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 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_ddb.h" 44 #include "opt_ipsec.h" 45 #include "opt_inet.h" 46 #include "opt_inet6.h" 47 #include "opt_ratelimit.h" 48 #include "opt_pcbgroup.h" 49 #include "opt_route.h" 50 #include "opt_rss.h" 51 52 #include <sys/param.h> 53 #include <sys/systm.h> 54 #include <sys/lock.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/callout.h> 58 #include <sys/eventhandler.h> 59 #include <sys/domain.h> 60 #include <sys/protosw.h> 61 #include <sys/rmlock.h> 62 #include <sys/smp.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/sockio.h> 66 #include <sys/priv.h> 67 #include <sys/proc.h> 68 #include <sys/refcount.h> 69 #include <sys/jail.h> 70 #include <sys/kernel.h> 71 #include <sys/sysctl.h> 72 73 #ifdef DDB 74 #include <ddb/ddb.h> 75 #endif 76 77 #include <vm/uma.h> 78 #include <vm/vm.h> 79 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/if_types.h> 83 #include <net/if_llatbl.h> 84 #include <net/route.h> 85 #include <net/rss_config.h> 86 #include <net/vnet.h> 87 88 #if defined(INET) || defined(INET6) 89 #include <netinet/in.h> 90 #include <netinet/in_pcb.h> 91 #ifdef INET 92 #include <netinet/in_var.h> 93 #include <netinet/in_fib.h> 94 #endif 95 #include <netinet/ip_var.h> 96 #include <netinet/tcp_var.h> 97 #ifdef TCPHPTS 98 #include <netinet/tcp_hpts.h> 99 #endif 100 #include <netinet/udp.h> 101 #include <netinet/udp_var.h> 102 #ifdef INET6 103 #include <netinet/ip6.h> 104 #include <netinet6/in6_pcb.h> 105 #include <netinet6/in6_var.h> 106 #include <netinet6/ip6_var.h> 107 #endif /* INET6 */ 108 #include <net/route/nhop.h> 109 #endif 110 111 #include <netipsec/ipsec_support.h> 112 113 #include <security/mac/mac_framework.h> 114 115 #define INPCBLBGROUP_SIZMIN 8 116 #define INPCBLBGROUP_SIZMAX 256 117 118 static struct callout ipport_tick_callout; 119 120 /* 121 * These configure the range of local port addresses assigned to 122 * "unspecified" outgoing connections/packets/whatever. 123 */ 124 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 125 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 126 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 127 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 128 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 129 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 130 131 /* 132 * Reserved ports accessible only to root. There are significant 133 * security considerations that must be accounted for when changing these, 134 * but the security benefits can be great. Please be careful. 135 */ 136 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 137 VNET_DEFINE(int, ipport_reservedlow); 138 139 /* Variables dealing with random ephemeral port allocation. */ 140 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ 141 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ 142 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ 143 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ 144 VNET_DEFINE(int, ipport_tcpallocs); 145 VNET_DEFINE_STATIC(int, ipport_tcplastcount); 146 147 #define V_ipport_tcplastcount VNET(ipport_tcplastcount) 148 149 static void in_pcbremlists(struct inpcb *inp); 150 #ifdef INET 151 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 152 struct in_addr faddr, u_int fport_arg, 153 struct in_addr laddr, u_int lport_arg, 154 int lookupflags, struct ifnet *ifp, 155 uint8_t numa_domain); 156 157 #define RANGECHK(var, min, max) \ 158 if ((var) < (min)) { (var) = (min); } \ 159 else if ((var) > (max)) { (var) = (max); } 160 161 static int 162 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 163 { 164 int error; 165 166 error = sysctl_handle_int(oidp, arg1, arg2, req); 167 if (error == 0) { 168 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 169 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 170 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 171 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 172 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 173 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 174 } 175 return (error); 176 } 177 178 #undef RANGECHK 179 180 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, 181 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 182 "IP Ports"); 183 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 186 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", 187 ""); 188 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 189 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 190 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", 191 ""); 192 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 193 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 194 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", 195 ""); 196 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 197 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 198 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", 199 ""); 200 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 201 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 202 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", 203 ""); 204 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 205 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 206 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", 207 ""); 208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 209 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 210 &VNET_NAME(ipport_reservedhigh), 0, ""); 211 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 212 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 213 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 214 CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 216 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, 217 CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " 219 "allocations before switching to a sequental one"); 220 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, 221 CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(ipport_randomtime), 0, 223 "Minimum time to keep sequental port " 224 "allocation before switching to a random one"); 225 226 #ifdef RATELIMIT 227 counter_u64_t rate_limit_new; 228 counter_u64_t rate_limit_chg; 229 counter_u64_t rate_limit_active; 230 counter_u64_t rate_limit_alloc_fail; 231 counter_u64_t rate_limit_set_ok; 232 233 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 234 "IP Rate Limiting"); 235 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, 236 &rate_limit_active, "Active rate limited connections"); 237 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, 238 &rate_limit_alloc_fail, "Rate limited connection failures"); 239 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, 240 &rate_limit_set_ok, "Rate limited setting succeeded"); 241 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD, 242 &rate_limit_new, "Total Rate limit new attempts"); 243 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD, 244 &rate_limit_chg, "Total Rate limited change attempts"); 245 246 #endif /* RATELIMIT */ 247 248 #endif /* INET */ 249 250 /* 251 * in_pcb.c: manage the Protocol Control Blocks. 252 * 253 * NOTE: It is assumed that most of these functions will be called with 254 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 255 * functions often modify hash chains or addresses in pcbs. 256 */ 257 258 static struct inpcblbgroup * 259 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag, 260 uint16_t port, const union in_dependaddr *addr, int size, 261 uint8_t numa_domain) 262 { 263 struct inpcblbgroup *grp; 264 size_t bytes; 265 266 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 267 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 268 if (!grp) 269 return (NULL); 270 grp->il_vflag = vflag; 271 grp->il_lport = port; 272 grp->il_numa_domain = numa_domain; 273 grp->il_dependladdr = *addr; 274 grp->il_inpsiz = size; 275 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 276 return (grp); 277 } 278 279 static void 280 in_pcblbgroup_free_deferred(epoch_context_t ctx) 281 { 282 struct inpcblbgroup *grp; 283 284 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 285 free(grp, M_PCB); 286 } 287 288 static void 289 in_pcblbgroup_free(struct inpcblbgroup *grp) 290 { 291 292 CK_LIST_REMOVE(grp, il_list); 293 NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); 294 } 295 296 static struct inpcblbgroup * 297 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 298 struct inpcblbgroup *old_grp, int size) 299 { 300 struct inpcblbgroup *grp; 301 int i; 302 303 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, 304 old_grp->il_lport, &old_grp->il_dependladdr, size, 305 old_grp->il_numa_domain); 306 if (grp == NULL) 307 return (NULL); 308 309 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 310 ("invalid new local group size %d and old local group count %d", 311 grp->il_inpsiz, old_grp->il_inpcnt)); 312 313 for (i = 0; i < old_grp->il_inpcnt; ++i) 314 grp->il_inp[i] = old_grp->il_inp[i]; 315 grp->il_inpcnt = old_grp->il_inpcnt; 316 in_pcblbgroup_free(old_grp); 317 return (grp); 318 } 319 320 /* 321 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 322 * and shrink group if possible. 323 */ 324 static void 325 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 326 int i) 327 { 328 struct inpcblbgroup *grp, *new_grp; 329 330 grp = *grpp; 331 for (; i + 1 < grp->il_inpcnt; ++i) 332 grp->il_inp[i] = grp->il_inp[i + 1]; 333 grp->il_inpcnt--; 334 335 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 336 grp->il_inpcnt <= grp->il_inpsiz / 4) { 337 /* Shrink this group. */ 338 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 339 if (new_grp != NULL) 340 *grpp = new_grp; 341 } 342 } 343 344 /* 345 * Add PCB to load balance group for SO_REUSEPORT_LB option. 346 */ 347 static int 348 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain) 349 { 350 const static struct timeval interval = { 60, 0 }; 351 static struct timeval lastprint; 352 struct inpcbinfo *pcbinfo; 353 struct inpcblbgrouphead *hdr; 354 struct inpcblbgroup *grp; 355 uint32_t idx; 356 357 pcbinfo = inp->inp_pcbinfo; 358 359 INP_WLOCK_ASSERT(inp); 360 INP_HASH_WLOCK_ASSERT(pcbinfo); 361 362 /* 363 * Don't allow jailed socket to join local group. 364 */ 365 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred)) 366 return (0); 367 368 #ifdef INET6 369 /* 370 * Don't allow IPv4 mapped INET6 wild socket. 371 */ 372 if ((inp->inp_vflag & INP_IPV4) && 373 inp->inp_laddr.s_addr == INADDR_ANY && 374 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 375 return (0); 376 } 377 #endif 378 379 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 380 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 381 CK_LIST_FOREACH(grp, hdr, il_list) { 382 if (grp->il_vflag == inp->inp_vflag && 383 grp->il_lport == inp->inp_lport && 384 grp->il_numa_domain == numa_domain && 385 memcmp(&grp->il_dependladdr, 386 &inp->inp_inc.inc_ie.ie_dependladdr, 387 sizeof(grp->il_dependladdr)) == 0) 388 break; 389 } 390 if (grp == NULL) { 391 /* Create new load balance group. */ 392 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, 393 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 394 INPCBLBGROUP_SIZMIN, numa_domain); 395 if (grp == NULL) 396 return (ENOBUFS); 397 } else if (grp->il_inpcnt == grp->il_inpsiz) { 398 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 399 if (ratecheck(&lastprint, &interval)) 400 printf("lb group port %d, limit reached\n", 401 ntohs(grp->il_lport)); 402 return (0); 403 } 404 405 /* Expand this local group. */ 406 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 407 if (grp == NULL) 408 return (ENOBUFS); 409 } 410 411 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 412 ("invalid local group size %d and count %d", grp->il_inpsiz, 413 grp->il_inpcnt)); 414 415 grp->il_inp[grp->il_inpcnt] = inp; 416 grp->il_inpcnt++; 417 return (0); 418 } 419 420 /* 421 * Remove PCB from load balance group. 