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 INP_WLOCK_ASSERT(inp); 658 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 659 660 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 661 return (EINVAL); 662 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 663 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 664 &inp->inp_lport, cred); 665 if (error) 666 return (error); 667 if (in_pcbinshash(inp) != 0) { 668 inp->inp_laddr.s_addr = INADDR_ANY; 669 inp->inp_lport = 0; 670 return (EAGAIN); 671 } 672 if (anonport) 673 inp->inp_flags |= INP_ANONPORT; 674 return (0); 675 } 676 #endif 677 678 #if defined(INET) || defined(INET6) 679 /* 680 * Assign a local port like in_pcb_lport(), but also used with connect() 681 * and a foreign address and port. If fsa is non-NULL, choose a local port 682 * that is unused with those, otherwise one that is completely unused. 683 * lsa can be NULL for IPv6. 684 */ 685 int 686 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, 687 struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags) 688 { 689 struct inpcbinfo *pcbinfo; 690 struct inpcb *tmpinp; 691 unsigned short *lastport; 692 int count, dorandom, error; 693 u_short aux, first, last, lport; 694 #ifdef INET 695 struct in_addr laddr, faddr; 696 #endif 697 #ifdef INET6 698 struct in6_addr *laddr6, *faddr6; 699 #endif 700 701 pcbinfo = inp->inp_pcbinfo; 702 703 /* 704 * Because no actual state changes occur here, a global write lock on 705 * the pcbinfo isn't required. 706 */ 707 INP_LOCK_ASSERT(inp); 708 INP_HASH_LOCK_ASSERT(pcbinfo); 709 710 if (inp->inp_flags & INP_HIGHPORT) { 711 first = V_ipport_hifirstauto; /* sysctl */ 712 last = V_ipport_hilastauto; 713 lastport = &pcbinfo->ipi_lasthi; 714 } else if (inp->inp_flags & INP_LOWPORT) { 715 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 716 if (error) 717 return (error); 718 first = V_ipport_lowfirstauto; /* 1023 */ 719 last = V_ipport_lowlastauto; /* 600 */ 720 lastport = &pcbinfo->ipi_lastlow; 721 } else { 722 first = V_ipport_firstauto; /* sysctl */ 723 last = V_ipport_lastauto; 724 lastport = &pcbinfo->ipi_lastport; 725 } 726 /* 727 * For UDP(-Lite), use random port allocation as long as the user 728 * allows it. For TCP (and as of yet unknown) connections, 729 * use random port allocation only if the user allows it AND 730 * ipport_tick() allows it. 731 */ 732 if (V_ipport_randomized && 733 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 734 pcbinfo == &V_ulitecbinfo)) 735 dorandom = 1; 736 else 737 dorandom = 0; 738 /* 739 * It makes no sense to do random port allocation if 740 * we have the only port available. 741 */ 742 if (first == last) 743 dorandom = 0; 744 /* Make sure to not include UDP(-Lite) packets in the count. */ 745 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 746 V_ipport_tcpallocs++; 747 /* 748 * Instead of having two loops further down counting up or down 749 * make sure that first is always <= last and go with only one 750 * code path implementing all logic. 751 */ 752 if (first > last) { 753 aux = first; 754 first = last; 755 last = aux; 756 } 757 758 #ifdef INET 759 laddr.s_addr = INADDR_ANY; 760 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 761 if (lsa != NULL) 762 laddr = ((struct sockaddr_in *)lsa)->sin_addr; 763 if (fsa != NULL) 764 faddr = ((struct sockaddr_in *)fsa)->sin_addr; 765 } 766 #endif 767 #ifdef INET6 768 laddr6 = NULL; 769 if ((inp->inp_vflag & INP_IPV6) != 0) { 770 if (lsa != NULL) 771 laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; 772 if (fsa != NULL) 773 faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; 774 } 775 #endif 776 777 tmpinp = NULL; 778 lport = *lportp; 779 780 if (dorandom) 781 *lastport = first + (arc4random() % (last - first)); 782 783 count = last - first; 784 785 do { 786 if (count-- < 0) /* completely used? */ 787 return (EADDRNOTAVAIL); 788 ++*lastport; 789 if (*lastport < first || *lastport > last) 790 *lastport = first; 791 lport = htons(*lastport); 792 793 if (fsa != NULL) { 794 #ifdef INET 795 if (lsa->sa_family == AF_INET) { 796 tmpinp = in_pcblookup_hash_locked(pcbinfo, 797 faddr, fport, laddr, lport, lookupflags, 798 NULL, M_NODOM); 799 } 800 #endif 801 #ifdef INET6 802 if (lsa->sa_family == AF_INET6) { 803 tmpinp = in6_pcblookup_hash_locked(pcbinfo, 804 faddr6, fport, laddr6, lport, lookupflags, 805 NULL, M_NODOM); 806 } 807 #endif 808 } else { 809 #ifdef INET6 810 if ((inp->inp_vflag & INP_IPV6) != 0) 811 tmpinp = in6_pcblookup_local(pcbinfo, 812 &inp->in6p_laddr, lport, lookupflags, cred); 813 #endif 814 #if defined(INET) && defined(INET6) 815 else 816 #endif 817 #ifdef INET 818 tmpinp = in_pcblookup_local(pcbinfo, laddr, 819 lport, lookupflags, cred); 820 #endif 821 } 822 } while (tmpinp != NULL); 823 824 *lportp = lport; 825 826 return (0); 827 } 828 829 /* 830 * Select a local port (number) to use. 831 */ 832 int 833 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 834 struct ucred *cred, int lookupflags) 835 { 836 struct sockaddr_in laddr; 837 838 if (laddrp) { 839 bzero(&laddr, sizeof(laddr)); 840 laddr.sin_family = AF_INET; 841 laddr.sin_addr = *laddrp; 842 } 843 return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : 844 NULL, lportp, NULL, 0, cred, lookupflags)); 845 } 846 847 /* 848 * Return cached socket options. 849 */ 850 int 851 inp_so_options(const struct inpcb *inp) 852 { 853 int so_options; 854 855 so_options = 0; 856 857 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 858 so_options |= SO_REUSEPORT_LB; 859 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 860 so_options |= SO_REUSEPORT; 861 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 862 so_options |= SO_REUSEADDR; 863 return (so_options); 864 } 865 #endif /* INET || INET6 */ 866 867 /* 868 * Check if a new BINDMULTI socket is allowed to be created. 869 * 870 * ni points to the new inp. 871 * oi points to the exisitng inp. 872 * 873 * This checks whether the existing inp also has BINDMULTI and 874 * whether the credentials match. 875 */ 876 int 877 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 878 { 879 /* Check permissions match */ 880 if ((ni->inp_flags2 & INP_BINDMULTI) && 881 (ni->inp_cred->cr_uid != 882 oi->inp_cred->cr_uid)) 883 return (0); 884 885 /* Check the existing inp has BINDMULTI set */ 886 if ((ni->inp_flags2 & INP_BINDMULTI) && 887 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 888 return (0); 889 890 /* 891 * We're okay - either INP_BINDMULTI isn't set on ni, or 892 * it is and it matches the checks. 