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 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1408 * For most protocols, this will be invoked immediately prior to calling 1409 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1410 * socket, in which case in_pcbfree() is deferred. 1411 */ 1412 void 1413 in_pcbdetach(struct inpcb *inp) 1414 { 1415 1416 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1417 1418 #ifdef RATELIMIT 1419 if (inp->inp_snd_tag != NULL) 1420 in_pcbdetach_txrtlmt(inp); 1421 #endif 1422 inp->inp_socket->so_pcb = NULL; 1423 inp->inp_socket = NULL; 1424 } 1425 1426 /* 1427 * inpcb hash lookups are protected by SMR section. 1428 * 1429 * Once desired pcb has been found, switching from SMR section to a pcb 1430 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK 1431 * here because SMR is a critical section. 1432 * In 99%+ cases inp_smr_lock() would obtain the lock immediately. 1433 */ 1434 void 1435 inp_lock(struct inpcb *inp, const inp_lookup_t lock) 1436 { 1437 1438 lock == INPLOOKUP_RLOCKPCB ? 1439 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock); 1440 } 1441 1442 void 1443 inp_unlock(struct inpcb *inp, const inp_lookup_t lock) 1444 { 1445 1446 lock == INPLOOKUP_RLOCKPCB ? 1447 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock); 1448 } 1449 1450 int 1451 inp_trylock(struct inpcb *inp, const inp_lookup_t lock) 1452 { 1453 1454 return (lock == INPLOOKUP_RLOCKPCB ? 1455 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock)); 1456 } 1457 1458 static inline bool 1459 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags) 1460 { 1461 1462 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB); 1463 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr); 1464 1465 if (__predict_true(inp_trylock(inp, lock))) { 1466 if (__predict_false(inp->inp_flags & ignflags)) { 1467 smr_exit(inp->inp_pcbinfo->ipi_smr); 1468 inp_unlock(inp, lock); 1469 return (false); 1470 } 1471 smr_exit(inp->inp_pcbinfo->ipi_smr); 1472 return (true); 1473 } 1474 1475 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1476 smr_exit(inp->inp_pcbinfo->ipi_smr); 1477 inp_lock(inp, lock); 1478 if (__predict_false(in_pcbrele(inp, lock))) 1479 return (false); 1480 /* 1481 * inp acquired through refcount & lock for sure didn't went 1482 * through uma_zfree(). However, it may have already went 1483 * through in_pcbfree() and has another reference, that 1484 * prevented its release by our in_pcbrele(). 1485 */ 1486 if (__predict_false(inp->inp_flags & ignflags)) { 1487 inp_unlock(inp, lock); 1488 return (false); 1489 } 1490 return (true); 1491 } else { 1492 smr_exit(inp->inp_pcbinfo->ipi_smr); 1493 return (false); 1494 } 1495 } 1496 1497 bool 1498 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock) 1499 { 1500 1501 /* 1502 * in_pcblookup() family of functions ignore not only freed entries, 1503 * that may be found due to lockless access to the hash, but dropped 1504 * entries, too. 1505 */ 1506 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED)); 1507 } 1508 1509 /* 1510 * inp_next() - inpcb hash/list traversal iterator 1511 * 1512 * Requires initialized struct inpcb_iterator for context. 1513 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR(). 1514 * 1515 * - Iterator can have either write-lock or read-lock semantics, that can not 1516 * be changed later. 1517 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through 1518 * a single hash slot. Note: only rip_input() does the latter. 1519 * - Iterator may have optional bool matching function. The matching function 1520 * will be executed for each inpcb in the SMR context, so it can not acquire 1521 * locks and can safely access only immutable fields of inpcb. 1522 * 1523 * A fresh initialized iterator has NULL inpcb in its context and that 1524 * means that inp_next() call would return the very first inpcb on the list 1525 * locked with desired semantic. In all following calls the context pointer 1526 * shall hold the current inpcb pointer. The KPI user is not supposed to 1527 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL 1528 * and write NULL to its context. After end of traversal an iterator can be 1529 * reused. 1530 * 1531 * List traversals have the following features/constraints: 1532 * - New entries won't be seen, as they are always added to the head of a list. 1533 * - Removed entries won't stop traversal as long as they are not added to 1534 * a different list. This is violated by in_pcbrehash(). 1535 */ 1536 #define II_LIST_FIRST(ipi, hash) \ 1537 (((hash) == INP_ALL_LIST) ? \ 1538 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \ 1539 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)])) 1540 #define II_LIST_NEXT(inp, hash) \ 1541 (((hash) == INP_ALL_LIST) ? \ 1542 CK_LIST_NEXT((inp), inp_list) : \ 1543 CK_LIST_NEXT((inp), inp_hash_exact)) 1544 #define II_LOCK_ASSERT(inp, lock) \ 1545 rw_assert(&(inp)->inp_lock, \ 1546 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED ) 1547 struct inpcb * 1548 inp_next(struct inpcb_iterator *ii) 1549 { 1550 const struct inpcbinfo *ipi = ii->ipi; 1551 inp_match_t *match = ii->match; 1552 void *ctx = ii->ctx; 1553 inp_lookup_t lock = ii->lock; 1554 int hash = ii->hash; 1555 struct inpcb *inp; 1556 1557 if (ii->inp == NULL) { /* First call. */ 1558 smr_enter(ipi->ipi_smr); 1559 /* This is unrolled CK_LIST_FOREACH(). */ 1560 for (inp = II_LIST_FIRST(ipi, hash); 1561 inp != NULL; 1562 inp = II_LIST_NEXT(inp, hash)) { 1563 if (match != NULL && (match)(inp, ctx) == false) 1564 continue; 1565 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED))) 1566 break; 1567 else { 1568 smr_enter(ipi->ipi_smr); 1569 MPASS(inp != II_LIST_FIRST(ipi, hash)); 1570 inp = II_LIST_FIRST(ipi, hash); 1571 if (inp == NULL) 1572 break; 1573 } 1574 } 1575 1576 if (inp == NULL) 1577 smr_exit(ipi->ipi_smr); 1578 else 1579 ii->inp = inp; 1580 1581 return (inp); 1582 } 1583 1584 /* Not a first call. */ 1585 smr_enter(ipi->ipi_smr); 1586 restart: 1587 inp = ii->inp; 1588 II_LOCK_ASSERT(inp, lock); 1589 next: 1590 inp = II_LIST_NEXT(inp, hash); 1591 if (inp == NULL) { 1592 smr_exit(ipi->ipi_smr); 1593 goto found; 1594 } 1595 1596 if (match != NULL && (match)(inp, ctx) == false) 1597 goto next; 1598 1599 if (__predict_true(inp_trylock(inp, lock))) { 1600 if (__predict_false(inp->inp_flags & INP_FREED)) { 1601 /* 1602 * Entries are never inserted in middle of a list, thus 1603 * as long as we are in SMR, we can continue traversal. 1604 * Jump to 'restart' should yield in the same result, 1605 * but could produce unnecessary looping. Could this 1606 * looping be unbound? 1607 */ 1608 inp_unlock(inp, lock); 1609 goto next; 1610 } else { 1611 smr_exit(ipi->ipi_smr); 1612 goto found; 1613 } 1614 } 1615 1616 /* 1617 * Can't obtain lock immediately, thus going hard. Once we exit the 1618 * SMR section we can no longer jump to 'next', and our only stable 1619 * anchoring point is ii->inp, which we keep locked for this case, so 1620 * we jump to 'restart'. 1621 */ 1622 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1623 smr_exit(ipi->ipi_smr); 1624 inp_lock(inp, lock); 1625 if (__predict_false(in_pcbrele(inp, lock))) { 1626 smr_enter(ipi->ipi_smr); 1627 goto restart; 1628 } 1629 /* 1630 * See comment in inp_smr_lock(). 1631 */ 1632 if (__predict_false(inp->inp_flags & INP_FREED)) { 1633 inp_unlock(inp, lock); 1634 smr_enter(ipi->ipi_smr); 1635 goto restart; 1636 } 1637 } else 1638 goto next; 1639 1640 found: 1641 inp_unlock(ii->inp, lock); 1642 ii->inp = inp; 1643 1644 return (ii->inp); 1645 } 1646 1647 /* 1648 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1649 * stability of an inpcb pointer despite the inpcb lock being released or 1650 * SMR section exited. 