422 */ 423 static void 424 in_pcbremlbgrouphash(struct inpcb *inp) 425 { 426 struct inpcbinfo *pcbinfo; 427 struct inpcblbgrouphead *hdr; 428 struct inpcblbgroup *grp; 429 int i; 430 431 pcbinfo = inp->inp_pcbinfo; 432 433 INP_WLOCK_ASSERT(inp); 434 INP_HASH_WLOCK_ASSERT(pcbinfo); 435 436 hdr = &pcbinfo->ipi_lbgrouphashbase[ 437 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 438 CK_LIST_FOREACH(grp, hdr, il_list) { 439 for (i = 0; i < grp->il_inpcnt; ++i) { 440 if (grp->il_inp[i] != inp) 441 continue; 442 443 if (grp->il_inpcnt == 1) { 444 /* We are the last, free this local group. */ 445 in_pcblbgroup_free(grp); 446 } else { 447 /* Pull up inpcbs, shrink group if possible. */ 448 in_pcblbgroup_reorder(hdr, &grp, i); 449 } 450 return; 451 } 452 } 453 } 454 455 int 456 in_pcblbgroup_numa(struct inpcb *inp, int arg) 457 { 458 struct inpcbinfo *pcbinfo; 459 struct inpcblbgrouphead *hdr; 460 struct inpcblbgroup *grp; 461 int err, i; 462 uint8_t numa_domain; 463 464 switch (arg) { 465 case TCP_REUSPORT_LB_NUMA_NODOM: 466 numa_domain = M_NODOM; 467 break; 468 case TCP_REUSPORT_LB_NUMA_CURDOM: 469 numa_domain = PCPU_GET(domain); 470 break; 471 default: 472 if (arg < 0 || arg >= vm_ndomains) 473 return (EINVAL); 474 numa_domain = arg; 475 } 476 477 err = 0; 478 pcbinfo = inp->inp_pcbinfo; 479 INP_WLOCK_ASSERT(inp); 480 INP_HASH_WLOCK(pcbinfo); 481 hdr = &pcbinfo->ipi_lbgrouphashbase[ 482 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 483 CK_LIST_FOREACH(grp, hdr, il_list) { 484 for (i = 0; i < grp->il_inpcnt; ++i) { 485 if (grp->il_inp[i] != inp) 486 continue; 487 488 if (grp->il_numa_domain == numa_domain) { 489 goto abort_with_hash_wlock; 490 } 491 492 /* Remove it from the old group. */ 493 in_pcbremlbgrouphash(inp); 494 495 /* Add it to the new group based on numa domain. */ 496 in_pcbinslbgrouphash(inp, numa_domain); 497 goto abort_with_hash_wlock; 498 } 499 } 500 err = ENOENT; 501 abort_with_hash_wlock: 502 INP_HASH_WUNLOCK(pcbinfo); 503 return (err); 504 } 505 506 /* 507 * Different protocols initialize their inpcbs differently - giving 508 * different name to the lock. But they all are disposed the same. 509 */ 510 static void 511 inpcb_fini(void *mem, int size) 512 { 513 struct inpcb *inp = mem; 514 515 INP_LOCK_DESTROY(inp); 516 } 517 518 /* 519 * Initialize an inpcbinfo -- we should be able to reduce the number of 520 * arguments in time. 521 */ 522 void 523 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 524 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 525 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) 526 { 527 528 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 529 530 INP_INFO_LOCK_INIT(pcbinfo, name); 531 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 532 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 533 #ifdef VIMAGE 534 pcbinfo->ipi_vnet = curvnet; 535 #endif 536 pcbinfo->ipi_listhead = listhead; 537 CK_LIST_INIT(pcbinfo->ipi_listhead); 538 pcbinfo->ipi_count = 0; 539 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 540 &pcbinfo->ipi_hashmask); 541 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 542 &pcbinfo->ipi_porthashmask); 543 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 544 &pcbinfo->ipi_lbgrouphashmask); 545 #ifdef PCBGROUP 546 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 547 #endif 548 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 549 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); 550 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 551 uma_zone_set_warning(pcbinfo->ipi_zone, 552 "kern.ipc.maxsockets limit reached"); 553 } 554 555 /* 556 * Destroy an inpcbinfo. 557 */ 558 void 559 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 560 { 561 562 KASSERT(pcbinfo->ipi_count == 0, 563 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 564 565 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 566 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 567 pcbinfo->ipi_porthashmask); 568 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 569 pcbinfo->ipi_lbgrouphashmask); 570 #ifdef PCBGROUP 571 in_pcbgroup_destroy(pcbinfo); 572 #endif 573 uma_zdestroy(pcbinfo->ipi_zone); 574 INP_LIST_LOCK_DESTROY(pcbinfo); 575 INP_HASH_LOCK_DESTROY(pcbinfo); 576 INP_INFO_LOCK_DESTROY(pcbinfo); 577 } 578 579 /* 580 * Allocate a PCB and associate it with the socket. 581 * On success return with the PCB locked. 582 */ 583 int 584 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 585 { 586 struct inpcb *inp; 587 int error; 588 589 error = 0; 590 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 591 if (inp == NULL) 592 return (ENOBUFS); 593 bzero(&inp->inp_start_zero, inp_zero_size); 594 #ifdef NUMA 595 inp->inp_numa_domain = M_NODOM; 596 #endif 597 inp->inp_pcbinfo = pcbinfo; 598 inp->inp_socket = so; 599 inp->inp_cred = crhold(so->so_cred); 600 inp->inp_inc.inc_fibnum = so->so_fibnum; 601 #ifdef MAC 602 error = mac_inpcb_init(inp, M_NOWAIT); 603 if (error != 0) 604 goto out; 605 mac_inpcb_create(so, inp); 606 #endif 607 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 608 error = ipsec_init_pcbpolicy(inp); 609 if (error != 0) { 610 #ifdef MAC 611 mac_inpcb_destroy(inp); 612 #endif 613 goto out; 614 } 615 #endif /*IPSEC*/ 616 #ifdef INET6 617 if (INP_SOCKAF(so) == AF_INET6) { 618 inp->inp_vflag |= INP_IPV6PROTO; 619 if (V_ip6_v6only) 620 inp->inp_flags |= IN6P_IPV6_V6ONLY; 621 } 622 #endif 623 INP_WLOCK(inp); 624 INP_LIST_WLOCK(pcbinfo); 625 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 626 pcbinfo->ipi_count++; 627 so->so_pcb = (caddr_t)inp; 628 #ifdef INET6 629 if (V_ip6_auto_flowlabel) 630 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 631 #endif 632 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 633 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 634 635 /* 636 * Routes in inpcb's can cache L2 as well; they are guaranteed 637 * to be cleaned up. 638 */ 639 inp->inp_route.ro_flags = RT_LLE_CACHE; 640 INP_LIST_WUNLOCK(pcbinfo); 641 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 642 out: 643 if (error != 0) { 644 crfree(inp->inp_cred); 645 uma_zfree(pcbinfo->ipi_zone, inp); 646 } 647 #endif 648 return (error); 649 } 650 651 #ifdef INET 652 int 653 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 654 { 655 int anonport, error; 656 657 KASSERT(nam == NULL || nam->sa_family == AF_INET, 658 ("%s: invalid address family for %p", __func__, nam)); 659 KASSERT(nam == NULL || nam->sa_len == sizeof(struct sockaddr_in), 660 ("%s: invalid address length for %p", __func__, nam)); 661 INP_WLOCK_ASSERT(inp); 662 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 663 664 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 665 return (EINVAL); 666 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 667 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 668 &inp->inp_lport, cred); 669 if (error) 670 return (error); 671 if (in_pcbinshash(inp) != 0) { 672 inp->inp_laddr.s_addr = INADDR_ANY; 673 inp->inp_lport = 0; 674 return (EAGAIN); 675 } 676 if (anonport) 677 inp->inp_flags |= INP_ANONPORT; 678 return (0); 679 } 680 #endif 681 682 #if defined(INET) || defined(INET6) 683 /* 684 * Assign a local port like in_pcb_lport(), but also used with connect() 685 * and a foreign address and port. If fsa is non-NULL, choose a local port 686 * that is unused with those, otherwise one that is completely unused. 687 * lsa can be NULL for IPv6. 688 */ 689 int 690 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 691 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 692 { 693 struct inpcbinfo *pcbinfo; 694 struct inpcb *tmpinp; 695 unsigned short *lastport; 696 int count, dorandom, error; 697 u_short aux, first, last, lport; 698 #ifdef INET 699 struct in_addr laddr, faddr; 700 #endif 701 #ifdef INET6 702 struct in6_addr *laddr6, *faddr6; 703 #endif 704 705 pcbinfo = inp->inp_pcbinfo; 706 707 /* 708 * Because no actual state changes occur here, a global write lock on 709 * the pcbinfo isn't required. 710 */ 711 INP_LOCK_ASSERT(inp); 712 INP_HASH_LOCK_ASSERT(pcbinfo); 713 714 if (inp->inp_flags & INP_HIGHPORT) { 715 first = V_ipport_hifirstauto; /* sysctl */ 716 last = V_ipport_hilastauto; 717 lastport = &pcbinfo->ipi_lasthi; 718 } else if (inp->inp_flags & INP_LOWPORT) { 719 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 720 if (error) 721 return (error); 722 first = V_ipport_lowfirstauto; /* 1023 */ 723 last = V_ipport_lowlastauto; /* 600 */ 724 lastport = &pcbinfo->ipi_lastlow; 725 } else { 726 first = V_ipport_firstauto; /* sysctl */ 727 last = V_ipport_lastauto; 728 lastport = &pcbinfo->ipi_lastport; 729 } 730 /* 731 * For UDP(-Lite), use random port allocation as long as the user 732 * allows it. For TCP (and as of yet unknown) connections, 733 * use random port allocation only if the user allows it AND 734 * ipport_tick() allows it. 735 */ 736 if (V_ipport_randomized && 737 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 738 pcbinfo == &V_ulitecbinfo)) 739 dorandom = 1; 740 else 741 dorandom = 0; 742 /* 743 * It makes no sense to do random port allocation if 744 * we have the only port available. 745 */ 746 if (first == last) 747 dorandom = 0; 748 /* Make sure to not include UDP(-Lite) packets in the count. */ 749 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 750 V_ipport_tcpallocs++; 751 /* 752 * Instead of having two loops further down counting up or down 753 * make sure that first is always <= last and go with only one 754 * code path implementing all logic. 755 */ 756 if (first > last) { 757 aux = first; 758 first = last; 759 last = aux; 760 } 761 762 #ifdef INET 763 laddr.s_addr = INADDR_ANY; 764 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 765 if (lsa != NULL) 766 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 767 if (fsa != NULL) 768 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 769 } 770 #endif 771 #ifdef INET6 772 laddr6 = NULL; 773 if ((inp->inp_vflag & INP_IPV6) != 0) { 774 if (lsa != NULL) 775 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 776 if (fsa != NULL) 777 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 778 } 779 #endif 780 781 tmpinp = NULL; 782 lport = *lportp; 783 784 if (dorandom) 785 *lastport = first + (arc4random() % (last - first)); 786 787 count = last - first; 788 789 do { 790 if (count-- < 0) /* completely used? */ 791 return (EADDRNOTAVAIL); 792 ++*lastport; 793 if (*lastport < first || *lastport > last) 794 *lastport = first; 795 lport = htons(*lastport); 796 797 if (fsa != NULL) { 798 #ifdef INET 799 if (lsa->sa_family == AF_INET) { 800 tmpinp = in_pcblookup_hash_locked(pcbinfo, 801 faddr, fport, laddr, lport, lookupflags, 802 NULL, M_NODOM); 803 } 804 #endif 805 #ifdef INET6 806 if (lsa->sa_family == AF_INET6) { 807 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 808 faddr6, fport, laddr6, lport, lookupflags, 809 NULL, M_NODOM); 810 } 811 #endif 812 } else { 813 #ifdef INET6 814 if ((inp->inp_vflag & INP_IPV6) != 0) 815 tmpinp = in6_pcblookup_local(pcbinfo, 816 &inp->in6p_laddr, lport, lookupflags, cred); 817 #endif 818 #if defined(INET) && defined(INET6) 819 else 820 #endif 821 #ifdef INET 822 tmpinp = in_pcblookup_local(pcbinfo, laddr, 823 lport, lookupflags, cred); 824 #endif 825 } 826 } while (tmpinp != NULL); 827 828 *lportp = lport; 829 830 return (0); 831 } 832 833 /* 834 * Select a local port (number) to use. 835 */ 836 int 837 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 838 struct ucred *cred, int lookupflags) 839 { 840 struct sockaddr_in laddr; 841 842 if (laddrp) { 843 bzero(&laddr, sizeof(laddr)); 844 laddr.sin_family = AF_INET; 845 laddr.sin_addr = *laddrp; 846 } 847 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 848 NULL, lportp, NULL, 0, cred, lookupflags)); 849 } 850 851 /* 852 * Return cached socket options. 853 */ 854 int 855 inp_so_options(const struct inpcb *inp) 856 { 857 int so_options; 858 859 so_options = 0; 860 861 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 862 so_options |= SO_REUSEPORT_LB; 863 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 864 so_options |= SO_REUSEPORT; 865 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 866 so_options |= SO_REUSEADDR; 867 return (so_options); 868 } 869 #endif /* INET || INET6 */ 870 871 /* 872 * Check if a new BINDMULTI socket is allowed to be created. 873 * 874 * ni points to the new inp. 875 * oi points to the exisitng inp. 876 * 877 * This checks whether the existing inp also has BINDMULTI and 878 * whether the credentials match. 879 */ 880 int 881 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 882 { 883 /* Check permissions match */ 884 if ((ni->inp_flags2 & INP_BINDMULTI) && 885 (ni->inp_cred->cr_uid != 886 oi->inp_cred->cr_uid)) 887 return (0); 888 889 /* Check the existing inp has BINDMULTI set */ 890 if ((ni->inp_flags2 & INP_BINDMULTI) && 891 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 892 return (0); 893 894 /* 895 * We're okay - either INP_BINDMULTI isn't set on ni, or 896 * it is and it matches the checks. 