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_rss.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/callout.h> 57 #include <sys/eventhandler.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/rmlock.h> 61 #include <sys/smp.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <sys/sockio.h> 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/refcount.h> 68 #include <sys/jail.h> 69 #include <sys/kernel.h> 70 #include <sys/sysctl.h> 71 72 #ifdef DDB 73 #include <ddb/ddb.h> 74 #endif 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/if_types.h> 81 #include <net/if_llatbl.h> 82 #include <net/route.h> 83 #include <net/rss_config.h> 84 #include <net/vnet.h> 85 86 #if defined(INET) || defined(INET6) 87 #include <netinet/in.h> 88 #include <netinet/in_pcb.h> 89 #ifdef INET 90 #include <netinet/in_var.h> 91 #endif 92 #include <netinet/ip_var.h> 93 #include <netinet/tcp_var.h> 94 #ifdef TCPHPTS 95 #include <netinet/tcp_hpts.h> 96 #endif 97 #include <netinet/udp.h> 98 #include <netinet/udp_var.h> 99 #ifdef INET6 100 #include <netinet/ip6.h> 101 #include <netinet6/in6_pcb.h> 102 #include <netinet6/in6_var.h> 103 #include <netinet6/ip6_var.h> 104 #endif /* INET6 */ 105 #endif 106 107 #include <netipsec/ipsec_support.h> 108 109 #include <security/mac/mac_framework.h> 110 111 #define INPCBLBGROUP_SIZMIN 8 112 #define INPCBLBGROUP_SIZMAX 256 113 114 static struct callout ipport_tick_callout; 115 116 /* 117 * These configure the range of local port addresses assigned to 118 * "unspecified" outgoing connections/packets/whatever. 119 */ 120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 126 127 /* 128 * Reserved ports accessible only to root. There are significant 129 * security considerations that must be accounted for when changing these, 130 * but the security benefits can be great. Please be careful. 131 */ 132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 133 VNET_DEFINE(int, ipport_reservedlow); 134 135 /* Variables dealing with random ephemeral port allocation. */ 136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ 137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ 138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ 139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ 140 VNET_DEFINE(int, ipport_tcpallocs); 141 VNET_DEFINE_STATIC(int, ipport_tcplastcount); 142 143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount) 144 145 static void in_pcbremlists(struct inpcb *inp); 146 #ifdef INET 147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 148 struct in_addr faddr, u_int fport_arg, 149 struct in_addr laddr, u_int lport_arg, 150 int lookupflags, struct ifnet *ifp); 151 152 #define RANGECHK(var, min, max) \ 153 if ((var) < (min)) { (var) = (min); } \ 154 else if ((var) > (max)) { (var) = (max); } 155 156 static int 157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 158 { 159 int error; 160 161 error = sysctl_handle_int(oidp, arg1, arg2, req); 162 if (error == 0) { 163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 169 } 170 return (error); 171 } 172 173 #undef RANGECHK 174 175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, 176 "IP Ports"); 177 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", ""); 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", ""); 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", ""); 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", ""); 190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", ""); 193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", ""); 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 198 &VNET_NAME(ipport_reservedhigh), 0, ""); 199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 202 CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, 205 CTLFLAG_VNET | CTLFLAG_RW, 206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " 207 "allocations before switching to a sequental one"); 208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, 209 CTLFLAG_VNET | CTLFLAG_RW, 210 &VNET_NAME(ipport_randomtime), 0, 211 "Minimum time to keep sequental port " 212 "allocation before switching to a random one"); 213 214 #ifdef RATELIMIT 215 counter_u64_t rate_limit_active; 216 counter_u64_t rate_limit_alloc_fail; 217 counter_u64_t rate_limit_set_ok; 218 219 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD, 0, 220 "IP Rate Limiting"); 221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, 222 &rate_limit_active, "Active rate limited connections"); 223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, 224 &rate_limit_alloc_fail, "Rate limited connection failures"); 225 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, 226 &rate_limit_set_ok, "Rate limited setting succeeded"); 227 #endif /* RATELIMIT */ 228 229 #endif /* INET */ 230 231 /* 232 * in_pcb.c: manage the Protocol Control Blocks. 233 * 234 * NOTE: It is assumed that most of these functions will be called with 235 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 236 * functions often modify hash chains or addresses in pcbs. 237 */ 238 239 static struct inpcblbgroup * 240 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag, 241 uint16_t port, const union in_dependaddr *addr, int size) 242 { 243 struct inpcblbgroup *grp; 244 size_t bytes; 245 246 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 247 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 248 if (!grp) 249 return (NULL); 250 grp->il_vflag = vflag; 251 grp->il_lport = port; 252 grp->il_dependladdr = *addr; 253 grp->il_inpsiz = size; 254 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 255 return (grp); 256 } 257 258 static void 259 in_pcblbgroup_free_deferred(epoch_context_t ctx) 260 { 261 struct inpcblbgroup *grp; 262 263 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 264 free(grp, M_PCB); 265 } 266 267 static void 268 in_pcblbgroup_free(struct inpcblbgroup *grp) 269 { 270 271 CK_LIST_REMOVE(grp, il_list); 272 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx, 273 in_pcblbgroup_free_deferred); 274 } 275 276 static struct inpcblbgroup * 277 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 278 struct inpcblbgroup *old_grp, int size) 279 { 280 struct inpcblbgroup *grp; 281 int i; 282 283 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, 284 old_grp->il_lport, &old_grp->il_dependladdr, size); 285 if (grp == NULL) 286 return (NULL); 287 288 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 289 ("invalid new local group size %d and old local group count %d", 290 grp->il_inpsiz, old_grp->il_inpcnt)); 291 292 for (i = 0; i < old_grp->il_inpcnt; ++i) 293 grp->il_inp[i] = old_grp->il_inp[i]; 294 grp->il_inpcnt = old_grp->il_inpcnt; 295 in_pcblbgroup_free(old_grp); 296 return (grp); 297 } 298 299 /* 300 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 301 * and shrink group if possible. 