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 #ifdef INET 238 VNET_DEFINE_STATIC(int, connect_inaddr_wild) = 1; 239 #define V_connect_inaddr_wild VNET(connect_inaddr_wild) 240 SYSCTL_INT(_net_inet_ip, OID_AUTO, connect_inaddr_wild, 241 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(connect_inaddr_wild), 0, 242 "Allow connecting to INADDR_ANY or INADDR_BROADCAST for connect(2)"); 243 #endif 244 245 static void in_pcbremhash(struct inpcb *); 246 247 /* 248 * in_pcb.c: manage the Protocol Control Blocks. 249 * 250 * NOTE: It is assumed that most of these functions will be called with 251 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 252 * functions often modify hash chains or addresses in pcbs. 253 */ 254 255 static struct inpcblbgroup * 256 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred, 257 u_char vflag, uint16_t port, const union in_dependaddr *addr, int size, 258 uint8_t numa_domain) 259 { 260 struct inpcblbgroup *grp; 261 size_t bytes; 262 263 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 264 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 265 if (grp == NULL) 266 return (NULL); 267 grp->il_cred = crhold(cred); 268 grp->il_vflag = vflag; 269 grp->il_lport = port; 270 grp->il_numa_domain = numa_domain; 271 grp->il_dependladdr = *addr; 272 grp->il_inpsiz = size; 273 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 274 return (grp); 275 } 276 277 static void 278 in_pcblbgroup_free_deferred(epoch_context_t ctx) 279 { 280 struct inpcblbgroup *grp; 281 282 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 283 crfree(grp->il_cred); 284 free(grp, M_PCB); 285 } 286 287 static void 288 in_pcblbgroup_free(struct inpcblbgroup *grp) 289 { 290 291 CK_LIST_REMOVE(grp, il_list); 292 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); 293 } 294 295 static struct inpcblbgroup * 296 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 297 struct inpcblbgroup *old_grp, int size) 298 { 299 struct inpcblbgroup *grp; 300 int i; 301 302 grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag, 303 old_grp->il_lport, &old_grp->il_dependladdr, size, 304 old_grp->il_numa_domain); 305 if (grp == NULL) 306 return (NULL); 307 308 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 309 ("invalid new local group size %d and old local group count %d", 310 grp->il_inpsiz, old_grp->il_inpcnt)); 311 312 for (i = 0; i < old_grp->il_inpcnt; ++i) 313 grp->il_inp[i] = old_grp->il_inp[i]; 314 grp->il_inpcnt = old_grp->il_inpcnt; 315 in_pcblbgroup_free(old_grp); 316 return (grp); 317 } 318 319 /* 320 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 321 * and shrink group if possible. 322 */ 323 static void 324 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 325 int i) 326 { 327 struct inpcblbgroup *grp, *new_grp; 328 329 grp = *grpp; 330 for (; i + 1 < grp->il_inpcnt; ++i) 331 grp->il_inp[i] = grp->il_inp[i + 1]; 332 grp->il_inpcnt--; 333 334 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 335 grp->il_inpcnt <= grp->il_inpsiz / 4) { 336 /* Shrink this group. */ 337 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 338 if (new_grp != NULL) 339 *grpp = new_grp; 340 } 341 } 342 343 /* 344 * Add PCB to load balance group for SO_REUSEPORT_LB option. 345 */ 346 static int 347 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain) 348 { 349 const static struct timeval interval = { 60, 0 }; 350 static struct timeval lastprint; 351 struct inpcbinfo *pcbinfo; 352 struct inpcblbgrouphead *hdr; 353 struct inpcblbgroup *grp; 354 uint32_t idx; 355 356 pcbinfo = inp->inp_pcbinfo; 357 358 INP_WLOCK_ASSERT(inp); 359 INP_HASH_WLOCK_ASSERT(pcbinfo); 360 361 #ifdef INET6 362 /* 363 * Don't allow IPv4 mapped INET6 wild socket. 364 */ 365 if ((inp->inp_vflag & INP_IPV4) && 366 inp->inp_laddr.s_addr == INADDR_ANY && 367 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 368 return (0); 369 } 370 #endif 371 372 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 373 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 374 CK_LIST_FOREACH(grp, hdr, il_list) { 375 if (grp->il_cred->cr_prison == inp->inp_cred->cr_prison && 376 grp->il_vflag == inp->inp_vflag && 377 grp->il_lport == inp->inp_lport && 378 grp->il_numa_domain == numa_domain && 379 memcmp(&grp->il_dependladdr, 380 &inp->inp_inc.inc_ie.ie_dependladdr, 381 sizeof(grp->il_dependladdr)) == 0) { 382 break; 383 } 384 } 385 if (grp == NULL) { 386 /* Create new load balance group. */ 387 grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag, 388 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 389 INPCBLBGROUP_SIZMIN, numa_domain); 390 if (grp == NULL) 391 return (ENOBUFS); 392 } else if (grp->il_inpcnt == grp->il_inpsiz) { 393 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 394 if (ratecheck(&lastprint, &interval)) 395 printf("lb group port %d, limit reached\n", 396 ntohs(grp->il_lport)); 397 return (0); 398 } 399 400 /* Expand this local group. */ 401 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 402 if (grp == NULL) 403 return (ENOBUFS); 404 } 405 406 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 407 ("invalid local group size %d and count %d", grp->il_inpsiz, 408 grp->il_inpcnt)); 409 410 grp->il_inp[grp->il_inpcnt] = inp; 411 grp->il_inpcnt++; 412 inp->inp_flags |= INP_INLBGROUP; 413 return (0); 414 } 415 416 /* 417 * Remove PCB from load balance group. 418 */ 419 static void 420 in_pcbremlbgrouphash(struct inpcb *inp) 421 { 422 struct inpcbinfo *pcbinfo; 423 struct inpcblbgrouphead *hdr; 424 struct inpcblbgroup *grp; 425 int i; 426 427 pcbinfo = inp->inp_pcbinfo; 428 429 INP_WLOCK_ASSERT(inp); 430 MPASS(inp->inp_flags & INP_INLBGROUP); 431 INP_HASH_WLOCK_ASSERT(pcbinfo); 432 433 hdr = &pcbinfo->ipi_lbgrouphashbase[ 434 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 435 CK_LIST_FOREACH(grp, hdr, il_list) { 436 for (i = 0; i < grp->il_inpcnt; ++i) { 437 if (grp->il_inp[i] != inp) 438 continue; 439 440 if (grp->il_inpcnt == 1) { 441 /* We are the last, free this local group. */ 442 in_pcblbgroup_free(grp); 443 } else { 444 /* Pull up inpcbs, shrink group if possible. */ 445 in_pcblbgroup_reorder(hdr, &grp, i); 446 } 447 inp->inp_flags &= ~INP_INLBGROUP; 448 return; 449 } 450 } 451 KASSERT(0, ("%s: did not find %p", __func__, inp)); 452 } 453 454 int 455 in_pcblbgroup_numa(struct inpcb *inp, int arg) 456 { 457 struct inpcbinfo *pcbinfo; 458 struct inpcblbgrouphead *hdr; 459 struct inpcblbgroup *grp; 460 int err, i; 461 uint8_t numa_domain; 462 463 switch (arg) { 464 case TCP_REUSPORT_LB_NUMA_NODOM: 465 numa_domain = M_NODOM; 466 break; 467 case TCP_REUSPORT_LB_NUMA_CURDOM: 468 numa_domain = PCPU_GET(domain); 469 break; 470 default: 471 if (arg < 0 || arg >= vm_ndomains) 472 return (EINVAL); 473 numa_domain = arg; 474 } 475 476 err = 0; 477 pcbinfo = inp->inp_pcbinfo; 478 INP_WLOCK_ASSERT(inp); 479 INP_HASH_WLOCK(pcbinfo); 480 hdr = &pcbinfo->ipi_lbgrouphashbase[ 481 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 482 CK_LIST_FOREACH(grp, hdr, il_list) { 483 for (i = 0; i < grp->il_inpcnt; ++i) { 484 if (grp->il_inp[i] != inp) 485 continue; 486 487 if (grp->il_numa_domain == numa_domain) { 488 goto abort_with_hash_wlock; 489 } 490 491 /* Remove it from the old group. */ 492 in_pcbremlbgrouphash(inp); 493 494 /* Add it to the new group based on numa domain. */ 495 in_pcbinslbgrouphash(inp, numa_domain); 496 goto abort_with_hash_wlock; 497 } 498 } 499 err = ENOENT; 500 abort_with_hash_wlock: 501 INP_HASH_WUNLOCK(pcbinfo); 502 return (err); 503 } 504 505 /* Make sure it is safe to use hashinit(9) on CK_LIST. */ 506 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb))); 507 508 /* 509 * Initialize an inpcbinfo - a per-VNET instance of connections db. 510 */ 511 void 512 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor, 513 u_int hash_nelements, u_int porthash_nelements) 514 { 515 516 mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF); 517 mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name, 518 NULL, MTX_DEF); 519 #ifdef VIMAGE 520 pcbinfo->ipi_vnet = curvnet; 521 #endif 522 CK_LIST_INIT(&pcbinfo->ipi_listhead); 523 pcbinfo->ipi_count = 0; 524 pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB, 525 &pcbinfo->ipi_hashmask); 526 pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB, 527 &pcbinfo->ipi_hashmask); 528 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 529 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 530 &pcbinfo->ipi_porthashmask); 531 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 532 &pcbinfo->ipi_lbgrouphashmask); 533 pcbinfo->ipi_zone = pcbstor->ips_zone; 534 pcbinfo->ipi_portzone = pcbstor->ips_portzone; 535 pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone); 536 } 537 538 /* 539 * Destroy an inpcbinfo. 540 */ 541 void 542 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 543 { 544 545 KASSERT(pcbinfo->ipi_count == 0, 546 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 547 548 hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask); 549 hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask); 550 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 551 pcbinfo->ipi_porthashmask); 552 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 553 pcbinfo->ipi_lbgrouphashmask); 554 mtx_destroy(&pcbinfo->ipi_hash_lock); 555 mtx_destroy(&pcbinfo->ipi_lock); 556 } 557 558 /* 559 * Initialize a pcbstorage - per protocol zones to allocate inpcbs. 560 */ 561 static void inpcb_fini(void *, int); 562 void 563 in_pcbstorage_init(void *arg) 564 { 565 struct inpcbstorage *pcbstor = arg; 566 567 pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name, 568 pcbstor->ips_size, NULL, NULL, pcbstor->ips_pcbinit, 569 inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR); 570 pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name, 571 sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 572 uma_zone_set_smr(pcbstor->ips_portzone, 573 uma_zone_get_smr(pcbstor->ips_zone)); 574 } 575 576 /* 577 * Destroy a pcbstorage - used by unloadable protocols. 578 */ 579 void 580 in_pcbstorage_destroy(void *arg) 581 { 582 struct inpcbstorage *pcbstor = arg; 583 584 uma_zdestroy(pcbstor->ips_zone); 585 uma_zdestroy(pcbstor->ips_portzone); 586 } 587 588 /* 589 * Allocate a PCB and associate it with the socket. 590 * On success return with the PCB locked. 