893 */ 894 return (1); 895 } 896 897 #ifdef INET 898 /* 899 * Set up a bind operation on a PCB, performing port allocation 900 * as required, but do not actually modify the PCB. Callers can 901 * either complete the bind by setting inp_laddr/inp_lport and 902 * calling in_pcbinshash(), or they can just use the resulting 903 * port and address to authorise the sending of a once-off packet. 904 * 905 * On error, the values of *laddrp and *lportp are not changed. 906 */ 907 int 908 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 909 u_short *lportp, struct ucred *cred) 910 { 911 struct socket *so = inp->inp_socket; 912 struct sockaddr_in *sin; 913 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 914 struct in_addr laddr; 915 u_short lport = 0; 916 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 917 int error; 918 919 /* 920 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 921 * so that we don't have to add to the (already messy) code below. 922 */ 923 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 924 925 /* 926 * No state changes, so read locks are sufficient here. 927 */ 928 INP_LOCK_ASSERT(inp); 929 INP_HASH_LOCK_ASSERT(pcbinfo); 930 931 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 932 return (EADDRNOTAVAIL); 933 laddr.s_addr = *laddrp; 934 if (nam != NULL && laddr.s_addr != INADDR_ANY) 935 return (EINVAL); 936 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 937 lookupflags = INPLOOKUP_WILDCARD; 938 if (nam == NULL) { 939 if ((error = prison_local_ip4(cred, &laddr)) != 0) 940 return (error); 941 } else { 942 sin = (struct sockaddr_in *)nam; 943 if (nam->sa_len != sizeof (*sin)) 944 return (EINVAL); 945 #ifdef notdef 946 /* 947 * We should check the family, but old programs 948 * incorrectly fail to initialize it. 949 */ 950 if (sin->sin_family != AF_INET) 951 return (EAFNOSUPPORT); 952 #endif 953 error = prison_local_ip4(cred, &sin->sin_addr); 954 if (error) 955 return (error); 956 if (sin->sin_port != *lportp) { 957 /* Don't allow the port to change. */ 958 if (*lportp != 0) 959 return (EINVAL); 960 lport = sin->sin_port; 961 } 962 /* NB: lport is left as 0 if the port isn't being changed. */ 963 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 964 /* 965 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 966 * allow complete duplication of binding if 967 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 968 * and a multicast address is bound on both 969 * new and duplicated sockets. 970 */ 971 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 972 reuseport = SO_REUSEADDR|SO_REUSEPORT; 973 /* 974 * XXX: How to deal with SO_REUSEPORT_LB here? 975 * Treat same as SO_REUSEPORT for now. 976 */ 977 if ((so->so_options & 978 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 979 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 980 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 981 sin->sin_port = 0; /* yech... */ 982 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 983 /* 984 * Is the address a local IP address? 985 * If INP_BINDANY is set, then the socket may be bound 986 * to any endpoint address, local or not. 987 */ 988 if ((inp->inp_flags & INP_BINDANY) == 0 && 989 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 990 return (EADDRNOTAVAIL); 991 } 992 laddr = sin->sin_addr; 993 if (lport) { 994 struct inpcb *t; 995 struct tcptw *tw; 996 997 /* GROSS */ 998 if (ntohs(lport) <= V_ipport_reservedhigh && 999 ntohs(lport) >= V_ipport_reservedlow && 1000 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 1001 return (EACCES); 1002 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 1003 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 1004 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 1005 lport, INPLOOKUP_WILDCARD, cred); 1006 /* 1007 * XXX 1008 * This entire block sorely needs a rewrite. 1009 */ 1010 if (t && 1011 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 1012 ((t->inp_flags & INP_TIMEWAIT) == 0) && 1013 (so->so_type != SOCK_STREAM || 1014 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 1015 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 1016 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 1017 (t->inp_flags2 & INP_REUSEPORT) || 1018 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 1019 (inp->inp_cred->cr_uid != 1020 t->inp_cred->cr_uid)) 1021 return (EADDRINUSE); 1022 1023 /* 1024 * If the socket is a BINDMULTI socket, then 1025 * the credentials need to match and the 1026 * original socket also has to have been bound 1027 * with BINDMULTI. 1028 */ 1029 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 1030 return (EADDRINUSE); 1031 } 1032 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 1033 lport, lookupflags, cred); 1034 if (t && (t->inp_flags & INP_TIMEWAIT)) { 1035 /* 1036 * XXXRW: If an incpb has had its timewait 1037 * state recycled, we treat the address as 1038 * being in use (for now). This is better 1039 * than a panic, but not desirable. 1040 */ 1041 tw = intotw(t); 1042 if (tw == NULL || 1043 ((reuseport & tw->tw_so_options) == 0 && 1044 (reuseport_lb & 1045 tw->tw_so_options) == 0)) { 1046 return (EADDRINUSE); 1047 } 1048 } else if (t && 1049 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 1050 (reuseport & inp_so_options(t)) == 0 && 1051 (reuseport_lb & inp_so_options(t)) == 0) { 1052 #ifdef INET6 1053 if (ntohl(sin->sin_addr.s_addr) != 1054 INADDR_ANY || 1055 ntohl(t->inp_laddr.s_addr) != 1056 INADDR_ANY || 1057 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 1058 (t->inp_vflag & INP_IPV6PROTO) == 0) 1059 #endif 1060 return (EADDRINUSE); 1061 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 1062 return (EADDRINUSE); 1063 } 1064 } 1065 } 1066 if (*lportp != 0) 1067 lport = *lportp; 1068 if (lport == 0) { 1069 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 1070 if (error != 0) 1071 return (error); 1072 } 1073 *laddrp = laddr.s_addr; 1074 *lportp = lport; 1075 return (0); 1076 } 1077 1078 /* 1079 * Connect from a socket to a specified address. 