302 */ 303 static void 304 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 305 int i) 306 { 307 struct inpcblbgroup *grp, *new_grp; 308 309 grp = *grpp; 310 for (; i + 1 < grp->il_inpcnt; ++i) 311 grp->il_inp[i] = grp->il_inp[i + 1]; 312 grp->il_inpcnt--; 313 314 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 315 grp->il_inpcnt <= grp->il_inpsiz / 4) { 316 /* Shrink this group. */ 317 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 318 if (new_grp != NULL) 319 *grpp = new_grp; 320 } 321 } 322 323 /* 324 * Add PCB to load balance group for SO_REUSEPORT_LB option. 325 */ 326 static int 327 in_pcbinslbgrouphash(struct inpcb *inp) 328 { 329 const static struct timeval interval = { 60, 0 }; 330 static struct timeval lastprint; 331 struct inpcbinfo *pcbinfo; 332 struct inpcblbgrouphead *hdr; 333 struct inpcblbgroup *grp; 334 uint32_t idx; 335 336 pcbinfo = inp->inp_pcbinfo; 337 338 INP_WLOCK_ASSERT(inp); 339 INP_HASH_WLOCK_ASSERT(pcbinfo); 340 341 /* 342 * Don't allow jailed socket to join local group. 343 */ 344 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred)) 345 return (0); 346 347 #ifdef INET6 348 /* 349 * Don't allow IPv4 mapped INET6 wild socket. 350 */ 351 if ((inp->inp_vflag & INP_IPV4) && 352 inp->inp_laddr.s_addr == INADDR_ANY && 353 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 354 return (0); 355 } 356 #endif 357 358 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 359 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 360 CK_LIST_FOREACH(grp, hdr, il_list) { 361 if (grp->il_vflag == inp->inp_vflag && 362 grp->il_lport == inp->inp_lport && 363 memcmp(&grp->il_dependladdr, 364 &inp->inp_inc.inc_ie.ie_dependladdr, 365 sizeof(grp->il_dependladdr)) == 0) 366 break; 367 } 368 if (grp == NULL) { 369 /* Create new load balance group. */ 370 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, 371 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 372 INPCBLBGROUP_SIZMIN); 373 if (grp == NULL) 374 return (ENOBUFS); 375 } else if (grp->il_inpcnt == grp->il_inpsiz) { 376 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 377 if (ratecheck(&lastprint, &interval)) 378 printf("lb group port %d, limit reached\n", 379 ntohs(grp->il_lport)); 380 return (0); 381 } 382 383 /* Expand this local group. */ 384 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 385 if (grp == NULL) 386 return (ENOBUFS); 387 } 388 389 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 390 ("invalid local group size %d and count %d", grp->il_inpsiz, 391 grp->il_inpcnt)); 392 393 grp->il_inp[grp->il_inpcnt] = inp; 394 grp->il_inpcnt++; 395 return (0); 396 } 397 398 /* 399 * Remove PCB from load balance group. 400 */ 401 static void 402 in_pcbremlbgrouphash(struct inpcb *inp) 403 { 404 struct inpcbinfo *pcbinfo; 405 struct inpcblbgrouphead *hdr; 406 struct inpcblbgroup *grp; 407 int i; 408 409 pcbinfo = inp->inp_pcbinfo; 410 411 INP_WLOCK_ASSERT(inp); 412 INP_HASH_WLOCK_ASSERT(pcbinfo); 413 414 hdr = &pcbinfo->ipi_lbgrouphashbase[ 415 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 416 CK_LIST_FOREACH(grp, hdr, il_list) { 417 for (i = 0; i < grp->il_inpcnt; ++i) { 418 if (grp->il_inp[i] != inp) 419 continue; 420 421 if (grp->il_inpcnt == 1) { 422 /* We are the last, free this local group. */ 423 in_pcblbgroup_free(grp); 424 } else { 425 /* Pull up inpcbs, shrink group if possible. */ 426 in_pcblbgroup_reorder(hdr, &grp, i); 427 } 428 return; 429 } 430 } 431 } 432 433 /* 434 * Different protocols initialize their inpcbs differently - giving 435 * different name to the lock. But they all are disposed the same. 436 */ 437 static void 438 inpcb_fini(void *mem, int size) 439 { 440 struct inpcb *inp = mem; 441 442 INP_LOCK_DESTROY(inp); 443 } 444 445 /* 446 * Initialize an inpcbinfo -- we should be able to reduce the number of 447 * arguments in time. 448 */ 449 void 450 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 451 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 452 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) 453 { 454 455 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 456 457 INP_INFO_LOCK_INIT(pcbinfo, name); 458 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 459 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 460 #ifdef VIMAGE 461 pcbinfo->ipi_vnet = curvnet; 462 #endif 463 pcbinfo->ipi_listhead = listhead; 464 CK_LIST_INIT(pcbinfo->ipi_listhead); 465 pcbinfo->ipi_count = 0; 466 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 467 &pcbinfo->ipi_hashmask); 468 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 469 &pcbinfo->ipi_porthashmask); 470 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 471 &pcbinfo->ipi_lbgrouphashmask); 472 #ifdef PCBGROUP 473 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 474 #endif 475 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 476 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); 477 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 478 uma_zone_set_warning(pcbinfo->ipi_zone, 479 "kern.ipc.maxsockets limit reached"); 480 } 481 482 /* 483 * Destroy an inpcbinfo. 484 */ 485 void 486 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 487 { 488 489 KASSERT(pcbinfo->ipi_count == 0, 490 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 491 492 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 493 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 494 pcbinfo->ipi_porthashmask); 495 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 496 pcbinfo->ipi_lbgrouphashmask); 497 #ifdef PCBGROUP 498 in_pcbgroup_destroy(pcbinfo); 499 #endif 500 uma_zdestroy(pcbinfo->ipi_zone); 501 INP_LIST_LOCK_DESTROY(pcbinfo); 502 INP_HASH_LOCK_DESTROY(pcbinfo); 503 INP_INFO_LOCK_DESTROY(pcbinfo); 504 } 505 506 /* 507 * Allocate a PCB and associate it with the socket. 508 * On success return with the PCB locked. 509 */ 510 int 511 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 512 { 513 struct inpcb *inp; 514 int error; 515 516 #ifdef INVARIANTS 517 if (pcbinfo == &V_tcbinfo) { 518 INP_INFO_RLOCK_ASSERT(pcbinfo); 519 } else { 520 INP_INFO_WLOCK_ASSERT(pcbinfo); 521 } 522 #endif 523 524 error = 0; 525 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 526 if (inp == NULL) 527 return (ENOBUFS); 528 bzero(&inp->inp_start_zero, inp_zero_size); 529 #ifdef NUMA 530 inp->inp_numa_domain = M_NODOM; 531 #endif 532 inp->inp_pcbinfo = pcbinfo; 533 inp->inp_socket = so; 534 inp->inp_cred = crhold(so->so_cred); 535 inp->inp_inc.inc_fibnum = so->so_fibnum; 536 #ifdef MAC 537 error = mac_inpcb_init(inp, M_NOWAIT); 538 if (error != 0) 539 goto out; 540 mac_inpcb_create(so, inp); 541 #endif 542 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 543 error = ipsec_init_pcbpolicy(inp); 544 if (error != 0) { 545 #ifdef MAC 546 mac_inpcb_destroy(inp); 547 #endif 548 goto out; 549 } 550 #endif /*IPSEC*/ 551 #ifdef INET6 552 if (INP_SOCKAF(so) == AF_INET6) { 553 inp->inp_vflag |= INP_IPV6PROTO; 554 if (V_ip6_v6only) 555 inp->inp_flags |= IN6P_IPV6_V6ONLY; 556 } 557 #endif 558 INP_WLOCK(inp); 559 INP_LIST_WLOCK(pcbinfo); 560 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 561 pcbinfo->ipi_count++; 562 so->so_pcb = (caddr_t)inp; 563 #ifdef INET6 564 if (V_ip6_auto_flowlabel) 565 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 566 #endif 567 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 