2815 } 2816 #endif 2817 2818 void 2819 inp_apply_all(struct inpcbinfo *pcbinfo, 2820 void (*func)(struct inpcb *, void *), void *arg) 2821 { 2822 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2823 INPLOOKUP_WLOCKPCB); 2824 struct inpcb *inp; 2825 2826 while ((inp = inp_next(&inpi)) != NULL) 2827 func(inp, arg); 2828 } 2829 2830 struct socket * 2831 inp_inpcbtosocket(struct inpcb *inp) 2832 { 2833 2834 INP_WLOCK_ASSERT(inp); 2835 return (inp->inp_socket); 2836 } 2837 2838 struct tcpcb * 2839 inp_inpcbtotcpcb(struct inpcb *inp) 2840 { 2841 2842 INP_WLOCK_ASSERT(inp); 2843 return ((struct tcpcb *)inp->inp_ppcb); 2844 } 2845 2846 int 2847 inp_ip_tos_get(const struct inpcb *inp) 2848 { 2849 2850 return (inp->inp_ip_tos); 2851 } 2852 2853 void 2854 inp_ip_tos_set(struct inpcb *inp, int val) 2855 { 2856 2857 inp->inp_ip_tos = val; 2858 } 2859 2860 void 2861 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2862 uint32_t *faddr, uint16_t *fp) 2863 { 2864 2865 INP_LOCK_ASSERT(inp); 2866 *laddr = inp->inp_laddr.s_addr; 2867 *faddr = inp->inp_faddr.s_addr; 2868 *lp = inp->inp_lport; 2869 *fp = inp->inp_fport; 2870 } 2871 2872 struct inpcb * 2873 so_sotoinpcb(struct socket *so) 2874 { 2875 2876 return (sotoinpcb(so)); 2877 } 2878 2879 /* 2880 * Create an external-format (``xinpcb'') structure using the information in 2881 * the kernel-format in_pcb structure pointed to by inp. This is done to 2882 * reduce the spew of irrelevant information over this interface, to isolate 2883 * user code from changes in the kernel structure, and potentially to provide 2884 * information-hiding if we decide that some of this information should be 2885 * hidden from users. 2886 */ 2887 void 2888 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2889 { 2890 2891 bzero(xi, sizeof(*xi)); 2892 xi->xi_len = sizeof(struct xinpcb); 2893 if (inp->inp_socket) 2894 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2895 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2896 xi->inp_gencnt = inp->inp_gencnt; 2897 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2898 xi->inp_flow = inp->inp_flow; 2899 xi->inp_flowid = inp->inp_flowid; 2900 xi->inp_flowtype = inp->inp_flowtype; 2901 xi->inp_flags = inp->inp_flags; 2902 xi->inp_flags2 = inp->inp_flags2; 2903 xi->in6p_cksum = inp->in6p_cksum; 2904 xi->in6p_hops = inp->in6p_hops; 2905 xi->inp_ip_tos = inp->inp_ip_tos; 2906 xi->inp_vflag = inp->inp_vflag; 2907 xi->inp_ip_ttl = inp->inp_ip_ttl; 2908 xi->inp_ip_p = inp->inp_ip_p; 2909 xi->inp_ip_minttl = inp->inp_ip_minttl; 2910 } 2911 2912 int 2913 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo, 2914 int (*ctloutput_set)(struct inpcb *, struct sockopt *)) 2915 { 2916 struct sockopt sopt; 2917 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2918 INPLOOKUP_WLOCKPCB); 2919 struct inpcb *inp; 2920 struct sockopt_parameters *params; 2921 struct socket *so; 2922 int error; 2923 char buf[1024]; 2924 2925 if (req->oldptr != NULL || req->oldlen != 0) 2926 return (EINVAL); 2927 if (req->newptr == NULL) 2928 return (EPERM); 2929 if (req->newlen > sizeof(buf)) 2930 return (ENOMEM); 2931 error = SYSCTL_IN(req, buf, req->newlen); 2932 if (error != 0) 2933 return (error); 2934 if (req->newlen < sizeof(struct sockopt_parameters)) 2935 return (EINVAL); 2936 params = (struct sockopt_parameters *)buf; 2937 sopt.sopt_level = params->sop_level; 2938 sopt.sopt_name = params->sop_optname; 2939 sopt.sopt_dir = SOPT_SET; 2940 sopt.sopt_val = params->sop_optval; 2941 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters); 2942 sopt.sopt_td = NULL; 2943 #ifdef INET6 2944 if (params->sop_inc.inc_flags & INC_ISIPV6) { 2945 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr)) 2946 params->sop_inc.inc6_laddr.s6_addr16[1] = 2947 htons(params->sop_inc.inc6_zoneid & 0xffff); 2948 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr)) 2949 params->sop_inc.inc6_faddr.s6_addr16[1] = 2950 htons(params->sop_inc.inc6_zoneid & 0xffff); 2951 } 2952 #endif 2953 if (params->sop_inc.inc_lport != htons(0) && 2954 params->sop_inc.inc_fport != htons(0)) { 2955 #ifdef INET6 2956 if (params->sop_inc.inc_flags & INC_ISIPV6) 2957 inpi.hash = INP6_PCBHASH( 2958 ¶ms->sop_inc.inc6_faddr, 2959 params->sop_inc.inc_lport, 2960 params->sop_inc.inc_fport, 2961 pcbinfo->ipi_hashmask); 2962 else 2963 #endif 2964 inpi.hash = INP_PCBHASH( 2965 ¶ms->sop_inc.inc_faddr, 2966 params->sop_inc.inc_lport, 2967 params->sop_inc.inc_fport, 2968 pcbinfo->ipi_hashmask); 2969 } 2970 while ((inp = inp_next(&inpi)) != NULL) 2971 if (inp->inp_gencnt == params->sop_id) { 2972 if (inp->inp_flags & INP_DROPPED) { 2973 INP_WUNLOCK(inp); 2974 return (ECONNRESET); 2975 } 2976 so = inp->inp_socket; 2977 KASSERT(so != NULL, ("inp_socket == NULL")); 2978 soref(so); 2979 if (params->sop_level == SOL_SOCKET) { 2980 INP_WUNLOCK(inp); 2981 error = sosetopt(so, &sopt); 2982 } else 2983 error = (*ctloutput_set)(inp, &sopt); 2984 sorele(so); 2985 break; 2986 } 2987 if (inp == NULL) 2988 error = ESRCH; 2989 return (error); 2990 } 2991 2992 #ifdef DDB 2993 static void 2994 db_print_indent(int indent) 2995 { 2996 int i; 2997 2998 for (i = 0; i < indent; i++) 2999 db_printf(" "); 3000 } 3001 3002 static void 3003 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 3004 { 3005 char faddr_str[48], laddr_str[48]; 3006 3007 db_print_indent(indent); 3008 db_printf("%s at %p\n", name, inc); 3009 3010 indent += 2; 3011 3012 #ifdef INET6 3013 if (inc->inc_flags & INC_ISIPV6) { 3014 /* IPv6. */ 3015 ip6_sprintf(laddr_str, &inc->inc6_laddr); 3016 ip6_sprintf(faddr_str, &inc->inc6_faddr); 3017 } else 3018 #endif 3019 { 3020 /* IPv4. */ 3021 inet_ntoa_r(inc->inc_laddr, laddr_str); 3022 inet_ntoa_r(inc->inc_faddr, faddr_str); 3023 } 3024 db_print_indent(indent); 3025 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3026 ntohs(inc->inc_lport)); 3027 db_print_indent(indent); 3028 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3029 ntohs(inc->inc_fport)); 3030 } 3031 3032 static void 3033 db_print_inpflags(int inp_flags) 3034 { 3035 int comma; 3036 3037 comma = 0; 3038 if (inp_flags & INP_RECVOPTS) { 3039 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3040 comma = 1; 3041 } 3042 if (inp_flags & INP_RECVRETOPTS) { 3043 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3044 comma = 1; 3045 } 3046 if (inp_flags & INP_RECVDSTADDR) { 3047 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3048 comma = 1; 3049 } 3050 if (inp_flags & INP_ORIGDSTADDR) { 3051 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3052 comma = 1; 3053 } 3054 if (inp_flags & INP_HDRINCL) { 3055 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3056 comma = 1; 3057 } 3058 if (inp_flags & INP_HIGHPORT) { 3059 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3060 comma = 1; 3061 } 3062 if (inp_flags & INP_LOWPORT) { 3063 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3064 comma = 1; 3065 } 3066 if (inp_flags & INP_ANONPORT) { 3067 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3068 comma = 1; 3069 } 3070 if (inp_flags & INP_RECVIF) { 3071 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3072 comma = 1; 3073 } 3074 if (inp_flags & INP_MTUDISC) { 3075 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3076 comma = 1; 3077 } 3078 if (inp_flags & INP_RECVTTL) { 3079 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3080 comma = 1; 3081 } 3082 if (inp_flags & INP_DONTFRAG) { 3083 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3084 comma = 1; 3085 } 3086 if (inp_flags & INP_RECVTOS) { 3087 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3088 comma = 1; 3089 } 3090 if (inp_flags & IN6P_IPV6_V6ONLY) { 3091 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3092 comma = 1; 3093 } 3094 if (inp_flags & IN6P_PKTINFO) { 3095 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3096 comma = 1; 3097 } 3098 if (inp_flags & IN6P_HOPLIMIT) { 3099 