1080 * Both address and port must be specified in argument sin. 1081 * If don't have a local address for this socket yet, 1082 * then pick one. 1083 */ 1084 int 1085 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 1086 struct ucred *cred, struct mbuf *m, bool rehash) 1087 { 1088 u_short lport, fport; 1089 in_addr_t laddr, faddr; 1090 int anonport, error; 1091 1092 INP_WLOCK_ASSERT(inp); 1093 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1094 1095 lport = inp->inp_lport; 1096 laddr = inp->inp_laddr.s_addr; 1097 anonport = (lport == 0); 1098 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 1099 NULL, cred); 1100 if (error) 1101 return (error); 1102 1103 /* Do the initial binding of the local address if required. */ 1104 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 1105 KASSERT(rehash == true, 1106 ("Rehashing required for unbound inps")); 1107 inp->inp_lport = lport; 1108 inp->inp_laddr.s_addr = laddr; 1109 if (in_pcbinshash(inp) != 0) { 1110 inp->inp_laddr.s_addr = INADDR_ANY; 1111 inp->inp_lport = 0; 1112 return (EAGAIN); 1113 } 1114 } 1115 1116 /* Commit the remaining changes. */ 1117 inp->inp_lport = lport; 1118 inp->inp_laddr.s_addr = laddr; 1119 inp->inp_faddr.s_addr = faddr; 1120 inp->inp_fport = fport; 1121 if (rehash) { 1122 in_pcbrehash_mbuf(inp, m); 1123 } else { 1124 in_pcbinshash_mbuf(inp, m); 1125 } 1126 1127 if (anonport) 1128 inp->inp_flags |= INP_ANONPORT; 1129 return (0); 1130 } 1131 1132 int 1133 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 1134 { 1135 1136 return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true)); 1137 } 1138 1139 /* 1140 * Do proper source address selection on an unbound socket in case 1141 * of connect. Take jails into account as well. 1142 */ 1143 int 1144 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1145 struct ucred *cred) 1146 { 1147 struct ifaddr *ifa; 1148 struct sockaddr *sa; 1149 struct sockaddr_in *sin, dst; 1150 struct nhop_object *nh; 1151 int error; 1152 1153 NET_EPOCH_ASSERT(); 1154 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1155 /* 1156 * Bypass source address selection and use the primary jail IP 1157 * if requested. 1158 */ 1159 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1160 return (0); 1161 1162 error = 0; 1163 1164 nh = NULL; 1165 bzero(&dst, sizeof(dst)); 1166 sin = &dst; 1167 sin->sin_family = AF_INET; 1168 sin->sin_len = sizeof(struct sockaddr_in); 1169 sin->sin_addr.s_addr = faddr->s_addr; 1170 1171 /* 1172 * If route is known our src addr is taken from the i/f, 1173 * else punt. 1174 * 1175 * Find out route to destination. 1176 */ 1177 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1178 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 1179 0, NHR_NONE, 0); 1180 1181 /* 1182 * If we found a route, use the address corresponding to 1183 * the outgoing interface. 1184 * 1185 * Otherwise assume faddr is reachable on a directly connected 1186 * network and try to find a corresponding interface to take 1187 * the source address from. 1188 */ 1189 if (nh == NULL || nh->nh_ifp == NULL) { 1190 struct in_ifaddr *ia; 1191 struct ifnet *ifp; 1192 1193 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1194 inp->inp_socket->so_fibnum)); 1195 if (ia == NULL) { 1196 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1197 inp->inp_socket->so_fibnum)); 1198 } 1199 if (ia == NULL) { 1200 error = ENETUNREACH; 1201 goto done; 1202 } 1203 1204 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1205 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1206 goto done; 1207 } 1208 1209 ifp = ia->ia_ifp; 1210 ia = NULL; 1211 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1212 sa = ifa->ifa_addr; 1213 if (sa->sa_family != AF_INET) 1214 continue; 1215 sin = (struct sockaddr_in *)sa; 1216 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1217 ia = (struct in_ifaddr *)ifa; 1218 break; 1219 } 1220 } 1221 if (ia != NULL) { 1222 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1223 goto done; 1224 } 1225 1226 /* 3. As a last resort return the 'default' jail address. */ 1227 error = prison_get_ip4(cred, laddr); 1228 goto done; 1229 } 1230 1231 /* 1232 * If the outgoing interface on the route found is not 1233 * a loopback interface, use the address from that interface. 1234 * In case of jails do those three steps: 1235 * 1. check if the interface address belongs to the jail. If so use it. 1236 * 2. check if we have any address on the outgoing interface 1237 * belonging to this jail. If so use it. 1238 * 3. as a last resort return the 'default' jail address. 1239 */ 1240 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) { 1241 struct in_ifaddr *ia; 1242 struct ifnet *ifp; 1243 1244 /* If not jailed, use the default returned. */ 1245 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1246 ia = (struct in_ifaddr *)nh->nh_ifa; 1247 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1248 goto done; 1249 } 1250 1251 /* Jailed. */ 1252 /* 1. Check if the iface address belongs to the jail. */ 1253 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr; 1254 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1255 ia = (struct in_ifaddr *)nh->nh_ifa; 1256 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1257 goto done; 1258 } 1259 1260 /* 1261 * 2. Check if we have any address on the outgoing interface 1262 * belonging to this jail. 1263 */ 1264 ia = NULL; 1265 ifp = nh->nh_ifp; 1266 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1267 sa = ifa->ifa_addr; 1268 if (sa->sa_family != AF_INET) 1269 continue; 1270 sin = (struct sockaddr_in *)sa; 1271 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1272 ia = (struct in_ifaddr *)ifa; 1273 break; 1274 } 1275 } 1276 if (ia != NULL) { 1277 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1278 goto done; 1279 } 1280 1281 /* 3. As a last resort return the 'default' jail address. */ 1282 error = prison_get_ip4(cred, laddr); 1283 goto done; 1284 } 1285 1286 /* 1287 * The outgoing interface is marked with 'loopback net', so a route 1288 * to ourselves is here. 1289 * Try to find the interface of the destination address and then 1290 * take the address from there. That interface is not necessarily 1291 * a loopback interface. 1292 * In case of jails, check that it is an address of the jail 1293 * and if we cannot find, fall back to the 'default' jail address. 