568 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 569 570 /* 571 * Routes in inpcb's can cache L2 as well; they are guaranteed 572 * to be cleaned up. 573 */ 574 inp->inp_route.ro_flags = RT_LLE_CACHE; 575 INP_LIST_WUNLOCK(pcbinfo); 576 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 577 out: 578 if (error != 0) { 579 crfree(inp->inp_cred); 580 uma_zfree(pcbinfo->ipi_zone, inp); 581 } 582 #endif 583 return (error); 584 } 585 586 #ifdef INET 587 int 588 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 589 { 590 int anonport, error; 591 592 INP_WLOCK_ASSERT(inp); 593 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 594 595 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 596 return (EINVAL); 597 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 598 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 599 &inp->inp_lport, cred); 600 if (error) 601 return (error); 602 if (in_pcbinshash(inp) != 0) { 603 inp->inp_laddr.s_addr = INADDR_ANY; 604 inp->inp_lport = 0; 605 return (EAGAIN); 606 } 607 if (anonport) 608 inp->inp_flags |= INP_ANONPORT; 609 return (0); 610 } 611 #endif 612 613 /* 614 * Select a local port (number) to use. 615 */ 616 #if defined(INET) || defined(INET6) 617 int 618 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 619 struct ucred *cred, int lookupflags) 620 { 621 struct inpcbinfo *pcbinfo; 622 struct inpcb *tmpinp; 623 unsigned short *lastport; 624 int count, dorandom, error; 625 u_short aux, first, last, lport; 626 #ifdef INET 627 struct in_addr laddr; 628 #endif 629 630 pcbinfo = inp->inp_pcbinfo; 631 632 /* 633 * Because no actual state changes occur here, a global write lock on 634 * the pcbinfo isn't required. 635 */ 636 INP_LOCK_ASSERT(inp); 637 INP_HASH_LOCK_ASSERT(pcbinfo); 638 639 if (inp->inp_flags & INP_HIGHPORT) { 640 first = V_ipport_hifirstauto; /* sysctl */ 641 last = V_ipport_hilastauto; 642 lastport = &pcbinfo->ipi_lasthi; 643 } else if (inp->inp_flags & INP_LOWPORT) { 644 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 645 if (error) 646 return (error); 647 first = V_ipport_lowfirstauto; /* 1023 */ 648 last = V_ipport_lowlastauto; /* 600 */ 649 lastport = &pcbinfo->ipi_lastlow; 650 } else { 651 first = V_ipport_firstauto; /* sysctl */ 652 last = V_ipport_lastauto; 653 lastport = &pcbinfo->ipi_lastport; 654 } 655 /* 656 * For UDP(-Lite), use random port allocation as long as the user 657 * allows it. For TCP (and as of yet unknown) connections, 658 * use random port allocation only if the user allows it AND 659 * ipport_tick() allows it. 660 */ 661 if (V_ipport_randomized && 662 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 663 pcbinfo == &V_ulitecbinfo)) 664 dorandom = 1; 665 else 666 dorandom = 0; 667 /* 668 * It makes no sense to do random port allocation if 669 * we have the only port available. 670 */ 671 if (first == last) 672 dorandom = 0; 673 /* Make sure to not include UDP(-Lite) packets in the count. */ 674 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 675 V_ipport_tcpallocs++; 676 /* 677 * Instead of having two loops further down counting up or down 678 * make sure that first is always <= last and go with only one 679 * code path implementing all logic. 680 */ 681 if (first > last) { 682 aux = first; 683 first = last; 684 last = aux; 685 } 686 687 #ifdef INET 688 /* Make the compiler happy. */ 689 laddr.s_addr = 0; 690 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 691 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", 692 __func__, inp)); 693 laddr = *laddrp; 694 } 695 #endif 696 tmpinp = NULL; /* Make compiler happy. */ 697 lport = *lportp; 698 699 if (dorandom) 700 *lastport = first + (arc4random() % (last - first)); 701 702 count = last - first; 703 704 do { 705 if (count-- < 0) /* completely used? */ 706 return (EADDRNOTAVAIL); 707 ++*lastport; 708 if (*lastport < first || *lastport > last) 709 *lastport = first; 710 lport = htons(*lastport); 711 712 #ifdef INET6 713 if ((inp->inp_vflag & INP_IPV6) != 0) 714 tmpinp = in6_pcblookup_local(pcbinfo, 715 &inp->in6p_laddr, lport, lookupflags, cred); 716 #endif 717 #if defined(INET) && defined(INET6) 718 else 719 #endif 720 #ifdef INET 721 tmpinp = in_pcblookup_local(pcbinfo, laddr, 722 lport, lookupflags, cred); 723 #endif 724 } while (tmpinp != NULL); 725 726 #ifdef INET 727 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) 728 laddrp->s_addr = laddr.s_addr; 729 #endif 730 *lportp = lport; 731 732 return (0); 733 } 734 735 /* 736 * Return cached socket options. 737 */ 738 int 739 inp_so_options(const struct inpcb *inp) 740 { 741 int so_options; 742 743 so_options = 0; 744 745 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 746 so_options |= SO_REUSEPORT_LB; 747 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 748 so_options |= SO_REUSEPORT; 749 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 750 so_options |= SO_REUSEADDR; 751 return (so_options); 752 } 753 #endif /* INET || INET6 */ 754 755 /* 756 * Check if a new BINDMULTI socket is allowed to be created. 757 * 758 * ni points to the new inp. 759 * oi points to the exisitng inp. 760 * 761 * This checks whether the existing inp also has BINDMULTI and 762 * whether the credentials match. 763 */ 764 int 765 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 766 { 767 /* Check permissions match */ 768 if ((ni->inp_flags2 & INP_BINDMULTI) && 769 (ni->inp_cred->cr_uid != 770 oi->inp_cred->cr_uid)) 771 return (0); 772 773 /* Check the existing inp has BINDMULTI set */ 774 if ((ni->inp_flags2 & INP_BINDMULTI) && 775 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 776 return (0); 777 778 /* 779 * We're okay - either INP_BINDMULTI isn't set on ni, or 780 * it is and it matches the checks. 781 */ 782 return (1); 783 } 784 785 #ifdef INET 786 /* 787 * Set up a bind operation on a PCB, performing port allocation 788 * as required, but do not actually modify the PCB. Callers can 789 * either complete the bind by setting inp_laddr/inp_lport and 790 * calling in_pcbinshash(), or they can just use the resulting 791 * port and address to authorise the sending of a once-off packet. 792 * 793 * On error, the values of *laddrp and *lportp are not changed. 794 */ 795 int 796 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 797 u_short *lportp, struct ucred *cred) 798 { 799 struct socket *so = inp->inp_socket; 800 struct sockaddr_in *sin; 801 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 802 struct in_addr laddr; 803 u_short lport = 0; 804 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 805 int error; 806 807 /* 808 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 809 * so that we don't have to add to the (already messy) code below. 