1651 * 1652 * To free a reference later in_pcbrele_(r|w)locked() must be performed. 1653 */ 1654 void 1655 in_pcbref(struct inpcb *inp) 1656 { 1657 u_int old __diagused; 1658 1659 old = refcount_acquire(&inp->inp_refcount); 1660 KASSERT(old > 0, ("%s: refcount 0", __func__)); 1661 } 1662 1663 /* 1664 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially 1665 * freeing the pcb, if the reference was very last. 1666 */ 1667 bool 1668 in_pcbrele_rlocked(struct inpcb *inp) 1669 { 1670 1671 INP_RLOCK_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_RUNLOCK(inp); 1683 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1684 return (true); 1685 } 1686 1687 bool 1688 in_pcbrele_wlocked(struct inpcb *inp) 1689 { 1690 1691 INP_WLOCK_ASSERT(inp); 1692 1693 if (!refcount_release(&inp->inp_refcount)) 1694 return (false); 1695 1696 MPASS(inp->inp_flags & INP_FREED); 1697 MPASS(inp->inp_socket == NULL); 1698 crfree(inp->inp_cred); 1699 #ifdef INVARIANTS 1700 inp->inp_cred = NULL; 1701 #endif 1702 INP_WUNLOCK(inp); 1703 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1704 return (true); 1705 } 1706 1707 bool 1708 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock) 1709 { 1710 1711 return (lock == INPLOOKUP_RLOCKPCB ? 1712 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp)); 1713 } 1714 1715 /* 1716 * Unconditionally schedule an inpcb to be freed by decrementing its 1717 * reference count, which should occur only after the inpcb has been detached 1718 * from its socket. If another thread holds a temporary reference (acquired 1719 * using in_pcbref()) then the free is deferred until that reference is 1720 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked. 1721 * Almost all work, including removal from global lists, is done in this 1722 * context, where the pcbinfo lock is held. 1723 */ 1724 void 1725 in_pcbfree(struct inpcb *inp) 1726 { 1727 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1728 #ifdef INET 1729 struct ip_moptions *imo; 1730 #endif 1731 #ifdef INET6 1732 struct ip6_moptions *im6o; 1733 #endif 1734 1735 INP_WLOCK_ASSERT(inp); 1736 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1737 KASSERT((inp->inp_flags & INP_FREED) == 0, 1738 ("%s: called twice for pcb %p", __func__, inp)); 1739 1740 inp->inp_flags |= INP_FREED; 1741 INP_INFO_WLOCK(pcbinfo); 1742 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1743 pcbinfo->ipi_count--; 1744 CK_LIST_REMOVE(inp, inp_list); 1745 INP_INFO_WUNLOCK(pcbinfo); 1746 1747 if (inp->inp_flags & INP_INHASHLIST) 1748 in_pcbremhash(inp); 1749 1750 RO_INVALIDATE_CACHE(&inp->inp_route); 1751 #ifdef MAC 1752 mac_inpcb_destroy(inp); 1753 #endif 1754 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1755 if (inp->inp_sp != NULL) 1756 ipsec_delete_pcbpolicy(inp); 1757 #endif 1758 #ifdef INET 1759 if (inp->inp_options) 1760 (void)m_free(inp->inp_options); 1761 imo = inp->inp_moptions; 1762 #endif 1763 #ifdef INET6 1764 if (inp->inp_vflag & INP_IPV6PROTO) { 1765 ip6_freepcbopts(inp->in6p_outputopts); 1766 im6o = inp->in6p_moptions; 1767 } else 1768 im6o = NULL; 1769 #endif 1770 1771 if (__predict_false(in_pcbrele_wlocked(inp) == false)) { 1772 INP_WUNLOCK(inp); 1773 } 1774 #ifdef INET6 1775 ip6_freemoptions(im6o); 1776 #endif 1777 #ifdef INET 1778 inp_freemoptions(imo); 1779 #endif 1780 } 1781 1782 /* 1783 * Different protocols initialize their inpcbs differently - giving 1784 * different name to the lock. But they all are disposed the same. 1785 */ 1786 static void 1787 inpcb_fini(void *mem, int size) 1788 { 1789 struct inpcb *inp = mem; 1790 1791 INP_LOCK_DESTROY(inp); 1792 } 1793 1794 /* 1795 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1796 * port reservation, and preventing it from being returned by inpcb lookups. 1797 * 1798 * It is used by TCP to mark an inpcb as unused and avoid future packet 1799 * delivery or event notification when a socket remains open but TCP has 1800 * closed. This might occur as a result of a shutdown()-initiated TCP close 1801 * or a RST on the wire, and allows the port binding to be reused while still 1802 * maintaining the invariant that so_pcb always points to a valid inpcb until 1803 * in_pcbdetach(). 1804 * 1805 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1806 * in_pcbpurgeif0()? 1807 */ 1808 void 1809 in_pcbdrop(struct inpcb *inp) 1810 { 1811 1812 INP_WLOCK_ASSERT(inp); 1813 #ifdef INVARIANTS 1814 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1815 MPASS(inp->inp_refcount > 1); 1816 #endif 1817 1818 inp->inp_flags |= INP_DROPPED; 1819 if (inp->inp_flags & INP_INHASHLIST) 1820 in_pcbremhash(inp); 1821 } 1822 1823 #ifdef INET 1824 /* 1825 * Common routines to return the socket addresses associated with inpcbs. 1826 */ 1827 int 1828 in_getsockaddr(struct socket *so, struct sockaddr *sa) 1829 { 1830 struct inpcb *inp; 1831 1832 inp = sotoinpcb(so); 1833 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1834 1835 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 1836 .sin_len = sizeof(struct sockaddr_in), 1837 .sin_family = AF_INET, 1838 .sin_port = inp->inp_lport, 1839 .sin_addr = inp->inp_laddr, 1840 }; 1841 1842 return (0); 1843 } 1844 1845 int 1846 in_getpeeraddr(struct socket *so, struct sockaddr *sa) 1847 { 1848 struct inpcb *inp; 1849 1850 inp = sotoinpcb(so); 1851 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1852 1853 *(struct sockaddr_in *)sa = (struct sockaddr_in ){ 1854 .sin_len = sizeof(struct sockaddr_in), 1855 .sin_family = AF_INET, 1856 .sin_port = inp->inp_fport, 1857 .sin_addr = inp->inp_faddr, 1858 }; 1859 1860 return (0); 1861 } 1862 1863 static bool 1864 inp_v4_multi_match(const struct inpcb *inp, void *v __unused) 1865 { 1866 1867 if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL) 1868 return (true); 1869 else 1870 return (false); 1871 } 1872 1873 void 1874 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1875 { 1876 struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB, 1877 inp_v4_multi_match, NULL); 1878 struct inpcb *inp; 1879 struct in_multi *inm; 1880 struct in_mfilter *imf; 1881 struct ip_moptions *imo; 1882 1883 IN_MULTI_LOCK_ASSERT(); 1884 1885 while ((inp = inp_next(&inpi)) != NULL) { 1886 INP_WLOCK_ASSERT(inp); 1887 1888 imo = inp->inp_moptions; 1889 /* 1890 * Unselect the outgoing interface if it is being 1891 * detached. 1892 */ 1893 if (imo->imo_multicast_ifp == ifp) 1894 imo->imo_multicast_ifp = NULL; 1895 1896 /* 1897 * Drop multicast group membership if we joined 1898 * through the interface being detached. 1899 * 1900 * XXX This can all be deferred to an epoch_call 1901 */ 1902 restart: 1903 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 1904 if ((inm = imf->imf_inm) == NULL) 1905 continue; 1906 if (inm->inm_ifp != ifp) 1907 continue; 1908 ip_mfilter_remove(&imo->imo_head, imf); 1909 in_leavegroup_locked(inm, NULL); 1910 ip_mfilter_free(imf); 1911 goto restart; 1912 } 1913 } 1914 } 1915 1916 /* 1917 * Lookup a PCB based on the local address and port. Caller must hold the 1918 * hash lock. No inpcb locks or references are acquired. 