1294 */ 1295 if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { 1296 struct in_ifaddr *ia; 1297 1298 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), 1299 inp->inp_socket->so_fibnum)); 1300 if (ia == NULL) 1301 ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, 1302 inp->inp_socket->so_fibnum)); 1303 if (ia == NULL) 1304 ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 1305 1306 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1307 if (ia == NULL) { 1308 error = ENETUNREACH; 1309 goto done; 1310 } 1311 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1312 goto done; 1313 } 1314 1315 /* Jailed. */ 1316 if (ia != NULL) { 1317 struct ifnet *ifp; 1318 1319 ifp = ia->ia_ifp; 1320 ia = NULL; 1321 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1322 sa = ifa->ifa_addr; 1323 if (sa->sa_family != AF_INET) 1324 continue; 1325 sin = (struct sockaddr_in *)sa; 1326 if (prison_check_ip4(cred, 1327 &sin->sin_addr) == 0) { 1328 ia = (struct in_ifaddr *)ifa; 1329 break; 1330 } 1331 } 1332 if (ia != NULL) { 1333 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1334 goto done; 1335 } 1336 } 1337 1338 /* 3. As a last resort return the 'default' jail address. */ 1339 error = prison_get_ip4(cred, laddr); 1340 goto done; 1341 } 1342 1343 done: 1344 return (error); 1345 } 1346 1347 /* 1348 * Set up for a connect from a socket to the specified address. 1349 * On entry, *laddrp and *lportp should contain the current local 1350 * address and port for the PCB; these are updated to the values 1351 * that should be placed in inp_laddr and inp_lport to complete 1352 * the connect. 1353 * 1354 * On success, *faddrp and *fportp will be set to the remote address 1355 * and port. These are not updated in the error case. 1356 * 1357 * If the operation fails because the connection already exists, 1358 * *oinpp will be set to the PCB of that connection so that the 1359 * caller can decide to override it. In all other cases, *oinpp 1360 * is set to NULL. 1361 */ 1362 int 1363 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1364 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1365 struct inpcb **oinpp, struct ucred *cred) 1366 { 1367 struct rm_priotracker in_ifa_tracker; 1368 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1369 struct in_ifaddr *ia; 1370 struct inpcb *oinp; 1371 struct in_addr laddr, faddr; 1372 u_short lport, fport; 1373 int error; 1374 1375 /* 1376 * Because a global state change doesn't actually occur here, a read 1377 * lock is sufficient. 1378 */ 1379 NET_EPOCH_ASSERT(); 1380 INP_LOCK_ASSERT(inp); 1381 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1382 1383 if (oinpp != NULL) 1384 *oinpp = NULL; 1385 if (nam->sa_len != sizeof (*sin)) 1386 return (EINVAL); 1387 if (sin->sin_family != AF_INET) 1388 return (EAFNOSUPPORT); 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_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp, 2522 numa_domain); 2523 if (inp != NULL) { 2524 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2525 INP_WLOCK(inp); 2526 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2527 INP_WUNLOCK(inp); 2528 inp = NULL; 2529 } 2530 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2531 INP_RLOCK(inp); 2532 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2533 INP_RUNLOCK(inp); 2534 inp = NULL; 2535 } 2536 } else 2537 panic("%s: locking bug", __func__); 2538 #ifdef INVARIANTS 2539 if (inp != NULL) { 2540 if (lookupflags & INPLOOKUP_WLOCKPCB) 2541 INP_WLOCK_ASSERT(inp); 2542 else 2543 INP_RLOCK_ASSERT(inp); 2544 } 2545 #endif 2546 } 2547 2548 return (inp); 2549 } 2550 2551 /* 2552 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2553 * from which a pre-calculated hash value may be extracted. 2554 * 2555 * Possibly more of this logic should be in in_pcbgroup.c. 2556 */ 2557 struct inpcb * 2558 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2559 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2560 { 2561 #if defined(PCBGROUP) && !defined(RSS) 2562 struct inpcbgroup *pcbgroup; 2563 #endif 2564 2565 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2566 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2567 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2568 ("%s: LOCKPCB not set", __func__)); 2569 2570 /* 2571 * When not using RSS, use connection groups in preference to the 2572 * reservation table when looking up 4-tuples. When using RSS, just 2573 * use the reservation table, due to the cost of the Toeplitz hash 2574 * in software. 2575 * 2576 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2577 * we could be doing RSS with a non-Toeplitz hash that is affordable 2578 * in software. 2579 */ 2580 #if defined(PCBGROUP) && !defined(RSS) 2581 if (in_pcbgroup_enabled(pcbinfo)) { 2582 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2583 fport); 2584 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2585 laddr, lport, lookupflags, ifp)); 2586 } 2587 #endif 2588 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2589 lookupflags, ifp, M_NODOM)); 2590 } 2591 2592 struct inpcb * 2593 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2594 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2595 struct ifnet *ifp, struct mbuf *m) 2596 { 2597 #ifdef PCBGROUP 2598 struct inpcbgroup *pcbgroup; 2599 #endif 2600 2601 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2602 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2603 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2604 ("%s: LOCKPCB not set", __func__)); 2605 2606 #ifdef PCBGROUP 2607 /* 2608 * If we can use a hardware-generated hash to look up the connection 2609 * group, use that connection group to find the inpcb. Otherwise 2610 * fall back on a software hash -- or the reservation table if we're 2611 * using RSS. 2612 * 2613 * XXXRW: As above, that policy belongs in the pcbgroup code. 2614 */ 2615 if (in_pcbgroup_enabled(pcbinfo) && 2616 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2617 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2618 m->m_pkthdr.flowid); 2619 if (pcbgroup != NULL) 2620 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2621 fport, laddr, lport, lookupflags, ifp)); 2622 #ifndef RSS 2623 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2624 fport); 2625 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2626 laddr, lport, lookupflags, ifp)); 2627 #endif 2628 } 2629 #endif 2630 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2631 lookupflags, ifp, m->m_pkthdr.numa_domain)); 2632 } 2633 #endif /* INET */ 2634 2635 /* 2636 * Insert PCB onto various hash lists. 