810 */ 811 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 812 813 /* 814 * No state changes, so read locks are sufficient here. 815 */ 816 INP_LOCK_ASSERT(inp); 817 INP_HASH_LOCK_ASSERT(pcbinfo); 818 819 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 820 return (EADDRNOTAVAIL); 821 laddr.s_addr = *laddrp; 822 if (nam != NULL && laddr.s_addr != INADDR_ANY) 823 return (EINVAL); 824 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 825 lookupflags = INPLOOKUP_WILDCARD; 826 if (nam == NULL) { 827 if ((error = prison_local_ip4(cred, &laddr)) != 0) 828 return (error); 829 } else { 830 sin = (struct sockaddr_in *)nam; 831 if (nam->sa_len != sizeof (*sin)) 832 return (EINVAL); 833 #ifdef notdef 834 /* 835 * We should check the family, but old programs 836 * incorrectly fail to initialize it. 837 */ 838 if (sin->sin_family != AF_INET) 839 return (EAFNOSUPPORT); 840 #endif 841 error = prison_local_ip4(cred, &sin->sin_addr); 842 if (error) 843 return (error); 844 if (sin->sin_port != *lportp) { 845 /* Don't allow the port to change. */ 846 if (*lportp != 0) 847 return (EINVAL); 848 lport = sin->sin_port; 849 } 850 /* NB: lport is left as 0 if the port isn't being changed. */ 851 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 852 /* 853 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 854 * allow complete duplication of binding if 855 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 856 * and a multicast address is bound on both 857 * new and duplicated sockets. 858 */ 859 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 860 reuseport = SO_REUSEADDR|SO_REUSEPORT; 861 /* 862 * XXX: How to deal with SO_REUSEPORT_LB here? 863 * Treat same as SO_REUSEPORT for now. 864 */ 865 if ((so->so_options & 866 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 867 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 868 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 869 sin->sin_port = 0; /* yech... */ 870 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 871 /* 872 * Is the address a local IP address? 873 * If INP_BINDANY is set, then the socket may be bound 874 * to any endpoint address, local or not. 875 */ 876 if ((inp->inp_flags & INP_BINDANY) == 0 && 877 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 878 return (EADDRNOTAVAIL); 879 } 880 laddr = sin->sin_addr; 881 if (lport) { 882 struct inpcb *t; 883 struct tcptw *tw; 884 885 /* GROSS */ 886 if (ntohs(lport) <= V_ipport_reservedhigh && 887 ntohs(lport) >= V_ipport_reservedlow && 888 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 889 return (EACCES); 890 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 891 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 892 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 893 lport, INPLOOKUP_WILDCARD, cred); 894 /* 895 * XXX 896 * This entire block sorely needs a rewrite. 897 */ 898 if (t && 899 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 900 ((t->inp_flags & INP_TIMEWAIT) == 0) && 901 (so->so_type != SOCK_STREAM || 902 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 903 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 904 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 905 (t->inp_flags2 & INP_REUSEPORT) || 906 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 907 (inp->inp_cred->cr_uid != 908 t->inp_cred->cr_uid)) 909 return (EADDRINUSE); 910 911 /* 912 * If the socket is a BINDMULTI socket, then 913 * the credentials need to match and the 914 * original socket also has to have been bound 915 * with BINDMULTI. 916 */ 917 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 918 return (EADDRINUSE); 919 } 920 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 921 lport, lookupflags, cred); 922 if (t && (t->inp_flags & INP_TIMEWAIT)) { 923 /* 924 * XXXRW: If an incpb has had its timewait 925 * state recycled, we treat the address as 926 * being in use (for now). This is better 927 * than a panic, but not desirable. 928 */ 929 tw = intotw(t); 930 if (tw == NULL || 931 ((reuseport & tw->tw_so_options) == 0 && 932 (reuseport_lb & 933 tw->tw_so_options) == 0)) { 934 return (EADDRINUSE); 935 } 936 } else if (t && 937 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 938 (reuseport & inp_so_options(t)) == 0 && 939 (reuseport_lb & inp_so_options(t)) == 0) { 940 #ifdef INET6 941 if (ntohl(sin->sin_addr.s_addr) != 942 INADDR_ANY || 943 ntohl(t->inp_laddr.s_addr) != 944 INADDR_ANY || 945 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 946 (t->inp_vflag & INP_IPV6PROTO) == 0) 947 #endif 948 return (EADDRINUSE); 949 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 950 return (EADDRINUSE); 951 } 952 } 953 } 954 if (*lportp != 0) 955 lport = *lportp; 956 if (lport == 0) { 957 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 958 if (error != 0) 959 return (error); 960 961 } 962 *laddrp = laddr.s_addr; 963 *lportp = lport; 964 return (0); 965 } 966 967 /* 968 * Connect from a socket to a specified address. 969 * Both address and port must be specified in argument sin. 970 * If don't have a local address for this socket yet, 971 * then pick one. 972 */ 973 int 974 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 975 struct ucred *cred, struct mbuf *m) 976 { 977 u_short lport, fport; 978 in_addr_t laddr, faddr; 979 int anonport, error; 980 981 INP_WLOCK_ASSERT(inp); 982 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 983 984 lport = inp->inp_lport; 985 laddr = inp->inp_laddr.s_addr; 986 anonport = (lport == 0); 987 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 988 NULL, cred); 989 if (error) 990 return (error); 991 992 /* Do the initial binding of the local address if required. */ 993 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 994 inp->inp_lport = lport; 995 inp->inp_laddr.s_addr = laddr; 996 if (in_pcbinshash(inp) != 0) { 997 inp->inp_laddr.s_addr = INADDR_ANY; 998 inp->inp_lport = 0; 999 return (EAGAIN); 1000 } 1001 } 1002 1003 /* Commit the remaining changes. */ 1004 inp->inp_lport = lport; 1005 inp->inp_laddr.s_addr = laddr; 1006 inp->inp_faddr.s_addr = faddr; 1007 inp->inp_fport = fport; 1008 in_pcbrehash_mbuf(inp, m); 1009 1010 if (anonport) 1011 inp->inp_flags |= INP_ANONPORT; 1012 return (0); 1013 } 1014 1015 int 1016 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 1017 { 1018 1019 return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); 1020 } 1021 1022 /* 1023 * Do proper source address selection on an unbound socket in case 1024 * of connect. Take jails into account as well. 1025 */ 1026 int 1027 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1028 struct ucred *cred) 1029 { 1030 struct ifaddr *ifa; 1031 struct sockaddr *sa; 1032 struct sockaddr_in *sin; 1033 struct route sro; 1034 struct epoch_tracker et; 1035 int error; 1036 1037 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1038 /* 1039 * Bypass source address selection and use the primary jail IP 1040 * if requested. 1041 */ 1042 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1043 return (0); 1044 1045 error = 0; 1046 bzero(&sro, sizeof(sro)); 1047 1048 sin = (struct sockaddr_in *)&sro.ro_dst; 1049 sin->sin_family = AF_INET; 1050 sin->sin_len = sizeof(struct sockaddr_in); 1051 sin->sin_addr.s_addr = faddr->s_addr; 1052 1053 /* 1054 * If route is known our src addr is taken from the i/f, 1055 * else punt. 1056 * 1057 * Find out route to destination. 