591 */ 592 int 593 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 594 { 595 struct inpcb *inp; 596 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 597 int error; 598 #endif 599 600 inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT); 601 if (inp == NULL) 602 return (ENOBUFS); 603 bzero(&inp->inp_start_zero, inp_zero_size); 604 #ifdef NUMA 605 inp->inp_numa_domain = M_NODOM; 606 #endif 607 inp->inp_pcbinfo = pcbinfo; 608 inp->inp_socket = so; 609 inp->inp_cred = crhold(so->so_cred); 610 inp->inp_inc.inc_fibnum = so->so_fibnum; 611 #ifdef MAC 612 error = mac_inpcb_init(inp, M_NOWAIT); 613 if (error != 0) 614 goto out; 615 mac_inpcb_create(so, inp); 616 #endif 617 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 618 error = ipsec_init_pcbpolicy(inp); 619 if (error != 0) { 620 #ifdef MAC 621 mac_inpcb_destroy(inp); 622 #endif 623 goto out; 624 } 625 #endif /*IPSEC*/ 626 #ifdef INET6 627 if (INP_SOCKAF(so) == AF_INET6) { 628 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6; 629 if (V_ip6_v6only) 630 inp->inp_flags |= IN6P_IPV6_V6ONLY; 631 #ifdef INET 632 else 633 inp->inp_vflag |= INP_IPV4; 634 #endif 635 if (V_ip6_auto_flowlabel) 636 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 637 inp->in6p_hops = -1; /* use kernel default */ 638 } 639 #endif 640 #if defined(INET) && defined(INET6) 641 else 642 #endif 643 #ifdef INET 644 inp->inp_vflag |= INP_IPV4; 645 #endif 646 inp->inp_smr = SMR_SEQ_INVALID; 647 648 /* 649 * Routes in inpcb's can cache L2 as well; they are guaranteed 650 * to be cleaned up. 651 */ 652 inp->inp_route.ro_flags = RT_LLE_CACHE; 653 refcount_init(&inp->inp_refcount, 1); /* Reference from socket. */ 654 INP_WLOCK(inp); 655 INP_INFO_WLOCK(pcbinfo); 656 pcbinfo->ipi_count++; 657 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 658 CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list); 659 INP_INFO_WUNLOCK(pcbinfo); 660 so->so_pcb = inp; 661 662 return (0); 663 664 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 665 out: 666 crfree(inp->inp_cred); 667 #ifdef INVARIANTS 668 inp->inp_cred = NULL; 669 #endif 670 uma_zfree_smr(pcbinfo->ipi_zone, inp); 671 return (error); 672 #endif 673 } 674 675 #ifdef INET 676 int 677 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred) 678 { 679 int anonport, error; 680 681 KASSERT(sin == NULL || sin->sin_family == AF_INET, 682 ("%s: invalid address family for %p", __func__, sin)); 683 KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in), 684 ("%s: invalid address length for %p", __func__, sin)); 685 INP_WLOCK_ASSERT(inp); 686 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 687 688 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 689 return (EINVAL); 690 anonport = sin == NULL || sin->sin_port == 0; 691 error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr, 692 &inp->inp_lport, cred); 693 if (error) 694 return (error); 695 if (in_pcbinshash(inp) != 0) { 696 inp->inp_laddr.s_addr = INADDR_ANY; 697 inp->inp_lport = 0; 698 return (EAGAIN); 699 } 700 if (anonport) 701 inp->inp_flags |= INP_ANONPORT; 702 return (0); 703 } 704 #endif 705 706 #if defined(INET) || defined(INET6) 707 /* 708 * Assign a local port like in_pcb_lport(), but also used with connect() 709 * and a foreign address and port. If fsa is non-NULL, choose a local port 710 * that is unused with those, otherwise one that is completely unused. 711 * lsa can be NULL for IPv6. 712 */ 713 int 714 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 715 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 716 { 717 struct inpcbinfo *pcbinfo; 718 struct inpcb *tmpinp; 719 unsigned short *lastport; 720 int count, error; 721 u_short aux, first, last, lport; 722 #ifdef INET 723 struct in_addr laddr, faddr; 724 #endif 725 #ifdef INET6 726 struct in6_addr *laddr6, *faddr6; 727 #endif 728 729 pcbinfo = inp->inp_pcbinfo; 730 731 /* 732 * Because no actual state changes occur here, a global write lock on 733 * the pcbinfo isn't required. 734 */ 735 INP_LOCK_ASSERT(inp); 736 INP_HASH_LOCK_ASSERT(pcbinfo); 737 738 if (inp->inp_flags & INP_HIGHPORT) { 739 first = V_ipport_hifirstauto; /* sysctl */ 740 last = V_ipport_hilastauto; 741 lastport = &pcbinfo->ipi_lasthi; 742 } else if (inp->inp_flags & INP_LOWPORT) { 743 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 744 if (error) 745 return (error); 746 first = V_ipport_lowfirstauto; /* 1023 */ 747 last = V_ipport_lowlastauto; /* 600 */ 748 lastport = &pcbinfo->ipi_lastlow; 749 } else { 750 first = V_ipport_firstauto; /* sysctl */ 751 last = V_ipport_lastauto; 752 lastport = &pcbinfo->ipi_lastport; 753 } 754 755 /* 756 * Instead of having two loops further down counting up or down 757 * make sure that first is always <= last and go with only one 758 * code path implementing all logic. 759 */ 760 if (first > last) { 761 aux = first; 762 first = last; 763 last = aux; 764 } 765 766 #ifdef INET 767 laddr.s_addr = INADDR_ANY; /* used by INET6+INET below too */ 768 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 769 if (lsa != NULL) 770 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 771 if (fsa != NULL) 772 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 773 } 774 #endif 775 #ifdef INET6 776 laddr6 = NULL; 777 if ((inp->inp_vflag & INP_IPV6) != 0) { 778 if (lsa != NULL) 779 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 780 if (fsa != NULL) 781 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 782 } 783 #endif 784 785 tmpinp = NULL; 786 lport = *lportp; 787 788 if (V_ipport_randomized) 789 *lastport = first + (arc4random() % (last - first)); 790 791 count = last - first; 792 793 do { 794 if (count-- < 0) /* completely used? */ 795 return (EADDRNOTAVAIL); 796 ++*lastport; 797 if (*lastport < first || *lastport > last) 798 *lastport = first; 799 lport = htons(*lastport); 800 801 if (fsa != NULL) { 802 #ifdef INET 803 if (lsa->sa_family == AF_INET) { 804 tmpinp = in_pcblookup_hash_locked(pcbinfo, 805 faddr, fport, laddr, lport, lookupflags, 806 M_NODOM); 807 } 808 #endif 809 #ifdef INET6 810 if (lsa->sa_family == AF_INET6) { 811 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 812 faddr6, fport, laddr6, lport, lookupflags, 813 M_NODOM); 814 } 815 #endif 816 } else { 817 #ifdef INET6 818 if ((inp->inp_vflag & INP_IPV6) != 0) { 819 tmpinp = in6_pcblookup_local(pcbinfo, 820 &inp->in6p_laddr, lport, lookupflags, cred); 821 #ifdef INET 822 if (tmpinp == NULL && 823 (inp->inp_vflag & INP_IPV4)) 824 tmpinp = in_pcblookup_local(pcbinfo, 825 laddr, lport, lookupflags, cred); 826 #endif 827 } 828 #endif 829 #if defined(INET) && defined(INET6) 830 else 831 #endif 832 #ifdef INET 833 tmpinp = in_pcblookup_local(pcbinfo, laddr, 834 lport, lookupflags, cred); 835 #endif 836 } 837 } while (tmpinp != NULL); 838 839 *lportp = lport; 840 841 return (0); 842 } 843 844 /* 845 * Select a local port (number) to use. 846 */ 847 int 848 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 849 struct ucred *cred, int lookupflags) 850 { 851 struct sockaddr_in laddr; 852 853 if (laddrp) { 854 bzero(&laddr, sizeof(laddr)); 855 laddr.sin_family = AF_INET; 856 laddr.sin_addr = *laddrp; 857 } 858 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 859 NULL, lportp, NULL, 0, cred, lookupflags)); 860 } 861 #endif /* INET || INET6 */ 862 863 #ifdef INET 864 /* 865 * Set up a bind operation on a PCB, performing port allocation 866 * as required, but do not actually modify the PCB. Callers can 867 * either complete the bind by setting inp_laddr/inp_lport and 868 * calling in_pcbinshash(), or they can just use the resulting 869 * port and address to authorise the sending of a once-off packet. 870 * 871 * On error, the values of *laddrp and *lportp are not changed. 872 */ 873 int 874 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp, 875 u_short *lportp, struct ucred *cred) 876 { 877 struct socket *so = inp->inp_socket; 878 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 879 struct in_addr laddr; 880 u_short lport = 0; 881 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 882 int error; 883 884 /* 885 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 886 * so that we don't have to add to the (already messy) code below. 887 */ 888 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 889 890 /* 891 * No state changes, so read locks are sufficient here. 892 */ 893 INP_LOCK_ASSERT(inp); 894 INP_HASH_LOCK_ASSERT(pcbinfo); 895 896 laddr.s_addr = *laddrp; 897 if (sin != NULL && laddr.s_addr != INADDR_ANY) 898 return (EINVAL); 899 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 900 lookupflags = INPLOOKUP_WILDCARD; 901 if (sin == NULL) { 902 if ((error = prison_local_ip4(cred, &laddr)) != 0) 903 return (error); 904 } else { 905 KASSERT(sin->sin_family == AF_INET, 906 ("%s: invalid family for address %p", __func__, sin)); 907 KASSERT(sin->sin_len == sizeof(*sin), 908 ("%s: invalid length for address %p", __func__, sin)); 909 910 error = prison_local_ip4(cred, &sin->sin_addr); 911 if (error) 912 return (error); 913 if (sin->sin_port != *lportp) { 914 /* Don't allow the port to change. */ 915 if (*lportp != 0) 916 return (EINVAL); 917 lport = sin->sin_port; 918 } 919 laddr = sin->sin_addr; 920 921 /* NB: lport is left as 0 if the port isn't being changed. */ 922 if (IN_MULTICAST(ntohl(laddr.s_addr))) { 923 /* 924 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 925 * allow complete duplication of binding if 926 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 927 * and a multicast address is bound on both 928 * new and duplicated sockets. 929 */ 930 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 931 reuseport = SO_REUSEADDR|SO_REUSEPORT; 932 /* 933 * XXX: How to deal with SO_REUSEPORT_LB here? 934 * Treat same as SO_REUSEPORT for now. 935 */ 936 if ((so->so_options & 937 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 938 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 939 } else if (!in_nullhost(laddr)) { 940 sin->sin_port = 0; /* yech... */ 941 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 942 /* 943 * Is the address a local IP address? 