897 */ 898 return (1); 899 } 900 901 #ifdef INET 902 /* 903 * Set up a bind operation on a PCB, performing port allocation 904 * as required, but do not actually modify the PCB. Callers can 905 * either complete the bind by setting inp_laddr/inp_lport and 906 * calling in_pcbinshash(), or they can just use the resulting 907 * port and address to authorise the sending of a once-off packet. 908 * 909 * On error, the values of *laddrp and *lportp are not changed. 910 */ 911 int 912 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 913 u_short *lportp, struct ucred *cred) 914 { 915 struct socket *so = inp->inp_socket; 916 struct sockaddr_in *sin; 917 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 918 struct in_addr laddr; 919 u_short lport = 0; 920 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 921 int error; 922 923 /* 924 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 925 * so that we don't have to add to the (already messy) code below. 926 */ 927 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 928 929 /* 930 * No state changes, so read locks are sufficient here. 931 */ 932 INP_LOCK_ASSERT(inp); 933 INP_HASH_LOCK_ASSERT(pcbinfo); 934 935 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 936 return (EADDRNOTAVAIL); 937 laddr.s_addr = *laddrp; 938 if (nam != NULL && laddr.s_addr != INADDR_ANY) 939 return (EINVAL); 940 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 941 lookupflags = INPLOOKUP_WILDCARD; 942 if (nam == NULL) { 943 if ((error = prison_local_ip4(cred, &laddr)) != 0) 944 return (error); 945 } else { 946 sin = (struct sockaddr_in *)nam; 947 KASSERT(sin->sin_family == AF_INET, 948 ("%s: invalid family for address %p", __func__, sin)); 949 KASSERT(sin->sin_len == sizeof(*sin), 950 ("%s: invalid length for address %p", __func__, sin)); 951 952 error = prison_local_ip4(cred, &sin->sin_addr); 953 if (error) 954 return (error); 955 if (sin->sin_port != *lportp) { 956 /* Don't allow the port to change. */ 957 if (*lportp != 0) 958 return (EINVAL); 959 lport = sin->sin_port; 960 } 961 /* NB: lport is left as 0 if the port isn't being changed. */ 962 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 963 /* 964 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 965 * allow complete duplication of binding if 966 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 967 * and a multicast address is bound on both 968 * new and duplicated sockets. 969 */ 970 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 971 reuseport = SO_REUSEADDR|SO_REUSEPORT; 972 /* 973 * XXX: How to deal with SO_REUSEPORT_LB here? 974 * Treat same as SO_REUSEPORT for now. 975 */ 976 if ((so->so_options & 977 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 978 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 979 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 980 sin->sin_port = 0; /* yech... */ 981 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 982 /* 983 * Is the address a local IP address? 984 * If INP_BINDANY is set, then the socket may be bound 985 * to any endpoint address, local or not. 986 */ 987 if ((inp->inp_flags & INP_BINDANY) == 0 && 988 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 989 return (EADDRNOTAVAIL); 990 } 991 laddr = sin->sin_addr; 992 if (lport) { 993 struct inpcb *t; 994 struct tcptw *tw; 995 996 /* GROSS */ 997 if (ntohs(lport) <= V_ipport_reservedhigh && 998 ntohs(lport) >= V_ipport_reservedlow && 999 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 1000 return (EACCES); 1001 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 1002 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 1003 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 1004 lport, INPLOOKUP_WILDCARD, cred); 1005 /* 1006 * XXX 1007 * This entire block sorely needs a rewrite. 1008 */ 1009 if (t && 1010 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 1011 ((t->inp_flags & INP_TIMEWAIT) == 0) && 1012 (so->so_type != SOCK_STREAM || 1013 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 1014 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 1015 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 1016 (t->inp_flags2 & INP_REUSEPORT) || 1017 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 1018 (inp->inp_cred->cr_uid != 1019 t->inp_cred->cr_uid)) 1020 return (EADDRINUSE); 1021 1022 /* 1023 * If the socket is a BINDMULTI socket, then 1024 * the credentials need to match and the 1025 * original socket also has to have been bound 1026 * with BINDMULTI. 1027 */ 1028 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 1029 return (EADDRINUSE); 1030 } 1031 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 1032 lport, lookupflags, cred); 1033 if (t && (t->inp_flags & INP_TIMEWAIT)) { 1034 /* 1035 * XXXRW: If an incpb has had its timewait 1036 * state recycled, we treat the address as 1037 * being in use (for now). This is better 1038 * than a panic, but not desirable. 1039 */ 1040 tw = intotw(t); 1041 if (tw == NULL || 1042 ((reuseport & tw->tw_so_options) == 0 && 1043 (reuseport_lb & 1044 tw->tw_so_options) == 0)) { 1045 return (EADDRINUSE); 1046 } 1047 } else if (t && 1048 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 1049 (reuseport & inp_so_options(t)) == 0 && 1050 (reuseport_lb & inp_so_options(t)) == 0) { 1051 #ifdef INET6 1052 if (ntohl(sin->sin_addr.s_addr) != 1053 INADDR_ANY || 1054 ntohl(t->inp_laddr.s_addr) != 1055 INADDR_ANY || 1056 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 1057 (t->inp_vflag & INP_IPV6PROTO) == 0) 1058 #endif 1059 return (EADDRINUSE); 1060 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 1061 return (EADDRINUSE); 1062 } 1063 } 1064 } 1065 if (*lportp != 0) 1066 lport = *lportp; 1067 if (lport == 0) { 1068 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 1069 if (error != 0) 1070 return (error); 1071 } 1072 *laddrp = laddr.s_addr; 1073 *lportp = lport; 1074 return (0); 1075 } 1076 1077 /* 1078 * Connect from a socket to a specified address. 1079 * Both address and port must be specified in argument sin. 1080 * If don't have a local address for this socket yet, 1081 * then pick one. 1082 */ 1083 int 1084 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 1085 struct ucred *cred, struct mbuf *m, bool rehash) 1086 { 1087 u_short lport, fport; 1088 in_addr_t laddr, faddr; 1089 int anonport, error; 1090 1091 INP_WLOCK_ASSERT(inp); 1092 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1093 1094 lport = inp->inp_lport; 1095 laddr = inp->inp_laddr.s_addr; 1096 anonport = (lport == 0); 1097 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 1098 NULL, cred); 1099 if (error) 1100 return (error); 1101 1102 /* Do the initial binding of the local address if required. */ 1103 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1104 KASSERT(rehash == true, 1105 ("Rehashing required for unbound inps")); 1106 inp->inp_lport = lport; 1107 inp->inp_laddr.s_addr = laddr; 1108 if (in_pcbinshash(inp) != 0) { 1109 inp->inp_laddr.s_addr = INADDR_ANY; 1110 inp->inp_lport = 0; 1111 return (EAGAIN); 1112 } 1113 } 1114 1115 /* Commit the remaining changes. */ 1116 inp->inp_lport = lport; 1117 inp->inp_laddr.s_addr = laddr; 1118 inp->inp_faddr.s_addr = faddr; 1119 inp->inp_fport = fport; 1120 if (rehash) { 1121 in_pcbrehash_mbuf(inp, m); 1122 } else { 1123 in_pcbinshash_mbuf(inp, m); 1124 } 1125 1126 if (anonport) 1127 inp->inp_flags |= INP_ANONPORT; 1128 return (0); 1129 } 1130 1131 int 1132 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 1133 { 1134 1135 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true)); 1136 } 1137 1138 /* 1139 * Do proper source address selection on an unbound socket in case 1140 * of connect. Take jails into account as well. 1141 */ 1142 int 1143 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1144 struct ucred *cred) 1145 { 1146 struct ifaddr *ifa; 1147 struct sockaddr *sa; 1148 struct sockaddr_in *sin, dst; 1149 struct nhop_object *nh; 1150 int error; 1151 1152 NET_EPOCH_ASSERT(); 1153 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1154 /* 1155 * Bypass source address selection and use the primary jail IP 1156 * if requested. 1157 */ 1158 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1159 return (0); 1160 1161 error = 0; 1162 1163 nh = NULL; 1164 bzero(&dst, sizeof(dst)); 1165 sin = &dst; 1166 sin->sin_family = AF_INET; 1167 sin->sin_len = sizeof(struct sockaddr_in); 1168 sin->sin_addr.s_addr = faddr->s_addr; 1169 1170 /* 1171 * If route is known our src addr is taken from the i/f, 1172 * else punt. 1173 * 1174 * Find out route to destination. 1175 */ 1176 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1177 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1178 0, NHR_NONE, 0); 1179 1180 /* 1181 * If we found a route, use the address corresponding to 1182 * the outgoing interface. 1183 * 1184 * Otherwise assume faddr is reachable on a directly connected 1185 * network and try to find a corresponding interface to take 1186 * the source address from. 1187 */ 1188 if (nh == NULL || nh->nh_ifp == NULL) { 1189 struct in_ifaddr *ia; 1190 struct ifnet *ifp; 1191 1192 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1193 inp->inp_socket->so_fibnum)); 1194 if (ia == NULL) { 1195 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1196 inp->inp_socket->so_fibnum)); 1197 } 1198 if (ia == NULL) { 1199 error = ENETUNREACH; 1200 goto done; 1201 } 1202 1203 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1204 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1205 goto done; 1206 } 1207 1208 ifp = ia->ia_ifp; 1209 ia = NULL; 1210 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1211 sa = ifa->ifa_addr; 1212 if (sa->sa_family != AF_INET) 1213 continue; 1214 sin = (struct sockaddr_in *)sa; 1215 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1216 ia = (struct in_ifaddr *)ifa; 1217 break; 1218 } 1219 } 1220 if (ia != NULL) { 1221 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1222 goto done; 1223 } 1224 1225 /* 3. As a last resort return the 'default' jail address. */ 1226 error = prison_get_ip4(cred, laddr); 1227 goto done; 1228 } 1229 1230 /* 1231 * If the outgoing interface on the route found is not 1232 * a loopback interface, use the address from that interface. 1233 * In case of jails do those three steps: 1234 * 1. check if the interface address belongs to the jail. If so use it. 1235 * 2. check if we have any address on the outgoing interface 1236 * belonging to this jail. If so use it. 1237 * 3. as a last resort return the 'default' jail address. 1238 */ 1239 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1240 struct in_ifaddr *ia; 1241 struct ifnet *ifp; 1242 1243 /* If not jailed, use the default returned. */ 1244 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1245 ia = (struct in_ifaddr *)nh->nh_ifa; 1246 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1247 goto done; 1248 } 1249 1250 /* Jailed. */ 1251 /* 1. Check if the iface address belongs to the jail. */ 1252 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1253 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1254 ia = (struct in_ifaddr *)nh->nh_ifa; 1255 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1256 goto done; 1257 } 1258 1259 /* 1260 * 2. Check if we have any address on the outgoing interface 1261 * belonging to this jail. 1262 */ 1263 ia = NULL; 1264 ifp = nh->nh_ifp; 1265 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1266 sa = ifa->ifa_addr; 1267 if (sa->sa_family != AF_INET) 1268 continue; 1269 sin = (struct sockaddr_in *)sa; 1270 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1271 ia = (struct in_ifaddr *)ifa; 1272 break; 1273 } 1274 } 1275 if (ia != NULL) { 1276 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1277 goto done; 1278 } 1279 1280 /* 3. As a last resort return the 'default' jail address. */ 1281 error = prison_get_ip4(cred, laddr); 1282 goto done; 1283 } 1284 1285 /* 1286 * The outgoing interface is marked with 'loopback net', so a route 1287 * to ourselves is here. 1288 * Try to find the interface of the destination address and then 1289 * take the address from there. That interface is not necessarily 1290 * a loopback interface. 1291 * In case of jails, check that it is an address of the jail 1292 * and if we cannot find, fall back to the 'default' jail address. 