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3100 comma = 1; 3101 } 3102 if (inp_flags & IN6P_HOPOPTS) { 3103 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3104 comma = 1; 3105 } 3106 if (inp_flags & IN6P_DSTOPTS) { 3107 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3108 comma = 1; 3109 } 3110 if (inp_flags & IN6P_RTHDR) { 3111 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3112 comma = 1; 3113 } 3114 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3115 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3116 comma = 1; 3117 } 3118 if (inp_flags & IN6P_TCLASS) { 3119 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3120 comma = 1; 3121 } 3122 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3123 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3124 comma = 1; 3125 } 3126 if (inp_flags & INP_ONESBCAST) { 3127 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3128 comma = 1; 3129 } 3130 if (inp_flags & INP_DROPPED) { 3131 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3132 comma = 1; 3133 } 3134 if (inp_flags & INP_SOCKREF) { 3135 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3136 comma = 1; 3137 } 3138 if (inp_flags & IN6P_RFC2292) { 3139 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3140 comma = 1; 3141 } 3142 if (inp_flags & IN6P_MTU) { 3143 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3144 comma = 1; 3145 } 3146 } 3147 3148 static void 3149 db_print_inpvflag(u_char inp_vflag) 3150 { 3151 int comma; 3152 3153 comma = 0; 3154 if (inp_vflag & INP_IPV4) { 3155 db_printf("%sINP_IPV4", comma ? ", " : ""); 3156 comma = 1; 3157 } 3158 if (inp_vflag & INP_IPV6) { 3159 db_printf("%sINP_IPV6", comma ? ", " : ""); 3160 comma = 1; 3161 } 3162 if (inp_vflag & INP_IPV6PROTO) { 3163 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3164 comma = 1; 3165 } 3166 } 3167 3168 static void 3169 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3170 { 3171 3172 db_print_indent(indent); 3173 db_printf("%s at %p\n", name, inp); 3174 3175 indent += 2; 3176 3177 db_print_indent(indent); 3178 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3179 3180 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3181 3182 db_print_indent(indent); 3183 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3184 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3185 3186 db_print_indent(indent); 3187 db_printf("inp_label: %p inp_flags: 0x%x (", 3188 inp->inp_label, inp->inp_flags); 3189 db_print_inpflags(inp->inp_flags); 3190 db_printf(")\n"); 3191 3192 db_print_indent(indent); 3193 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3194 inp->inp_vflag); 3195 db_print_inpvflag(inp->inp_vflag); 3196 db_printf(")\n"); 3197 3198 db_print_indent(indent); 3199 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3200 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3201 3202 db_print_indent(indent); 3203 #ifdef INET6 3204 if (inp->inp_vflag & INP_IPV6) { 3205 db_printf("in6p_options: %p in6p_outputopts: %p " 3206 "in6p_moptions: %p\n", inp->in6p_options, 3207 inp->in6p_outputopts, inp->in6p_moptions); 3208 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3209 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3210 inp->in6p_hops); 3211 } else 3212 #endif 3213 { 3214 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3215 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3216 inp->inp_options, inp->inp_moptions); 3217 } 3218 3219 db_print_indent(indent); 3220 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3221 (uintmax_t)inp->inp_gencnt); 3222 } 3223 3224 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3225 { 3226 struct inpcb *inp; 3227 3228 if (!have_addr) { 3229 db_printf("usage: show inpcb <addr>\n"); 3230 return; 3231 } 3232 inp = (struct inpcb *)addr; 3233 3234 db_print_inpcb(inp, "inpcb", 0); 3235 } 3236 #endif /* DDB */ 3237 3238 #ifdef RATELIMIT 3239 /* 3240 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3241 * if any. 3242 */ 3243 int 3244 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3245 { 3246 union if_snd_tag_modify_params params = { 3247 .rate_limit.max_rate = max_pacing_rate, 3248 .rate_limit.flags = M_NOWAIT, 3249 }; 3250 struct m_snd_tag *mst; 3251 int error; 3252 3253 mst = inp->inp_snd_tag; 3254 if (mst == NULL) 3255 return (EINVAL); 3256 3257 if (mst->sw->snd_tag_modify == NULL) { 3258 error = EOPNOTSUPP; 3259 } else { 3260 error = mst->sw->snd_tag_modify(mst, ¶ms); 3261 } 3262 return (error); 3263 } 3264 3265 /* 3266 * Query existing TX rate limit based on the existing 3267 * "inp->inp_snd_tag", if any. 3268 */ 3269 int 3270 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3271 { 3272 union if_snd_tag_query_params params = { }; 3273 struct m_snd_tag *mst; 3274 int error; 3275 3276 mst = inp->inp_snd_tag; 3277 if (mst == NULL) 3278 return (EINVAL); 3279 3280 if (mst->sw->snd_tag_query == NULL) { 3281 error = EOPNOTSUPP; 3282 } else { 3283 error = mst->sw->snd_tag_query(mst, ¶ms); 3284 if (error == 0 && p_max_pacing_rate != NULL) 3285 *p_max_pacing_rate = params.rate_limit.max_rate; 3286 } 3287 return (error); 3288 } 3289 3290 /* 3291 * Query existing TX queue level based on the existing 3292 * "inp->inp_snd_tag", if any. 3293 */ 3294 int 3295 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3296 { 3297 union if_snd_tag_query_params params = { }; 3298 struct m_snd_tag *mst; 3299 int error; 3300 3301 mst = inp->inp_snd_tag; 3302 if (mst == NULL) 3303 return (EINVAL); 3304 3305 if (mst->sw->snd_tag_query == NULL) 3306 return (EOPNOTSUPP); 3307 3308 error = mst->sw->snd_tag_query(mst, ¶ms); 3309 if (error == 0 && p_txqueue_level != NULL) 3310 *p_txqueue_level = params.rate_limit.queue_level; 3311 return (error); 3312 } 3313 3314 /* 3315 * Allocate a new TX rate limit send tag from the network interface 3316 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3317 */ 3318 int 3319 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3320 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3321 3322 { 3323 union if_snd_tag_alloc_params params = { 3324 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3325 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3326 .rate_limit.hdr.flowid = flowid, 3327 .rate_limit.hdr.flowtype = flowtype, 3328 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3329 .rate_limit.max_rate = max_pacing_rate, 3330 .rate_limit.flags = M_NOWAIT, 3331 }; 3332 int error; 3333 3334 INP_WLOCK_ASSERT(inp); 3335 3336 /* 3337 * If there is already a send tag, or the INP is being torn 3338 * down, allocating a new send tag is not allowed. Else send 3339 * tags may leak. 3340 */ 3341 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0) 3342 return (EINVAL); 3343 3344 error = m_snd_tag_alloc(ifp, ¶ms, st); 3345 #ifdef INET 3346 if (error == 0) { 3347 counter_u64_add(rate_limit_set_ok, 1); 3348 counter_u64_add(rate_limit_active, 1); 3349 } else if (error != EOPNOTSUPP) 3350 counter_u64_add(rate_limit_alloc_fail, 1); 3351 #endif 3352 return (error); 3353 } 3354 3355 void 3356 in_pcbdetach_tag(struct m_snd_tag *mst) 3357 { 3358 3359 m_snd_tag_rele(mst); 3360 #ifdef INET 3361 counter_u64_add(rate_limit_active, -1); 3362 #endif 3363 } 3364 3365 /* 3366 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3367 * if any: 3368 */ 3369 void 3370 in_pcbdetach_txrtlmt(struct inpcb *inp) 3371 { 3372 struct m_snd_tag *mst; 3373 3374 INP_WLOCK_ASSERT(inp); 3375 3376 mst = inp->inp_snd_tag; 3377 inp->inp_snd_tag = NULL; 3378 3379 if (mst == NULL) 3380 return; 3381 3382 m_snd_tag_rele(mst); 3383 #ifdef INET 3384 counter_u64_add(rate_limit_active, -1); 3385 #endif 3386 } 3387 3388 int 3389 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3390 { 3391 int error; 3392 3393 /* 3394 * If the existing send tag is for the wrong interface due to 3395 * a route change, first drop the existing tag. Set the 3396 * CHANGED flag so that we will keep trying to allocate