944 * If INP_BINDANY is set, then the socket may be bound 945 * to any endpoint address, local or not. 946 */ 947 if ((inp->inp_flags & INP_BINDANY) == 0 && 948 ifa_ifwithaddr_check( 949 (const struct sockaddr *)sin) == 0) 950 return (EADDRNOTAVAIL); 951 } 952 if (lport) { 953 struct inpcb *t; 954 955 if (ntohs(lport) <= V_ipport_reservedhigh && 956 ntohs(lport) >= V_ipport_reservedlow && 957 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 958 return (EACCES); 959 960 if (!IN_MULTICAST(ntohl(laddr.s_addr)) && 961 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 962 t = in_pcblookup_local(pcbinfo, laddr, lport, 963 INPLOOKUP_WILDCARD, cred); 964 if (t != NULL && 965 (so->so_type != SOCK_STREAM || 966 in_nullhost(t->inp_faddr)) && 967 (!in_nullhost(laddr) || 968 !in_nullhost(t->inp_laddr) || 969 (t->inp_socket->so_options & SO_REUSEPORT) || 970 (t->inp_socket->so_options & SO_REUSEPORT_LB) == 0) && 971 (inp->inp_cred->cr_uid != 972 t->inp_cred->cr_uid)) 973 return (EADDRINUSE); 974 } 975 t = in_pcblookup_local(pcbinfo, laddr, lport, 976 lookupflags, cred); 977 if (t != NULL && ((reuseport | reuseport_lb) & 978 t->inp_socket->so_options) == 0) { 979 #ifdef INET6 980 if (!in_nullhost(laddr) || 981 !in_nullhost(t->inp_laddr) || 982 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 983 (t->inp_vflag & INP_IPV6PROTO) == 0) 984 #endif 985 return (EADDRINUSE); 986 } 987 } 988 } 989 if (*lportp != 0) 990 lport = *lportp; 991 if (lport == 0) { 992 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 993 if (error != 0) 994 return (error); 995 } 996 *laddrp = laddr.s_addr; 997 *lportp = lport; 998 return (0); 999 } 1000 1001 /* 1002 * Connect from a socket to a specified address. 1003 * Both address and port must be specified in argument sin. 1004 * If don't have a local address for this socket yet, 1005 * then pick one. 1006 */ 1007 int 1008 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred) 1009 { 1010 u_short lport, fport; 1011 in_addr_t laddr, faddr; 1012 int anonport, error; 1013 1014 INP_WLOCK_ASSERT(inp); 1015 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1016 KASSERT(in_nullhost(inp->inp_faddr), 1017 ("%s: inp is already connected", __func__)); 1018 1019 lport = inp->inp_lport; 1020 laddr = inp->inp_laddr.s_addr; 1021 anonport = (lport == 0); 1022 error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport, 1023 cred); 1024 if (error) 1025 return (error); 1026 1027 inp->inp_faddr.s_addr = faddr; 1028 inp->inp_fport = fport; 1029 1030 /* Do the initial binding of the local address if required. */ 1031 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1032 inp->inp_lport = lport; 1033 inp->inp_laddr.s_addr = laddr; 1034 if (in_pcbinshash(inp) != 0) { 1035 inp->inp_laddr.s_addr = inp->inp_faddr.s_addr = 1036 INADDR_ANY; 1037 inp->inp_lport = inp->inp_fport = 0; 1038 return (EAGAIN); 1039 } 1040 } else { 1041 inp->inp_lport = lport; 1042 inp->inp_laddr.s_addr = laddr; 1043 if ((inp->inp_flags & INP_INHASHLIST) != 0) 1044 in_pcbrehash(inp); 1045 else 1046 in_pcbinshash(inp); 1047 } 1048 1049 if (anonport) 1050 inp->inp_flags |= INP_ANONPORT; 1051 return (0); 1052 } 1053 1054 /* 1055 * Do proper source address selection on an unbound socket in case 1056 * of connect. Take jails into account as well. 1057 */ 1058 int 1059 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1060 struct ucred *cred) 1061 { 1062 struct ifaddr *ifa; 1063 struct sockaddr *sa; 1064 struct sockaddr_in *sin, dst; 1065 struct nhop_object *nh; 1066 int error; 1067 1068 NET_EPOCH_ASSERT(); 1069 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1070 1071 /* 1072 * Bypass source address selection and use the primary jail IP 1073 * if requested. 1074 */ 1075 if (!prison_saddrsel_ip4(cred, laddr)) 1076 return (0); 1077 1078 error = 0; 1079 1080 nh = NULL; 1081 bzero(&dst, sizeof(dst)); 1082 sin = &dst; 1083 sin->sin_family = AF_INET; 1084 sin->sin_len = sizeof(struct sockaddr_in); 1085 sin->sin_addr.s_addr = faddr->s_addr; 1086 1087 /* 1088 * If route is known our src addr is taken from the i/f, 1089 * else punt. 1090 * 1091 * Find out route to destination. 1092 */ 1093 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1094 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1095 0, NHR_NONE, 0); 1096 1097 /* 1098 * If we found a route, use the address corresponding to 1099 * the outgoing interface. 1100 * 1101 * Otherwise assume faddr is reachable on a directly connected 1102 * network and try to find a corresponding interface to take 1103 * the source address from. 1104 */ 1105 if (nh == NULL || nh->nh_ifp == NULL) { 1106 struct in_ifaddr *ia; 1107 struct ifnet *ifp; 1108 1109 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1110 inp->inp_socket->so_fibnum)); 1111 if (ia == NULL) { 1112 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1113 inp->inp_socket->so_fibnum)); 1114 } 1115 if (ia == NULL) { 1116 error = ENETUNREACH; 1117 goto done; 1118 } 1119 1120 if (!prison_flag(cred, PR_IP4)) { 1121 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1122 goto done; 1123 } 1124 1125 ifp = ia->ia_ifp; 1126 ia = NULL; 1127 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1128 sa = ifa->ifa_addr; 1129 if (sa->sa_family != AF_INET) 1130 continue; 1131 sin = (struct sockaddr_in *)sa; 1132 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1133 ia = (struct in_ifaddr *)ifa; 1134 break; 1135 } 1136 } 1137 if (ia != NULL) { 1138 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1139 goto done; 1140 } 1141 1142 /* 3. As a last resort return the 'default' jail address. */ 1143 error = prison_get_ip4(cred, laddr); 1144 goto done; 1145 } 1146 1147 /* 1148 * If the outgoing interface on the route found is not 1149 * a loopback interface, use the address from that interface. 1150 * In case of jails do those three steps: 1151 * 1. check if the interface address belongs to the jail. If so use it. 1152 * 2. check if we have any address on the outgoing interface 1153 * belonging to this jail. If so use it. 1154 * 3. as a last resort return the 'default' jail address. 1155 */ 1156 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1157 struct in_ifaddr *ia; 1158 struct ifnet *ifp; 1159 1160 /* If not jailed, use the default returned. */ 1161 if (!prison_flag(cred, PR_IP4)) { 1162 ia = (struct in_ifaddr *)nh->nh_ifa; 1163 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1164 goto done; 1165 } 1166 1167 /* Jailed. */ 1168 /* 1. Check if the iface address belongs to the jail. */ 1169 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1170 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1171 ia = (struct in_ifaddr *)nh->nh_ifa; 1172 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1173 goto done; 1174 } 1175 1176 /* 1177 * 2. Check if we have any address on the outgoing interface 1178 * belonging to this jail. 1179 */ 1180 ia = NULL; 1181 ifp = nh->nh_ifp; 1182 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1183 sa = ifa->ifa_addr; 1184 if (sa->sa_family != AF_INET) 1185 continue; 1186 sin = (struct sockaddr_in *)sa; 1187 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1188 ia = (struct in_ifaddr *)ifa; 1189 break; 1190 } 1191 } 1192 if (ia != NULL) { 1193 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1194 goto done; 1195 } 1196 1197 /* 3. As a last resort return the 'default' jail address. */ 1198 error = prison_get_ip4(cred, laddr); 1199 goto done; 1200 } 1201 1202 /* 1203 * The outgoing interface is marked with 'loopback net', so a route 1204 * to ourselves is here. 1205 * Try to find the interface of the destination address and then 1206 * take the address from there. That interface is not necessarily 1207 * a loopback interface. 1208 * In case of jails, check that it is an address of the jail 1209 * and if we cannot find, fall back to the 'default' jail address. 1210 */ 1211 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1212 struct in_ifaddr *ia; 1213 1214 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1215 inp->inp_socket->so_fibnum)); 1216 if (ia == NULL) 1217 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1218 inp->inp_socket->so_fibnum)); 1219 if (ia == NULL) 1220 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1221 1222 if (!prison_flag(cred, PR_IP4)) { 1223 if (ia == NULL) { 1224 error = ENETUNREACH; 1225 goto done; 1226 } 1227 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1228 goto done; 1229 } 1230 1231 /* Jailed. */ 1232 if (ia != NULL) { 1233 struct ifnet *ifp; 1234 1235 ifp = ia->ia_ifp; 1236 ia = NULL; 1237 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1238 sa = ifa->ifa_addr; 1239 if (sa->sa_family != AF_INET) 1240 continue; 1241 sin = (struct sockaddr_in *)sa; 1242 if (prison_check_ip4(cred, 1243 &sin->sin_addr) == 0) { 1244 ia = (struct in_ifaddr *)ifa; 1245 break; 1246 } 1247 } 1248 if (ia != NULL) { 1249 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1250 goto done; 1251 } 1252 } 1253 1254 /* 3. As a last resort return the 'default' jail address. */ 1255 error = prison_get_ip4(cred, laddr); 1256 goto done; 1257 } 1258 1259 done: 1260 if (error == 0 && laddr->s_addr == INADDR_ANY) 1261 return (EHOSTUNREACH); 1262 return (error); 1263 } 1264 1265 /* 1266 * Set up for a connect from a socket to the specified address. 1267 * On entry, *laddrp and *lportp should contain the current local 1268 * address and port for the PCB; these are updated to the values 1269 * that should be placed in inp_laddr and inp_lport to complete 1270 * the connect. 1271 * 1272 * On success, *faddrp and *fportp will be set to the remote address 1273 * and port. These are not updated in the error case. 1274 */ 1275 int 1276 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin, 1277 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1278 struct ucred *cred) 1279 { 1280 struct in_ifaddr *ia; 1281 struct in_addr laddr, faddr; 1282 u_short lport, fport; 1283 int error; 1284 1285 KASSERT(sin->sin_family == AF_INET, 1286 ("%s: invalid address family for %p", __func__, sin)); 1287 KASSERT(sin->sin_len == sizeof(*sin), 1288 ("%s: invalid address length for %p", __func__, sin)); 1289 1290 /* 1291 * Because a global state change doesn't actually occur here, a read 1292 * lock is sufficient. 1293 */ 1294 NET_EPOCH_ASSERT(); 1295 INP_LOCK_ASSERT(inp); 1296 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1297 1298 if (sin->sin_port == 0) 1299 return (EADDRNOTAVAIL); 1300 laddr.s_addr = *laddrp; 1301 lport = *lportp; 1302 faddr = sin->sin_addr; 1303 fport = sin->sin_port; 1304 #ifdef ROUTE_MPATH 1305 if (CALC_FLOWID_OUTBOUND) { 1306 uint32_t hash_val, hash_type; 1307 1308 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, 1309 inp->inp_socket->so_proto->pr_protocol, &hash_type); 1310 1311 inp->inp_flowid = hash_val; 1312 inp->inp_flowtype = hash_type; 1313 } 1314 #endif 1315 if (V_connect_inaddr_wild && !CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1316 /* 1317 * If the destination address is INADDR_ANY, 1318 * use the primary local address. 