2637 */ 2638 static int 2639 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m) 2640 { 2641 struct inpcbhead *pcbhash; 2642 struct inpcbporthead *pcbporthash; 2643 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2644 struct inpcbport *phd; 2645 u_int32_t hashkey_faddr; 2646 int so_options; 2647 2648 INP_WLOCK_ASSERT(inp); 2649 INP_HASH_WLOCK_ASSERT(pcbinfo); 2650 2651 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2652 ("in_pcbinshash: INP_INHASHLIST")); 2653 2654 #ifdef INET6 2655 if (inp->inp_vflag & INP_IPV6) 2656 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2657 else 2658 #endif 2659 hashkey_faddr = inp->inp_faddr.s_addr; 2660 2661 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2662 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2663 2664 pcbporthash = &pcbinfo->ipi_porthashbase[ 2665 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2666 2667 /* 2668 * Add entry to load balance group. 2669 * Only do this if SO_REUSEPORT_LB is set. 2670 */ 2671 so_options = inp_so_options(inp); 2672 if (so_options & SO_REUSEPORT_LB) { 2673 int ret = in_pcbinslbgrouphash(inp, M_NODOM); 2674 if (ret) { 2675 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2676 return (ret); 2677 } 2678 } 2679 2680 /* 2681 * Go through port list and look for a head for this lport. 2682 */ 2683 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2684 if (phd->phd_port == inp->inp_lport) 2685 break; 2686 } 2687 /* 2688 * If none exists, malloc one and tack it on. 2689 */ 2690 if (phd == NULL) { 2691 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2692 if (phd == NULL) { 2693 return (ENOBUFS); /* XXX */ 2694 } 2695 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2696 phd->phd_port = inp->inp_lport; 2697 CK_LIST_INIT(&phd->phd_pcblist); 2698 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2699 } 2700 inp->inp_phd = phd; 2701 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2702 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2703 inp->inp_flags |= INP_INHASHLIST; 2704 #ifdef PCBGROUP 2705 if (m != NULL) { 2706 in_pcbgroup_update_mbuf(inp, m); 2707 } else { 2708 in_pcbgroup_update(inp); 2709 } 2710 #endif 2711 return (0); 2712 } 2713 2714 int 2715 in_pcbinshash(struct inpcb *inp) 2716 { 2717 2718 return (in_pcbinshash_internal(inp, NULL)); 2719 } 2720 2721 int 2722 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m) 2723 { 2724 2725 return (in_pcbinshash_internal(inp, m)); 2726 } 2727 2728 /* 2729 * Move PCB to the proper hash bucket when { faddr, fport } have been 2730 * changed. NOTE: This does not handle the case of the lport changing (the 2731 * hashed port list would have to be updated as well), so the lport must 2732 * not change after in_pcbinshash() has been called. 2733 */ 2734 void 2735 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2736 { 2737 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2738 struct inpcbhead *head; 2739 u_int32_t hashkey_faddr; 2740 2741 INP_WLOCK_ASSERT(inp); 2742 INP_HASH_WLOCK_ASSERT(pcbinfo); 2743 2744 KASSERT(inp->inp_flags & INP_INHASHLIST, 2745 ("in_pcbrehash: !INP_INHASHLIST")); 2746 2747 #ifdef INET6 2748 if (inp->inp_vflag & INP_IPV6) 2749 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2750 else 2751 #endif 2752 hashkey_faddr = inp->inp_faddr.s_addr; 2753 2754 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2755 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2756 2757 CK_LIST_REMOVE(inp, inp_hash); 2758 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2759 2760 #ifdef PCBGROUP 2761 if (m != NULL) 2762 in_pcbgroup_update_mbuf(inp, m); 2763 else 2764 in_pcbgroup_update(inp); 2765 #endif 2766 } 2767 2768 void 2769 in_pcbrehash(struct inpcb *inp) 2770 { 2771 2772 in_pcbrehash_mbuf(inp, NULL); 2773 } 2774 2775 /* 2776 * Remove PCB from various lists. 2777 */ 2778 static void 2779 in_pcbremlists(struct inpcb *inp) 2780 { 2781 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2782 2783 INP_WLOCK_ASSERT(inp); 2784 INP_LIST_WLOCK_ASSERT(pcbinfo); 2785 2786 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2787 if (inp->inp_flags & INP_INHASHLIST) { 2788 struct inpcbport *phd = inp->inp_phd; 2789 2790 INP_HASH_WLOCK(pcbinfo); 2791 2792 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2793 in_pcbremlbgrouphash(inp); 2794 2795 CK_LIST_REMOVE(inp, inp_hash); 2796 CK_LIST_REMOVE(inp, inp_portlist); 2797 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2798 CK_LIST_REMOVE(phd, phd_hash); 2799 NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); 2800 } 2801 INP_HASH_WUNLOCK(pcbinfo); 2802 inp->inp_flags &= ~INP_INHASHLIST; 2803 } 2804 CK_LIST_REMOVE(inp, inp_list); 2805 pcbinfo->ipi_count--; 2806 #ifdef PCBGROUP 2807 in_pcbgroup_remove(inp); 2808 #endif 2809 } 2810 2811 /* 2812 * Check for alternatives when higher level complains 2813 * about service problems. For now, invalidate cached 2814 * routing information. If the route was created dynamically 2815 * (by a redirect), time to try a default gateway again. 2816 */ 2817 void 2818 in_losing(struct inpcb *inp) 2819 { 2820 2821 RO_INVALIDATE_CACHE(&inp->inp_route); 2822 return; 2823 } 2824 2825 /* 2826 * A set label operation has occurred at the socket layer, propagate the 2827 * label change into the in_pcb for the socket. 2828 */ 2829 void 2830 in_pcbsosetlabel(struct socket *so) 2831 { 2832 #ifdef MAC 2833 struct inpcb *inp; 2834 2835 inp = sotoinpcb(so); 2836 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2837 2838 INP_WLOCK(inp); 2839 SOCK_LOCK(so); 2840 mac_inpcb_sosetlabel(so, inp); 2841 SOCK_UNLOCK(so); 2842 INP_WUNLOCK(inp); 2843 #endif 2844 } 2845 2846 /* 2847 * ipport_tick runs once per second, determining if random port allocation 2848 * should be continued. If more than ipport_randomcps ports have been 2849 * allocated in the last second, then we return to sequential port 2850 * allocation. We return to random allocation only once we drop below 2851 * ipport_randomcps for at least ipport_randomtime seconds. 2852 */ 2853 static void 2854 ipport_tick(void *xtp) 2855 { 2856 VNET_ITERATOR_DECL(vnet_iter); 2857 2858 VNET_LIST_RLOCK_NOSLEEP(); 2859 VNET_FOREACH(vnet_iter) { 2860 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2861 if (V_ipport_tcpallocs <= 2862 V_ipport_tcplastcount + V_ipport_randomcps) { 2863 if (V_ipport_stoprandom > 0) 2864 V_ipport_stoprandom--; 2865 } else 2866 V_ipport_stoprandom = V_ipport_randomtime; 2867 V_ipport_tcplastcount = V_ipport_tcpallocs; 2868 CURVNET_RESTORE(); 2869 } 2870 VNET_LIST_RUNLOCK_NOSLEEP(); 2871 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2872 } 2873 2874 static void 2875 ip_fini(void *xtp) 2876 { 2877 2878 callout_stop(&ipport_tick_callout); 2879 } 2880 2881 /* 2882 * The ipport_callout should start running at about the time we attach the 2883 * inet or inet6 domains. 