1293 */ 1294 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1295 struct in_ifaddr *ia; 1296 1297 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1298 inp->inp_socket->so_fibnum)); 1299 if (ia == NULL) 1300 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1301 inp->inp_socket->so_fibnum)); 1302 if (ia == NULL) 1303 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1304 1305 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1306 if (ia == NULL) { 1307 error = ENETUNREACH; 1308 goto done; 1309 } 1310 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1311 goto done; 1312 } 1313 1314 /* Jailed. */ 1315 if (ia != NULL) { 1316 struct ifnet *ifp; 1317 1318 ifp = ia->ia_ifp; 1319 ia = NULL; 1320 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1321 sa = ifa->ifa_addr; 1322 if (sa->sa_family != AF_INET) 1323 continue; 1324 sin = (struct sockaddr_in *)sa; 1325 if (prison_check_ip4(cred, 1326 &sin->sin_addr) == 0) { 1327 ia = (struct in_ifaddr *)ifa; 1328 break; 1329 } 1330 } 1331 if (ia != NULL) { 1332 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1333 goto done; 1334 } 1335 } 1336 1337 /* 3. As a last resort return the 'default' jail address. */ 1338 error = prison_get_ip4(cred, laddr); 1339 goto done; 1340 } 1341 1342 done: 1343 return (error); 1344 } 1345 1346 /* 1347 * Set up for a connect from a socket to the specified address. 1348 * On entry, *laddrp and *lportp should contain the current local 1349 * address and port for the PCB; these are updated to the values 1350 * that should be placed in inp_laddr and inp_lport to complete 1351 * the connect. 1352 * 1353 * On success, *faddrp and *fportp will be set to the remote address 1354 * and port. These are not updated in the error case. 1355 * 1356 * If the operation fails because the connection already exists, 1357 * *oinpp will be set to the PCB of that connection so that the 1358 * caller can decide to override it. In all other cases, *oinpp 1359 * is set to NULL. 1360 */ 1361 int 1362 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1363 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1364 struct inpcb **oinpp, struct ucred *cred) 1365 { 1366 struct rm_priotracker in_ifa_tracker; 1367 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1368 struct in_ifaddr *ia; 1369 struct inpcb *oinp; 1370 struct in_addr laddr, faddr; 1371 u_short lport, fport; 1372 int error; 1373 1374 KASSERT(sin->sin_family == AF_INET, 1375 ("%s: invalid address family for %p", __func__, sin)); 1376 KASSERT(sin->sin_len == sizeof(*sin), 1377 ("%s: invalid address length for %p", __func__, sin)); 1378 1379 /* 1380 * Because a global state change doesn't actually occur here, a read 1381 * lock is sufficient. 1382 */ 1383 NET_EPOCH_ASSERT(); 1384 INP_LOCK_ASSERT(inp); 1385 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1386 1387 if (oinpp != NULL) 1388 *oinpp = NULL; 1389 if (sin->sin_port == 0) 1390 return (EADDRNOTAVAIL); 1391 laddr.s_addr = *laddrp; 1392 lport = *lportp; 1393 faddr = sin->sin_addr; 1394 fport = sin->sin_port; 1395 #ifdef ROUTE_MPATH 1396 if (CALC_FLOWID_OUTBOUND) { 1397 uint32_t hash_val, hash_type; 1398 1399 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, 1400 inp->inp_socket->so_proto->pr_protocol, &hash_type); 1401 1402 inp->inp_flowid = hash_val; 1403 inp->inp_flowtype = hash_type; 1404 } 1405 #endif 1406 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1407 /* 1408 * If the destination address is INADDR_ANY, 1409 * use the primary local address. 1410 * If the supplied address is INADDR_BROADCAST, 1411 * and the primary interface supports broadcast, 1412 * choose the broadcast address for that interface. 1413 */ 1414 if (faddr.s_addr == INADDR_ANY) { 1415 IN_IFADDR_RLOCK(&in_ifa_tracker); 1416 faddr = 1417 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1418 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1419 if (cred != NULL && 1420 (error = prison_get_ip4(cred, &faddr)) != 0) 1421 return (error); 1422 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1423 IN_IFADDR_RLOCK(&in_ifa_tracker); 1424 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1425 IFF_BROADCAST) 1426 faddr = satosin(&CK_STAILQ_FIRST( 1427 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1428 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1429 } 1430 } 1431 if (laddr.s_addr == INADDR_ANY) { 1432 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1433 /* 1434 * If the destination address is multicast and an outgoing 1435 * interface has been set as a multicast option, prefer the 1436 * address of that interface as our source address. 1437 */ 1438 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1439 inp->inp_moptions != NULL) { 1440 struct ip_moptions *imo; 1441 struct ifnet *ifp; 1442 1443 imo = inp->inp_moptions; 1444 if (imo->imo_multicast_ifp != NULL) { 1445 ifp = imo->imo_multicast_ifp; 1446 IN_IFADDR_RLOCK(&in_ifa_tracker); 1447 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1448 if ((ia->ia_ifp == ifp) && 1449 (cred == NULL || 1450 prison_check_ip4(cred, 1451 &ia->ia_addr.sin_addr) == 0)) 1452 break; 1453 } 1454 if (ia == NULL) 1455 error = EADDRNOTAVAIL; 1456 else { 1457 laddr = ia->ia_addr.sin_addr; 1458 error = 0; 1459 } 1460 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1461 } 1462 } 1463 if (error) 1464 return (error); 1465 } 1466 1467 if (lport != 0) { 1468 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, 1469 fport, laddr, lport, 0, NULL, M_NODOM); 1470 if (oinp != NULL) { 1471 if (oinpp != NULL) 1472 *oinpp = oinp; 1473 return (EADDRINUSE); 1474 } 1475 } else { 1476 struct sockaddr_in lsin, fsin; 1477 1478 bzero(&lsin, sizeof(lsin)); 1479 bzero(&fsin, sizeof(fsin)); 1480 lsin.sin_family = AF_INET; 1481 lsin.sin_addr = laddr; 1482 fsin.sin_family = AF_INET; 1483 fsin.sin_addr = faddr; 1484 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, 1485 &lport, (struct sockaddr *)& fsin, fport, cred, 1486 INPLOOKUP_WILDCARD); 1487 if (error) 1488 return (error); 1489 } 1490 *laddrp = laddr.s_addr; 1491 *lportp = lport; 1492 *faddrp = faddr.s_addr; 1493 *fportp = fport; 1494 return (0); 1495 } 1496 1497 void 1498 in_pcbdisconnect(struct inpcb *inp) 1499 { 1500 1501 INP_WLOCK_ASSERT(inp); 1502 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1503 1504 inp->inp_faddr.s_addr = INADDR_ANY; 1505 inp->inp_fport = 0; 1506 in_pcbrehash(inp); 1507 } 1508 #endif /* INET */ 1509 1510 /* 1511 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1512 * For most protocols, this will be invoked immediately prior to calling 1513 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1514 * socket, in which case in_pcbfree() is deferred. 1515 */ 1516 void 1517 in_pcbdetach(struct inpcb *inp) 1518 { 1519 1520 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1521 1522 #ifdef RATELIMIT 1523 if (inp->inp_snd_tag != NULL) 1524 in_pcbdetach_txrtlmt(inp); 1525 #endif 1526 inp->inp_socket->so_pcb = NULL; 1527 inp->inp_socket = NULL; 1528 } 1529 1530 /* 1531 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1532 * stability of an inpcb pointer despite the inpcb lock being released. This 1533 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1534 * but where the inpcb lock may already held, or when acquiring a reference 1535 * via a pcbgroup. 1536 * 1537 * in_pcbref() should be used only to provide brief memory stability, and 1538 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1539 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1540 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1541 * lock and rele are the *only* safe operations that may be performed on the 1542 * inpcb. 1543 * 1544 * While the inpcb will not be freed, releasing the inpcb lock means that the 1545 * connection's state may change, so the caller should be careful to 1546 * revalidate any cached state on reacquiring the lock. Drop the reference 1547 * using in_pcbrele(). 1548 */ 1549 void 1550 in_pcbref(struct inpcb *inp) 1551 { 1552 1553 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1554 1555 refcount_acquire(&inp->inp_refcount); 1556 } 1557 1558 /* 1559 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1560 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1561 * return a flag indicating whether or not the inpcb remains valid. If it is 1562 * valid, we return with the inpcb lock held. 1563 * 1564 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1565 * reference on an inpcb. Historically more work was done here (actually, in 1566 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1567 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1568 * about memory stability (and continued use of the write lock). 1569 */ 1570 int 1571 in_pcbrele_rlocked(struct inpcb *inp) 1572 { 1573 struct inpcbinfo *pcbinfo; 1574 1575 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1576 1577 INP_RLOCK_ASSERT(inp); 1578 1579 if (refcount_release(&inp->inp_refcount) == 0) { 1580 /* 1581 * If the inpcb has been freed, let the caller know, even if 1582 * this isn't the last reference. 1583 */ 1584 if (inp->inp_flags2 & INP_FREED) { 1585 INP_RUNLOCK(inp); 1586 return (1); 1587 } 1588 return (0); 1589 } 1590 1591 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1592 #ifdef TCPHPTS 1593 if (inp->inp_in_hpts || inp->inp_in_input) { 1594 struct tcp_hpts_entry *hpts; 1595 /* 1596 * We should not be on the hpts at 1597 * this point in any form. we must 1598 * get the lock to be sure. 1599 */ 1600 hpts = tcp_hpts_lock(inp); 1601 if (inp->inp_in_hpts) 1602 panic("Hpts:%p inp:%p at free still on hpts", 1603 hpts, inp); 1604 mtx_unlock(&hpts->p_mtx); 1605 hpts = tcp_input_lock(inp); 1606 if (inp->inp_in_input) 1607 panic("Hpts:%p inp:%p at free still on input hpts", 1608 hpts, inp); 1609 mtx_unlock(&hpts->p_mtx); 1610 } 1611 #endif 1612 INP_RUNLOCK(inp); 1613 pcbinfo = inp->inp_pcbinfo; 1614 uma_zfree(pcbinfo->ipi_zone, inp); 1615 return (1); 1616 } 1617 1618 int 1619 in_pcbrele_wlocked(struct inpcb *inp) 1620 { 1621 struct inpcbinfo *pcbinfo; 1622 1623 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1624 1625 INP_WLOCK_ASSERT(inp); 1626 1627 if (refcount_release(&inp->inp_refcount) == 0) { 1628 /* 1629 * If the inpcb has been freed, let the caller know, even if 1630 * this isn't the last reference. 1631 */ 1632 if (inp->inp_flags2 & INP_FREED) { 1633 INP_WUNLOCK(inp); 1634 return (1); 1635 } 1636 return (0); 1637 } 1638 1639 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1640 #ifdef TCPHPTS 1641 if (inp->inp_in_hpts || inp->inp_in_input) { 1642 struct tcp_hpts_entry *hpts; 1643 /* 1644 * We should not be on the hpts at 1645 * this point in any form. we must 1646 * get the lock to be sure. 1647 */ 1648 hpts = tcp_hpts_lock(inp); 1649 if (inp->inp_in_hpts) 1650 panic("Hpts:%p inp:%p at free still on hpts", 1651 hpts, inp); 1652 mtx_unlock(&hpts->p_mtx); 1653 hpts = tcp_input_lock(inp); 1654 if (inp->inp_in_input) 1655 panic("Hpts:%p inp:%p at free still on input hpts", 1656 hpts, inp); 1657 mtx_unlock(&hpts->p_mtx); 1658 } 1659 #endif 1660 INP_WUNLOCK(inp); 1661 pcbinfo = inp->inp_pcbinfo; 1662 uma_zfree(pcbinfo->ipi_zone, inp); 1663 return (1); 1664 } 1665 1666 /* 1667 * Temporary wrapper. 1668 */ 1669 int 1670 in_pcbrele(struct inpcb *inp) 1671 { 1672 1673 return (in_pcbrele_wlocked(inp)); 1674 } 1675 1676 void 1677 in_pcblist_rele_rlocked(epoch_context_t ctx) 1678 { 1679 struct in_pcblist *il; 1680 struct inpcb *inp; 1681 struct inpcbinfo *pcbinfo; 1682 int i, n; 1683 1684 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1685 pcbinfo = il->il_pcbinfo; 1686 n = il->il_count; 1687 INP_INFO_WLOCK(pcbinfo); 1688 for (i = 0; i < n; i++) { 1689 inp = il->il_inp_list[i]; 1690 INP_RLOCK(inp); 1691 if (!in_pcbrele_rlocked(inp)) 1692 INP_RUNLOCK(inp); 1693 } 1694 INP_INFO_WUNLOCK(pcbinfo); 1695 free(il, M_TEMP); 1696 } 1697 1698 static void 1699 inpcbport_free(epoch_context_t ctx) 1700 { 1701 struct inpcbport *phd; 1702 1703 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); 1704 free(phd, M_PCB); 1705 } 1706 1707 static void 1708 in_pcbfree_deferred(epoch_context_t ctx) 1709 { 1710 struct inpcb *inp; 1711 int released __unused; 1712 1713 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); 1714 1715 INP_WLOCK(inp); 1716 CURVNET_SET(inp->inp_vnet); 1717 #ifdef INET 1718 struct ip_moptions *imo = inp->inp_moptions; 1719 inp->inp_moptions = NULL; 1720 #endif 1721 /* XXXRW: Do as much as possible here. */ 1722 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1723 if (inp->inp_sp != NULL) 1724 ipsec_delete_pcbpolicy(inp); 1725 #endif 1726 #ifdef INET6 1727 struct ip6_moptions *im6o = NULL; 1728 if (inp->inp_vflag & INP_IPV6PROTO) { 1729 ip6_freepcbopts(inp->in6p_outputopts); 1730 im6o = inp->in6p_moptions; 1731 inp->in6p_moptions = NULL; 1732 } 1733 #endif 1734 if (inp->inp_options) 1735 (void)m_free(inp->inp_options); 1736 inp->inp_vflag = 0; 1737 crfree(inp->inp_cred); 1738 #ifdef MAC 1739 mac_inpcb_destroy(inp); 1740 #endif 1741 released = in_pcbrele_wlocked(inp); 1742 MPASS(released); 1743 #ifdef INET6 1744 ip6_freemoptions(im6o); 1745 #endif 1746 #ifdef INET 1747 inp_freemoptions(imo); 1748 #endif 1749 CURVNET_RESTORE(); 1750 } 1751 1752 /* 1753 * Unconditionally schedule an inpcb to be freed by decrementing its 1754 * reference count, which should occur only after the inpcb has been detached 1755 * from its socket. If another thread holds a temporary reference (acquired 1756 * using in_pcbref()) then the free is deferred until that reference is 1757 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1758 * work, including removal from global lists, is done in this context, where 1759 * the pcbinfo lock is held. 1760 */ 1761 void 1762 in_pcbfree(struct inpcb *inp) 1763 { 1764 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1765 1766 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1767 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1768 ("%s: called twice for pcb %p", __func__, inp)); 1769 if (inp->inp_flags2 & INP_FREED) { 1770 INP_WUNLOCK(inp); 1771 return; 1772 } 1773 1774 INP_WLOCK_ASSERT(inp); 1775 INP_LIST_WLOCK(pcbinfo); 1776 in_pcbremlists(inp); 1777 INP_LIST_WUNLOCK(pcbinfo); 1778 RO_INVALIDATE_CACHE(&inp->inp_route); 1779 /* mark as destruction in progress */ 1780 inp->inp_flags2 |= INP_FREED; 1781 INP_WUNLOCK(inp); 1782 NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx); 1783 } 1784 1785 /* 1786 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1787 * port reservation, and preventing it from being returned by inpcb lookups. 