1919 */ 1920 #define INP_LOOKUP_MAPPED_PCB_COST 3 1921 struct inpcb * 1922 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1923 u_short lport, int lookupflags, struct ucred *cred) 1924 { 1925 struct inpcb *inp; 1926 #ifdef INET6 1927 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1928 #else 1929 int matchwild = 3; 1930 #endif 1931 int wildcard; 1932 1933 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1934 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1935 INP_HASH_LOCK_ASSERT(pcbinfo); 1936 1937 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1938 struct inpcbhead *head; 1939 /* 1940 * Look for an unconnected (wildcard foreign addr) PCB that 1941 * matches the local address and port we're looking for. 1942 */ 1943 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 1944 pcbinfo->ipi_hashmask)]; 1945 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 1946 #ifdef INET6 1947 /* XXX inp locking */ 1948 if ((inp->inp_vflag & INP_IPV4) == 0) 1949 continue; 1950 #endif 1951 if (inp->inp_faddr.s_addr == INADDR_ANY && 1952 inp->inp_laddr.s_addr == laddr.s_addr && 1953 inp->inp_lport == lport) { 1954 /* 1955 * Found? 1956 */ 1957 if (prison_equal_ip4(cred->cr_prison, 1958 inp->inp_cred->cr_prison)) 1959 return (inp); 1960 } 1961 } 1962 /* 1963 * Not found. 1964 */ 1965 return (NULL); 1966 } else { 1967 struct inpcbporthead *porthash; 1968 struct inpcbport *phd; 1969 struct inpcb *match = NULL; 1970 /* 1971 * Best fit PCB lookup. 1972 * 1973 * First see if this local port is in use by looking on the 1974 * port hash list. 1975 */ 1976 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1977 pcbinfo->ipi_porthashmask)]; 1978 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1979 if (phd->phd_port == lport) 1980 break; 1981 } 1982 if (phd != NULL) { 1983 /* 1984 * Port is in use by one or more PCBs. Look for best 1985 * fit. 1986 */ 1987 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1988 wildcard = 0; 1989 if (!prison_equal_ip4(inp->inp_cred->cr_prison, 1990 cred->cr_prison)) 1991 continue; 1992 #ifdef INET6 1993 /* XXX inp locking */ 1994 if ((inp->inp_vflag & INP_IPV4) == 0) 1995 continue; 1996 /* 1997 * We never select the PCB that has 1998 * INP_IPV6 flag and is bound to :: if 1999 * we have another PCB which is bound 2000 * to 0.0.0.0. If a PCB has the 2001 * INP_IPV6 flag, then we set its cost 2002 * higher than IPv4 only PCBs. 2003 * 2004 * Note that the case only happens 2005 * when a socket is bound to ::, under 2006 * the condition that the use of the 2007 * mapped address is allowed. 2008 */ 2009 if ((inp->inp_vflag & INP_IPV6) != 0) 2010 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 2011 #endif 2012 if (inp->inp_faddr.s_addr != INADDR_ANY) 2013 wildcard++; 2014 if (inp->inp_laddr.s_addr != INADDR_ANY) { 2015 if (laddr.s_addr == INADDR_ANY) 2016 wildcard++; 2017 else if (inp->inp_laddr.s_addr != laddr.s_addr) 2018 continue; 2019 } else { 2020 if (laddr.s_addr != INADDR_ANY) 2021 wildcard++; 2022 } 2023 if (wildcard < matchwild) { 2024 match = inp; 2025 matchwild = wildcard; 2026 if (matchwild == 0) 2027 break; 2028 } 2029 } 2030 } 2031 return (match); 2032 } 2033 } 2034 #undef INP_LOOKUP_MAPPED_PCB_COST 2035 2036 static bool 2037 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain) 2038 { 2039 return (domain == M_NODOM || domain == grp->il_numa_domain); 2040 } 2041 2042 static struct inpcb * 2043 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 2044 const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr, 2045 uint16_t lport, int domain) 2046 { 2047 const struct inpcblbgrouphead *hdr; 2048 struct inpcblbgroup *grp; 2049 struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild; 2050 2051 INP_HASH_LOCK_ASSERT(pcbinfo); 2052 2053 hdr = &pcbinfo->ipi_lbgrouphashbase[ 2054 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 2055 2056 /* 2057 * Search for an LB group match based on the following criteria: 2058 * - prefer jailed groups to non-jailed groups 2059 * - prefer exact source address matches to wildcard matches 2060 * - prefer groups bound to the specified NUMA domain 2061 */ 2062 jail_exact = jail_wild = local_exact = local_wild = NULL; 2063 CK_LIST_FOREACH(grp, hdr, il_list) { 2064 bool injail; 2065 2066 #ifdef INET6 2067 if (!(grp->il_vflag & INP_IPV4)) 2068 continue; 2069 #endif 2070 if (grp->il_lport != lport) 2071 continue; 2072 2073 injail = prison_flag(grp->il_cred, PR_IP4) != 0; 2074 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison, 2075 laddr) != 0) 2076 continue; 2077 2078 if (grp->il_laddr.s_addr == laddr->s_addr) { 2079 if (injail) { 2080 jail_exact = grp; 2081 if (in_pcblookup_lb_numa_match(grp, domain)) 2082 /* This is a perfect match. */ 2083 goto out; 2084 } else if (local_exact == NULL || 2085 in_pcblookup_lb_numa_match(grp, domain)) { 2086 local_exact = grp; 2087 } 2088 } else if (grp->il_laddr.s_addr == INADDR_ANY) { 2089 if (injail) { 2090 if (jail_wild == NULL || 2091 in_pcblookup_lb_numa_match(grp, domain)) 2092 jail_wild = grp; 2093 } else if (local_wild == NULL || 2094 in_pcblookup_lb_numa_match(grp, domain)) { 2095 local_wild = grp; 2096 } 2097 } 2098 } 2099 2100 if (jail_exact != NULL) 2101 grp = jail_exact; 2102 else if (jail_wild != NULL) 2103 grp = jail_wild; 2104 else if (local_exact != NULL) 2105 grp = local_exact; 2106 else 2107 grp = local_wild; 2108 if (grp == NULL) 2109 return (NULL); 2110 out: 2111 return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) % 2112 grp->il_inpcnt]); 2113 } 2114 2115 static bool 2116 in_pcblookup_exact_match(const struct inpcb *inp, struct in_addr faddr, 2117 u_short fport, struct in_addr laddr, u_short lport) 2118 { 2119 #ifdef INET6 2120 /* XXX inp locking */ 2121 if ((inp->inp_vflag & INP_IPV4) == 0) 2122 return (false); 2123 #endif 2124 if (inp->inp_faddr.s_addr == faddr.s_addr && 2125 inp->inp_laddr.s_addr == laddr.s_addr && 2126 inp->inp_fport == fport && 2127 inp->inp_lport == lport) 2128 return (true); 2129 return (false); 2130 } 2131 2132 static struct inpcb * 2133 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2134 u_short fport, struct in_addr laddr, u_short lport) 2135 { 2136 struct inpcbhead *head; 2137 struct inpcb *inp; 2138 2139 INP_HASH_LOCK_ASSERT(pcbinfo); 2140 2141 head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport, 2142 pcbinfo->ipi_hashmask)]; 2143 CK_LIST_FOREACH(inp, head, inp_hash_exact) { 2144 if (in_pcblookup_exact_match(inp, faddr, fport, laddr, lport)) 2145 return (inp); 2146 } 2147 return (NULL); 2148 } 2149 2150 typedef enum { 2151 INPLOOKUP_MATCH_NONE = 0, 2152 INPLOOKUP_MATCH_WILD = 1, 2153 INPLOOKUP_MATCH_LADDR = 2, 2154 } inp_lookup_match_t; 2155 2156 static inp_lookup_match_t 2157 in_pcblookup_wild_match(const struct inpcb *inp, struct in_addr laddr, 2158 u_short lport) 2159 { 2160 #ifdef INET6 2161 /* XXX inp locking */ 2162 if ((inp->inp_vflag & INP_IPV4) == 0) 2163 return (INPLOOKUP_MATCH_NONE); 2164 #endif 2165 if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) 2166 return (INPLOOKUP_MATCH_NONE); 2167 if (inp->inp_laddr.s_addr == INADDR_ANY) 2168 return (INPLOOKUP_MATCH_WILD); 2169 if (inp->inp_laddr.s_addr == laddr.s_addr) 2170 return (INPLOOKUP_MATCH_LADDR); 2171 return (INPLOOKUP_MATCH_NONE); 2172 } 2173 2174 #define INP_LOOKUP_AGAIN ((struct inpcb *)(uintptr_t)-1) 2175 2176 static struct inpcb * 2177 in_pcblookup_hash_wild_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2178 u_short fport, struct in_addr laddr, u_short lport, 2179 const inp_lookup_t lockflags) 2180 { 2181 struct inpcbhead *head; 2182 struct inpcb *inp; 2183 2184 KASSERT(SMR_ENTERED(pcbinfo->ipi_smr), 2185 ("%s: not in SMR read section", __func__)); 2186 2187 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2188 pcbinfo->ipi_hashmask)]; 2189 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2190 inp_lookup_match_t match; 2191 2192 match = in_pcblookup_wild_match(inp, laddr, lport); 2193 if (match == INPLOOKUP_MATCH_NONE) 2194 continue; 2195 2196 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2197 match = in_pcblookup_wild_match(inp, laddr, lport); 2198 if (match != INPLOOKUP_MATCH_NONE && 2199 prison_check_ip4_locked(inp->inp_cred->cr_prison, 2200 &laddr) == 0) 2201 return (inp); 2202 inp_unlock(inp, lockflags); 2203 } 2204 2205 /* 2206 * The matching socket disappeared out from under us. Fall back 2207 * to a serialized lookup. 