1058 */ 1059 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1060 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 1061 1062 /* 1063 * If we found a route, use the address corresponding to 1064 * the outgoing interface. 1065 * 1066 * Otherwise assume faddr is reachable on a directly connected 1067 * network and try to find a corresponding interface to take 1068 * the source address from. 1069 */ 1070 NET_EPOCH_ENTER(et); 1071 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 1072 struct in_ifaddr *ia; 1073 struct ifnet *ifp; 1074 1075 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1076 inp->inp_socket->so_fibnum)); 1077 if (ia == NULL) { 1078 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1079 inp->inp_socket->so_fibnum)); 1080 1081 } 1082 if (ia == NULL) { 1083 error = ENETUNREACH; 1084 goto done; 1085 } 1086 1087 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1088 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1089 goto done; 1090 } 1091 1092 ifp = ia->ia_ifp; 1093 ia = NULL; 1094 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1095 1096 sa = ifa->ifa_addr; 1097 if (sa->sa_family != AF_INET) 1098 continue; 1099 sin = (struct sockaddr_in *)sa; 1100 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1101 ia = (struct in_ifaddr *)ifa; 1102 break; 1103 } 1104 } 1105 if (ia != NULL) { 1106 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1107 goto done; 1108 } 1109 1110 /* 3. As a last resort return the 'default' jail address. */ 1111 error = prison_get_ip4(cred, laddr); 1112 goto done; 1113 } 1114 1115 /* 1116 * If the outgoing interface on the route found is not 1117 * a loopback interface, use the address from that interface. 1118 * In case of jails do those three steps: 1119 * 1. check if the interface address belongs to the jail. If so use it. 1120 * 2. check if we have any address on the outgoing interface 1121 * belonging to this jail. If so use it. 1122 * 3. as a last resort return the 'default' jail address. 1123 */ 1124 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 1125 struct in_ifaddr *ia; 1126 struct ifnet *ifp; 1127 1128 /* If not jailed, use the default returned. */ 1129 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1130 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1131 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1132 goto done; 1133 } 1134 1135 /* Jailed. */ 1136 /* 1. Check if the iface address belongs to the jail. */ 1137 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 1138 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1139 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1140 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1141 goto done; 1142 } 1143 1144 /* 1145 * 2. Check if we have any address on the outgoing interface 1146 * belonging to this jail. 1147 */ 1148 ia = NULL; 1149 ifp = sro.ro_rt->rt_ifp; 1150 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1151 sa = ifa->ifa_addr; 1152 if (sa->sa_family != AF_INET) 1153 continue; 1154 sin = (struct sockaddr_in *)sa; 1155 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1156 ia = (struct in_ifaddr *)ifa; 1157 break; 1158 } 1159 } 1160 if (ia != NULL) { 1161 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1162 goto done; 1163 } 1164 1165 /* 3. As a last resort return the 'default' jail address. */ 1166 error = prison_get_ip4(cred, laddr); 1167 goto done; 1168 } 1169 1170 /* 1171 * The outgoing interface is marked with 'loopback net', so a route 1172 * to ourselves is here. 1173 * Try to find the interface of the destination address and then 1174 * take the address from there. That interface is not necessarily 1175 * a loopback interface. 1176 * In case of jails, check that it is an address of the jail 1177 * and if we cannot find, fall back to the 'default' jail address. 1178 */ 1179 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 1180 struct sockaddr_in sain; 1181 struct in_ifaddr *ia; 1182 1183 bzero(&sain, sizeof(struct sockaddr_in)); 1184 sain.sin_family = AF_INET; 1185 sain.sin_len = sizeof(struct sockaddr_in); 1186 sain.sin_addr.s_addr = faddr->s_addr; 1187 1188 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), 1189 inp->inp_socket->so_fibnum)); 1190 if (ia == NULL) 1191 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, 1192 inp->inp_socket->so_fibnum)); 1193 if (ia == NULL) 1194 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 1195 1196 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1197 if (ia == NULL) { 1198 error = ENETUNREACH; 1199 goto done; 1200 } 1201 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1202 goto done; 1203 } 1204 1205 /* Jailed. */ 1206 if (ia != NULL) { 1207 struct ifnet *ifp; 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, 1217 &sin->sin_addr) == 0) { 1218 ia = (struct in_ifaddr *)ifa; 1219 break; 1220 } 1221 } 1222 if (ia != NULL) { 1223 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1224 goto done; 1225 } 1226 } 1227 1228 /* 3. As a last resort return the 'default' jail address. */ 1229 error = prison_get_ip4(cred, laddr); 1230 goto done; 1231 } 1232 1233 done: 1234 NET_EPOCH_EXIT(et); 1235 if (sro.ro_rt != NULL) 1236 RTFREE(sro.ro_rt); 1237 return (error); 1238 } 1239 1240 /* 1241 * Set up for a connect from a socket to the specified address. 1242 * On entry, *laddrp and *lportp should contain the current local 1243 * address and port for the PCB; these are updated to the values 1244 * that should be placed in inp_laddr and inp_lport to complete 1245 * the connect. 1246 * 1247 * On success, *faddrp and *fportp will be set to the remote address 1248 * and port. These are not updated in the error case. 1249 * 1250 * If the operation fails because the connection already exists, 1251 * *oinpp will be set to the PCB of that connection so that the 1252 * caller can decide to override it. In all other cases, *oinpp 1253 * is set to NULL. 1254 */ 1255 int 1256 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1257 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1258 struct inpcb **oinpp, struct ucred *cred) 1259 { 1260 struct rm_priotracker in_ifa_tracker; 1261 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1262 struct in_ifaddr *ia; 1263 struct inpcb *oinp; 1264 struct in_addr laddr, faddr; 1265 u_short lport, fport; 1266 int error; 1267 1268 /* 1269 * Because a global state change doesn't actually occur here, a read 1270 * lock is sufficient. 1271 */ 1272 INP_LOCK_ASSERT(inp); 1273 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1274 1275 if (oinpp != NULL) 1276 *oinpp = NULL; 1277 if (nam->sa_len != sizeof (*sin)) 1278 return (EINVAL); 1279 if (sin->sin_family != AF_INET) 1280 return (EAFNOSUPPORT); 1281 if (sin->sin_port == 0) 1282 return (EADDRNOTAVAIL); 1283 laddr.s_addr = *laddrp; 1284 lport = *lportp; 1285 faddr = sin->sin_addr; 1286 fport = sin->sin_port; 1287 1288 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1289 /* 1290 * If the destination address is INADDR_ANY, 1291 * use the primary local address. 1292 * If the supplied address is INADDR_BROADCAST, 1293 * and the primary interface supports broadcast, 1294 * choose the broadcast address for that interface. 