1788 * 1789 * It is used by TCP to mark an inpcb as unused and avoid future packet 1790 * delivery or event notification when a socket remains open but TCP has 1791 * closed. This might occur as a result of a shutdown()-initiated TCP close 1792 * or a RST on the wire, and allows the port binding to be reused while still 1793 * maintaining the invariant that so_pcb always points to a valid inpcb until 1794 * in_pcbdetach(). 1795 * 1796 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1797 * in_pcbnotifyall() and in_pcbpurgeif0()? 1798 */ 1799 void 1800 in_pcbdrop(struct inpcb *inp) 1801 { 1802 1803 INP_WLOCK_ASSERT(inp); 1804 #ifdef INVARIANTS 1805 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1806 MPASS(inp->inp_refcount > 1); 1807 #endif 1808 1809 /* 1810 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1811 * the hash lock...? 1812 */ 1813 inp->inp_flags |= INP_DROPPED; 1814 if (inp->inp_flags & INP_INHASHLIST) { 1815 struct inpcbport *phd = inp->inp_phd; 1816 1817 INP_HASH_WLOCK(inp->inp_pcbinfo); 1818 in_pcbremlbgrouphash(inp); 1819 CK_LIST_REMOVE(inp, inp_hash); 1820 CK_LIST_REMOVE(inp, inp_portlist); 1821 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 1822 CK_LIST_REMOVE(phd, phd_hash); 1823 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); 1824 } 1825 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1826 inp->inp_flags &= ~INP_INHASHLIST; 1827 #ifdef PCBGROUP 1828 in_pcbgroup_remove(inp); 1829 #endif 1830 } 1831 } 1832 1833 #ifdef INET 1834 /* 1835 * Common routines to return the socket addresses associated with inpcbs. 1836 */ 1837 struct sockaddr * 1838 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1839 { 1840 struct sockaddr_in *sin; 1841 1842 sin = malloc(sizeof *sin, M_SONAME, 1843 M_WAITOK | M_ZERO); 1844 sin->sin_family = AF_INET; 1845 sin->sin_len = sizeof(*sin); 1846 sin->sin_addr = *addr_p; 1847 sin->sin_port = port; 1848 1849 return (struct sockaddr *)sin; 1850 } 1851 1852 int 1853 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1854 { 1855 struct inpcb *inp; 1856 struct in_addr addr; 1857 in_port_t port; 1858 1859 inp = sotoinpcb(so); 1860 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1861 1862 INP_RLOCK(inp); 1863 port = inp->inp_lport; 1864 addr = inp->inp_laddr; 1865 INP_RUNLOCK(inp); 1866 1867 *nam = in_sockaddr(port, &addr); 1868 return 0; 1869 } 1870 1871 int 1872 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1873 { 1874 struct inpcb *inp; 1875 struct in_addr addr; 1876 in_port_t port; 1877 1878 inp = sotoinpcb(so); 1879 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1880 1881 INP_RLOCK(inp); 1882 port = inp->inp_fport; 1883 addr = inp->inp_faddr; 1884 INP_RUNLOCK(inp); 1885 1886 *nam = in_sockaddr(port, &addr); 1887 return 0; 1888 } 1889 1890 void 1891 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1892 struct inpcb *(*notify)(struct inpcb *, int)) 1893 { 1894 struct inpcb *inp, *inp_temp; 1895 1896 INP_INFO_WLOCK(pcbinfo); 1897 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1898 INP_WLOCK(inp); 1899 #ifdef INET6 1900 if ((inp->inp_vflag & INP_IPV4) == 0) { 1901 INP_WUNLOCK(inp); 1902 continue; 1903 } 1904 #endif 1905 if (inp->inp_faddr.s_addr != faddr.s_addr || 1906 inp->inp_socket == NULL) { 1907 INP_WUNLOCK(inp); 1908 continue; 1909 } 1910 if ((*notify)(inp, errno)) 1911 INP_WUNLOCK(inp); 1912 } 1913 INP_INFO_WUNLOCK(pcbinfo); 1914 } 1915 1916 void 1917 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1918 { 1919 struct inpcb *inp; 1920 struct in_multi *inm; 1921 struct in_mfilter *imf; 1922 struct ip_moptions *imo; 1923 1924 INP_INFO_WLOCK(pcbinfo); 1925 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1926 INP_WLOCK(inp); 1927 imo = inp->inp_moptions; 1928 if ((inp->inp_vflag & INP_IPV4) && 1929 imo != NULL) { 1930 /* 1931 * Unselect the outgoing interface if it is being 1932 * detached. 1933 */ 1934 if (imo->imo_multicast_ifp == ifp) 1935 imo->imo_multicast_ifp = NULL; 1936 1937 /* 1938 * Drop multicast group membership if we joined 1939 * through the interface being detached. 1940 * 1941 * XXX This can all be deferred to an epoch_call 1942 */ 1943 restart: 1944 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 1945 if ((inm = imf->imf_inm) == NULL) 1946 continue; 1947 if (inm->inm_ifp != ifp) 1948 continue; 1949 ip_mfilter_remove(&imo->imo_head, imf); 1950 IN_MULTI_LOCK_ASSERT(); 1951 in_leavegroup_locked(inm, NULL); 1952 ip_mfilter_free(imf); 1953 goto restart; 1954 } 1955 } 1956 INP_WUNLOCK(inp); 1957 } 1958 INP_INFO_WUNLOCK(pcbinfo); 1959 } 1960 1961 /* 1962 * Lookup a PCB based on the local address and port. Caller must hold the 1963 * hash lock. No inpcb locks or references are acquired. 1964 */ 1965 #define INP_LOOKUP_MAPPED_PCB_COST 3 1966 struct inpcb * 1967 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1968 u_short lport, int lookupflags, struct ucred *cred) 1969 { 1970 struct inpcb *inp; 1971 #ifdef INET6 1972 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1973 #else 1974 int matchwild = 3; 1975 #endif 1976 int wildcard; 1977 1978 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1979 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1980 1981 INP_HASH_LOCK_ASSERT(pcbinfo); 1982 1983 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1984 struct inpcbhead *head; 1985 /* 1986 * Look for an unconnected (wildcard foreign addr) PCB that 1987 * matches the local address and port we're looking for. 1988 */ 1989 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1990 0, pcbinfo->ipi_hashmask)]; 1991 CK_LIST_FOREACH(inp, head, inp_hash) { 1992 #ifdef INET6 1993 /* XXX inp locking */ 1994 if ((inp->inp_vflag & INP_IPV4) == 0) 1995 continue; 1996 #endif 1997 if (inp->inp_faddr.s_addr == INADDR_ANY && 1998 inp->inp_laddr.s_addr == laddr.s_addr && 1999 inp->inp_lport == lport) { 2000 /* 2001 * Found? 2002 */ 2003 if (cred == NULL || 2004 prison_equal_ip4(cred->cr_prison, 2005 inp->inp_cred->cr_prison)) 2006 return (inp); 2007 } 2008 } 2009 /* 2010 * Not found. 2011 */ 2012 return (NULL); 2013 } else { 2014 struct inpcbporthead *porthash; 2015 struct inpcbport *phd; 2016 struct inpcb *match = NULL; 2017 /* 2018 * Best fit PCB lookup. 2019 * 2020 * First see if this local port is in use by looking on the 2021 * port hash list. 2022 */ 2023 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 2024 pcbinfo->ipi_porthashmask)]; 2025 CK_LIST_FOREACH(phd, porthash, phd_hash) { 2026 if (phd->phd_port == lport) 2027 break; 2028 } 2029 if (phd != NULL) { 2030 /* 2031 * Port is in use by one or more PCBs. Look for best 2032 * fit. 2033 */ 2034 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 2035 wildcard = 0; 2036 if (cred != NULL && 2037 !prison_equal_ip4(inp->inp_cred->cr_prison, 2038 cred->cr_prison)) 2039 continue; 2040 #ifdef INET6 2041 /* XXX inp locking */ 2042 if ((inp->inp_vflag & INP_IPV4) == 0) 2043 continue; 2044 /* 2045 * We never select the PCB that has 2046 * INP_IPV6 flag and is bound to :: if 2047 * we have another PCB which is bound 2048 * to 0.0.0.0. If a PCB has the 2049 * INP_IPV6 flag, then we set its cost 2050 * higher than IPv4 only PCBs. 2051 * 2052 * Note that the case only happens 2053 * when a socket is bound to ::, under 2054 * the condition that the use of the 2055 * mapped address is allowed. 2056 */ 2057 if ((inp->inp_vflag & INP_IPV6) != 0) 2058 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 2059 #endif 2060 if (inp->inp_faddr.s_addr != INADDR_ANY) 2061 wildcard++; 2062 if (inp->inp_laddr.s_addr != INADDR_ANY) { 2063 if (laddr.s_addr == INADDR_ANY) 2064 wildcard++; 2065 else if (inp->inp_laddr.s_addr != laddr.s_addr) 2066 continue; 2067 } else { 2068 if (laddr.s_addr != INADDR_ANY) 2069 wildcard++; 2070 } 2071 if (wildcard < matchwild) { 2072 match = inp; 2073 matchwild = wildcard; 2074 if (matchwild == 0) 2075 break; 2076 } 2077 } 2078 } 2079 return (match); 2080 } 2081 } 2082 #undef INP_LOOKUP_MAPPED_PCB_COST 2083 2084 static struct inpcb * 2085 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 2086 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, 2087 uint16_t fport, int lookupflags, int numa_domain) 2088 { 2089 struct inpcb *local_wild, *numa_wild; 2090 const struct inpcblbgrouphead *hdr; 2091 struct inpcblbgroup *grp; 2092 uint32_t idx; 2093 2094 INP_HASH_LOCK_ASSERT(pcbinfo); 2095 2096 hdr = &pcbinfo->ipi_lbgrouphashbase[ 2097 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 2098 2099 /* 2100 * Order of socket selection: 2101 * 1. non-wild. 2102 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). 2103 * 2104 * NOTE: 2105 * - Load balanced group does not contain jailed sockets 2106 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets 2107 */ 2108 local_wild = NULL; 2109 numa_wild = NULL; 2110 CK_LIST_FOREACH(grp, hdr, il_list) { 2111 #ifdef INET6 2112 if (!(grp->il_vflag & INP_IPV4)) 2113 continue; 2114 #endif 2115 if (grp->il_lport != lport) 2116 continue; 2117 2118 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % 2119 grp->il_inpcnt; 2120 if (grp->il_laddr.s_addr == laddr->s_addr) { 2121 if (numa_domain == M_NODOM || 2122 grp->il_numa_domain == numa_domain) { 2123 return (grp->il_inp[idx]); 2124 } else { 2125 numa_wild = grp->il_inp[idx]; 2126 } 2127 } 2128 if (grp->il_laddr.s_addr == INADDR_ANY && 2129 (lookupflags & INPLOOKUP_WILDCARD) != 0 && 2130 (local_wild == NULL || numa_domain == M_NODOM || 2131 grp->il_numa_domain == numa_domain)) { 2132 local_wild = grp->il_inp[idx]; 2133 } 2134 } 2135 if (numa_wild != NULL) 2136 return (numa_wild); 2137 2138 return (local_wild); 2139 } 2140 2141 #ifdef PCBGROUP 2142 /* 2143 * Lookup PCB in hash list, using pcbgroup tables. 2144 */ 2145 static struct inpcb * 2146 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 2147 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 2148 u_int lport_arg, int lookupflags, struct ifnet *ifp) 2149 { 2150 struct inpcbhead *head; 2151 struct inpcb *inp, *tmpinp; 2152 u_short fport = fport_arg, lport = lport_arg; 2153 bool locked; 2154 2155 /* 2156 * First look for an exact match. 2157 */ 2158 tmpinp = NULL; 2159 INP_GROUP_LOCK(pcbgroup); 2160 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2161 pcbgroup->ipg_hashmask)]; 2162 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2163 #ifdef INET6 2164 /* XXX inp locking */ 2165 if ((inp->inp_vflag & INP_IPV4) == 0) 2166 continue; 2167 #endif 2168 if (inp->inp_faddr.s_addr == faddr.s_addr && 2169 inp->inp_laddr.s_addr == laddr.s_addr && 2170 inp->inp_fport == fport && 2171 inp->inp_lport == lport) { 2172 /* 2173 * XXX We should be able to directly return 2174 * the inp here, without any checks. 2175 * Well unless both bound with SO_REUSEPORT? 2176 */ 2177 if (prison_flag(inp->inp_cred, PR_IP4)) 2178 goto found; 2179 if (tmpinp == NULL) 2180 tmpinp = inp; 2181 } 2182 } 2183 if (tmpinp != NULL) { 2184 inp = tmpinp; 2185 goto found; 2186 } 2187 2188 #ifdef RSS 2189 /* 2190 * For incoming connections, we may wish to do a wildcard 2191 * match for an RSS-local socket. 