2208 */ 2209 return (INP_LOOKUP_AGAIN); 2210 } 2211 return (NULL); 2212 } 2213 2214 static struct inpcb * 2215 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2216 u_short fport, struct in_addr laddr, u_short lport) 2217 { 2218 struct inpcbhead *head; 2219 struct inpcb *inp, *local_wild, *local_exact, *jail_wild; 2220 #ifdef INET6 2221 struct inpcb *local_wild_mapped; 2222 #endif 2223 2224 INP_HASH_LOCK_ASSERT(pcbinfo); 2225 2226 /* 2227 * Order of socket selection - we always prefer jails. 2228 * 1. jailed, non-wild. 2229 * 2. jailed, wild. 2230 * 3. non-jailed, non-wild. 2231 * 4. non-jailed, wild. 2232 */ 2233 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2234 pcbinfo->ipi_hashmask)]; 2235 local_wild = local_exact = jail_wild = NULL; 2236 #ifdef INET6 2237 local_wild_mapped = NULL; 2238 #endif 2239 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2240 inp_lookup_match_t match; 2241 bool injail; 2242 2243 match = in_pcblookup_wild_match(inp, laddr, lport); 2244 if (match == INPLOOKUP_MATCH_NONE) 2245 continue; 2246 2247 injail = prison_flag(inp->inp_cred, PR_IP4) != 0; 2248 if (injail) { 2249 if (prison_check_ip4_locked(inp->inp_cred->cr_prison, 2250 &laddr) != 0) 2251 continue; 2252 } else { 2253 if (local_exact != NULL) 2254 continue; 2255 } 2256 2257 if (match == INPLOOKUP_MATCH_LADDR) { 2258 if (injail) 2259 return (inp); 2260 local_exact = inp; 2261 } else { 2262 #ifdef INET6 2263 /* XXX inp locking, NULL check */ 2264 if (inp->inp_vflag & INP_IPV6PROTO) 2265 local_wild_mapped = inp; 2266 else 2267 #endif 2268 if (injail) 2269 jail_wild = inp; 2270 else 2271 local_wild = inp; 2272 } 2273 } 2274 if (jail_wild != NULL) 2275 return (jail_wild); 2276 if (local_exact != NULL) 2277 return (local_exact); 2278 if (local_wild != NULL) 2279 return (local_wild); 2280 #ifdef INET6 2281 if (local_wild_mapped != NULL) 2282 return (local_wild_mapped); 2283 #endif 2284 return (NULL); 2285 } 2286 2287 /* 2288 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2289 * that the caller has either locked the hash list, which usually happens 2290 * for bind(2) operations, or is in SMR section, which happens when sorting 2291 * out incoming packets. 2292 */ 2293 static struct inpcb * 2294 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2295 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2296 uint8_t numa_domain) 2297 { 2298 struct inpcb *inp; 2299 const u_short fport = fport_arg, lport = lport_arg; 2300 2301 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0, 2302 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2303 KASSERT(faddr.s_addr != INADDR_ANY, 2304 ("%s: invalid foreign address", __func__)); 2305 KASSERT(laddr.s_addr != INADDR_ANY, 2306 ("%s: invalid local address", __func__)); 2307 INP_HASH_WLOCK_ASSERT(pcbinfo); 2308 2309 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2310 if (inp != NULL) 2311 return (inp); 2312 2313 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2314 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2315 &laddr, lport, numa_domain); 2316 if (inp == NULL) { 2317 inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr, 2318 fport, laddr, lport); 2319 } 2320 } 2321 2322 return (inp); 2323 } 2324 2325 static struct inpcb * 2326 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2327 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2328 uint8_t numa_domain) 2329 { 2330 struct inpcb *inp; 2331 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2332 2333 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2334 ("%s: LOCKPCB not set", __func__)); 2335 2336 INP_HASH_WLOCK(pcbinfo); 2337 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2338 lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain); 2339 if (inp != NULL && !inp_trylock(inp, lockflags)) { 2340 in_pcbref(inp); 2341 INP_HASH_WUNLOCK(pcbinfo); 2342 inp_lock(inp, lockflags); 2343 if (in_pcbrele(inp, lockflags)) 2344 /* XXX-MJ or retry until we get a negative match? */ 2345 inp = NULL; 2346 } else { 2347 INP_HASH_WUNLOCK(pcbinfo); 2348 } 2349 return (inp); 2350 } 2351 2352 static struct inpcb * 2353 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2354 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2355 uint8_t numa_domain) 2356 { 2357 struct inpcb *inp; 2358 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2359 const u_short fport = fport_arg, lport = lport_arg; 2360 2361 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2362 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2363 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2364 ("%s: LOCKPCB not set", __func__)); 2365 2366 smr_enter(pcbinfo->ipi_smr); 2367 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2368 if (inp != NULL) { 2369 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2370 /* 2371 * Revalidate the 4-tuple, the socket could have been 2372 * disconnected. 2373 */ 2374 if (__predict_true(in_pcblookup_exact_match(inp, 2375 faddr, fport, laddr, lport))) 2376 return (inp); 2377 inp_unlock(inp, lockflags); 2378 } 2379 2380 /* 2381 * We failed to lock the inpcb, or its connection state changed 2382 * out from under us. Fall back to a precise search. 2383 */ 2384 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2385 lookupflags, numa_domain)); 2386 } 2387 2388 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2389 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2390 &laddr, lport, numa_domain); 2391 if (inp != NULL) { 2392 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2393 if (__predict_true(in_pcblookup_wild_match(inp, 2394 laddr, lport) != INPLOOKUP_MATCH_NONE)) 2395 return (inp); 2396 inp_unlock(inp, lockflags); 2397 } 2398 inp = INP_LOOKUP_AGAIN; 2399 } else { 2400 inp = in_pcblookup_hash_wild_smr(pcbinfo, faddr, fport, 2401 laddr, lport, lockflags); 2402 } 2403 if (inp == INP_LOOKUP_AGAIN) { 2404 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, 2405 lport, lookupflags, numa_domain)); 2406 } 2407 } 2408 2409 if (inp == NULL) 2410 smr_exit(pcbinfo->ipi_smr); 2411 2412 return (inp); 2413 } 2414 2415 /* 2416 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2417 * from which a pre-calculated hash value may be extracted. 2418 */ 2419 struct inpcb * 2420 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2421 struct in_addr laddr, u_int lport, int lookupflags, 2422 struct ifnet *ifp __unused) 2423 { 2424 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2425 lookupflags, M_NODOM)); 2426 } 2427 2428 struct inpcb * 2429 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2430 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2431 struct ifnet *ifp __unused, struct mbuf *m) 2432 { 2433 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2434 lookupflags, m->m_pkthdr.numa_domain)); 2435 } 2436 #endif /* INET */ 2437 2438 static bool 2439 in_pcbjailed(const struct inpcb *inp, unsigned int flag) 2440 { 2441 return (prison_flag(inp->inp_cred, flag) != 0); 2442 } 2443 2444 /* 2445 * Insert the PCB into a hash chain using ordering rules which ensure that 2446 * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first. 