2884 */ 2885 static void 2886 ipport_tick_init(const void *unused __unused) 2887 { 2888 2889 /* Start ipport_tick. */ 2890 callout_init(&ipport_tick_callout, 1); 2891 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2892 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2893 SHUTDOWN_PRI_DEFAULT); 2894 } 2895 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2896 ipport_tick_init, NULL); 2897 2898 void 2899 inp_wlock(struct inpcb *inp) 2900 { 2901 2902 INP_WLOCK(inp); 2903 } 2904 2905 void 2906 inp_wunlock(struct inpcb *inp) 2907 { 2908 2909 INP_WUNLOCK(inp); 2910 } 2911 2912 void 2913 inp_rlock(struct inpcb *inp) 2914 { 2915 2916 INP_RLOCK(inp); 2917 } 2918 2919 void 2920 inp_runlock(struct inpcb *inp) 2921 { 2922 2923 INP_RUNLOCK(inp); 2924 } 2925 2926 #ifdef INVARIANT_SUPPORT 2927 void 2928 inp_lock_assert(struct inpcb *inp) 2929 { 2930 2931 INP_WLOCK_ASSERT(inp); 2932 } 2933 2934 void 2935 inp_unlock_assert(struct inpcb *inp) 2936 { 2937 2938 INP_UNLOCK_ASSERT(inp); 2939 } 2940 #endif 2941 2942 void 2943 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2944 { 2945 struct inpcb *inp; 2946 2947 INP_INFO_WLOCK(&V_tcbinfo); 2948 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2949 INP_WLOCK(inp); 2950 func(inp, arg); 2951 INP_WUNLOCK(inp); 2952 } 2953 INP_INFO_WUNLOCK(&V_tcbinfo); 2954 } 2955 2956 struct socket * 2957 inp_inpcbtosocket(struct inpcb *inp) 2958 { 2959 2960 INP_WLOCK_ASSERT(inp); 2961 return (inp->inp_socket); 2962 } 2963 2964 struct tcpcb * 2965 inp_inpcbtotcpcb(struct inpcb *inp) 2966 { 2967 2968 INP_WLOCK_ASSERT(inp); 2969 return ((struct tcpcb *)inp->inp_ppcb); 2970 } 2971 2972 int 2973 inp_ip_tos_get(const struct inpcb *inp) 2974 { 2975 2976 return (inp->inp_ip_tos); 2977 } 2978 2979 void 2980 inp_ip_tos_set(struct inpcb *inp, int val) 2981 { 2982 2983 inp->inp_ip_tos = val; 2984 } 2985 2986 void 2987 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2988 uint32_t *faddr, uint16_t *fp) 2989 { 2990 2991 INP_LOCK_ASSERT(inp); 2992 *laddr = inp->inp_laddr.s_addr; 2993 *faddr = inp->inp_faddr.s_addr; 2994 *lp = inp->inp_lport; 2995 *fp = inp->inp_fport; 2996 } 2997 2998 struct inpcb * 2999 so_sotoinpcb(struct socket *so) 3000 { 3001 3002 return (sotoinpcb(so)); 3003 } 3004 3005 struct tcpcb * 3006 so_sototcpcb(struct socket *so) 3007 { 3008 3009 return (sototcpcb(so)); 3010 } 3011 3012 /* 3013 * Create an external-format (``xinpcb'') structure using the information in 3014 * the kernel-format in_pcb structure pointed to by inp. This is done to 3015 * reduce the spew of irrelevant information over this interface, to isolate 3016 * user code from changes in the kernel structure, and potentially to provide 3017 * information-hiding if we decide that some of this information should be 3018 * hidden from users. 3019 */ 3020 void 3021 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 3022 { 3023 3024 bzero(xi, sizeof(*xi)); 3025 xi->xi_len = sizeof(struct xinpcb); 3026 if (inp->inp_socket) 3027 sotoxsocket(inp->inp_socket, &xi->xi_socket); 3028 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 3029 xi->inp_gencnt = inp->inp_gencnt; 3030 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 3031 xi->inp_flow = inp->inp_flow; 3032 xi->inp_flowid = inp->inp_flowid; 3033 xi->inp_flowtype = inp->inp_flowtype; 3034 xi->inp_flags = inp->inp_flags; 3035 xi->inp_flags2 = inp->inp_flags2; 3036 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 3037 xi->in6p_cksum = inp->in6p_cksum; 3038 xi->in6p_hops = inp->in6p_hops; 3039 xi->inp_ip_tos = inp->inp_ip_tos; 3040 xi->inp_vflag = inp->inp_vflag; 3041 xi->inp_ip_ttl = inp->inp_ip_ttl; 3042 xi->inp_ip_p = inp->inp_ip_p; 3043 xi->inp_ip_minttl = inp->inp_ip_minttl; 3044 } 3045 3046 #ifdef DDB 3047 static void 3048 db_print_indent(int indent) 3049 { 3050 int i; 3051 3052 for (i = 0; i < indent; i++) 3053 db_printf(" "); 3054 } 3055 3056 static void 3057 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 3058 { 3059 char faddr_str[48], laddr_str[48]; 3060 3061 db_print_indent(indent); 3062 db_printf("%s at %p\n", name, inc); 3063 3064 indent += 2; 3065 3066 #ifdef INET6 3067 if (inc->inc_flags & INC_ISIPV6) { 3068 /* IPv6. */ 3069 ip6_sprintf(laddr_str, &inc->inc6_laddr); 3070 ip6_sprintf(faddr_str, &inc->inc6_faddr); 3071 } else 3072 #endif 3073 { 3074 /* IPv4. */ 3075 inet_ntoa_r(inc->inc_laddr, laddr_str); 3076 inet_ntoa_r(inc->inc_faddr, faddr_str); 3077 } 3078 db_print_indent(indent); 3079 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 3080 ntohs(inc->inc_lport)); 3081 db_print_indent(indent); 3082 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 3083 ntohs(inc->inc_fport)); 3084 } 3085 3086 static void 3087 db_print_inpflags(int inp_flags) 3088 { 3089 int comma; 3090 3091 comma = 0; 3092 if (inp_flags & INP_RECVOPTS) { 3093 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 3094 comma = 1; 3095 } 3096 if (inp_flags & INP_RECVRETOPTS) { 3097 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 3098 comma = 1; 3099 } 3100 if (inp_flags & INP_RECVDSTADDR) { 3101 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 3102 comma = 1; 3103 } 3104 if (inp_flags & INP_ORIGDSTADDR) { 3105 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 3106 comma = 1; 3107 } 3108 if (inp_flags & INP_HDRINCL) { 3109 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 3110 comma = 1; 3111 } 3112 if (inp_flags & INP_HIGHPORT) { 3113 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 3114 comma = 1; 3115 } 3116 if (inp_flags & INP_LOWPORT) { 3117 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3118 comma = 1; 3119 } 3120 if (inp_flags & INP_ANONPORT) { 3121 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3122 comma = 1; 3123 } 3124 if (inp_flags & INP_RECVIF) { 3125 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3126 comma = 1; 3127 } 3128 if (inp_flags & INP_MTUDISC) { 3129 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3130 comma = 1; 3131 } 3132 if (inp_flags & INP_RECVTTL) { 3133 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3134 comma = 1; 3135 } 3136 