2192 */ 2193 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2194 struct inpcb *local_wild = NULL, *local_exact = NULL; 2195 #ifdef INET6 2196 struct inpcb *local_wild_mapped = NULL; 2197 #endif 2198 struct inpcb *jail_wild = NULL; 2199 struct inpcbhead *head; 2200 int injail; 2201 2202 /* 2203 * Order of socket selection - we always prefer jails. 2204 * 1. jailed, non-wild. 2205 * 2. jailed, wild. 2206 * 3. non-jailed, non-wild. 2207 * 4. non-jailed, wild. 2208 */ 2209 2210 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 2211 lport, 0, pcbgroup->ipg_hashmask)]; 2212 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2213 #ifdef INET6 2214 /* XXX inp locking */ 2215 if ((inp->inp_vflag & INP_IPV4) == 0) 2216 continue; 2217 #endif 2218 if (inp->inp_faddr.s_addr != INADDR_ANY || 2219 inp->inp_lport != lport) 2220 continue; 2221 2222 injail = prison_flag(inp->inp_cred, PR_IP4); 2223 if (injail) { 2224 if (prison_check_ip4(inp->inp_cred, 2225 &laddr) != 0) 2226 continue; 2227 } else { 2228 if (local_exact != NULL) 2229 continue; 2230 } 2231 2232 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2233 if (injail) 2234 goto found; 2235 else 2236 local_exact = inp; 2237 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2238 #ifdef INET6 2239 /* XXX inp locking, NULL check */ 2240 if (inp->inp_vflag & INP_IPV6PROTO) 2241 local_wild_mapped = inp; 2242 else 2243 #endif 2244 if (injail) 2245 jail_wild = inp; 2246 else 2247 local_wild = inp; 2248 } 2249 } /* LIST_FOREACH */ 2250 2251 inp = jail_wild; 2252 if (inp == NULL) 2253 inp = local_exact; 2254 if (inp == NULL) 2255 inp = local_wild; 2256 #ifdef INET6 2257 if (inp == NULL) 2258 inp = local_wild_mapped; 2259 #endif 2260 if (inp != NULL) 2261 goto found; 2262 } 2263 #endif 2264 2265 /* 2266 * Then look for a wildcard match, if requested. 2267 */ 2268 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2269 struct inpcb *local_wild = NULL, *local_exact = NULL; 2270 #ifdef INET6 2271 struct inpcb *local_wild_mapped = NULL; 2272 #endif 2273 struct inpcb *jail_wild = NULL; 2274 struct inpcbhead *head; 2275 int injail; 2276 2277 /* 2278 * Order of socket selection - we always prefer jails. 2279 * 1. jailed, non-wild. 2280 * 2. jailed, wild. 2281 * 3. non-jailed, non-wild. 2282 * 4. non-jailed, wild. 2283 */ 2284 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 2285 0, pcbinfo->ipi_wildmask)]; 2286 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 2287 #ifdef INET6 2288 /* XXX inp locking */ 2289 if ((inp->inp_vflag & INP_IPV4) == 0) 2290 continue; 2291 #endif 2292 if (inp->inp_faddr.s_addr != INADDR_ANY || 2293 inp->inp_lport != lport) 2294 continue; 2295 2296 injail = prison_flag(inp->inp_cred, PR_IP4); 2297 if (injail) { 2298 if (prison_check_ip4(inp->inp_cred, 2299 &laddr) != 0) 2300 continue; 2301 } else { 2302 if (local_exact != NULL) 2303 continue; 2304 } 2305 2306 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2307 if (injail) 2308 goto found; 2309 else 2310 local_exact = inp; 2311 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2312 #ifdef INET6 2313 /* XXX inp locking, NULL check */ 2314 if (inp->inp_vflag & INP_IPV6PROTO) 2315 local_wild_mapped = inp; 2316 else 2317 #endif 2318 if (injail) 2319 jail_wild = inp; 2320 else 2321 local_wild = inp; 2322 } 2323 } /* LIST_FOREACH */ 2324 inp = jail_wild; 2325 if (inp == NULL) 2326 inp = local_exact; 2327 if (inp == NULL) 2328 inp = local_wild; 2329 #ifdef INET6 2330 if (inp == NULL) 2331 inp = local_wild_mapped; 2332 #endif 2333 if (inp != NULL) 2334 goto found; 2335 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 2336 INP_GROUP_UNLOCK(pcbgroup); 2337 return (NULL); 2338 2339 found: 2340 if (lookupflags & INPLOOKUP_WLOCKPCB) 2341 locked = INP_TRY_WLOCK(inp); 2342 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2343 locked = INP_TRY_RLOCK(inp); 2344 else 2345 panic("%s: locking bug", __func__); 2346 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { 2347 if (lookupflags & INPLOOKUP_WLOCKPCB) 2348 INP_WUNLOCK(inp); 2349 else 2350 INP_RUNLOCK(inp); 2351 return (NULL); 2352 } else if (!locked) 2353 in_pcbref(inp); 2354 INP_GROUP_UNLOCK(pcbgroup); 2355 if (!locked) { 2356 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2357 INP_WLOCK(inp); 2358 if (in_pcbrele_wlocked(inp)) 2359 return (NULL); 2360 } else { 2361 INP_RLOCK(inp); 2362 if (in_pcbrele_rlocked(inp)) 2363 return (NULL); 2364 } 2365 } 2366 #ifdef INVARIANTS 2367 if (lookupflags & INPLOOKUP_WLOCKPCB) 2368 INP_WLOCK_ASSERT(inp); 2369 else 2370 INP_RLOCK_ASSERT(inp); 2371 #endif 2372 return (inp); 2373 } 2374 #endif /* PCBGROUP */ 2375 2376 /* 2377 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2378 * that the caller has locked the hash list, and will not perform any further 2379 * locking or reference operations on either the hash list or the connection. 2380 */ 2381 static struct inpcb * 2382 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2383 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2384 struct ifnet *ifp, uint8_t numa_domain) 2385 { 2386 struct inpcbhead *head; 2387 struct inpcb *inp, *tmpinp; 2388 u_short fport = fport_arg, lport = lport_arg; 2389 2390 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2391 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2392 INP_HASH_LOCK_ASSERT(pcbinfo); 2393 2394 /* 2395 * First look for an exact match. 2396 */ 2397 tmpinp = NULL; 2398 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2399 pcbinfo->ipi_hashmask)]; 2400 CK_LIST_FOREACH(inp, head, inp_hash) { 2401 #ifdef INET6 2402 /* XXX inp locking */ 2403 if ((inp->inp_vflag & INP_IPV4) == 0) 2404 continue; 2405 #endif 2406 if (inp->inp_faddr.s_addr == faddr.s_addr && 2407 inp->inp_laddr.s_addr == laddr.s_addr && 2408 inp->inp_fport == fport && 2409 inp->inp_lport == lport) { 2410 /* 2411 * XXX We should be able to directly return 2412 * the inp here, without any checks. 2413 * Well unless both bound with SO_REUSEPORT? 2414 */ 2415 if (prison_flag(inp->inp_cred, PR_IP4)) 2416 return (inp); 2417 if (tmpinp == NULL) 2418 tmpinp = inp; 2419 } 2420 } 2421 if (tmpinp != NULL) 2422 return (tmpinp); 2423 2424 /* 2425 * Then look in lb group (for wildcard match). 2426 */ 2427 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2428 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, 2429 fport, lookupflags, numa_domain); 2430 if (inp != NULL) 2431 return (inp); 2432 } 2433 2434 /* 2435 * Then look for a wildcard match, if requested. 2436 */ 2437 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2438 struct inpcb *local_wild = NULL, *local_exact = NULL; 2439 #ifdef INET6 2440 struct inpcb *local_wild_mapped = NULL; 2441 #endif 2442 struct inpcb *jail_wild = NULL; 2443 int injail; 2444 2445 /* 2446 * Order of socket selection - we always prefer jails. 2447 * 1. jailed, non-wild. 2448 * 2. jailed, wild. 2449 * 3. non-jailed, non-wild. 2450 * 4. non-jailed, wild. 2451 */ 2452 2453 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2454 0, pcbinfo->ipi_hashmask)]; 2455 CK_LIST_FOREACH(inp, head, inp_hash) { 2456 #ifdef INET6 2457 /* XXX inp locking */ 2458 if ((inp->inp_vflag & INP_IPV4) == 0) 2459 continue; 2460 #endif 2461 if (inp->inp_faddr.s_addr != INADDR_ANY || 2462 inp->inp_lport != lport) 2463 continue; 2464 2465 injail = prison_flag(inp->inp_cred, PR_IP4); 2466 if (injail) { 2467 if (prison_check_ip4(inp->inp_cred, 2468 &laddr) != 0) 2469 continue; 2470 } else { 2471 if (local_exact != NULL) 2472 continue; 2473 } 2474 2475 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2476 if (injail) 2477 return (inp); 2478 else 2479 local_exact = inp; 2480 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2481 #ifdef INET6 2482 /* XXX inp locking, NULL check */ 2483 if (inp->inp_vflag & INP_IPV6PROTO) 2484 local_wild_mapped = inp; 2485 else 2486 #endif 2487 if (injail) 2488 jail_wild = inp; 2489 else 2490 local_wild = inp; 2491 } 2492 } /* LIST_FOREACH */ 2493 if (jail_wild != NULL) 2494 return (jail_wild); 2495 if (local_exact != NULL) 2496 return (local_exact); 2497 if (local_wild != NULL) 2498 return (local_wild); 2499 #ifdef INET6 2500 if (local_wild_mapped != NULL) 2501 return (local_wild_mapped); 2502 #endif 2503 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2504 2505 return (NULL); 2506 } 2507 2508 /* 2509 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2510 * hash list lock, and will return the inpcb locked (i.e., requires 2511 * INPLOOKUP_LOCKPCB). 2512 */ 2513 static struct inpcb * 2514 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2515 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2516 struct ifnet *ifp, uint8_t numa_domain) 2517 { 2518 struct inpcb *inp; 2519 2520 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2521 lookupflags & INPLOOKUP_WILDCARD, ifp, numa_domain); 2522 if (inp != NULL) { 2523 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2524 INP_WLOCK(inp); 2525 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2526 INP_RLOCK(inp); 2527 } else 2528 panic("%s: locking bug", __func__); 2529 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2530 INP_UNLOCK(inp); 2531 inp = NULL; 2532 } 2533 } 2534 2535 return (inp); 2536 } 2537 2538 /* 2539 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2540 * from which a pre-calculated hash value may be extracted. 2541 * 2542 * Possibly more of this logic should be in in_pcbgroup.c. 2543 */ 2544 struct inpcb * 2545 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2546 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2547 { 2548 #if defined(PCBGROUP) && !defined(RSS) 2549 struct inpcbgroup *pcbgroup; 2550 #endif 2551 2552 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2553 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2554 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2555 ("%s: LOCKPCB not set", __func__)); 2556 2557 /* 2558 * When not using RSS, use connection groups in preference to the 2559 * reservation table when looking up 4-tuples. When using RSS, just 2560 * use the reservation table, due to the cost of the Toeplitz hash 2561 * in software. 2562 * 2563 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2564 * we could be doing RSS with a non-Toeplitz hash that is affordable 2565 * in software. 2566 */ 2567 #if defined(PCBGROUP) && !defined(RSS) 2568 if (in_pcbgroup_enabled(pcbinfo)) { 2569 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2570 fport); 2571 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2572 laddr, lport, lookupflags, ifp)); 2573 } 2574 #endif 2575 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2576 lookupflags, ifp, M_NODOM)); 2577 } 2578 2579 struct inpcb * 2580 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2581 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2582 struct ifnet *ifp, struct mbuf *m) 2583 { 2584 #ifdef PCBGROUP 2585 struct inpcbgroup *pcbgroup; 2586 #endif 2587 2588 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2589 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2590 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2591 ("%s: LOCKPCB not set", __func__)); 2592 2593 #ifdef PCBGROUP 2594 /* 2595 * If we can use a hardware-generated hash to look up the connection 2596 * group, use that connection group to find the inpcb. Otherwise 2597 * fall back on a software hash -- or the reservation table if we're 2598 * using RSS. 2599 * 2600 * XXXRW: As above, that policy belongs in the pcbgroup code. 2601 */ 2602 if (in_pcbgroup_enabled(pcbinfo) && 2603 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2604 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2605 m->m_pkthdr.flowid); 2606 if (pcbgroup != NULL) 2607 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2608 fport, laddr, lport, lookupflags, ifp)); 2609 #ifndef RSS 2610 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2611 fport); 2612 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2613 laddr, lport, lookupflags, ifp)); 2614 #endif 2615 } 2616 #endif 2617 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2618 lookupflags, ifp, m->m_pkthdr.numa_domain)); 2619 } 2620 #endif /* INET */ 2621 2622 /* 2623 * Insert PCB onto various hash lists. 