2447 * 2448 * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs 2449 * with exact local addresses ahead of wildcard PCBs. Unbound v4-mapped v6 PCBs 2450 * always appear last no matter whether they are jailed. 2451 */ 2452 static void 2453 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2454 { 2455 struct inpcb *last; 2456 bool bound, injail; 2457 2458 INP_LOCK_ASSERT(inp); 2459 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2460 2461 last = NULL; 2462 bound = inp->inp_laddr.s_addr != INADDR_ANY; 2463 if (!bound && (inp->inp_vflag & INP_IPV6PROTO) != 0) { 2464 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2465 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2466 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2467 return; 2468 } 2469 } 2470 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2471 return; 2472 } 2473 2474 injail = in_pcbjailed(inp, PR_IP4); 2475 if (!injail) { 2476 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2477 if (!in_pcbjailed(last, PR_IP4)) 2478 break; 2479 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2480 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2481 return; 2482 } 2483 } 2484 } else if (!CK_LIST_EMPTY(pcbhash) && 2485 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) { 2486 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2487 return; 2488 } 2489 if (!bound) { 2490 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2491 if (last->inp_laddr.s_addr == INADDR_ANY) 2492 break; 2493 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2494 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2495 return; 2496 } 2497 } 2498 } 2499 if (last == NULL) 2500 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2501 else 2502 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2503 } 2504 2505 #ifdef INET6 2506 /* 2507 * See the comment above _in_pcbinshash_wild(). 2508 */ 2509 static void 2510 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2511 { 2512 struct inpcb *last; 2513 bool bound, injail; 2514 2515 INP_LOCK_ASSERT(inp); 2516 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2517 2518 last = NULL; 2519 bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr); 2520 injail = in_pcbjailed(inp, PR_IP6); 2521 if (!injail) { 2522 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2523 if (!in_pcbjailed(last, PR_IP6)) 2524 break; 2525 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2526 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2527 return; 2528 } 2529 } 2530 } else if (!CK_LIST_EMPTY(pcbhash) && 2531 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) { 2532 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2533 return; 2534 } 2535 if (!bound) { 2536 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2537 if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr)) 2538 break; 2539 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2540 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2541 return; 2542 } 2543 } 2544 } 2545 if (last == NULL) 2546 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2547 else 2548 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2549 } 2550 #endif 2551 2552 /* 2553 * Insert PCB onto various hash lists. 2554 */ 2555 int 2556 in_pcbinshash(struct inpcb *inp) 2557 { 2558 struct inpcbhead *pcbhash; 2559 struct inpcbporthead *pcbporthash; 2560 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2561 struct inpcbport *phd; 2562 uint32_t hash; 2563 bool connected; 2564 2565 INP_WLOCK_ASSERT(inp); 2566 INP_HASH_WLOCK_ASSERT(pcbinfo); 2567 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2568 ("in_pcbinshash: INP_INHASHLIST")); 2569 2570 #ifdef INET6 2571 if (inp->inp_vflag & INP_IPV6) { 2572 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2573 inp->inp_fport, pcbinfo->ipi_hashmask); 2574 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2575 } else 2576 #endif 2577 { 2578 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2579 inp->inp_fport, pcbinfo->ipi_hashmask); 2580 connected = !in_nullhost(inp->inp_faddr); 2581 } 2582 2583 if (connected) 2584 pcbhash = &pcbinfo->ipi_hash_exact[hash]; 2585 else 2586 pcbhash = &pcbinfo->ipi_hash_wild[hash]; 2587 2588 pcbporthash = &pcbinfo->ipi_porthashbase[ 2589 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2590 2591 /* 2592 * Add entry to load balance group. 2593 * Only do this if SO_REUSEPORT_LB is set. 2594 */ 2595 if ((inp->inp_socket->so_options & SO_REUSEPORT_LB) != 0) { 2596 int error = in_pcbinslbgrouphash(inp, M_NODOM); 2597 if (error != 0) 2598 return (error); 2599 } 2600 2601 /* 2602 * Go through port list and look for a head for this lport. 2603 */ 2604 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2605 if (phd->phd_port == inp->inp_lport) 2606 break; 2607 } 2608 2609 /* 2610 * If none exists, malloc one and tack it on. 2611 */ 2612 if (phd == NULL) { 2613 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT); 2614 if (phd == NULL) { 2615 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2616 in_pcbremlbgrouphash(inp); 2617 return (ENOMEM); 2618 } 2619 phd->phd_port = inp->inp_lport; 2620 CK_LIST_INIT(&phd->phd_pcblist); 2621 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2622 } 2623 inp->inp_phd = phd; 2624 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2625 2626 /* 2627 * The PCB may have been disconnected in the past. Before we can safely 2628 * make it visible in the hash table, we must wait for all readers which 2629 * may be traversing this PCB to finish. 2630 */ 2631 if (inp->inp_smr != SMR_SEQ_INVALID) { 2632 smr_wait(pcbinfo->ipi_smr, inp->inp_smr); 2633 inp->inp_smr = SMR_SEQ_INVALID; 2634 } 2635 2636 if (connected) 2637 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact); 2638 else { 2639 #ifdef INET6 2640 if ((inp->inp_vflag & INP_IPV6) != 0) 2641 _in6_pcbinshash_wild(pcbhash, inp); 2642 else 2643 #endif 2644 _in_pcbinshash_wild(pcbhash, inp); 2645 } 2646 inp->inp_flags |= INP_INHASHLIST; 2647 2648 return (0); 2649 } 2650 2651 void 2652 in_pcbremhash_locked(struct inpcb *inp) 2653 { 2654 struct inpcbport *phd = inp->inp_phd; 2655 2656 INP_WLOCK_ASSERT(inp); 2657 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2658 MPASS(inp->inp_flags & INP_INHASHLIST); 2659 2660 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2661 in_pcbremlbgrouphash(inp); 2662 #ifdef INET6 2663 if (inp->inp_vflag & INP_IPV6) { 2664 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) 2665 CK_LIST_REMOVE(inp, inp_hash_wild); 2666 else 2667 CK_LIST_REMOVE(inp, inp_hash_exact); 2668 } else 2669 #endif 2670 { 2671 if (in_nullhost(inp->inp_faddr)) 2672 CK_LIST_REMOVE(inp, inp_hash_wild); 2673 else 2674 CK_LIST_REMOVE(inp, inp_hash_exact); 2675 } 2676 CK_LIST_REMOVE(inp, inp_portlist); 2677 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2678 CK_LIST_REMOVE(phd, phd_hash); 2679 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd); 2680 } 2681 inp->inp_flags &= ~INP_INHASHLIST; 2682 } 2683 2684 static void 2685 in_pcbremhash(struct inpcb *inp) 2686 { 2687 INP_HASH_WLOCK(inp->inp_pcbinfo); 2688 in_pcbremhash_locked(inp); 2689 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 2690 } 2691 2692 /* 2693 * Move PCB to the proper hash bucket when { faddr, fport } have been 2694 * changed. NOTE: This does not handle the case of the lport changing (the 2695 * hashed port list would have to be updated as well), so the lport must 2696 * not change after in_pcbinshash() has been called. 