a new 3397 * tag if we fail to allocate one this time. 3398 */ 3399 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3400 in_pcbdetach_txrtlmt(inp); 3401 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3402 } 3403 3404 /* 3405 * NOTE: When attaching to a network interface a reference is 3406 * made to ensure the network interface doesn't go away until 3407 * all ratelimit connections are gone. The network interface 3408 * pointers compared below represent valid network interfaces, 3409 * except when comparing towards NULL. 3410 */ 3411 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3412 error = 0; 3413 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3414 if (inp->inp_snd_tag != NULL) 3415 in_pcbdetach_txrtlmt(inp); 3416 error = 0; 3417 } else if (inp->inp_snd_tag == NULL) { 3418 /* 3419 * In order to utilize packet pacing with RSS, we need 3420 * to wait until there is a valid RSS hash before we 3421 * can proceed: 3422 */ 3423 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3424 error = EAGAIN; 3425 } else { 3426 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3427 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3428 } 3429 } else { 3430 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3431 } 3432 if (error == 0 || error == EOPNOTSUPP) 3433 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3434 3435 return (error); 3436 } 3437 3438 /* 3439 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3440 * is set in the fast path and will attach/detach/modify the TX rate 3441 * limit send tag based on the socket's so_max_pacing_rate value. 3442 */ 3443 void 3444 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3445 { 3446 struct socket *socket; 3447 uint32_t max_pacing_rate; 3448 bool did_upgrade; 3449 3450 if (inp == NULL) 3451 return; 3452 3453 socket = inp->inp_socket; 3454 if (socket == NULL) 3455 return; 3456 3457 if (!INP_WLOCKED(inp)) { 3458 /* 3459 * NOTE: If the write locking fails, we need to bail 3460 * out and use the non-ratelimited ring for the 3461 * transmit until there is a new chance to get the 3462 * write lock. 3463 */ 3464 if (!INP_TRY_UPGRADE(inp)) 3465 return; 3466 did_upgrade = 1; 3467 } else { 3468 did_upgrade = 0; 3469 } 3470 3471 /* 3472 * NOTE: The so_max_pacing_rate value is read unlocked, 3473 * because atomic updates are not required since the variable 3474 * is checked at every mbuf we send. It is assumed that the 3475 * variable read itself will be atomic. 3476 */ 3477 max_pacing_rate = socket->so_max_pacing_rate; 3478 3479 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3480 3481 if (did_upgrade) 3482 INP_DOWNGRADE(inp); 3483 } 3484 3485 /* 3486 * Track route changes for TX rate limiting. 3487 */ 3488 void 3489 in_pcboutput_eagain(struct inpcb *inp) 3490 { 3491 bool did_upgrade; 3492 3493 if (inp == NULL) 3494 return; 3495 3496 if (inp->inp_snd_tag == NULL) 3497 return; 3498 3499 if (!INP_WLOCKED(inp)) { 3500 /* 3501 * NOTE: If the write locking fails, we need to bail 3502 * out and use the non-ratelimited ring for the 3503 * transmit until there is a new chance to get the 3504 * write lock. 3505 */ 3506 if (!INP_TRY_UPGRADE(inp)) 3507 return; 3508 did_upgrade = 1; 3509 } else { 3510 did_upgrade = 0; 3511 } 3512 3513 /* detach rate limiting */ 3514 in_pcbdetach_txrtlmt(inp); 3515 3516 /* make sure new mbuf send tag allocation is made */ 3517 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3518 3519 if (did_upgrade) 3520 INP_DOWNGRADE(inp); 3521 } 3522 3523 #ifdef INET 3524 static void 3525 rl_init(void *st) 3526 { 3527 rate_limit_new = counter_u64_alloc(M_WAITOK); 3528 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3529 rate_limit_active = counter_u64_alloc(M_WAITOK); 3530 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3531 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3532 } 3533 3534 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3535 #endif 3536 #endif /* RATELIMIT */ 3537