1319 * If the supplied address is INADDR_BROADCAST, 1320 * and the primary interface supports broadcast, 1321 * choose the broadcast address for that interface. 1322 */ 1323 if (faddr.s_addr == INADDR_ANY) { 1324 faddr = 1325 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1326 if ((error = prison_get_ip4(cred, &faddr)) != 0) 1327 return (error); 1328 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1329 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1330 IFF_BROADCAST) 1331 faddr = satosin(&CK_STAILQ_FIRST( 1332 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1333 } 1334 } else if (faddr.s_addr == INADDR_ANY) { 1335 return (ENETUNREACH); 1336 } 1337 if (laddr.s_addr == INADDR_ANY) { 1338 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1339 /* 1340 * If the destination address is multicast and an outgoing 1341 * interface has been set as a multicast option, prefer the 1342 * address of that interface as our source address. 1343 */ 1344 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1345 inp->inp_moptions != NULL) { 1346 struct ip_moptions *imo; 1347 struct ifnet *ifp; 1348 1349 imo = inp->inp_moptions; 1350 if (imo->imo_multicast_ifp != NULL) { 1351 ifp = imo->imo_multicast_ifp; 1352 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1353 if (ia->ia_ifp == ifp && 1354 prison_check_ip4(cred, 1355 &ia->ia_addr.sin_addr) == 0) 1356 break; 1357 } 1358 if (ia == NULL) 1359 error = EADDRNOTAVAIL; 1360 else { 1361 laddr = ia->ia_addr.sin_addr; 1362 error = 0; 1363 } 1364 } 1365 } 1366 if (error) 1367 return (error); 1368 } 1369 1370 if (lport != 0) { 1371 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, 1372 fport, laddr, lport, 0, M_NODOM) != NULL) 1373 return (EADDRINUSE); 1374 } else { 1375 struct sockaddr_in lsin, fsin; 1376 1377 bzero(&lsin, sizeof(lsin)); 1378 bzero(&fsin, sizeof(fsin)); 1379 lsin.sin_family = AF_INET; 1380 lsin.sin_addr = laddr; 1381 fsin.sin_family = AF_INET; 1382 fsin.sin_addr = faddr; 1383 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, 1384 &lport, (struct sockaddr *)& fsin, fport, cred, 1385 INPLOOKUP_WILDCARD); 1386 if (error) 1387 return (error); 1388 } 1389 *laddrp = laddr.s_addr; 1390 *lportp = lport; 1391 *faddrp = faddr.s_addr; 1392 *fportp = fport; 1393 return (0); 1394 } 1395 1396 void 1397 in_pcbdisconnect(struct inpcb *inp) 1398 { 1399 1400 INP_WLOCK_ASSERT(inp); 1401 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1402 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 1403 ("%s: inp %p was already disconnected", __func__, inp)); 1404 1405 in_pcbremhash_locked(inp); 1406 1407 /* See the comment in in_pcbinshash(). */ 1408 inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr); 1409 inp->inp_laddr.s_addr = INADDR_ANY; 1410 inp->inp_faddr.s_addr = INADDR_ANY; 1411 inp->inp_fport = 0; 1412 } 1413 #endif /* INET */ 1414 1415 /* 1416 * inpcb hash lookups are protected by SMR section. 1417 * 1418 * Once desired pcb has been found, switching from SMR section to a pcb 1419 * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK 1420 * here because SMR is a critical section. 1421 * In 99%+ cases inp_smr_lock() would obtain the lock immediately. 1422 */ 1423 void 1424 inp_lock(struct inpcb *inp, const inp_lookup_t lock) 1425 { 1426 1427 lock == INPLOOKUP_RLOCKPCB ? 1428 rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock); 1429 } 1430 1431 void 1432 inp_unlock(struct inpcb *inp, const inp_lookup_t lock) 1433 { 1434 1435 lock == INPLOOKUP_RLOCKPCB ? 1436 rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock); 1437 } 1438 1439 int 1440 inp_trylock(struct inpcb *inp, const inp_lookup_t lock) 1441 { 1442 1443 return (lock == INPLOOKUP_RLOCKPCB ? 1444 rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock)); 1445 } 1446 1447 static inline bool 1448 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags) 1449 { 1450 1451 MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB); 1452 SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr); 1453 1454 if (__predict_true(inp_trylock(inp, lock))) { 1455 if (__predict_false(inp->inp_flags & ignflags)) { 1456 smr_exit(inp->inp_pcbinfo->ipi_smr); 1457 inp_unlock(inp, lock); 1458 return (false); 1459 } 1460 smr_exit(inp->inp_pcbinfo->ipi_smr); 1461 return (true); 1462 } 1463 1464 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1465 smr_exit(inp->inp_pcbinfo->ipi_smr); 1466 inp_lock(inp, lock); 1467 if (__predict_false(in_pcbrele(inp, lock))) 1468 return (false); 1469 /* 1470 * inp acquired through refcount & lock for sure didn't went 1471 * through uma_zfree(). However, it may have already went 1472 * through in_pcbfree() and has another reference, that 1473 * prevented its release by our in_pcbrele(). 1474 */ 1475 if (__predict_false(inp->inp_flags & ignflags)) { 1476 inp_unlock(inp, lock); 1477 return (false); 1478 } 1479 return (true); 1480 } else { 1481 smr_exit(inp->inp_pcbinfo->ipi_smr); 1482 return (false); 1483 } 1484 } 1485 1486 bool 1487 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock) 1488 { 1489 1490 /* 1491 * in_pcblookup() family of functions ignore not only freed entries, 1492 * that may be found due to lockless access to the hash, but dropped 1493 * entries, too. 1494 */ 1495 return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED)); 1496 } 1497 1498 /* 1499 * inp_next() - inpcb hash/list traversal iterator 1500 * 1501 * Requires initialized struct inpcb_iterator for context. 1502 * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR(). 1503 * 1504 * - Iterator can have either write-lock or read-lock semantics, that can not 1505 * be changed later. 1506 * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through 1507 * a single hash slot. Note: only rip_input() does the latter. 1508 * - Iterator may have optional bool matching function. The matching function 1509 * will be executed for each inpcb in the SMR context, so it can not acquire 1510 * locks and can safely access only immutable fields of inpcb. 1511 * 1512 * A fresh initialized iterator has NULL inpcb in its context and that 1513 * means that inp_next() call would return the very first inpcb on the list 1514 * locked with desired semantic. In all following calls the context pointer 1515 * shall hold the current inpcb pointer. The KPI user is not supposed to 1516 * unlock the current inpcb! Upon end of traversal inp_next() will return NULL 1517 * and write NULL to its context. After end of traversal an iterator can be 1518 * reused. 1519 * 1520 * List traversals have the following features/constraints: 1521 * - New entries won't be seen, as they are always added to the head of a list. 1522 * - Removed entries won't stop traversal as long as they are not added to 1523 * a different list. This is violated by in_pcbrehash(). 1524 */ 1525 #define II_LIST_FIRST(ipi, hash) \ 1526 (((hash) == INP_ALL_LIST) ? \ 1527 CK_LIST_FIRST(&(ipi)->ipi_listhead) : \ 1528 CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)])) 1529 #define II_LIST_NEXT(inp, hash) \ 1530 (((hash) == INP_ALL_LIST) ? \ 1531 CK_LIST_NEXT((inp), inp_list) : \ 1532 CK_LIST_NEXT((inp), inp_hash_exact)) 1533 #define II_LOCK_ASSERT(inp, lock) \ 1534 rw_assert(&(inp)->inp_lock, \ 1535 (lock) == INPLOOKUP_RLOCKPCB ? RA_RLOCKED : RA_WLOCKED ) 1536 struct inpcb * 1537 inp_next(struct inpcb_iterator *ii) 1538 { 1539 const struct inpcbinfo *ipi = ii->ipi; 1540 inp_match_t *match = ii->match; 1541 void *ctx = ii->ctx; 1542 inp_lookup_t lock = ii->lock; 1543 int hash = ii->hash; 1544 struct inpcb *inp; 1545 1546 if (ii->inp == NULL) { /* First call. */ 1547 smr_enter(ipi->ipi_smr); 1548 /* This is unrolled CK_LIST_FOREACH(). */ 1549 for (inp = II_LIST_FIRST(ipi, hash); 1550 inp != NULL; 1551 inp = II_LIST_NEXT(inp, hash)) { 1552 if (match != NULL && (match)(inp, ctx) == false) 1553 continue; 1554 if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED))) 1555 break; 1556 else { 1557 smr_enter(ipi->ipi_smr); 1558 MPASS(inp != II_LIST_FIRST(ipi, hash)); 1559 inp = II_LIST_FIRST(ipi, hash); 1560 if (inp == NULL) 1561 break; 1562 } 1563 } 1564 1565 if (inp == NULL) 1566 smr_exit(ipi->ipi_smr); 1567 else 1568 ii->inp = inp; 1569 1570 return (inp); 1571 } 1572 1573 /* Not a first call. */ 1574 smr_enter(ipi->ipi_smr); 1575 restart: 1576 inp = ii->inp; 1577 II_LOCK_ASSERT(inp, lock); 1578 next: 1579 inp = II_LIST_NEXT(inp, hash); 1580 if (inp == NULL) { 1581 smr_exit(ipi->ipi_smr); 1582 goto found; 1583 } 1584 1585 if (match != NULL && (match)(inp, ctx) == false) 1586 goto next; 1587 1588 if (__predict_true(inp_trylock(inp, lock))) { 1589 if (__predict_false(inp->inp_flags & INP_FREED)) { 1590 /* 1591 * Entries are never inserted in middle of a list, thus 1592 * as long as we are in SMR, we can continue traversal. 1593 * Jump to 'restart' should yield in the same result, 1594 * but could produce unnecessary looping. Could this 1595 * looping be unbound? 1596 */ 1597 inp_unlock(inp, lock); 1598 goto next; 1599 } else { 1600 smr_exit(ipi->ipi_smr); 1601 goto found; 1602 } 1603 } 1604 1605 /* 1606 * Can't obtain lock immediately, thus going hard. Once we exit the 1607 * SMR section we can no longer jump to 'next', and our only stable 1608 * anchoring point is ii->inp, which we keep locked for this case, so 1609 * we jump to 'restart'. 1610 */ 1611 if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) { 1612 smr_exit(ipi->ipi_smr); 1613 inp_lock(inp, lock); 1614 if (__predict_false(in_pcbrele(inp, lock))) { 1615 smr_enter(ipi->ipi_smr); 1616 goto restart; 1617 } 1618 /* 1619 * See comment in inp_smr_lock(). 1620 */ 1621 if (__predict_false(inp->inp_flags & INP_FREED)) { 1622 inp_unlock(inp, lock); 1623 smr_enter(ipi->ipi_smr); 1624 goto restart; 1625 } 1626 } else 1627 goto next; 1628 1629 found: 1630 inp_unlock(ii->inp, lock); 1631 ii->inp = inp; 1632 1633 return (ii->inp); 1634 } 1635 1636 /* 1637 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1638 * stability of an inpcb pointer despite the inpcb lock being released or 1639 * SMR section exited. 1640 * 1641 * To free a reference later in_pcbrele_(r|w)locked() must be performed. 