2697 */ 2698 void 2699 in_pcbrehash(struct inpcb *inp) 2700 { 2701 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2702 struct inpcbhead *head; 2703 uint32_t hash; 2704 bool connected; 2705 2706 INP_WLOCK_ASSERT(inp); 2707 INP_HASH_WLOCK_ASSERT(pcbinfo); 2708 KASSERT(inp->inp_flags & INP_INHASHLIST, 2709 ("%s: !INP_INHASHLIST", __func__)); 2710 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 2711 ("%s: inp was disconnected", __func__)); 2712 2713 #ifdef INET6 2714 if (inp->inp_vflag & INP_IPV6) { 2715 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2716 inp->inp_fport, pcbinfo->ipi_hashmask); 2717 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2718 } else 2719 #endif 2720 { 2721 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2722 inp->inp_fport, pcbinfo->ipi_hashmask); 2723 connected = !in_nullhost(inp->inp_faddr); 2724 } 2725 2726 /* 2727 * When rehashing, the caller must ensure that either the new or the old 2728 * foreign address was unspecified. 2729 */ 2730 if (connected) 2731 CK_LIST_REMOVE(inp, inp_hash_wild); 2732 else 2733 CK_LIST_REMOVE(inp, inp_hash_exact); 2734 2735 if (connected) { 2736 head = &pcbinfo->ipi_hash_exact[hash]; 2737 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact); 2738 } else { 2739 head = &pcbinfo->ipi_hash_wild[hash]; 2740 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild); 2741 } 2742 } 2743 2744 /* 2745 * Check for alternatives when higher level complains 2746 * about service problems. For now, invalidate cached 2747 * routing information. If the route was created dynamically 2748 * (by a redirect), time to try a default gateway again. 2749 */ 2750 void 2751 in_losing(struct inpcb *inp) 2752 { 2753 2754 RO_INVALIDATE_CACHE(&inp->inp_route); 2755 return; 2756 } 2757 2758 /* 2759 * A set label operation has occurred at the socket layer, propagate the 2760 * label change into the in_pcb for the socket. 2761 */ 2762 void 2763 in_pcbsosetlabel(struct socket *so) 2764 { 2765 #ifdef MAC 2766 struct inpcb *inp; 2767 2768 inp = sotoinpcb(so); 2769 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2770 2771 INP_WLOCK(inp); 2772 SOCK_LOCK(so); 2773 mac_inpcb_sosetlabel(so, inp); 2774 SOCK_UNLOCK(so); 2775 INP_WUNLOCK(inp); 2776 #endif 2777 } 2778 2779 void 2780 inp_wlock(struct inpcb *inp) 2781 { 2782 2783 INP_WLOCK(inp); 2784 } 2785 2786 void 2787 inp_wunlock(struct inpcb *inp) 2788 { 2789 2790 INP_WUNLOCK(inp); 2791 } 2792 2793 void 2794 inp_rlock(struct inpcb *inp) 2795 { 2796 2797 INP_RLOCK(inp); 2798 } 2799 2800 void 2801 inp_runlock(struct inpcb *inp) 2802 { 2803 2804 INP_RUNLOCK(inp); 2805 } 2806 2807 #ifdef INVARIANT_SUPPORT 2808 void 2809 inp_lock_assert(struct inpcb *inp) 2810 { 2811 2812 INP_WLOCK_ASSERT(inp); 2813 } 2814 2815 void 2816 inp_unlock_assert(struct inpcb *inp) 2817 { 2818 2819 INP_UNLOCK_ASSERT(inp); 2820 } 2821 #endif 2822 2823 void 2824 inp_apply_all(struct inpcbinfo *pcbinfo, 2825 void (*func)(struct inpcb *, void *), void *arg) 2826 { 2827 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2828 INPLOOKUP_WLOCKPCB); 2829 struct inpcb *inp; 2830 2831 while ((inp = inp_next(&inpi)) != NULL) 2832 func(inp, arg); 2833 } 2834 2835 struct socket * 2836 inp_inpcbtosocket(struct inpcb *inp) 2837 { 2838 2839 INP_WLOCK_ASSERT(inp); 2840 return (inp->inp_socket); 2841 } 2842 2843 struct tcpcb * 2844 inp_inpcbtotcpcb(struct inpcb *inp) 2845 { 2846 2847 INP_WLOCK_ASSERT(inp); 2848 return ((struct tcpcb *)inp->inp_ppcb); 2849 } 2850 2851 int 2852 inp_ip_tos_get(const struct inpcb *inp) 2853 { 2854 2855 return (inp->inp_ip_tos); 2856 } 2857 2858 void 2859 inp_ip_tos_set(struct inpcb *inp, int val) 2860 { 2861 2862 inp->inp_ip_tos = val; 2863 } 2864 2865 void 2866 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2867 uint32_t *faddr, uint16_t *fp) 2868 { 2869 2870 INP_LOCK_ASSERT(inp); 2871 *laddr = inp->inp_laddr.s_addr; 2872 *faddr = inp->inp_faddr.s_addr; 2873 *lp = inp->inp_lport; 2874 *fp = inp->inp_fport; 2875 } 2876 2877 struct inpcb * 2878 so_sotoinpcb(struct socket *so) 2879 { 2880 2881 return (sotoinpcb(so)); 2882 } 2883 2884 /* 2885 * Create an external-format (``xinpcb'') structure using the information in 2886 * the kernel-format in_pcb structure pointed to by inp. This is done to 2887 * reduce the spew of irrelevant information over this interface, to isolate 2888 * user code from changes in the kernel structure, and potentially to provide 2889 * information-hiding if we decide that some of this information should be 2890 * hidden from users. 2891 */ 2892 void 2893 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2894 { 2895 2896 bzero(xi, sizeof(*xi)); 2897 xi->xi_len = sizeof(struct xinpcb); 2898 if (inp->inp_socket) 2899 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2900 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2901 xi->inp_gencnt = inp->inp_gencnt; 2902 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2903 xi->inp_flow = inp->inp_flow; 2904 xi->inp_flowid = inp->inp_flowid; 2905 xi->inp_flowtype = inp->inp_flowtype; 2906 xi->inp_flags = inp->inp_flags; 2907 xi->inp_flags2 = inp->inp_flags2; 2908 xi->in6p_cksum = inp->in6p_cksum; 2909 xi->in6p_hops = inp->in6p_hops; 2910 xi->inp_ip_tos = inp->inp_ip_tos; 2911 xi->inp_vflag = inp->inp_vflag; 2912 xi->inp_ip_ttl = inp->inp_ip_ttl; 2913 xi->inp_ip_p = inp->inp_ip_p; 2914 xi->inp_ip_minttl = inp->inp_ip_minttl; 2915 } 2916 2917 int 2918 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo, 2919 int (*ctloutput_set)(struct inpcb *, struct sockopt *)) 2920 { 2921 struct sockopt sopt; 2922 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2923 INPLOOKUP_WLOCKPCB); 2924 struct inpcb *inp; 2925 struct sockopt_parameters *params; 2926 struct socket *so; 2927 int error; 2928 char buf[1024]; 2929 2930 if (req->oldptr != NULL || req->oldlen != 0) 2931 return (EINVAL); 2932 if (req->newptr == NULL) 2933 return (EPERM); 2934 if (req->newlen > sizeof(buf)) 2935 return (ENOMEM); 2936 error = SYSCTL_IN(req, buf, req->newlen); 2937 if (error != 0) 2938 return (error); 2939 if (req->newlen < sizeof(struct sockopt_parameters)) 2940 return (EINVAL); 2941 params = (struct sockopt_parameters *)buf; 2942 sopt.sopt_level = params->sop_level; 2943 sopt.sopt_name = params->sop_optname; 2944 sopt.sopt_dir = SOPT_SET; 2945 sopt.sopt_val = params->sop_optval; 2946 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters); 2947 sopt.sopt_td = NULL; 2948 #ifdef INET6 2949 if (params->sop_inc.inc_flags & INC_ISIPV6) { 2950 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr)) 2951 params->sop_inc.inc6_laddr.s6_addr16[1] = 2952 htons(params->sop_inc.inc6_zoneid & 0xffff); 2953 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr)) 2954 params->sop_inc.inc6_faddr.s6_addr16[1] = 2955 htons(params->sop_inc.inc6_zoneid & 0xffff); 2956 } 2957 #endif 2958 if (params->sop_inc.inc_lport != htons(0)) { 2959 if (params->sop_inc.inc_fport == htons(0)) 2960 inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport, 2961 pcbinfo->ipi_hashmask); 2962 else 2963 #ifdef INET6 2964 if (params->sop_inc.inc_flags & INC_ISIPV6) 2965 inpi.hash = INP6_PCBHASH( 2966 ¶ms->sop_inc.inc6_faddr, 2967 params->sop_inc.inc_lport, 2968 params->sop_inc.inc_fport, 2969 pcbinfo->ipi_hashmask); 2970 else 2971 #endif 2972 inpi.hash = INP_PCBHASH( 2973 ¶ms->sop_inc.inc_faddr, 2974 params->sop_inc.inc_lport, 2975 params->sop_inc.inc_fport, 2976 pcbinfo->ipi_hashmask); 2977 } 2978 while ((inp = inp_next(&inpi)) != NULL) 2979 if (inp->inp_gencnt == params->sop_id) { 2980 if (inp->inp_flags & INP_DROPPED) { 2981 INP_WUNLOCK(inp); 2982 return (ECONNRESET); 2983 } 2984 so = inp->inp_socket; 2985 KASSERT(so != NULL, ("inp_socket == NULL")); 2986 soref(so); 2987 if (params->sop_level == SOL_SOCKET) { 2988 INP_WUNLOCK(inp); 2989 error = sosetopt(so, &sopt); 2990 } else 2991 error = (*ctloutput_set)(inp, &sopt); 2992 sorele(so); 2993 break; 2994 } 2995 if (inp == NULL) 2996 error = ESRCH; 2997 return (error); 2998 } 2999 3000 #ifdef DDB 3001 static void 3002 db_print_indent(int indent) 3003 { 3004 int i; 3005 3006 for (i = 0; i < indent; i++) 3007 db_printf(" "); 3008 } 3009 3010 static void 3011 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 3012 { 3013 char faddr_str[48], laddr_str[48]; 3014 3015 db_print_indent(indent); 3016 db_printf("%s at %p\n", name, inc); 3017 3018 indent += 2; 3019 3020 #ifdef INET6 3021 if (inc->inc_flags & INC_ISIPV6) { 3022 /* IPv6. */ 3023 ip6_sprintf(laddr_str, &inc->inc6_laddr); 3024 ip6_sprintf(faddr_str, &inc->inc6_faddr); 3025 } else 3026 #endif 3027 { 3028 /* IPv4. */ 3029 inet_ntoa_r(inc->inc_laddr, laddr_str); 3030 inet_ntoa_r(inc->inc_faddr, faddr_str); 3031 } 3032 db_print_indent(indent); 3033 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3034 ntohs(inc->inc_lport)); 3035 db_print_indent(indent); 3036 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3037 ntohs(inc->inc_fport)); 3038 } 3039 3040 static void 3041 db_print_inpflags(int inp_flags) 3042 { 3043 int comma; 3044 3045 comma = 0; 3046 if (inp_flags & INP_RECVOPTS) { 3047 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3048 comma = 1; 3049 } 3050 if (inp_flags & INP_RECVRETOPTS) { 3051 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3052 comma = 1; 3053 } 3054 if (inp_flags & INP_RECVDSTADDR) { 3055 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3056 comma = 1; 3057 } 3058 if (inp_flags & INP_ORIGDSTADDR) { 3059 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3060 comma = 1; 3061 } 3062 if (inp_flags & INP_HDRINCL) { 3063 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3064 comma = 1; 3065 } 3066 if (inp_flags & INP_HIGHPORT) { 3067 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3068 comma = 1; 3069 } 3070 if (inp_flags & INP_LOWPORT) { 3071 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3072 comma = 1; 3073 } 3074 if (inp_flags & INP_ANONPORT) { 3075 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3076 comma = 1; 3077 } 3078 if (inp_flags & INP_RECVIF) { 3079 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3080 comma = 1; 3081 } 3082 if (inp_flags & INP_MTUDISC) { 3083 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3084 comma = 1; 3085 } 3086 if (inp_flags & INP_RECVTTL) { 3087 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3088 comma = 1; 3089 } 3090 if (inp_flags & INP_DONTFRAG) { 3091 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3092 comma = 1; 3093 } 3094 if (inp_flags & INP_RECVTOS) { 3095 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3096 comma = 1; 3097 } 3098 if (inp_flags & IN6P_IPV6_V6ONLY) { 3099 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3100 comma = 1; 3101 } 3102 if (inp_flags & IN6P_PKTINFO) { 3103 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3104 comma = 1; 3105 } 3106 if (inp_flags & IN6P_HOPLIMIT) { 3107 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3108 comma = 1; 3109 } 3110 if (inp_flags & IN6P_HOPOPTS) { 3111 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3112 comma = 1; 3113 } 3114 if (inp_flags & IN6P_DSTOPTS) { 3115 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3116 comma = 1; 3117 } 3118 if (inp_flags & IN6P_RTHDR) { 3119 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3120 comma = 1; 3121 } 3122 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3123 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3124 comma = 1; 3125 } 3126 if (inp_flags & IN6P_TCLASS) { 3127 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3128 comma = 1; 3129 } 3130 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3131 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3132 comma = 1; 3133 } 3134 if (inp_flags & INP_ONESBCAST) { 3135 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3136 comma = 1; 3137 } 3138 if (inp_flags & INP_DROPPED) { 3139 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3140 comma = 1; 3141 } 3142 if (inp_flags & INP_SOCKREF) { 3143 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3144 comma = 1; 3145 } 3146 if (inp_flags & IN6P_RFC2292) { 3147 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3148 comma = 1; 3149 } 3150 if (inp_flags & IN6P_MTU) { 3151 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3152 comma = 1; 3153 } 3154 } 3155 3156 static void 3157 db_print_inpvflag(u_char inp_vflag) 3158 { 3159 int comma; 3160 3161 comma = 0; 3162 if (inp_vflag & INP_IPV4) { 3163 db_printf("%sINP_IPV4", comma ? ", " : ""); 3164 comma = 1; 3165 } 3166 if (inp_vflag & INP_IPV6) { 3167 db_printf("%sINP_IPV6", comma ? ", " : ""); 3168 comma = 1; 3169 } 3170 if (inp_vflag & INP_IPV6PROTO) { 3171 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3172 comma = 1; 3173 } 3174 } 3175 3176 static void 3177 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3178 { 3179 3180 db_print_indent(indent); 3181 db_printf("%s at %p\n", name, inp); 3182 3183 indent += 2; 3184 3185 db_print_indent(indent); 3186 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3187 3188 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3189 3190 db_print_indent(indent); 3191 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3192 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3193 3194 db_print_indent(indent); 3195 db_printf("inp_label: %p inp_flags: 0x%x (", 3196 inp->inp_label, inp->inp_flags); 3197 db_print_inpflags(inp->inp_flags); 3198 db_printf(")\n"); 3199 3200 db_print_indent(indent); 3201 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3202 inp->inp_vflag); 3203 db_print_inpvflag(inp->inp_vflag); 3204 db_printf(")\n"); 3205 3206 db_print_indent(indent); 3207 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3208 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3209 3210 db_print_indent(indent); 3211 #ifdef INET6 3212 if (inp->inp_vflag & INP_IPV6) { 3213 db_printf("in6p_options: %p in6p_outputopts: %p " 3214 "in6p_moptions: %p\n", inp->in6p_options, 3215 inp->in6p_outputopts, inp->in6p_moptions); 3216 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3217 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3218 inp->in6p_hops); 3219 } else 3220 #endif 3221 { 3222 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3223 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3224 inp->inp_options, inp->inp_moptions); 3225 } 3226 3227 db_print_indent(indent); 3228 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3229 (uintmax_t)inp->inp_gencnt); 3230 } 3231 3232 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3233 { 3234 struct inpcb *inp; 3235 3236 if (!have_addr) { 3237 db_printf("usage: show inpcb <addr>\n"); 3238 return; 3239 } 3240 inp = (struct inpcb *)addr; 3241 3242 db_print_inpcb(inp, "inpcb", 0); 3243 } 3244 #endif /* DDB */ 3245 3246 #ifdef RATELIMIT 3247 /* 3248 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3249 * if any. 3250 */ 3251 int 3252 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3253 { 3254 union if_snd_tag_modify_params params = { 3255 .rate_limit.max_rate = max_pacing_rate, 3256 .rate_limit.flags = M_NOWAIT, 3257 }; 3258 struct m_snd_tag *mst; 3259 int error; 3260 3261 mst = inp->inp_snd_tag; 3262 if (mst == NULL) 3263 return (EINVAL); 