if (inp_flags & INP_DONTFRAG) { 3137 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3138 comma = 1; 3139 } 3140 if (inp_flags & INP_RECVTOS) { 3141 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3142 comma = 1; 3143 } 3144 if (inp_flags & IN6P_IPV6_V6ONLY) { 3145 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3146 comma = 1; 3147 } 3148 if (inp_flags & IN6P_PKTINFO) { 3149 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3150 comma = 1; 3151 } 3152 if (inp_flags & IN6P_HOPLIMIT) { 3153 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3154 comma = 1; 3155 } 3156 if (inp_flags & IN6P_HOPOPTS) { 3157 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3158 comma = 1; 3159 } 3160 if (inp_flags & IN6P_DSTOPTS) { 3161 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3162 comma = 1; 3163 } 3164 if (inp_flags & IN6P_RTHDR) { 3165 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3166 comma = 1; 3167 } 3168 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3169 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3170 comma = 1; 3171 } 3172 if (inp_flags & IN6P_TCLASS) { 3173 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3174 comma = 1; 3175 } 3176 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3177 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3178 comma = 1; 3179 } 3180 if (inp_flags & INP_TIMEWAIT) { 3181 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3182 comma = 1; 3183 } 3184 if (inp_flags & INP_ONESBCAST) { 3185 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3186 comma = 1; 3187 } 3188 if (inp_flags & INP_DROPPED) { 3189 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3190 comma = 1; 3191 } 3192 if (inp_flags & INP_SOCKREF) { 3193 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3194 comma = 1; 3195 } 3196 if (inp_flags & IN6P_RFC2292) { 3197 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3198 comma = 1; 3199 } 3200 if (inp_flags & IN6P_MTU) { 3201 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3202 comma = 1; 3203 } 3204 } 3205 3206 static void 3207 db_print_inpvflag(u_char inp_vflag) 3208 { 3209 int comma; 3210 3211 comma = 0; 3212 if (inp_vflag & INP_IPV4) { 3213 db_printf("%sINP_IPV4", comma ? ", " : ""); 3214 comma = 1; 3215 } 3216 if (inp_vflag & INP_IPV6) { 3217 db_printf("%sINP_IPV6", comma ? ", " : ""); 3218 comma = 1; 3219 } 3220 if (inp_vflag & INP_IPV6PROTO) { 3221 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3222 comma = 1; 3223 } 3224 } 3225 3226 static void 3227 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3228 { 3229 3230 db_print_indent(indent); 3231 db_printf("%s at %p\n", name, inp); 3232 3233 indent += 2; 3234 3235 db_print_indent(indent); 3236 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3237 3238 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3239 3240 db_print_indent(indent); 3241 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3242 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3243 3244 db_print_indent(indent); 3245 db_printf("inp_label: %p inp_flags: 0x%x (", 3246 inp->inp_label, inp->inp_flags); 3247 db_print_inpflags(inp->inp_flags); 3248 db_printf(")\n"); 3249 3250 db_print_indent(indent); 3251 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3252 inp->inp_vflag); 3253 db_print_inpvflag(inp->inp_vflag); 3254 db_printf(")\n"); 3255 3256 db_print_indent(indent); 3257 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3258 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3259 3260 db_print_indent(indent); 3261 #ifdef INET6 3262 if (inp->inp_vflag & INP_IPV6) { 3263 db_printf("in6p_options: %p in6p_outputopts: %p " 3264 "in6p_moptions: %p\n", inp->in6p_options, 3265 inp->in6p_outputopts, inp->in6p_moptions); 3266 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3267 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3268 inp->in6p_hops); 3269 } else 3270 #endif 3271 { 3272 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3273 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3274 inp->inp_options, inp->inp_moptions); 3275 } 3276 3277 db_print_indent(indent); 3278 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3279 (uintmax_t)inp->inp_gencnt); 3280 } 3281 3282 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3283 { 3284 struct inpcb *inp; 3285 3286 if (!have_addr) { 3287 db_printf("usage: show inpcb <addr>\n"); 3288 return; 3289 } 3290 inp = (struct inpcb *)addr; 3291 3292 db_print_inpcb(inp, "inpcb", 0); 3293 } 3294 #endif /* DDB */ 3295 3296 #ifdef RATELIMIT 3297 /* 3298 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3299 * if any. 3300 */ 3301 int 3302 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3303 { 3304 union if_snd_tag_modify_params params = { 3305 .rate_limit.max_rate = max_pacing_rate, 3306 .rate_limit.flags = M_NOWAIT, 3307 }; 3308 struct m_snd_tag *mst; 3309 struct ifnet *ifp; 3310 int error; 3311 3312 mst = inp->inp_snd_tag; 3313 if (mst == NULL) 3314 return (EINVAL); 3315 3316 ifp = mst->ifp; 3317 if (ifp == NULL) 3318 return (EINVAL); 3319 3320 if (ifp->if_snd_tag_modify == NULL) { 3321 error = EOPNOTSUPP; 3322 } else { 3323 error = ifp->if_snd_tag_modify(mst, ¶ms); 3324 } 3325 return (error); 3326 } 3327 3328 /* 3329 * Query existing TX rate limit based on the existing 3330 * "inp->inp_snd_tag", if any. 3331 */ 3332 int 3333 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3334 { 3335 union if_snd_tag_query_params params = { }; 3336 struct m_snd_tag *mst; 3337 struct ifnet *ifp; 3338 int error; 3339 3340 mst = inp->inp_snd_tag; 3341 if (mst == NULL) 3342 return (EINVAL); 3343 3344 ifp = mst->ifp; 3345 if (ifp == NULL) 3346 return (EINVAL); 3347 3348 if (ifp->if_snd_tag_query == NULL) { 3349 error = EOPNOTSUPP; 3350 } else { 3351 error = ifp->if_snd_tag_query(mst, ¶ms); 3352 if (error == 0 && p_max_pacing_rate != NULL) 3353 *p_max_pacing_rate = params.rate_limit.max_rate; 3354 } 3355 return (error); 3356 } 3357 3358 /* 3359 * Query existing TX queue level based on the existing 3360 * "inp->inp_snd_tag", if any. 3361 */ 3362 int 3363 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3364 { 3365 union if_snd_tag_query_params params = { }; 3366 struct m_snd_tag *mst; 3367 struct ifnet *ifp; 3368 int error; 3369 3370 mst = inp->inp_snd_tag; 3371 if (mst == NULL) 3372 return (EINVAL); 3373 3374 ifp = mst->ifp; 3375 if (ifp == NULL) 3376 return (EINVAL); 3377 3378 if (ifp->if_snd_tag_query == NULL) 3379 return (EOPNOTSUPP); 3380 3381 error = ifp->if_snd_tag_query(mst, ¶ms); 3382 if (error == 0 && p_txqueue_level != NULL) 3383 *p_txqueue_level = params.rate_limit.queue_level; 3384 return (error); 3385 } 3386 3387 /* 3388 * Allocate a new TX rate limit send tag from the network interface 3389 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3390 */ 3391 int 3392 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3393 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3394 3395 { 3396 union if_snd_tag_alloc_params params = { 3397 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3398 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3399 .rate_limit.hdr.flowid = flowid, 3400 .rate_limit.hdr.flowtype = flowtype, 3401 .rate_limit.hdr.numa_domain = inp->inp_numa_domain, 3402 .rate_limit.max_rate = max_pacing_rate, 3403 .rate_limit.flags = M_NOWAIT, 3404 }; 3405 int error; 3406 3407 INP_WLOCK_ASSERT(inp); 3408 3409 /* 3410 * If there is already a send tag, or the INP is being torn 3411 * down, allocating a new send tag is not allowed. Else send 3412 * tags may leak. 3413 */ 3414 if (*st != NULL || (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0) 3415 return (EINVAL); 3416 3417 error = m_snd_tag_alloc(ifp, ¶ms, st); 3418 #ifdef INET 3419 if (error == 0) { 3420 counter_u64_add(rate_limit_set_ok, 1); 3421 counter_u64_add(rate_limit_active, 1); 3422 } else if (error != EOPNOTSUPP) 3423 counter_u64_add(rate_limit_alloc_fail, 1); 3424 #endif 3425 return (error); 3426 } 3427 3428 void 3429 in_pcbdetach_tag(struct m_snd_tag *mst) 3430 { 3431 3432 m_snd_tag_rele(mst); 3433 #ifdef INET 3434 counter_u64_add(rate_limit_active, -1); 3435 #endif 3436 } 3437 3438 /* 3439 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3440 * if any: 3441 */ 3442 void 3443 in_pcbdetach_txrtlmt(struct inpcb *inp) 3444 { 3445 struct m_snd_tag *mst; 3446 3447 INP_WLOCK_ASSERT(inp); 3448 3449 mst = inp->inp_snd_tag; 3450 inp->inp_snd_tag = NULL; 3451 3452 if (mst == NULL) 3453 return; 3454 3455 m_snd_tag_rele(mst); 3456 #ifdef INET 3457 counter_u64_add(rate_limit_active, -1); 3458 #endif 3459 } 3460 3461 int 3462 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3463 { 3464 int error; 3465 3466 /* 3467 * If the existing send tag is for the wrong interface due to 3468 * a route change, first drop the existing tag. Set the 3469 * CHANGED flag so that we will keep trying to allocate a new 3470 * tag if we fail to allocate one this time. 3471 */ 3472 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3473 in_pcbdetach_txrtlmt(inp); 3474 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3475 } 3476 3477 /* 3478 * NOTE: When attaching to a network interface a reference is 3479 * made to ensure the network interface doesn't go away until 3480 * all ratelimit connections are gone. The network interface 3481 * pointers compared below represent valid network interfaces, 3482 * except when comparing towards NULL. 3483 */ 3484 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3485 error = 0; 3486 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3487 if (inp->inp_snd_tag != NULL) 3488 in_pcbdetach_txrtlmt(inp); 3489 error = 0; 3490 } else if (inp->inp_snd_tag == NULL) { 3491 /* 3492 * In order to utilize packet pacing with RSS, we need 3493 * to wait until there is a valid RSS hash before we 3494 * can proceed: 3495 */ 3496 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3497 error = EAGAIN; 3498 } else { 3499 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3500 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3501 } 3502 } else { 3503 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3504 } 3505 if (error == 0 || error == EOPNOTSUPP) 3506 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3507 3508 return (error); 3509 } 3510 3511 /* 3512 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3513 * is set in the fast path and will attach/detach/modify the TX rate 3514 * limit send tag based on the socket's so_max_pacing_rate value. 3515 */ 3516 void 3517 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3518 { 3519 struct socket *socket; 3520 uint32_t max_pacing_rate; 3521 bool did_upgrade; 3522 int error; 3523 3524 if (inp == NULL) 3525 return; 3526 3527 socket = inp->inp_socket; 3528 if (socket == NULL) 3529 return; 3530 3531 if (!INP_WLOCKED(inp)) { 3532 /* 3533 * NOTE: If the write locking fails, we need to bail 3534 * out and use the non-ratelimited ring for the 3535 * transmit until there is a new chance to get the 3536 * write lock. 3537 */ 3538 if (!INP_TRY_UPGRADE(inp)) 3539 return; 3540 did_upgrade = 1; 3541 } else { 3542 did_upgrade = 0; 3543 } 3544 3545 /* 3546 * NOTE: The so_max_pacing_rate value is read unlocked, 3547 * because atomic updates are not required since the variable 3548 * is checked at every mbuf we send. It is assumed that the 3549 * variable read itself will be atomic. 3550 */ 3551 max_pacing_rate = socket->so_max_pacing_rate; 3552 3553 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3554 3555 if (did_upgrade) 3556 INP_DOWNGRADE(inp); 3557 } 3558 3559 /* 3560 * Track route changes for TX rate limiting. 3561 */ 3562 void 3563 in_pcboutput_eagain(struct inpcb *inp) 3564 { 3565 bool did_upgrade; 3566 3567 if (inp == NULL) 3568 return; 3569 3570 if (inp->inp_snd_tag == NULL) 3571 return; 3572 3573 if (!INP_WLOCKED(inp)) { 3574 /* 3575 * NOTE: If the write locking fails, we need to bail 3576 * out and use the non-ratelimited ring for the 3577 * transmit until there is a new chance to get the 3578 * write lock. 3579 */ 3580 if (!INP_TRY_UPGRADE(inp)) 3581 return; 3582 did_upgrade = 1; 3583 } else { 3584 did_upgrade = 0; 3585 } 3586 3587 /* detach rate limiting */ 3588 in_pcbdetach_txrtlmt(inp); 3589 3590 /* make sure new mbuf send tag allocation is made */ 3591 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3592 3593 if (did_upgrade) 3594 INP_DOWNGRADE(inp); 3595 } 3596 3597 #ifdef INET 3598 static void 3599 rl_init(void *st) 3600 { 3601 rate_limit_new = counter_u64_alloc(M_WAITOK); 3602 rate_limit_chg = counter_u64_alloc(M_WAITOK); 3603 rate_limit_active = counter_u64_alloc(M_WAITOK); 3604 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3605 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3606 } 3607 3608 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3609 #endif 3610 #endif /* RATELIMIT */ 3611