1295 */ 1296 if (faddr.s_addr == INADDR_ANY) { 1297 IN_IFADDR_RLOCK(&in_ifa_tracker); 1298 faddr = 1299 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1300 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1301 if (cred != NULL && 1302 (error = prison_get_ip4(cred, &faddr)) != 0) 1303 return (error); 1304 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1305 IN_IFADDR_RLOCK(&in_ifa_tracker); 1306 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1307 IFF_BROADCAST) 1308 faddr = satosin(&CK_STAILQ_FIRST( 1309 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1310 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1311 } 1312 } 1313 if (laddr.s_addr == INADDR_ANY) { 1314 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1315 /* 1316 * If the destination address is multicast and an outgoing 1317 * interface has been set as a multicast option, prefer the 1318 * address of that interface as our source address. 1319 */ 1320 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1321 inp->inp_moptions != NULL) { 1322 struct ip_moptions *imo; 1323 struct ifnet *ifp; 1324 1325 imo = inp->inp_moptions; 1326 if (imo->imo_multicast_ifp != NULL) { 1327 ifp = imo->imo_multicast_ifp; 1328 IN_IFADDR_RLOCK(&in_ifa_tracker); 1329 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1330 if ((ia->ia_ifp == ifp) && 1331 (cred == NULL || 1332 prison_check_ip4(cred, 1333 &ia->ia_addr.sin_addr) == 0)) 1334 break; 1335 } 1336 if (ia == NULL) 1337 error = EADDRNOTAVAIL; 1338 else { 1339 laddr = ia->ia_addr.sin_addr; 1340 error = 0; 1341 } 1342 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1343 } 1344 } 1345 if (error) 1346 return (error); 1347 } 1348 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1349 laddr, lport, 0, NULL); 1350 if (oinp != NULL) { 1351 if (oinpp != NULL) 1352 *oinpp = oinp; 1353 return (EADDRINUSE); 1354 } 1355 if (lport == 0) { 1356 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1357 cred); 1358 if (error) 1359 return (error); 1360 } 1361 *laddrp = laddr.s_addr; 1362 *lportp = lport; 1363 *faddrp = faddr.s_addr; 1364 *fportp = fport; 1365 return (0); 1366 } 1367 1368 void 1369 in_pcbdisconnect(struct inpcb *inp) 1370 { 1371 1372 INP_WLOCK_ASSERT(inp); 1373 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1374 1375 inp->inp_faddr.s_addr = INADDR_ANY; 1376 inp->inp_fport = 0; 1377 in_pcbrehash(inp); 1378 } 1379 #endif /* INET */ 1380 1381 /* 1382 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1383 * For most protocols, this will be invoked immediately prior to calling 1384 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1385 * socket, in which case in_pcbfree() is deferred. 1386 */ 1387 void 1388 in_pcbdetach(struct inpcb *inp) 1389 { 1390 1391 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1392 1393 #ifdef RATELIMIT 1394 if (inp->inp_snd_tag != NULL) 1395 in_pcbdetach_txrtlmt(inp); 1396 #endif 1397 inp->inp_socket->so_pcb = NULL; 1398 inp->inp_socket = NULL; 1399 } 1400 1401 /* 1402 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1403 * stability of an inpcb pointer despite the inpcb lock being released. This 1404 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1405 * but where the inpcb lock may already held, or when acquiring a reference 1406 * via a pcbgroup. 1407 * 1408 * in_pcbref() should be used only to provide brief memory stability, and 1409 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1410 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1411 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1412 * lock and rele are the *only* safe operations that may be performed on the 1413 * inpcb. 1414 * 1415 * While the inpcb will not be freed, releasing the inpcb lock means that the 1416 * connection's state may change, so the caller should be careful to 1417 * revalidate any cached state on reacquiring the lock. Drop the reference 1418 * using in_pcbrele(). 1419 */ 1420 void 1421 in_pcbref(struct inpcb *inp) 1422 { 1423 1424 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1425 1426 refcount_acquire(&inp->inp_refcount); 1427 } 1428 1429 /* 1430 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1431 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1432 * return a flag indicating whether or not the inpcb remains valid. If it is 1433 * valid, we return with the inpcb lock held. 1434 * 1435 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1436 * reference on an inpcb. Historically more work was done here (actually, in 1437 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1438 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1439 * about memory stability (and continued use of the write lock). 1440 */ 1441 int 1442 in_pcbrele_rlocked(struct inpcb *inp) 1443 { 1444 struct inpcbinfo *pcbinfo; 1445 1446 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1447 1448 INP_RLOCK_ASSERT(inp); 1449 1450 if (refcount_release(&inp->inp_refcount) == 0) { 1451 /* 1452 * If the inpcb has been freed, let the caller know, even if 1453 * this isn't the last reference. 1454 */ 1455 if (inp->inp_flags2 & INP_FREED) { 1456 INP_RUNLOCK(inp); 1457 return (1); 1458 } 1459 return (0); 1460 } 1461 1462 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1463 #ifdef TCPHPTS 1464 if (inp->inp_in_hpts || inp->inp_in_input) { 1465 struct tcp_hpts_entry *hpts; 1466 /* 1467 * We should not be on the hpts at 1468 * this point in any form. we must 1469 * get the lock to be sure. 1470 */ 1471 hpts = tcp_hpts_lock(inp); 1472 if (inp->inp_in_hpts) 1473 panic("Hpts:%p inp:%p at free still on hpts", 1474 hpts, inp); 1475 mtx_unlock(&hpts->p_mtx); 1476 hpts = tcp_input_lock(inp); 1477 if (inp->inp_in_input) 1478 panic("Hpts:%p inp:%p at free still on input hpts", 1479 hpts, inp); 1480 mtx_unlock(&hpts->p_mtx); 1481 } 1482 #endif 1483 INP_RUNLOCK(inp); 1484 pcbinfo = inp->inp_pcbinfo; 1485 uma_zfree(pcbinfo->ipi_zone, inp); 1486 return (1); 1487 } 1488 1489 int 1490 in_pcbrele_wlocked(struct inpcb *inp) 1491 { 1492 struct inpcbinfo *pcbinfo; 1493 1494 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1495 1496 INP_WLOCK_ASSERT(inp); 1497 1498 if (refcount_release(&inp->inp_refcount) == 0) { 1499 /* 1500 * If the inpcb has been freed, let the caller know, even if 1501 * this isn't the last reference. 1502 */ 1503 if (inp->inp_flags2 & INP_FREED) { 1504 INP_WUNLOCK(inp); 1505 return (1); 1506 } 1507 return (0); 1508 } 1509 1510 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1511 #ifdef TCPHPTS 1512 if (inp->inp_in_hpts || inp->inp_in_input) { 1513 struct tcp_hpts_entry *hpts; 1514 /* 1515 * We should not be on the hpts at 1516 * this point in any form. we must 1517 * get the lock to be sure. 1518 */ 1519 hpts = tcp_hpts_lock(inp); 1520 if (inp->inp_in_hpts) 1521 panic("Hpts:%p inp:%p at free still on hpts", 1522 hpts, inp); 1523 mtx_unlock(&hpts->p_mtx); 1524 hpts = tcp_input_lock(inp); 1525 if (inp->inp_in_input) 1526 panic("Hpts:%p inp:%p at free still on input hpts", 1527 hpts, inp); 1528 mtx_unlock(&hpts->p_mtx); 1529 } 1530 #endif 1531 INP_WUNLOCK(inp); 1532 pcbinfo = inp->inp_pcbinfo; 1533 uma_zfree(pcbinfo->ipi_zone, inp); 1534 return (1); 1535 } 1536 1537 /* 1538 * Temporary wrapper. 