3264 3265 if (mst->sw->snd_tag_modify == NULL) { 3266 error = EOPNOTSUPP; 3267 } else { 3268 error = mst->sw->snd_tag_modify(mst, ¶ms); 3269 } 3270 return (error); 3271 } 3272 3273 /* 3274 * Query existing TX rate limit based on the existing 3275 * "inp->inp_snd_tag", if any. 3276 */ 3277 int 3278 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3279 { 3280 union if_snd_tag_query_params params = { }; 3281 struct m_snd_tag *mst; 3282 int error; 3283 3284 mst = inp->inp_snd_tag; 3285 if (mst == NULL) 3286 return (EINVAL); 3287 3288 if (mst->sw->snd_tag_query == NULL) { 3289 error = EOPNOTSUPP; 3290 } else { 3291 error = mst->sw->snd_tag_query(mst, ¶ms); 3292 if (error == 0 && p_max_pacing_rate != NULL) 3293 *p_max_pacing_rate = params.rate_limit.max_rate; 3294 } 3295 return (error); 3296 } 3297 3298 /* 3299 * Query existing TX queue level based on the existing 3300 * "inp->inp_snd_tag", if any. 3301 */ 3302 int 3303 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3304 { 3305 union if_snd_tag_query_params params = { }; 3306 struct m_snd_tag *mst; 3307 int error; 3308 3309 mst = inp->inp_snd_tag; 3310 if (mst == NULL) 3311 return (EINVAL); 3312 3313 if (mst->sw->snd_tag_query == NULL) 3314 return (EOPNOTSUPP); 3315 3316 error = mst->sw->snd_tag_query(mst, ¶ms); 3317 if (error == 0 && p_txqueue_level != NULL) 3318 *p_txqueue_level = params.rate_limit.queue_level; 3319 return (error); 3320 } 3321 3322 /* 3323 * Allocate a new TX rate limit send tag from the network interface 3324 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3325 */ 3326 int 3327 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3328 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3329 3330 { 3331 union if_snd_tag_alloc_params params = { 3332 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3333 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3334 .rate_limit.hdr.flowid = flowid, 3335 .rate_limit.hdr.flowtype = flowtype, 3336 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3337 .rate_limit.max_rate = max_pacing_rate, 3338 .rate_limit.flags = M_NOWAIT, 3339 }; 3340 int error; 3341 3342 INP_WLOCK_ASSERT(inp); 3343 3344 /* 3345 * If there is already a send tag, or the INP is being torn 3346 * down, allocating a new send tag is not allowed. Else send 3347 * tags may leak. 3348 */ 3349 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0) 3350 return (EINVAL); 3351 3352 error = m_snd_tag_alloc(ifp, ¶ms, st); 3353 #ifdef INET 3354 if (error == 0) { 3355 counter_u64_add(rate_limit_set_ok, 1); 3356 counter_u64_add(rate_limit_active, 1); 3357 } else if (error != EOPNOTSUPP) 3358 counter_u64_add(rate_limit_alloc_fail, 1); 3359 #endif 3360 return (error); 3361 } 3362 3363 void 3364 in_pcbdetach_tag(struct m_snd_tag *mst) 3365 { 3366 3367 m_snd_tag_rele(mst); 3368 #ifdef INET 3369 counter_u64_add(rate_limit_active, -1); 3370 #endif 3371 } 3372 3373 /* 3374 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3375 * if any: 3376 */ 3377 void 3378 in_pcbdetach_txrtlmt(struct inpcb *inp) 3379 { 3380 struct m_snd_tag *mst; 3381 3382 INP_WLOCK_ASSERT(inp); 3383 3384 mst = inp->inp_snd_tag; 3385 inp->inp_snd_tag = NULL; 3386 3387 if (mst == NULL) 3388 return; 3389 3390 m_snd_tag_rele(mst); 3391 #ifdef INET 3392 counter_u64_add(rate_limit_active, -1); 3393 #endif 3394 } 3395 3396 int 3397 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3398 { 3399 int error; 3400 3401 /* 3402 * If the existing send tag is for the wrong interface due to 3403 * a route change, first drop the existing tag. Set the 3404 * CHANGED flag so that we will keep trying to allocate a new 3405 * tag if we fail to allocate one this time. 3406 */ 3407 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3408 in_pcbdetach_txrtlmt(inp); 3409 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3410 } 3411 3412 /* 3413 * NOTE: When attaching to a network interface a reference is 3414 * made to ensure the network interface doesn't go away until 3415 * all ratelimit connections are gone. The network interface 3416 * pointers compared below represent valid network interfaces, 3417 * except when comparing towards NULL. 3418 */ 3419 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3420 error = 0; 3421 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3422 if (inp->inp_snd_tag != NULL) 3423 in_pcbdetach_txrtlmt(inp); 3424 error = 0; 3425 } else if (inp->inp_snd_tag == NULL) { 3426 /* 3427 * In order to utilize packet pacing with RSS, we need 3428 * to wait until there is a valid RSS hash before we 3429 * can proceed: 3430 */ 3431 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3432 error = EAGAIN; 3433 } else { 3434 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3435 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3436 } 3437 } else { 3438 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3439 } 3440 if (error == 0 || error == EOPNOTSUPP) 3441 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3442 3443 return (error); 3444 } 3445 3446 /* 3447 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3448 * is set in the fast path and will attach/detach/modify the TX rate 3449 * limit send tag based on the socket's so_max_pacing_rate value. 3450 */ 3451 void 3452 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3453 { 3454 struct socket *socket; 3455 uint32_t max_pacing_rate; 3456 bool did_upgrade; 3457 3458 if (inp == NULL) 3459 return; 3460 3461 socket = inp->inp_socket; 3462 if (socket == NULL) 3463 return; 3464 3465 if (!INP_WLOCKED(inp)) { 3466 /* 3467 * NOTE: If the write locking fails, we need to bail 3468 * out and use the non-ratelimited ring for the 3469 * transmit until there is a new chance to get the 3470 * write lock. 3471 */ 3472 if (!INP_TRY_UPGRADE(inp)) 3473 return; 3474 did_upgrade = 1; 3475 } else { 3476 did_upgrade = 0; 3477 } 3478 3479 /* 3480 * NOTE: The so_max_pacing_rate value is read unlocked, 3481 * because atomic updates are not required since the variable 3482 * is checked at every mbuf we send. It is assumed that the 3483 * variable read itself will be atomic. 3484 */ 3485 max_pacing_rate = socket->so_max_pacing_rate; 3486 3487 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3488 3489 if (did_upgrade) 3490 INP_DOWNGRADE(inp); 3491 } 3492 3493 /* 3494 * Track route changes for TX rate limiting. 3495 */ 3496 void 3497 in_pcboutput_eagain(struct inpcb *inp) 3498 { 3499 bool did_upgrade; 3500 3501 if (inp == NULL) 3502 return; 3503 3504 if (inp->inp_snd_tag == NULL) 3505 return; 3506 3507 if (!INP_WLOCKED(inp)) { 3508 /* 3509 * NOTE: If the write locking fails, we need to bail 3510 * out and use the non-ratelimited ring for the 3511 * transmit until there is a new chance to get the 3512 * write lock. 3513 */ 3514 if (!INP_TRY_UPGRADE(inp)) 3515 return; 3516 did_upgrade = 1; 3517 } else { 3518 did_upgrade = 0; 3519 } 3520 3521 /* detach rate limiting */ 3522 in_pcbdetach_txrtlmt(inp); 3523 3524 /* make sure new mbuf send tag allocation is made */ 3525 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3526 3527 if (did_upgrade) 3528 INP_DOWNGRADE(inp); 3529 } 3530 3531 #ifdef INET 3532 static void 3533 rl_init(void *st) 3534 { 3535 rate_limit_new = counter_u64_alloc(M_WAITOK); 3536 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3537 rate_limit_active = counter_u64_alloc(M_WAITOK); 3538 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3539 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3540 } 3541 3542 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3543 #endif 3544 #endif /* RATELIMIT */ 3545