1642 */ 1643 void 1644 in_pcbref(struct inpcb *inp) 1645 { 1646 u_int old __diagused; 1647 1648 old = refcount_acquire(&inp->inp_refcount); 1649 KASSERT(old > 0, ("%s: refcount 0", __func__)); 1650 } 1651 1652 /* 1653 * Drop a refcount on an inpcb elevated using in_pcbref(), potentially 1654 * freeing the pcb, if the reference was very last. 1655 */ 1656 bool 1657 in_pcbrele_rlocked(struct inpcb *inp) 1658 { 1659 1660 INP_RLOCK_ASSERT(inp); 1661 1662 if (!refcount_release(&inp->inp_refcount)) 1663 return (false); 1664 1665 MPASS(inp->inp_flags & INP_FREED); 1666 MPASS(inp->inp_socket == NULL); 1667 crfree(inp->inp_cred); 1668 #ifdef INVARIANTS 1669 inp->inp_cred = NULL; 1670 #endif 1671 INP_RUNLOCK(inp); 1672 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1673 return (true); 1674 } 1675 1676 bool 1677 in_pcbrele_wlocked(struct inpcb *inp) 1678 { 1679 1680 INP_WLOCK_ASSERT(inp); 1681 1682 if (!refcount_release(&inp->inp_refcount)) 1683 return (false); 1684 1685 MPASS(inp->inp_flags & INP_FREED); 1686 MPASS(inp->inp_socket == NULL); 1687 crfree(inp->inp_cred); 1688 #ifdef INVARIANTS 1689 inp->inp_cred = NULL; 1690 #endif 1691 INP_WUNLOCK(inp); 1692 uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp); 1693 return (true); 1694 } 1695 1696 bool 1697 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock) 1698 { 1699 1700 return (lock == INPLOOKUP_RLOCKPCB ? 1701 in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp)); 1702 } 1703 1704 /* 1705 * Unconditionally schedule an inpcb to be freed by decrementing its 1706 * reference count, which should occur only after the inpcb has been detached 1707 * from its socket. If another thread holds a temporary reference (acquired 1708 * using in_pcbref()) then the free is deferred until that reference is 1709 * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked. 1710 * Almost all work, including removal from global lists, is done in this 1711 * context, where the pcbinfo lock is held. 1712 */ 1713 void 1714 in_pcbfree(struct inpcb *inp) 1715 { 1716 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1717 #ifdef INET 1718 struct ip_moptions *imo; 1719 #endif 1720 #ifdef INET6 1721 struct ip6_moptions *im6o; 1722 #endif 1723 1724 INP_WLOCK_ASSERT(inp); 1725 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1726 KASSERT((inp->inp_flags & INP_FREED) == 0, 1727 ("%s: called twice for pcb %p", __func__, inp)); 1728 1729 /* 1730 * in_pcblookup_local() and in6_pcblookup_local() may return an inpcb 1731 * from the hash without acquiring inpcb lock, they rely on the hash 1732 * lock, thus in_pcbremhash() should be the first action. 1733 */ 1734 if (inp->inp_flags & INP_INHASHLIST) 1735 in_pcbremhash(inp); 1736 INP_INFO_WLOCK(pcbinfo); 1737 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 1738 pcbinfo->ipi_count--; 1739 CK_LIST_REMOVE(inp, inp_list); 1740 INP_INFO_WUNLOCK(pcbinfo); 1741 1742 #ifdef RATELIMIT 1743 if (inp->inp_snd_tag != NULL) 1744 in_pcbdetach_txrtlmt(inp); 1745 #endif 1746 inp->inp_flags |= INP_FREED; 1747 inp->inp_socket->so_pcb = NULL; 1748 inp->inp_socket = NULL; 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 DEBUG_POISON_POINTER(inp->inp_options); 1762 imo = inp->inp_moptions; 1763 DEBUG_POISON_POINTER(inp->inp_moptions); 1764 #endif 1765 #ifdef INET6 1766 if (inp->inp_vflag & INP_IPV6PROTO) { 1767 ip6_freepcbopts(inp->in6p_outputopts); 1768 DEBUG_POISON_POINTER(inp->in6p_outputopts); 1769 im6o = inp->in6p_moptions; 1770 DEBUG_POISON_POINTER(inp->in6p_moptions); 1771 } else 1772 im6o = NULL; 1773 #endif 1774 1775 if (__predict_false(in_pcbrele_wlocked(inp) == false)) { 1776 INP_WUNLOCK(inp); 1777 } 1778 #ifdef INET6 1779 ip6_freemoptions(im6o); 1780 #endif 1781 #ifdef INET 1782 inp_freemoptions(imo); 1783 #endif 1784 } 1785 1786 /* 1787 * Different protocols initialize their inpcbs differently - giving 1788 * different name to the lock. But they all are disposed the same. 1789 */ 1790 static void 1791 inpcb_fini(void *mem, int size) 1792 { 1793 struct inpcb *inp = mem; 1794 1795 INP_LOCK_DESTROY(inp); 1796 } 1797 1798 /* 1799 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1800 * port reservation, and preventing it from being returned by inpcb lookups. 1801 * 1802 * It is used by TCP to mark an inpcb as unused and avoid future packet 1803 * delivery or event notification when a socket remains open but TCP has 1804 * closed. This might occur as a result of a shutdown()-initiated TCP close 1805 * or a RST on the wire, and allows the port binding to be reused while still 1806 * maintaining the invariant that so_pcb always points to a valid inpcb until 1807 * in_pcbdetach(). 1808 * 1809 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1810 * in_pcbpurgeif0()? 1811 */ 1812 void 1813 in_pcbdrop(struct inpcb *inp) 1814 { 1815 1816 INP_WLOCK_ASSERT(inp); 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 laddr, 2178 u_short lport, const inp_lookup_t lockflags) 2179 { 2180 struct inpcbhead *head; 2181 struct inpcb *inp; 2182 2183 KASSERT(SMR_ENTERED(pcbinfo->ipi_smr), 2184 ("%s: not in SMR read section", __func__)); 2185 2186 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2187 pcbinfo->ipi_hashmask)]; 2188 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2189 inp_lookup_match_t match; 2190 2191 match = in_pcblookup_wild_match(inp, laddr, lport); 2192 if (match == INPLOOKUP_MATCH_NONE) 2193 continue; 2194 2195 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2196 match = in_pcblookup_wild_match(inp, laddr, lport); 2197 if (match != INPLOOKUP_MATCH_NONE && 2198 prison_check_ip4_locked(inp->inp_cred->cr_prison, 2199 &laddr) == 0) 2200 return (inp); 2201 inp_unlock(inp, lockflags); 2202 } 2203 2204 /* 2205 * The matching socket disappeared out from under us. Fall back 2206 * to a serialized lookup. 2207 */ 2208 return (INP_LOOKUP_AGAIN); 2209 } 2210 return (NULL); 2211 } 2212 2213 static struct inpcb * 2214 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr laddr, 2215 u_short lport) 2216 { 2217 struct inpcbhead *head; 2218 struct inpcb *inp, *local_wild, *local_exact, *jail_wild; 2219 #ifdef INET6 2220 struct inpcb *local_wild_mapped; 2221 #endif 2222 2223 INP_HASH_LOCK_ASSERT(pcbinfo); 2224 2225 /* 2226 * Order of socket selection - we always prefer jails. 2227 * 1. jailed, non-wild. 2228 * 2. jailed, wild. 2229 * 3. non-jailed, non-wild. 2230 * 4. non-jailed, wild. 2231 */ 2232 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport, 2233 pcbinfo->ipi_hashmask)]; 2234 local_wild = local_exact = jail_wild = NULL; 2235 #ifdef INET6 2236 local_wild_mapped = NULL; 2237 #endif 2238 CK_LIST_FOREACH(inp, head, inp_hash_wild) { 2239 inp_lookup_match_t match; 2240 bool injail; 2241 2242 match = in_pcblookup_wild_match(inp, laddr, lport); 2243 if (match == INPLOOKUP_MATCH_NONE) 2244 continue; 2245 2246 injail = prison_flag(inp->inp_cred, PR_IP4) != 0; 2247 if (injail) { 2248 if (prison_check_ip4_locked(inp->inp_cred->cr_prison, 2249 &laddr) != 0) 2250 continue; 2251 } else { 2252 if (local_exact != NULL) 2253 continue; 2254 } 2255 2256 if (match == INPLOOKUP_MATCH_LADDR) { 2257 if (injail) 2258 return (inp); 2259 local_exact = inp; 2260 } else { 2261 #ifdef INET6 2262 /* XXX inp locking, NULL check */ 2263 if (inp->inp_vflag & INP_IPV6PROTO) 2264 local_wild_mapped = inp; 2265 else 2266 #endif 2267 if (injail) 2268 jail_wild = inp; 2269 else 2270 local_wild = inp; 2271 } 2272 } 2273 if (jail_wild != NULL) 2274 return (jail_wild); 2275 if (local_exact != NULL) 2276 return (local_exact); 2277 if (local_wild != NULL) 2278 return (local_wild); 2279 #ifdef INET6 2280 if (local_wild_mapped != NULL) 2281 return (local_wild_mapped); 2282 #endif 2283 return (NULL); 2284 } 2285 2286 /* 2287 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2288 * that the caller has either locked the hash list, which usually happens 2289 * for bind(2) operations, or is in SMR section, which happens when sorting 2290 * out incoming packets. 2291 */ 2292 static struct inpcb * 2293 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2294 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2295 uint8_t numa_domain) 2296 { 2297 struct inpcb *inp; 2298 const u_short fport = fport_arg, lport = lport_arg; 2299 2300 KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0, 2301 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2302 KASSERT(faddr.s_addr != INADDR_ANY, 2303 ("%s: invalid foreign address", __func__)); 2304 KASSERT(laddr.s_addr != INADDR_ANY, 2305 ("%s: invalid local address", __func__)); 2306 INP_HASH_WLOCK_ASSERT(pcbinfo); 2307 2308 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2309 if (inp != NULL) 2310 return (inp); 2311 2312 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2313 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2314 &laddr, lport, numa_domain); 2315 if (inp == NULL) { 2316 inp = in_pcblookup_hash_wild_locked(pcbinfo, laddr, 2317 lport); 2318 } 2319 } 2320 2321 return (inp); 2322 } 2323 2324 static struct inpcb * 2325 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2326 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2327 uint8_t numa_domain) 2328 { 2329 struct inpcb *inp; 2330 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2331 2332 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2333 ("%s: LOCKPCB not set", __func__)); 2334 2335 INP_HASH_WLOCK(pcbinfo); 2336 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2337 lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain); 2338 if (inp != NULL && !inp_trylock(inp, lockflags)) { 2339 in_pcbref(inp); 2340 INP_HASH_WUNLOCK(pcbinfo); 2341 inp_lock(inp, lockflags); 2342 if (in_pcbrele(inp, lockflags)) 2343 /* XXX-MJ or retry until we get a negative match? */ 2344 inp = NULL; 2345 } else { 2346 INP_HASH_WUNLOCK(pcbinfo); 2347 } 2348 return (inp); 2349 } 2350 2351 static struct inpcb * 2352 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2353 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2354 uint8_t numa_domain) 2355 { 2356 struct inpcb *inp; 2357 const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK; 2358 const u_short fport = fport_arg, lport = lport_arg; 2359 2360 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2361 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2362 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2363 ("%s: LOCKPCB not set", __func__)); 2364 2365 smr_enter(pcbinfo->ipi_smr); 2366 inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport); 2367 if (inp != NULL) { 2368 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2369 /* 2370 * Revalidate the 4-tuple, the socket could have been 2371 * disconnected. 2372 */ 2373 if (__predict_true(in_pcblookup_exact_match(inp, 2374 faddr, fport, laddr, lport))) 2375 return (inp); 2376 inp_unlock(inp, lockflags); 2377 } 2378 2379 /* 2380 * We failed to lock the inpcb, or its connection state changed 2381 * out from under us. Fall back to a precise search. 