1539 */ 1540 int 1541 in_pcbrele(struct inpcb *inp) 1542 { 1543 1544 return (in_pcbrele_wlocked(inp)); 1545 } 1546 1547 void 1548 in_pcblist_rele_rlocked(epoch_context_t ctx) 1549 { 1550 struct in_pcblist *il; 1551 struct inpcb *inp; 1552 struct inpcbinfo *pcbinfo; 1553 int i, n; 1554 1555 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1556 pcbinfo = il->il_pcbinfo; 1557 n = il->il_count; 1558 INP_INFO_WLOCK(pcbinfo); 1559 for (i = 0; i < n; i++) { 1560 inp = il->il_inp_list[i]; 1561 INP_RLOCK(inp); 1562 if (!in_pcbrele_rlocked(inp)) 1563 INP_RUNLOCK(inp); 1564 } 1565 INP_INFO_WUNLOCK(pcbinfo); 1566 free(il, M_TEMP); 1567 } 1568 1569 static void 1570 inpcbport_free(epoch_context_t ctx) 1571 { 1572 struct inpcbport *phd; 1573 1574 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); 1575 free(phd, M_PCB); 1576 } 1577 1578 static void 1579 in_pcbfree_deferred(epoch_context_t ctx) 1580 { 1581 struct inpcb *inp; 1582 int released __unused; 1583 1584 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); 1585 1586 INP_WLOCK(inp); 1587 CURVNET_SET(inp->inp_vnet); 1588 #ifdef INET 1589 struct ip_moptions *imo = inp->inp_moptions; 1590 inp->inp_moptions = NULL; 1591 #endif 1592 /* XXXRW: Do as much as possible here. */ 1593 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1594 if (inp->inp_sp != NULL) 1595 ipsec_delete_pcbpolicy(inp); 1596 #endif 1597 #ifdef INET6 1598 struct ip6_moptions *im6o = NULL; 1599 if (inp->inp_vflag & INP_IPV6PROTO) { 1600 ip6_freepcbopts(inp->in6p_outputopts); 1601 im6o = inp->in6p_moptions; 1602 inp->in6p_moptions = NULL; 1603 } 1604 #endif 1605 if (inp->inp_options) 1606 (void)m_free(inp->inp_options); 1607 inp->inp_vflag = 0; 1608 crfree(inp->inp_cred); 1609 #ifdef MAC 1610 mac_inpcb_destroy(inp); 1611 #endif 1612 released = in_pcbrele_wlocked(inp); 1613 MPASS(released); 1614 #ifdef INET6 1615 ip6_freemoptions(im6o); 1616 #endif 1617 #ifdef INET 1618 inp_freemoptions(imo); 1619 #endif 1620 CURVNET_RESTORE(); 1621 } 1622 1623 /* 1624 * Unconditionally schedule an inpcb to be freed by decrementing its 1625 * reference count, which should occur only after the inpcb has been detached 1626 * from its socket. If another thread holds a temporary reference (acquired 1627 * using in_pcbref()) then the free is deferred until that reference is 1628 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1629 * work, including removal from global lists, is done in this context, where 1630 * the pcbinfo lock is held. 1631 */ 1632 void 1633 in_pcbfree(struct inpcb *inp) 1634 { 1635 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1636 1637 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1638 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1639 ("%s: called twice for pcb %p", __func__, inp)); 1640 if (inp->inp_flags2 & INP_FREED) { 1641 INP_WUNLOCK(inp); 1642 return; 1643 } 1644 1645 #ifdef INVARIANTS 1646 if (pcbinfo == &V_tcbinfo) { 1647 INP_INFO_LOCK_ASSERT(pcbinfo); 1648 } else { 1649 INP_INFO_WLOCK_ASSERT(pcbinfo); 1650 } 1651 #endif 1652 INP_WLOCK_ASSERT(inp); 1653 INP_LIST_WLOCK(pcbinfo); 1654 in_pcbremlists(inp); 1655 INP_LIST_WUNLOCK(pcbinfo); 1656 RO_INVALIDATE_CACHE(&inp->inp_route); 1657 /* mark as destruction in progress */ 1658 inp->inp_flags2 |= INP_FREED; 1659 INP_WUNLOCK(inp); 1660 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred); 1661 } 1662 1663 /* 1664 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1665 * port reservation, and preventing it from being returned by inpcb lookups. 1666 * 1667 * It is used by TCP to mark an inpcb as unused and avoid future packet 1668 * delivery or event notification when a socket remains open but TCP has 1669 * closed. This might occur as a result of a shutdown()-initiated TCP close 1670 * or a RST on the wire, and allows the port binding to be reused while still 1671 * maintaining the invariant that so_pcb always points to a valid inpcb until 1672 * in_pcbdetach(). 1673 * 1674 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1675 * in_pcbnotifyall() and in_pcbpurgeif0()? 1676 */ 1677 void 1678 in_pcbdrop(struct inpcb *inp) 1679 { 1680 1681 INP_WLOCK_ASSERT(inp); 1682 #ifdef INVARIANTS 1683 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1684 MPASS(inp->inp_refcount > 1); 1685 #endif 1686 1687 /* 1688 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1689 * the hash lock...? 1690 */ 1691 inp->inp_flags |= INP_DROPPED; 1692 if (inp->inp_flags & INP_INHASHLIST) { 1693 struct inpcbport *phd = inp->inp_phd; 1694 1695 INP_HASH_WLOCK(inp->inp_pcbinfo); 1696 in_pcbremlbgrouphash(inp); 1697 CK_LIST_REMOVE(inp, inp_hash); 1698 CK_LIST_REMOVE(inp, inp_portlist); 1699 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 1700 CK_LIST_REMOVE(phd, phd_hash); 1701 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 1702 } 1703 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1704 inp->inp_flags &= ~INP_INHASHLIST; 1705 #ifdef PCBGROUP 1706 in_pcbgroup_remove(inp); 1707 #endif 1708 } 1709 } 1710 1711 #ifdef INET 1712 /* 1713 * Common routines to return the socket addresses associated with inpcbs. 1714 */ 1715 struct sockaddr * 1716 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1717 { 1718 struct sockaddr_in *sin; 1719 1720 sin = malloc(sizeof *sin, M_SONAME, 1721 M_WAITOK | M_ZERO); 1722 sin->sin_family = AF_INET; 1723 sin->sin_len = sizeof(*sin); 1724 sin->sin_addr = *addr_p; 1725 sin->sin_port = port; 1726 1727 return (struct sockaddr *)sin; 1728 } 1729 1730 int 1731 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1732 { 1733 struct inpcb *inp; 1734 struct in_addr addr; 1735 in_port_t port; 1736 1737 inp = sotoinpcb(so); 1738 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1739 1740 INP_RLOCK(inp); 1741 port = inp->inp_lport; 1742 addr = inp->inp_laddr; 1743 INP_RUNLOCK(inp); 1744 1745 *nam = in_sockaddr(port, &addr); 1746 return 0; 1747 } 1748 1749 int 1750 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1751 { 1752 struct inpcb *inp; 1753 struct in_addr addr; 1754 in_port_t port; 1755 1756 inp = sotoinpcb(so); 1757 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1758 1759 INP_RLOCK(inp); 1760 port = inp->inp_fport; 1761 addr = inp->inp_faddr; 1762 INP_RUNLOCK(inp); 1763 1764 *nam = in_sockaddr(port, &addr); 1765 return 0; 1766 } 1767 1768 void 1769 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1770 struct inpcb *(*notify)(struct inpcb *, int)) 1771 { 1772 struct inpcb *inp, *inp_temp; 1773 1774 INP_INFO_WLOCK(pcbinfo); 1775 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1776 INP_WLOCK(inp); 1777 #ifdef INET6 1778 if ((inp->inp_vflag & INP_IPV4) == 0) { 1779 INP_WUNLOCK(inp); 1780 continue; 1781 } 1782 #endif 1783 if (inp->inp_faddr.s_addr != faddr.s_addr || 1784 inp->inp_socket == NULL) { 1785 INP_WUNLOCK(inp); 1786 continue; 1787 } 1788 if ((*notify)(inp, errno)) 1789 INP_WUNLOCK(inp); 1790 } 1791 INP_INFO_WUNLOCK(pcbinfo); 1792 } 1793 1794 void 1795 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1796 { 1797 struct inpcb *inp; 1798 struct in_multi *inm; 1799 struct in_mfilter *imf; 1800 struct ip_moptions *imo; 1801 1802 INP_INFO_WLOCK(pcbinfo); 1803 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1804 INP_WLOCK(inp); 1805 imo = inp->inp_moptions; 1806 if ((inp->inp_vflag & INP_IPV4) && 1807 imo != NULL) { 1808 /* 1809 * Unselect the outgoing interface if it is being 1810 * detached. 