2624 */ 2625 static int 2626 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m) 2627 { 2628 struct inpcbhead *pcbhash; 2629 struct inpcbporthead *pcbporthash; 2630 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2631 struct inpcbport *phd; 2632 u_int32_t hashkey_faddr; 2633 int so_options; 2634 2635 INP_WLOCK_ASSERT(inp); 2636 INP_HASH_WLOCK_ASSERT(pcbinfo); 2637 2638 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2639 ("in_pcbinshash: INP_INHASHLIST")); 2640 2641 #ifdef INET6 2642 if (inp->inp_vflag & INP_IPV6) 2643 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2644 else 2645 #endif 2646 hashkey_faddr = inp->inp_faddr.s_addr; 2647 2648 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2649 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2650 2651 pcbporthash = &pcbinfo->ipi_porthashbase[ 2652 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2653 2654 /* 2655 * Add entry to load balance group. 2656 * Only do this if SO_REUSEPORT_LB is set. 2657 */ 2658 so_options = inp_so_options(inp); 2659 if (so_options & SO_REUSEPORT_LB) { 2660 int ret = in_pcbinslbgrouphash(inp, M_NODOM); 2661 if (ret) { 2662 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2663 return (ret); 2664 } 2665 } 2666 2667 /* 2668 * Go through port list and look for a head for this lport. 2669 */ 2670 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2671 if (phd->phd_port == inp->inp_lport) 2672 break; 2673 } 2674 /* 2675 * If none exists, malloc one and tack it on. 2676 */ 2677 if (phd == NULL) { 2678 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2679 if (phd == NULL) { 2680 return (ENOBUFS); /* XXX */ 2681 } 2682 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2683 phd->phd_port = inp->inp_lport; 2684 CK_LIST_INIT(&phd->phd_pcblist); 2685 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2686 } 2687 inp->inp_phd = phd; 2688 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2689 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2690 inp->inp_flags |= INP_INHASHLIST; 2691 #ifdef PCBGROUP 2692 if (m != NULL) { 2693 in_pcbgroup_update_mbuf(inp, m); 2694 } else { 2695 in_pcbgroup_update(inp); 2696 } 2697 #endif 2698 return (0); 2699 } 2700 2701 int 2702 in_pcbinshash(struct inpcb *inp) 2703 { 2704 2705 return (in_pcbinshash_internal(inp, NULL)); 2706 } 2707 2708 int 2709 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m) 2710 { 2711 2712 return (in_pcbinshash_internal(inp, m)); 2713 } 2714 2715 /* 2716 * Move PCB to the proper hash bucket when { faddr, fport } have been 2717 * changed. NOTE: This does not handle the case of the lport changing (the 2718 * hashed port list would have to be updated as well), so the lport must 2719 * not change after in_pcbinshash() has been called. 2720 */ 2721 void 2722 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2723 { 2724 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2725 struct inpcbhead *head; 2726 u_int32_t hashkey_faddr; 2727 2728 INP_WLOCK_ASSERT(inp); 2729 INP_HASH_WLOCK_ASSERT(pcbinfo); 2730 2731 KASSERT(inp->inp_flags & INP_INHASHLIST, 2732 ("in_pcbrehash: !INP_INHASHLIST")); 2733 2734 #ifdef INET6 2735 if (inp->inp_vflag & INP_IPV6) 2736 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2737 else 2738 #endif 2739 hashkey_faddr = inp->inp_faddr.s_addr; 2740 2741 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2742 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2743 2744 CK_LIST_REMOVE(inp, inp_hash); 2745 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2746 2747 #ifdef PCBGROUP 2748 if (m != NULL) 2749 in_pcbgroup_update_mbuf(inp, m); 2750 else 2751 in_pcbgroup_update(inp); 2752 #endif 2753 } 2754 2755 void 2756 in_pcbrehash(struct inpcb *inp) 2757 { 2758 2759 in_pcbrehash_mbuf(inp, NULL); 2760 } 2761 2762 /* 2763 * Remove PCB from various lists. 2764 */ 2765 static void 2766 in_pcbremlists(struct inpcb *inp) 2767 { 2768 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2769 2770 INP_WLOCK_ASSERT(inp); 2771 INP_LIST_WLOCK_ASSERT(pcbinfo); 2772 2773 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2774 if (inp->inp_flags & INP_INHASHLIST) { 2775 struct inpcbport *phd = inp->inp_phd; 2776 2777 INP_HASH_WLOCK(pcbinfo); 2778 2779 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2780 in_pcbremlbgrouphash(inp); 2781 2782 CK_LIST_REMOVE(inp, inp_hash); 2783 CK_LIST_REMOVE(inp, inp_portlist); 2784 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2785 CK_LIST_REMOVE(phd, phd_hash); 2786 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); 2787 } 2788 INP_HASH_WUNLOCK(pcbinfo); 2789 inp->inp_flags &= ~INP_INHASHLIST; 2790 } 2791 CK_LIST_REMOVE(inp, inp_list); 2792 pcbinfo->ipi_count--; 2793 #ifdef PCBGROUP 2794 in_pcbgroup_remove(inp); 2795 #endif 2796 } 2797 2798 /* 2799 * Check for alternatives when higher level complains 2800 * about service problems. For now, invalidate cached 2801 * routing information. If the route was created dynamically 2802 * (by a redirect), time to try a default gateway again. 2803 */ 2804 void 2805 in_losing(struct inpcb *inp) 2806 { 2807 2808 RO_INVALIDATE_CACHE(&inp->inp_route); 2809 return; 2810 } 2811 2812 /* 2813 * A set label operation has occurred at the socket layer, propagate the 2814 * label change into the in_pcb for the socket. 2815 */ 2816 void 2817 in_pcbsosetlabel(struct socket *so) 2818 { 2819 #ifdef MAC 2820 struct inpcb *inp; 2821 2822 inp = sotoinpcb(so); 2823 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2824 2825 INP_WLOCK(inp); 2826 SOCK_LOCK(so); 2827 mac_inpcb_sosetlabel(so, inp); 2828 SOCK_UNLOCK(so); 2829 INP_WUNLOCK(inp); 2830 #endif 2831 } 2832 2833 /* 2834 * ipport_tick runs once per second, determining if random port allocation 2835 * should be continued. If more than ipport_randomcps ports have been 2836 * allocated in the last second, then we return to sequential port 2837 * allocation. We return to random allocation only once we drop below 2838 * ipport_randomcps for at least ipport_randomtime seconds. 2839 */ 2840 static void 2841 ipport_tick(void *xtp) 2842 { 2843 VNET_ITERATOR_DECL(vnet_iter); 2844 2845 VNET_LIST_RLOCK_NOSLEEP(); 2846 VNET_FOREACH(vnet_iter) { 2847 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2848 if (V_ipport_tcpallocs <= 2849 V_ipport_tcplastcount + V_ipport_randomcps) { 2850 if (V_ipport_stoprandom > 0) 2851 V_ipport_stoprandom--; 2852 } else 2853 V_ipport_stoprandom = V_ipport_randomtime; 2854 V_ipport_tcplastcount = V_ipport_tcpallocs; 2855 CURVNET_RESTORE(); 2856 } 2857 VNET_LIST_RUNLOCK_NOSLEEP(); 2858 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2859 } 2860 2861 static void 2862 ip_fini(void *xtp) 2863 { 2864 2865 callout_stop(&ipport_tick_callout); 2866 } 2867 2868 /* 2869 * The ipport_callout should start running at about the time we attach the 2870 * inet or inet6 domains. 2871 */ 2872 static void 2873 ipport_tick_init(const void *unused __unused) 2874 { 2875 2876 /* Start ipport_tick. */ 2877 callout_init(&ipport_tick_callout, 1); 2878 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2879 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2880 SHUTDOWN_PRI_DEFAULT); 2881 } 2882 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2883 ipport_tick_init, NULL); 2884 2885 void 2886 inp_wlock(struct inpcb *inp) 2887 { 2888 2889 INP_WLOCK(inp); 2890 } 2891 2892 void 2893 inp_wunlock(struct inpcb *inp) 2894 { 2895 2896 INP_WUNLOCK(inp); 2897 } 2898 2899 void 2900 inp_rlock(struct inpcb *inp) 2901 { 2902 2903 INP_RLOCK(inp); 2904 } 2905 2906 void 2907 inp_runlock(struct inpcb *inp) 2908 { 2909 2910 INP_RUNLOCK(inp); 2911 } 2912 2913 #ifdef INVARIANT_SUPPORT 2914 void 2915 inp_lock_assert(struct inpcb *inp) 2916 { 2917 2918 INP_WLOCK_ASSERT(inp); 2919 } 2920 2921 void 2922 inp_unlock_assert(struct inpcb *inp) 2923 { 2924 2925 INP_UNLOCK_ASSERT(inp); 2926 } 2927 #endif 2928 2929 void 2930 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2931 { 2932 struct inpcb *inp; 2933 2934 INP_INFO_WLOCK(&V_tcbinfo); 2935 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2936 INP_WLOCK(inp); 2937 func(inp, arg); 2938 INP_WUNLOCK(inp); 2939 } 2940 INP_INFO_WUNLOCK(&V_tcbinfo); 2941 } 2942 2943 struct socket * 2944 inp_inpcbtosocket(struct inpcb *inp) 2945 { 2946 2947 INP_WLOCK_ASSERT(inp); 2948 return (inp->inp_socket); 2949 } 2950 2951 struct tcpcb * 2952 inp_inpcbtotcpcb(struct inpcb *inp) 2953 { 2954 2955 INP_WLOCK_ASSERT(inp); 2956 return ((struct tcpcb *)inp->inp_ppcb); 2957 } 2958 2959 int 2960 inp_ip_tos_get(const struct inpcb *inp) 2961 { 2962 2963 return (inp->inp_ip_tos); 2964 } 2965 2966 void 2967 inp_ip_tos_set(struct inpcb *inp, int val) 2968 { 2969 2970 inp->inp_ip_tos = val; 2971 } 2972 2973 void 2974 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2975 uint32_t *faddr, uint16_t *fp) 2976 { 2977 2978 INP_LOCK_ASSERT(inp); 2979 *laddr = inp->inp_laddr.s_addr; 2980 *faddr = inp->inp_faddr.s_addr; 2981 *lp = inp->inp_lport; 2982 *fp = inp->inp_fport; 2983 } 2984 2985 struct inpcb * 2986 so_sotoinpcb(struct socket *so) 2987 { 2988 2989 return (sotoinpcb(so)); 2990 } 2991 2992 struct tcpcb * 2993 so_sototcpcb(struct socket *so) 2994 { 2995 2996 return (sototcpcb(so)); 2997 } 2998 2999 /* 3000 * Create an external-format (``xinpcb'') structure using the information in 3001 * the kernel-format in_pcb structure pointed to by inp. This is done to 3002 * reduce the spew of irrelevant information over this interface, to isolate 3003 * user code from changes in the kernel structure, and potentially to provide 3004 * information-hiding if we decide that some of this information should be 3005 * hidden from users. 3006 */ 3007 void 3008 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 3009 { 3010 3011 bzero(xi, sizeof(*xi)); 3012 xi->xi_len = sizeof(struct xinpcb); 3013 if (inp->inp_socket) 3014 sotoxsocket(inp->inp_socket, &xi->xi_socket); 3015 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 3016 xi->inp_gencnt = inp->inp_gencnt; 3017 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 3018 xi->inp_flow = inp->inp_flow; 3019 xi->inp_flowid = inp->inp_flowid; 3020 xi->inp_flowtype = inp->inp_flowtype; 3021 xi->inp_flags = inp->inp_flags; 3022 xi->inp_flags2 = inp->inp_flags2; 3023 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 3024 xi->in6p_cksum = inp->in6p_cksum; 3025 xi->in6p_hops = inp->in6p_hops; 3026 xi->inp_ip_tos = inp->inp_ip_tos; 3027 xi->inp_vflag = inp->inp_vflag; 3028 xi->inp_ip_ttl = inp->inp_ip_ttl; 3029 xi->inp_ip_p = inp->inp_ip_p; 3030 xi->inp_ip_minttl = inp->inp_ip_minttl; 3031 } 3032 3033 #ifdef DDB 3034 static void 3035 db_print_indent(int indent) 3036 { 3037 int i; 3038 3039 for (i = 0; i < indent; i++) 3040 db_printf(" "); 3041 } 3042 3043 static void 3044 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 3045 { 3046 char faddr_str[48], laddr_str[48]; 3047 3048 db_print_indent(indent); 3049 db_printf("%s at %p\n", name, inc); 3050 3051 indent += 2; 3052 3053 #ifdef INET6 3054 if (inc->inc_flags & INC_ISIPV6) { 3055 /* IPv6. */ 3056 ip6_sprintf(laddr_str, &inc->inc6_laddr); 3057 ip6_sprintf(faddr_str, &inc->inc6_faddr); 3058 } else 3059 #endif 3060 { 3061 /* IPv4. */ 3062 inet_ntoa_r(inc->inc_laddr, laddr_str); 3063 inet_ntoa_r(inc->inc_faddr, faddr_str); 3064 } 3065 db_print_indent(indent); 3066 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3067 ntohs(inc->inc_lport)); 3068 db_print_indent(indent); 3069 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3070 ntohs(inc->inc_fport)); 3071 } 3072 3073 static void 3074 db_print_inpflags(int inp_flags) 3075 { 3076 int comma; 3077 3078 comma = 0; 3079 if (inp_flags & INP_RECVOPTS) { 3080 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3081 comma = 1; 3082 } 3083 if (inp_flags & INP_RECVRETOPTS) { 3084 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3085 comma = 1; 3086 } 3087 if (inp_flags & INP_RECVDSTADDR) { 3088 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3089 comma = 1; 3090 } 3091 if (inp_flags & INP_ORIGDSTADDR) { 3092 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3093 comma = 1; 3094 } 3095 if (inp_flags & INP_HDRINCL) { 3096 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3097 comma = 1; 3098 } 3099 if (inp_flags & INP_HIGHPORT) { 3100 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3101 comma = 1; 3102 } 3103 if (inp_flags & INP_LOWPORT) { 3104 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3105 comma = 1; 3106 } 3107 if (inp_flags & INP_ANONPORT) { 3108 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3109 comma = 1; 3110 } 3111 if (inp_flags & INP_RECVIF) { 3112 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3113 comma = 1; 3114 } 3115 if (inp_flags & INP_MTUDISC) { 3116 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3117 comma = 1; 3118 } 3119 if (inp_flags & INP_RECVTTL) { 3120 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3121 