2382 */ 2383 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2384 lookupflags, numa_domain)); 2385 } 2386 2387 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2388 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport, 2389 &laddr, lport, numa_domain); 2390 if (inp != NULL) { 2391 if (__predict_true(inp_smr_lock(inp, lockflags))) { 2392 if (__predict_true(in_pcblookup_wild_match(inp, 2393 laddr, lport) != INPLOOKUP_MATCH_NONE)) 2394 return (inp); 2395 inp_unlock(inp, lockflags); 2396 } 2397 inp = INP_LOOKUP_AGAIN; 2398 } else { 2399 inp = in_pcblookup_hash_wild_smr(pcbinfo, laddr, lport, 2400 lockflags); 2401 } 2402 if (inp == INP_LOOKUP_AGAIN) { 2403 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, 2404 lport, lookupflags, numa_domain)); 2405 } 2406 } 2407 2408 if (inp == NULL) 2409 smr_exit(pcbinfo->ipi_smr); 2410 2411 return (inp); 2412 } 2413 2414 /* 2415 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2416 * from which a pre-calculated hash value may be extracted. 2417 */ 2418 struct inpcb * 2419 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2420 struct in_addr laddr, u_int lport, int lookupflags, 2421 struct ifnet *ifp __unused) 2422 { 2423 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2424 lookupflags, M_NODOM)); 2425 } 2426 2427 struct inpcb * 2428 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2429 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2430 struct ifnet *ifp __unused, struct mbuf *m) 2431 { 2432 return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport, 2433 lookupflags, m->m_pkthdr.numa_domain)); 2434 } 2435 #endif /* INET */ 2436 2437 static bool 2438 in_pcbjailed(const struct inpcb *inp, unsigned int flag) 2439 { 2440 return (prison_flag(inp->inp_cred, flag) != 0); 2441 } 2442 2443 /* 2444 * Insert the PCB into a hash chain using ordering rules which ensure that 2445 * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first. 2446 * 2447 * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs 2448 * with exact local addresses ahead of wildcard PCBs. Unbound v4-mapped v6 PCBs 2449 * always appear last no matter whether they are jailed. 2450 */ 2451 static void 2452 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2453 { 2454 struct inpcb *last; 2455 bool bound, injail; 2456 2457 INP_LOCK_ASSERT(inp); 2458 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2459 2460 last = NULL; 2461 bound = inp->inp_laddr.s_addr != INADDR_ANY; 2462 if (!bound && (inp->inp_vflag & INP_IPV6PROTO) != 0) { 2463 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2464 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2465 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2466 return; 2467 } 2468 } 2469 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2470 return; 2471 } 2472 2473 injail = in_pcbjailed(inp, PR_IP4); 2474 if (!injail) { 2475 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2476 if (!in_pcbjailed(last, PR_IP4)) 2477 break; 2478 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2479 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2480 return; 2481 } 2482 } 2483 } else if (!CK_LIST_EMPTY(pcbhash) && 2484 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) { 2485 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2486 return; 2487 } 2488 if (!bound) { 2489 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2490 if (last->inp_laddr.s_addr == INADDR_ANY) 2491 break; 2492 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2493 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2494 return; 2495 } 2496 } 2497 } 2498 if (last == NULL) 2499 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2500 else 2501 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2502 } 2503 2504 #ifdef INET6 2505 /* 2506 * See the comment above _in_pcbinshash_wild(). 2507 */ 2508 static void 2509 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp) 2510 { 2511 struct inpcb *last; 2512 bool bound, injail; 2513 2514 INP_LOCK_ASSERT(inp); 2515 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2516 2517 last = NULL; 2518 bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr); 2519 injail = in_pcbjailed(inp, PR_IP6); 2520 if (!injail) { 2521 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) { 2522 if (!in_pcbjailed(last, PR_IP6)) 2523 break; 2524 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2525 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2526 return; 2527 } 2528 } 2529 } else if (!CK_LIST_EMPTY(pcbhash) && 2530 !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) { 2531 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2532 return; 2533 } 2534 if (!bound) { 2535 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) { 2536 if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr)) 2537 break; 2538 if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) { 2539 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild); 2540 return; 2541 } 2542 } 2543 } 2544 if (last == NULL) 2545 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild); 2546 else 2547 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild); 2548 } 2549 #endif 2550 2551 /* 2552 * Insert PCB onto various hash lists. 2553 */ 2554 int 2555 in_pcbinshash(struct inpcb *inp) 2556 { 2557 struct inpcbhead *pcbhash; 2558 struct inpcbporthead *pcbporthash; 2559 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2560 struct inpcbport *phd; 2561 uint32_t hash; 2562 bool connected; 2563 2564 INP_WLOCK_ASSERT(inp); 2565 INP_HASH_WLOCK_ASSERT(pcbinfo); 2566 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2567 ("in_pcbinshash: INP_INHASHLIST")); 2568 2569 #ifdef INET6 2570 if (inp->inp_vflag & INP_IPV6) { 2571 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2572 inp->inp_fport, pcbinfo->ipi_hashmask); 2573 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2574 } else 2575 #endif 2576 { 2577 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2578 inp->inp_fport, pcbinfo->ipi_hashmask); 2579 connected = !in_nullhost(inp->inp_faddr); 2580 } 2581 2582 if (connected) 2583 pcbhash = &pcbinfo->ipi_hash_exact[hash]; 2584 else 2585 pcbhash = &pcbinfo->ipi_hash_wild[hash]; 2586 2587 pcbporthash = &pcbinfo->ipi_porthashbase[ 2588 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2589 2590 /* 2591 * Add entry to load balance group. 2592 * Only do this if SO_REUSEPORT_LB is set. 2593 */ 2594 if ((inp->inp_socket->so_options & SO_REUSEPORT_LB) != 0) { 2595 int error = in_pcbinslbgrouphash(inp, M_NODOM); 2596 if (error != 0) 2597 return (error); 2598 } 2599 2600 /* 2601 * Go through port list and look for a head for this lport. 2602 */ 2603 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2604 if (phd->phd_port == inp->inp_lport) 2605 break; 2606 } 2607 2608 /* 2609 * If none exists, malloc one and tack it on. 2610 */ 2611 if (phd == NULL) { 2612 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT); 2613 if (phd == NULL) { 2614 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2615 in_pcbremlbgrouphash(inp); 2616 return (ENOMEM); 2617 } 2618 phd->phd_port = inp->inp_lport; 2619 CK_LIST_INIT(&phd->phd_pcblist); 2620 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2621 } 2622 inp->inp_phd = phd; 2623 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2624 2625 /* 2626 * The PCB may have been disconnected in the past. Before we can safely 2627 * make it visible in the hash table, we must wait for all readers which 2628 * may be traversing this PCB to finish. 2629 */ 2630 if (inp->inp_smr != SMR_SEQ_INVALID) { 2631 smr_wait(pcbinfo->ipi_smr, inp->inp_smr); 2632 inp->inp_smr = SMR_SEQ_INVALID; 2633 } 2634 2635 if (connected) 2636 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact); 2637 else { 2638 #ifdef INET6 2639 if ((inp->inp_vflag & INP_IPV6) != 0) 2640 _in6_pcbinshash_wild(pcbhash, inp); 2641 else 2642 #endif 2643 _in_pcbinshash_wild(pcbhash, inp); 2644 } 2645 inp->inp_flags |= INP_INHASHLIST; 2646 2647 return (0); 2648 } 2649 2650 void 2651 in_pcbremhash_locked(struct inpcb *inp) 2652 { 2653 struct inpcbport *phd = inp->inp_phd; 2654 2655 INP_WLOCK_ASSERT(inp); 2656 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 2657 MPASS(inp->inp_flags & INP_INHASHLIST); 2658 2659 if ((inp->inp_flags & INP_INLBGROUP) != 0) 2660 in_pcbremlbgrouphash(inp); 2661 #ifdef INET6 2662 if (inp->inp_vflag & INP_IPV6) { 2663 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) 2664 CK_LIST_REMOVE(inp, inp_hash_wild); 2665 else 2666 CK_LIST_REMOVE(inp, inp_hash_exact); 2667 } else 2668 #endif 2669 { 2670 if (in_nullhost(inp->inp_faddr)) 2671 CK_LIST_REMOVE(inp, inp_hash_wild); 2672 else 2673 CK_LIST_REMOVE(inp, inp_hash_exact); 2674 } 2675 CK_LIST_REMOVE(inp, inp_portlist); 2676 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2677 CK_LIST_REMOVE(phd, phd_hash); 2678 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd); 2679 } 2680 inp->inp_flags &= ~INP_INHASHLIST; 2681 } 2682 2683 static void 2684 in_pcbremhash(struct inpcb *inp) 2685 { 2686 INP_HASH_WLOCK(inp->inp_pcbinfo); 2687 in_pcbremhash_locked(inp); 2688 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 2689 } 2690 2691 /* 2692 * Move PCB to the proper hash bucket when { faddr, fport } have been 2693 * changed. NOTE: This does not handle the case of the lport changing (the 2694 * hashed port list would have to be updated as well), so the lport must 2695 * not change after in_pcbinshash() has been called. 2696 */ 2697 void 2698 in_pcbrehash(struct inpcb *inp) 2699 { 2700 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2701 struct inpcbhead *head; 2702 uint32_t hash; 2703 bool connected; 2704 2705 INP_WLOCK_ASSERT(inp); 2706 INP_HASH_WLOCK_ASSERT(pcbinfo); 2707 KASSERT(inp->inp_flags & INP_INHASHLIST, 2708 ("%s: !INP_INHASHLIST", __func__)); 2709 KASSERT(inp->inp_smr == SMR_SEQ_INVALID, 2710 ("%s: inp was disconnected", __func__)); 2711 2712 #ifdef INET6 2713 if (inp->inp_vflag & INP_IPV6) { 2714 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport, 2715 inp->inp_fport, pcbinfo->ipi_hashmask); 2716 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr); 2717 } else 2718 #endif 2719 { 2720 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport, 2721 inp->inp_fport, pcbinfo->ipi_hashmask); 2722 connected = !in_nullhost(inp->inp_faddr); 2723 } 2724 2725 /* 2726 * When rehashing, the caller must ensure that either the new or the old 2727 * foreign address was unspecified. 