comma = 1; 3122 } 3123 if (inp_flags & INP_DONTFRAG) { 3124 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3125 comma = 1; 3126 } 3127 if (inp_flags & INP_RECVTOS) { 3128 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3129 comma = 1; 3130 } 3131 if (inp_flags & IN6P_IPV6_V6ONLY) { 3132 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3133 comma = 1; 3134 } 3135 if (inp_flags & IN6P_PKTINFO) { 3136 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3137 comma = 1; 3138 } 3139 if (inp_flags & IN6P_HOPLIMIT) { 3140 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3141 comma = 1; 3142 } 3143 if (inp_flags & IN6P_HOPOPTS) { 3144 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3145 comma = 1; 3146 } 3147 if (inp_flags & IN6P_DSTOPTS) { 3148 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3149 comma = 1; 3150 } 3151 if (inp_flags & IN6P_RTHDR) { 3152 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3153 comma = 1; 3154 } 3155 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3156 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3157 comma = 1; 3158 } 3159 if (inp_flags & IN6P_TCLASS) { 3160 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3161 comma = 1; 3162 } 3163 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3164 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3165 comma = 1; 3166 } 3167 if (inp_flags & INP_TIMEWAIT) { 3168 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3169 comma = 1; 3170 } 3171 if (inp_flags & INP_ONESBCAST) { 3172 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3173 comma = 1; 3174 } 3175 if (inp_flags & INP_DROPPED) { 3176 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3177 comma = 1; 3178 } 3179 if (inp_flags & INP_SOCKREF) { 3180 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3181 comma = 1; 3182 } 3183 if (inp_flags & IN6P_RFC2292) { 3184 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3185 comma = 1; 3186 } 3187 if (inp_flags & IN6P_MTU) { 3188 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3189 comma = 1; 3190 } 3191 } 3192 3193 static void 3194 db_print_inpvflag(u_char inp_vflag) 3195 { 3196 int comma; 3197 3198 comma = 0; 3199 if (inp_vflag & INP_IPV4) { 3200 db_printf("%sINP_IPV4", comma ? ", " : ""); 3201 comma = 1; 3202 } 3203 if (inp_vflag & INP_IPV6) { 3204 db_printf("%sINP_IPV6", comma ? ", " : ""); 3205 comma = 1; 3206 } 3207 if (inp_vflag & INP_IPV6PROTO) { 3208 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3209 comma = 1; 3210 } 3211 } 3212 3213 static void 3214 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3215 { 3216 3217 db_print_indent(indent); 3218 db_printf("%s at %p\n", name, inp); 3219 3220 indent += 2; 3221 3222 db_print_indent(indent); 3223 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3224 3225 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3226 3227 db_print_indent(indent); 3228 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3229 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3230 3231 db_print_indent(indent); 3232 db_printf("inp_label: %p inp_flags: 0x%x (", 3233 inp->inp_label, inp->inp_flags); 3234 db_print_inpflags(inp->inp_flags); 3235 db_printf(")\n"); 3236 3237 db_print_indent(indent); 3238 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3239 inp->inp_vflag); 3240 db_print_inpvflag(inp->inp_vflag); 3241 db_printf(")\n"); 3242 3243 db_print_indent(indent); 3244 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3245 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3246 3247 db_print_indent(indent); 3248 #ifdef INET6 3249 if (inp->inp_vflag & INP_IPV6) { 3250 db_printf("in6p_options: %p in6p_outputopts: %p " 3251 "in6p_moptions: %p\n", inp->in6p_options, 3252 inp->in6p_outputopts, inp->in6p_moptions); 3253 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3254 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3255 inp->in6p_hops); 3256 } else 3257 #endif 3258 { 3259 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3260 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3261 inp->inp_options, inp->inp_moptions); 3262 } 3263 3264 db_print_indent(indent); 3265 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3266 (uintmax_t)inp->inp_gencnt); 3267 } 3268 3269 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3270 { 3271 struct inpcb *inp; 3272 3273 if (!have_addr) { 3274 db_printf("usage: show inpcb <addr>\n"); 3275 return; 3276 } 3277 inp = (struct inpcb *)addr; 3278 3279 db_print_inpcb(inp, "inpcb", 0); 3280 } 3281 #endif /* DDB */ 3282 3283 #ifdef RATELIMIT 3284 /* 3285 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3286 * if any. 3287 */ 3288 int 3289 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3290 { 3291 union if_snd_tag_modify_params params = { 3292 .rate_limit.max_rate = max_pacing_rate, 3293 .rate_limit.flags = M_NOWAIT, 3294 }; 3295 struct m_snd_tag *mst; 3296 int error; 3297 3298 mst = inp->inp_snd_tag; 3299 if (mst == NULL) 3300 return (EINVAL); 3301 3302 if (mst->sw->snd_tag_modify == NULL) { 3303 error = EOPNOTSUPP; 3304 } else { 3305 error = mst->sw->snd_tag_modify(mst, ¶ms); 3306 } 3307 return (error); 3308 } 3309 3310 /* 3311 * Query existing TX rate limit based on the existing 3312 * "inp->inp_snd_tag", if any. 3313 */ 3314 int 3315 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3316 { 3317 union if_snd_tag_query_params params = { }; 3318 struct m_snd_tag *mst; 3319 int error; 3320 3321 mst = inp->inp_snd_tag; 3322 if (mst == NULL) 3323 return (EINVAL); 3324 3325 if (mst->sw->snd_tag_query == NULL) { 3326 error = EOPNOTSUPP; 3327 } else { 3328 error = mst->sw->snd_tag_query(mst, ¶ms); 3329 if (error == 0 && p_max_pacing_rate != NULL) 3330 *p_max_pacing_rate = params.rate_limit.max_rate; 3331 } 3332 return (error); 3333 } 3334 3335 /* 3336 * Query existing TX queue level based on the existing 3337 * "inp->inp_snd_tag", if any. 3338 */ 3339 int 3340 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3341 { 3342 union if_snd_tag_query_params params = { }; 3343 struct m_snd_tag *mst; 3344 int error; 3345 3346 mst = inp->inp_snd_tag; 3347 if (mst == NULL) 3348 return (EINVAL); 3349 3350 if (mst->sw->snd_tag_query == NULL) 3351 return (EOPNOTSUPP); 3352 3353 error = mst->sw->snd_tag_query(mst, ¶ms); 3354 if (error == 0 && p_txqueue_level != NULL) 3355 *p_txqueue_level = params.rate_limit.queue_level; 3356 return (error); 3357 } 3358 3359 /* 3360 * Allocate a new TX rate limit send tag from the network interface 3361 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3362 */ 3363 int 3364 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3365 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3366 3367 { 3368 union if_snd_tag_alloc_params params = { 3369 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3370 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3371 .rate_limit.hdr.flowid = flowid, 3372 .rate_limit.hdr.flowtype = flowtype, 3373 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3374 .rate_limit.max_rate = max_pacing_rate, 3375 .rate_limit.flags = M_NOWAIT, 3376 }; 3377 int error; 3378 3379 INP_WLOCK_ASSERT(inp); 3380 3381 /* 3382 * If there is already a send tag, or the INP is being torn 3383 * down, allocating a new send tag is not allowed. Else send 3384 * tags may leak. 3385 */ 3386 if (*st != NULL || (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0) 3387 return (EINVAL); 3388 3389 error = m_snd_tag_alloc(ifp, ¶ms, st); 3390 #ifdef INET 3391 if (error == 0) { 3392 counter_u64_add(rate_limit_set_ok, 1); 3393 counter_u64_add(rate_limit_active, 1); 3394 } else if (error != EOPNOTSUPP) 3395 counter_u64_add(rate_limit_alloc_fail, 1); 3396 #endif 3397 return (error); 3398 } 3399 3400 void 3401 in_pcbdetach_tag(struct m_snd_tag *mst) 3402 { 3403 3404 m_snd_tag_rele(mst); 3405 #ifdef INET 3406 counter_u64_add(rate_limit_active, -1); 3407 #endif 3408 } 3409 3410 /* 3411 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3412 * if any: 3413 */ 3414 void 3415 in_pcbdetach_txrtlmt(struct inpcb *inp) 3416 { 3417 struct m_snd_tag *mst; 3418 3419 INP_WLOCK_ASSERT(inp); 3420 3421 mst = inp->inp_snd_tag; 3422 inp->inp_snd_tag = NULL; 3423 3424 if (mst == NULL) 3425 return; 3426 3427 m_snd_tag_rele(mst); 3428 #ifdef INET 3429 counter_u64_add(rate_limit_active, -1); 3430 #endif 3431 } 3432 3433 int 3434 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3435 { 3436 int error; 3437 3438 /* 3439 * If the existing send tag is for the wrong interface due to 3440 * a route change, first drop the existing tag. Set the 3441 * CHANGED flag so that we will keep trying to allocate a new 3442 * tag if we fail to allocate one this time. 3443 */ 3444 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3445 in_pcbdetach_txrtlmt(inp); 3446 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3447 } 3448 3449 /* 3450 * NOTE: When attaching to a network interface a reference is 3451 * made to ensure the network interface doesn't go away until 3452 * all ratelimit connections are gone. The network interface 3453 * pointers compared below represent valid network interfaces, 3454 * except when comparing towards NULL. 3455 */ 3456 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3457 error = 0; 3458 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3459 if (inp->inp_snd_tag != NULL) 3460 in_pcbdetach_txrtlmt(inp); 3461 error = 0; 3462 } else if (inp->inp_snd_tag == NULL) { 3463 /* 3464 * In order to utilize packet pacing with RSS, we need 3465 * to wait until there is a valid RSS hash before we 3466 * can proceed: 3467 */ 3468 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3469 error = EAGAIN; 3470 } else { 3471 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3472 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3473 } 3474 } else { 3475 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3476 } 3477 if (error == 0 || error == EOPNOTSUPP) 3478 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3479 3480 return (error); 3481 } 3482 3483 /* 3484 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3485 * is set in the fast path and will attach/detach/modify the TX rate 3486 * limit send tag based on the socket's so_max_pacing_rate value. 3487 */ 3488 void 3489 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3490 { 3491 struct socket *socket; 3492 uint32_t max_pacing_rate; 3493 bool did_upgrade; 3494 int error; 3495 3496 if (inp == NULL) 3497 return; 3498 3499 socket = inp->inp_socket; 3500 if (socket == NULL) 3501 return; 3502 3503 if (!INP_WLOCKED(inp)) { 3504 /* 3505 * NOTE: If the write locking fails, we need to bail 3506 * out and use the non-ratelimited ring for the 3507 * transmit until there is a new chance to get the 3508 * write lock. 3509 */ 3510 if (!INP_TRY_UPGRADE(inp)) 3511 return; 3512 did_upgrade = 1; 3513 } else { 3514 did_upgrade = 0; 3515 } 3516 3517 /* 3518 * NOTE: The so_max_pacing_rate value is read unlocked, 3519 * because atomic updates are not required since the variable 3520 * is checked at every mbuf we send. It is assumed that the 3521 * variable read itself will be atomic. 3522 */ 3523 max_pacing_rate = socket->so_max_pacing_rate; 3524 3525 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3526 3527 if (did_upgrade) 3528 INP_DOWNGRADE(inp); 3529 } 3530 3531 /* 3532 * Track route changes for TX rate limiting. 3533 */ 3534 void 3535 in_pcboutput_eagain(struct inpcb *inp) 3536 { 3537 bool did_upgrade; 3538 3539 if (inp == NULL) 3540 return; 3541 3542 if (inp->inp_snd_tag == NULL) 3543 return; 3544 3545 if (!INP_WLOCKED(inp)) { 3546 /* 3547 * NOTE: If the write locking fails, we need to bail 3548 * out and use the non-ratelimited ring for the 3549 * transmit until there is a new chance to get the 3550 * write lock. 3551 */ 3552 if (!INP_TRY_UPGRADE(inp)) 3553 return; 3554 did_upgrade = 1; 3555 } else { 3556 did_upgrade = 0; 3557 } 3558 3559 /* detach rate limiting */ 3560 in_pcbdetach_txrtlmt(inp); 3561 3562 /* make sure new mbuf send tag allocation is made */ 3563 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3564 3565 if (did_upgrade) 3566 INP_DOWNGRADE(inp); 3567 } 3568 3569 #ifdef INET 3570 static void 3571 rl_init(void *st) 3572 { 3573 rate_limit_new = counter_u64_alloc(M_WAITOK); 3574 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3575 rate_limit_active = counter_u64_alloc(M_WAITOK); 3576 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3577 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3578 } 3579 3580 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3581 #endif 3582 #endif /* RATELIMIT */ 3583