2728 */ 2729 if (connected) 2730 CK_LIST_REMOVE(inp, inp_hash_wild); 2731 else 2732 CK_LIST_REMOVE(inp, inp_hash_exact); 2733 2734 if (connected) { 2735 head = &pcbinfo->ipi_hash_exact[hash]; 2736 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact); 2737 } else { 2738 head = &pcbinfo->ipi_hash_wild[hash]; 2739 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild); 2740 } 2741 } 2742 2743 /* 2744 * Check for alternatives when higher level complains 2745 * about service problems. For now, invalidate cached 2746 * routing information. If the route was created dynamically 2747 * (by a redirect), time to try a default gateway again. 2748 */ 2749 void 2750 in_losing(struct inpcb *inp) 2751 { 2752 2753 RO_INVALIDATE_CACHE(&inp->inp_route); 2754 return; 2755 } 2756 2757 /* 2758 * A set label operation has occurred at the socket layer, propagate the 2759 * label change into the in_pcb for the socket. 2760 */ 2761 void 2762 in_pcbsosetlabel(struct socket *so) 2763 { 2764 #ifdef MAC 2765 struct inpcb *inp; 2766 2767 inp = sotoinpcb(so); 2768 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2769 2770 INP_WLOCK(inp); 2771 SOCK_LOCK(so); 2772 mac_inpcb_sosetlabel(so, inp); 2773 SOCK_UNLOCK(so); 2774 INP_WUNLOCK(inp); 2775 #endif 2776 } 2777 2778 void 2779 inp_wlock(struct inpcb *inp) 2780 { 2781 2782 INP_WLOCK(inp); 2783 } 2784 2785 void 2786 inp_wunlock(struct inpcb *inp) 2787 { 2788 2789 INP_WUNLOCK(inp); 2790 } 2791 2792 void 2793 inp_rlock(struct inpcb *inp) 2794 { 2795 2796 INP_RLOCK(inp); 2797 } 2798 2799 void 2800 inp_runlock(struct inpcb *inp) 2801 { 2802 2803 INP_RUNLOCK(inp); 2804 } 2805 2806 #ifdef INVARIANT_SUPPORT 2807 void 2808 inp_lock_assert(struct inpcb *inp) 2809 { 2810 2811 INP_WLOCK_ASSERT(inp); 2812 } 2813 2814 void 2815 inp_unlock_assert(struct inpcb *inp) 2816 { 2817 2818 INP_UNLOCK_ASSERT(inp); 2819 } 2820 #endif 2821 2822 void 2823 inp_apply_all(struct inpcbinfo *pcbinfo, 2824 void (*func)(struct inpcb *, void *), void *arg) 2825 { 2826 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2827 INPLOOKUP_WLOCKPCB); 2828 struct inpcb *inp; 2829 2830 while ((inp = inp_next(&inpi)) != NULL) 2831 func(inp, arg); 2832 } 2833 2834 struct socket * 2835 inp_inpcbtosocket(struct inpcb *inp) 2836 { 2837 2838 INP_WLOCK_ASSERT(inp); 2839 return (inp->inp_socket); 2840 } 2841 2842 void 2843 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2844 uint32_t *faddr, uint16_t *fp) 2845 { 2846 2847 INP_LOCK_ASSERT(inp); 2848 *laddr = inp->inp_laddr.s_addr; 2849 *faddr = inp->inp_faddr.s_addr; 2850 *lp = inp->inp_lport; 2851 *fp = inp->inp_fport; 2852 } 2853 2854 /* 2855 * Create an external-format (``xinpcb'') structure using the information in 2856 * the kernel-format in_pcb structure pointed to by inp. This is done to 2857 * reduce the spew of irrelevant information over this interface, to isolate 2858 * user code from changes in the kernel structure, and potentially to provide 2859 * information-hiding if we decide that some of this information should be 2860 * hidden from users. 2861 */ 2862 void 2863 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2864 { 2865 2866 bzero(xi, sizeof(*xi)); 2867 xi->xi_len = sizeof(struct xinpcb); 2868 if (inp->inp_socket) 2869 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2870 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2871 xi->inp_gencnt = inp->inp_gencnt; 2872 xi->inp_flow = inp->inp_flow; 2873 xi->inp_flowid = inp->inp_flowid; 2874 xi->inp_flowtype = inp->inp_flowtype; 2875 xi->inp_flags = inp->inp_flags; 2876 xi->inp_flags2 = inp->inp_flags2; 2877 xi->in6p_cksum = inp->in6p_cksum; 2878 xi->in6p_hops = inp->in6p_hops; 2879 xi->inp_ip_tos = inp->inp_ip_tos; 2880 xi->inp_vflag = inp->inp_vflag; 2881 xi->inp_ip_ttl = inp->inp_ip_ttl; 2882 xi->inp_ip_p = inp->inp_ip_p; 2883 xi->inp_ip_minttl = inp->inp_ip_minttl; 2884 } 2885 2886 int 2887 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo, 2888 int (*ctloutput_set)(struct inpcb *, struct sockopt *)) 2889 { 2890 struct sockopt sopt; 2891 struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo, 2892 INPLOOKUP_WLOCKPCB); 2893 struct inpcb *inp; 2894 struct sockopt_parameters *params; 2895 struct socket *so; 2896 int error; 2897 char buf[1024]; 2898 2899 if (req->oldptr != NULL || req->oldlen != 0) 2900 return (EINVAL); 2901 if (req->newptr == NULL) 2902 return (EPERM); 2903 if (req->newlen > sizeof(buf)) 2904 return (ENOMEM); 2905 error = SYSCTL_IN(req, buf, req->newlen); 2906 if (error != 0) 2907 return (error); 2908 if (req->newlen < sizeof(struct sockopt_parameters)) 2909 return (EINVAL); 2910 params = (struct sockopt_parameters *)buf; 2911 sopt.sopt_level = params->sop_level; 2912 sopt.sopt_name = params->sop_optname; 2913 sopt.sopt_dir = SOPT_SET; 2914 sopt.sopt_val = params->sop_optval; 2915 sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters); 2916 sopt.sopt_td = NULL; 2917 #ifdef INET6 2918 if (params->sop_inc.inc_flags & INC_ISIPV6) { 2919 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_laddr)) 2920 params->sop_inc.inc6_laddr.s6_addr16[1] = 2921 htons(params->sop_inc.inc6_zoneid & 0xffff); 2922 if (IN6_IS_SCOPE_LINKLOCAL(¶ms->sop_inc.inc6_faddr)) 2923 params->sop_inc.inc6_faddr.s6_addr16[1] = 2924 htons(params->sop_inc.inc6_zoneid & 0xffff); 2925 } 2926 #endif 2927 if (params->sop_inc.inc_lport != htons(0) && 2928 params->sop_inc.inc_fport != htons(0)) { 2929 #ifdef INET6 2930 if (params->sop_inc.inc_flags & INC_ISIPV6) 2931 inpi.hash = INP6_PCBHASH( 2932 ¶ms->sop_inc.inc6_faddr, 2933 params->sop_inc.inc_lport, 2934 params->sop_inc.inc_fport, 2935 pcbinfo->ipi_hashmask); 2936 else 2937 #endif 2938 inpi.hash = INP_PCBHASH( 2939 ¶ms->sop_inc.inc_faddr, 2940 params->sop_inc.inc_lport, 2941 params->sop_inc.inc_fport, 2942 pcbinfo->ipi_hashmask); 2943 } 2944 while ((inp = inp_next(&inpi)) != NULL) 2945 if (inp->inp_gencnt == params->sop_id) { 2946 if (inp->inp_flags & INP_DROPPED) { 2947 INP_WUNLOCK(inp); 2948 return (ECONNRESET); 2949 } 2950 so = inp->inp_socket; 2951 KASSERT(so != NULL, ("inp_socket == NULL")); 2952 soref(so); 2953 if (params->sop_level == SOL_SOCKET) { 2954 INP_WUNLOCK(inp); 2955 error = sosetopt(so, &sopt); 2956 } else 2957 error = (*ctloutput_set)(inp, &sopt); 2958 sorele(so); 2959 break; 2960 } 2961 if (inp == NULL) 2962 error = ESRCH; 2963 return (error); 2964 } 2965 2966 #ifdef DDB 2967 static void 2968 db_print_indent(int indent) 2969 { 2970 int i; 2971 2972 for (i = 0; i < indent; i++) 2973 db_printf(" "); 2974 } 2975 2976 static void 2977 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2978 { 2979 char faddr_str[48], laddr_str[48]; 2980 2981 db_print_indent(indent); 2982 db_printf("%s at %p\n", name, inc); 2983 2984 indent += 2; 2985 2986 #ifdef INET6 2987 if (inc->inc_flags & INC_ISIPV6) { 2988 /* IPv6. */ 2989 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2990 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2991 } else 2992 #endif 2993 { 2994 /* IPv4. */ 2995 inet_ntoa_r(inc->inc_laddr, laddr_str); 2996 inet_ntoa_r(inc->inc_faddr, faddr_str); 2997 } 2998 db_print_indent(indent); 2999 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3000 ntohs(inc->inc_lport)); 3001 db_print_indent(indent); 3002 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3003 ntohs(inc->inc_fport)); 3004 } 3005 3006 static void 3007 db_print_inpflags(int inp_flags) 3008 { 3009 int comma; 3010 3011 comma = 0; 3012 if (inp_flags & INP_RECVOPTS) { 3013 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3014 comma = 1; 3015 } 3016 if (inp_flags & INP_RECVRETOPTS) { 3017 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3018 comma = 1; 3019 } 3020 if (inp_flags & INP_RECVDSTADDR) { 3021 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3022 comma = 1; 3023 } 3024 if (inp_flags & INP_ORIGDSTADDR) { 3025 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3026 comma = 1; 3027 } 3028 if (inp_flags & INP_HDRINCL) { 3029 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3030 comma = 1; 3031 } 3032 if (inp_flags & INP_HIGHPORT) { 3033 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3034 comma = 1; 3035 } 3036 if (inp_flags & INP_LOWPORT) { 3037 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3038 comma = 1; 3039 } 3040 if (inp_flags & INP_ANONPORT) { 3041 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3042 comma = 1; 3043 } 3044 if (inp_flags & INP_RECVIF) { 3045 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3046 comma = 1; 3047 } 3048 if (inp_flags & INP_MTUDISC) { 3049 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3050 comma = 1; 3051 } 3052 if (inp_flags & INP_RECVTTL) { 3053 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3054 comma = 1; 3055 } 3056 if (inp_flags & INP_DONTFRAG) { 3057 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3058 comma = 1; 3059 } 3060 if (inp_flags & INP_RECVTOS) { 3061 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3062 comma = 1; 3063 } 3064 if (inp_flags & IN6P_IPV6_V6ONLY) { 3065 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3066 comma = 1; 3067 } 3068 if (inp_flags & IN6P_PKTINFO) { 3069 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3070 comma = 1; 3071 } 3072 if (inp_flags & IN6P_HOPLIMIT) { 3073 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3074 comma = 1; 3075 } 3076 if (inp_flags & IN6P_HOPOPTS) { 3077 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3078 comma = 1; 3079 } 3080 if (inp_flags & IN6P_DSTOPTS) { 3081 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3082 comma = 1; 3083 } 3084 if (inp_flags & IN6P_RTHDR) { 3085 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3086 comma = 1; 3087 } 3088 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3089 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3090 comma = 1; 3091 } 3092 if (inp_flags & IN6P_TCLASS) { 3093 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3094 comma = 1; 3095 } 3096 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3097 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3098 comma = 1; 3099 } 3100 if (inp_flags & INP_ONESBCAST) { 3101 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3102 comma = 1; 3103 } 3104 if (inp_flags & INP_DROPPED) { 3105 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3106 comma = 1; 3107 } 3108 if (inp_flags & INP_SOCKREF) { 3109 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3110 comma = 1; 3111 } 3112 if (inp_flags & IN6P_RFC2292) { 3113 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3114 comma = 1; 3115 } 3116 if (inp_flags & IN6P_MTU) { 3117 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3118 comma = 1; 3119 } 3120 } 3121 3122 static void 3123 db_print_inpvflag(u_char inp_vflag) 3124 { 3125 int comma; 3126 3127 comma = 0; 3128 if (inp_vflag & INP_IPV4) { 3129 db_printf("%sINP_IPV4", comma ? ", " : ""); 3130 comma = 1; 3131 } 3132 if (inp_vflag & INP_IPV6) { 3133 db_printf("%sINP_IPV6", comma ? ", " : ""); 3134 comma = 1; 3135 } 3136 if (inp_vflag & INP_IPV6PROTO) { 3137 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3138 comma = 1; 3139 } 3140 } 3141 3142 static void 3143 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3144 { 3145 3146 db_print_indent(indent); 3147 db_printf("%s at %p\n", name, inp); 3148 3149 indent += 2; 3150 3151 db_print_indent(indent); 3152 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3153 3154 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3155 3156 db_print_indent(indent); 3157 db_printf("inp_label: %p inp_flags: 0x%x (", 3158 inp->inp_label, inp->inp_flags); 3159 db_print_inpflags(inp->inp_flags); 3160 db_printf(")\n"); 3161 3162 db_print_indent(indent); 3163 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3164 inp->inp_vflag); 3165 db_print_inpvflag(inp->inp_vflag); 3166 db_printf(")\n"); 3167 3168 db_print_indent(indent); 3169 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3170 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3171 3172 db_print_indent(indent); 3173 #ifdef INET6 3174 if (inp->inp_vflag & INP_IPV6) { 3175 db_printf("in6p_options: %p in6p_outputopts: %p " 3176 "in6p_moptions: %p\n", inp->in6p_options, 3177 inp->in6p_outputopts, inp->in6p_moptions); 3178 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3179 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3180 inp->in6p_hops); 3181 } else 3182 #endif 3183 { 3184 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3185 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3186 inp->inp_options, inp->inp_moptions); 3187 } 3188 3189 db_print_indent(indent); 3190 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3191 (uintmax_t)inp->inp_gencnt); 3192 } 3193 3194 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3195 { 3196 struct inpcb *inp; 3197 3198 if (!have_addr) { 3199 db_printf("usage: show inpcb <addr>\n"); 3200 return; 3201 } 3202 inp = (struct inpcb *)addr; 3203 3204 db_print_inpcb(inp, "inpcb", 0); 3205 } 3206 #endif /* DDB */ 3207 3208 #ifdef RATELIMIT 3209 /* 3210 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3211 * if any. 3212 */ 3213 int 3214 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3215 { 3216 union if_snd_tag_modify_params params = { 3217 .rate_limit.max_rate = max_pacing_rate, 3218 .rate_limit.flags = M_NOWAIT, 3219 }; 3220 struct m_snd_tag *mst; 3221 int error; 3222 3223 mst = inp->inp_snd_tag; 3224 if (mst == NULL) 3225 return (EINVAL); 3226 3227 if (mst->sw->snd_tag_modify == NULL) { 3228 error = EOPNOTSUPP; 3229 } else { 3230 error = mst->sw->snd_tag_modify(mst, ¶ms); 3231 } 3232 return (error); 3233 } 3234 3235 /* 3236 * Query existing TX rate limit based on the existing 3237 * "inp->inp_snd_tag", if any. 3238 */ 3239 int 3240 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3241 { 3242 union if_snd_tag_query_params params = { }; 3243 struct m_snd_tag *mst; 3244 int error; 3245 3246 mst = inp->inp_snd_tag; 3247 if (mst == NULL) 3248 return (EINVAL); 3249 3250 if (mst->sw->snd_tag_query == NULL) { 3251 error = EOPNOTSUPP; 3252 } else { 3253 error = mst->sw->snd_tag_query(mst, ¶ms); 3254 if (error == 0 && p_max_pacing_rate != NULL) 3255 *p_max_pacing_rate = params.rate_limit.max_rate; 3256 } 3257 return (error); 3258 } 3259 3260 /* 3261 * Query existing TX queue level based on the existing 3262 * "inp->inp_snd_tag", if any. 3263 */ 3264 int 3265 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3266 { 3267 union if_snd_tag_query_params params = { }; 3268 struct m_snd_tag *mst; 3269 int error; 3270 3271 mst = inp->inp_snd_tag; 3272 if (mst == NULL) 3273 return (EINVAL); 3274 3275 if (mst->sw->snd_tag_query == NULL) 3276 return (EOPNOTSUPP); 3277 3278 error = mst->sw->snd_tag_query(mst, ¶ms); 3279 if (error == 0 && p_txqueue_level != NULL) 3280 *p_txqueue_level = params.rate_limit.queue_level; 3281 return (error); 3282 } 3283 3284 /* 3285 * Allocate a new TX rate limit send tag from the network interface 3286 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3287 */ 3288 int 3289 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3290 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3291 3292 { 3293 union if_snd_tag_alloc_params params = { 3294 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3295 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3296 .rate_limit.hdr.flowid = flowid, 3297 .rate_limit.hdr.flowtype = flowtype, 3298 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3299 .rate_limit.max_rate = max_pacing_rate, 3300 .rate_limit.flags = M_NOWAIT, 3301 }; 3302 int error; 3303 3304 INP_WLOCK_ASSERT(inp); 3305 3306 /* 3307 * If there is already a send tag, or the INP is being torn 3308 * down, allocating a new send tag is not allowed. Else send 3309 * tags may leak. 3310 */ 3311 if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0) 3312 return (EINVAL); 3313 3314 error = m_snd_tag_alloc(ifp, ¶ms, st); 3315 #ifdef INET 3316 if (error == 0) { 3317 counter_u64_add(rate_limit_set_ok, 1); 3318 counter_u64_add(rate_limit_active, 1); 3319 } else if (error != EOPNOTSUPP) 3320 counter_u64_add(rate_limit_alloc_fail, 1); 3321 #endif 3322 return (error); 3323 } 3324 3325 void 3326 in_pcbdetach_tag(struct m_snd_tag *mst) 3327 { 3328 3329 m_snd_tag_rele(mst); 3330 #ifdef INET 3331 counter_u64_add(rate_limit_active, -1); 3332 #endif 3333 } 3334 3335 /* 3336 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3337 * if any: 3338 */ 3339 void 3340 in_pcbdetach_txrtlmt(struct inpcb *inp) 3341 { 3342 struct m_snd_tag *mst; 3343 3344 INP_WLOCK_ASSERT(inp); 3345 3346 mst = inp->inp_snd_tag; 3347 inp->inp_snd_tag = NULL; 3348 3349 if (mst == NULL) 3350 return; 3351 3352 m_snd_tag_rele(mst); 3353 #ifdef INET 3354 counter_u64_add(rate_limit_active, -1); 3355 #endif 3356 } 3357 3358 int 3359 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3360 { 3361 int error; 3362 3363 /* 3364 * If the existing send tag is for the wrong interface due to 3365 * a route change, first drop the existing tag. Set the 3366 * CHANGED flag so that we will keep trying to allocate a new 3367 * tag if we fail to allocate one this time. 3368 */ 3369 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3370 in_pcbdetach_txrtlmt(inp); 3371 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3372 } 3373 3374 /* 3375 * NOTE: When attaching to a network interface a reference is 3376 * made to ensure the network interface doesn't go away until 3377 * all ratelimit connections are gone. The network interface 3378 * pointers compared below represent valid network interfaces, 3379 * except when comparing towards NULL. 3380 */ 3381 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3382 error = 0; 3383 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3384 if (inp->inp_snd_tag != NULL) 3385 in_pcbdetach_txrtlmt(inp); 3386 error = 0; 3387 } else if (inp->inp_snd_tag == NULL) { 3388 /* 3389 * In order to utilize packet pacing with RSS, we need 3390 * to wait until there is a valid RSS hash before we 3391 * can proceed: 3392 */ 3393 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3394 error = EAGAIN; 3395 } else { 3396 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3397 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3398 } 3399 } else { 3400 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3401 } 3402 if (error == 0 || error == EOPNOTSUPP) 3403 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3404 3405 return (error); 3406 } 3407 3408 /* 3409 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3410 * is set in the fast path and will attach/detach/modify the TX rate 3411 * limit send tag based on the socket's so_max_pacing_rate value. 3412 */ 3413 void 3414 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3415 { 3416 struct socket *socket; 3417 uint32_t max_pacing_rate; 3418 bool did_upgrade; 3419 3420 if (inp == NULL) 3421 return; 3422 3423 socket = inp->inp_socket; 3424 if (socket == NULL) 3425 return; 3426 3427 if (!INP_WLOCKED(inp)) { 3428 /* 3429 * NOTE: If the write locking fails, we need to bail 3430 * out and use the non-ratelimited ring for the 3431 * transmit until there is a new chance to get the 3432 * write lock. 3433 */ 3434 if (!INP_TRY_UPGRADE(inp)) 3435 return; 3436 did_upgrade = 1; 3437 } else { 3438 did_upgrade = 0; 3439 } 3440 3441 /* 3442 * NOTE: The so_max_pacing_rate value is read unlocked, 3443 * because atomic updates are not required since the variable 3444 * is checked at every mbuf we send. It is assumed that the 3445 * variable read itself will be atomic. 3446 */ 3447 max_pacing_rate = socket->so_max_pacing_rate; 3448 3449 in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3450 3451 if (did_upgrade) 3452 INP_DOWNGRADE(inp); 3453 } 3454 3455 /* 3456 * Track route changes for TX rate limiting. 3457 */ 3458 void 3459 in_pcboutput_eagain(struct inpcb *inp) 3460 { 3461 bool did_upgrade; 3462 3463 if (inp == NULL) 3464 return; 3465 3466 if (inp->inp_snd_tag == NULL) 3467 return; 3468 3469 if (!INP_WLOCKED(inp)) { 3470 /* 3471 * NOTE: If the write locking fails, we need to bail 3472 * out and use the non-ratelimited ring for the 3473 * transmit until there is a new chance to get the 3474 * write lock. 3475 */ 3476 if (!INP_TRY_UPGRADE(inp)) 3477 return; 3478 did_upgrade = 1; 3479 } else { 3480 did_upgrade = 0; 3481 } 3482 3483 /* detach rate limiting */ 3484 in_pcbdetach_txrtlmt(inp); 3485 3486 /* make sure new mbuf send tag allocation is made */ 3487 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3488 3489 if (did_upgrade) 3490 INP_DOWNGRADE(inp); 3491 } 3492 3493 #ifdef INET 3494 static void 3495 rl_init(void *st) 3496 { 3497 rate_limit_new = counter_u64_alloc(M_WAITOK); 3498 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3499 rate_limit_active = counter_u64_alloc(M_WAITOK); 3500 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3501 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3502 } 3503 3504 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3505 #endif 3506 #endif /* RATELIMIT */ 3507