1811 */ 1812 if (imo->imo_multicast_ifp == ifp) 1813 imo->imo_multicast_ifp = NULL; 1814 1815 /* 1816 * Drop multicast group membership if we joined 1817 * through the interface being detached. 1818 * 1819 * XXX This can all be deferred to an epoch_call 1820 */ 1821 restart: 1822 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 1823 if ((inm = imf->imf_inm) == NULL) 1824 continue; 1825 if (inm->inm_ifp != ifp) 1826 continue; 1827 ip_mfilter_remove(&imo->imo_head, imf); 1828 IN_MULTI_LOCK_ASSERT(); 1829 in_leavegroup_locked(inm, NULL); 1830 ip_mfilter_free(imf); 1831 goto restart; 1832 } 1833 } 1834 INP_WUNLOCK(inp); 1835 } 1836 INP_INFO_WUNLOCK(pcbinfo); 1837 } 1838 1839 /* 1840 * Lookup a PCB based on the local address and port. Caller must hold the 1841 * hash lock. No inpcb locks or references are acquired. 1842 */ 1843 #define INP_LOOKUP_MAPPED_PCB_COST 3 1844 struct inpcb * 1845 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1846 u_short lport, int lookupflags, struct ucred *cred) 1847 { 1848 struct inpcb *inp; 1849 #ifdef INET6 1850 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1851 #else 1852 int matchwild = 3; 1853 #endif 1854 int wildcard; 1855 1856 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1857 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1858 1859 INP_HASH_LOCK_ASSERT(pcbinfo); 1860 1861 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1862 struct inpcbhead *head; 1863 /* 1864 * Look for an unconnected (wildcard foreign addr) PCB that 1865 * matches the local address and port we're looking for. 1866 */ 1867 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1868 0, pcbinfo->ipi_hashmask)]; 1869 CK_LIST_FOREACH(inp, head, inp_hash) { 1870 #ifdef INET6 1871 /* XXX inp locking */ 1872 if ((inp->inp_vflag & INP_IPV4) == 0) 1873 continue; 1874 #endif 1875 if (inp->inp_faddr.s_addr == INADDR_ANY && 1876 inp->inp_laddr.s_addr == laddr.s_addr && 1877 inp->inp_lport == lport) { 1878 /* 1879 * Found? 1880 */ 1881 if (cred == NULL || 1882 prison_equal_ip4(cred->cr_prison, 1883 inp->inp_cred->cr_prison)) 1884 return (inp); 1885 } 1886 } 1887 /* 1888 * Not found. 1889 */ 1890 return (NULL); 1891 } else { 1892 struct inpcbporthead *porthash; 1893 struct inpcbport *phd; 1894 struct inpcb *match = NULL; 1895 /* 1896 * Best fit PCB lookup. 1897 * 1898 * First see if this local port is in use by looking on the 1899 * port hash list. 1900 */ 1901 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1902 pcbinfo->ipi_porthashmask)]; 1903 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1904 if (phd->phd_port == lport) 1905 break; 1906 } 1907 if (phd != NULL) { 1908 /* 1909 * Port is in use by one or more PCBs. Look for best 1910 * fit. 1911 */ 1912 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1913 wildcard = 0; 1914 if (cred != NULL && 1915 !prison_equal_ip4(inp->inp_cred->cr_prison, 1916 cred->cr_prison)) 1917 continue; 1918 #ifdef INET6 1919 /* XXX inp locking */ 1920 if ((inp->inp_vflag & INP_IPV4) == 0) 1921 continue; 1922 /* 1923 * We never select the PCB that has 1924 * INP_IPV6 flag and is bound to :: if 1925 * we have another PCB which is bound 1926 * to 0.0.0.0. If a PCB has the 1927 * INP_IPV6 flag, then we set its cost 1928 * higher than IPv4 only PCBs. 1929 * 1930 * Note that the case only happens 1931 * when a socket is bound to ::, under 1932 * the condition that the use of the 1933 * mapped address is allowed. 1934 */ 1935 if ((inp->inp_vflag & INP_IPV6) != 0) 1936 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1937 #endif 1938 if (inp->inp_faddr.s_addr != INADDR_ANY) 1939 wildcard++; 1940 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1941 if (laddr.s_addr == INADDR_ANY) 1942 wildcard++; 1943 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1944 continue; 1945 } else { 1946 if (laddr.s_addr != INADDR_ANY) 1947 wildcard++; 1948 } 1949 if (wildcard < matchwild) { 1950 match = inp; 1951 matchwild = wildcard; 1952 if (matchwild == 0) 1953 break; 1954 } 1955 } 1956 } 1957 return (match); 1958 } 1959 } 1960 #undef INP_LOOKUP_MAPPED_PCB_COST 1961 1962 static struct inpcb * 1963 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 1964 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, 1965 uint16_t fport, int lookupflags) 1966 { 1967 struct inpcb *local_wild; 1968 const struct inpcblbgrouphead *hdr; 1969 struct inpcblbgroup *grp; 1970 uint32_t idx; 1971 1972 INP_HASH_LOCK_ASSERT(pcbinfo); 1973 1974 hdr = &pcbinfo->ipi_lbgrouphashbase[ 1975 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 1976 1977 /* 1978 * Order of socket selection: 1979 * 1. non-wild. 1980 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). 1981 * 1982 * NOTE: 1983 * - Load balanced group does not contain jailed sockets 1984 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets 1985 */ 1986 local_wild = NULL; 1987 CK_LIST_FOREACH(grp, hdr, il_list) { 1988 #ifdef INET6 1989 if (!(grp->il_vflag & INP_IPV4)) 1990 continue; 1991 #endif 1992 if (grp->il_lport != lport) 1993 continue; 1994 1995 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % 1996 grp->il_inpcnt; 1997 if (grp->il_laddr.s_addr == laddr->s_addr) 1998 return (grp->il_inp[idx]); 1999 if (grp->il_laddr.s_addr == INADDR_ANY && 2000 (lookupflags & INPLOOKUP_WILDCARD) != 0) 2001 local_wild = grp->il_inp[idx]; 2002 } 2003 return (local_wild); 2004 } 2005 2006 #ifdef PCBGROUP 2007 /* 2008 * Lookup PCB in hash list, using pcbgroup tables. 2009 */ 2010 static struct inpcb * 2011 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 2012 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 2013 u_int lport_arg, int lookupflags, struct ifnet *ifp) 2014 { 2015 struct inpcbhead *head; 2016 struct inpcb *inp, *tmpinp; 2017 u_short fport = fport_arg, lport = lport_arg; 2018 bool locked; 2019 2020 /* 2021 * First look for an exact match. 2022 */ 2023 tmpinp = NULL; 2024 INP_GROUP_LOCK(pcbgroup); 2025 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2026 pcbgroup->ipg_hashmask)]; 2027 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2028 #ifdef INET6 2029 /* XXX inp locking */ 2030 if ((inp->inp_vflag & INP_IPV4) == 0) 2031 continue; 2032 #endif 2033 if (inp->inp_faddr.s_addr == faddr.s_addr && 2034 inp->inp_laddr.s_addr == laddr.s_addr && 2035 inp->inp_fport == fport && 2036 inp->inp_lport == lport) { 2037 /* 2038 * XXX We should be able to directly return 2039 * the inp here, without any checks. 2040 * Well unless both bound with SO_REUSEPORT? 2041 */ 2042 if (prison_flag(inp->inp_cred, PR_IP4)) 2043 goto found; 2044 if (tmpinp == NULL) 2045 tmpinp = inp; 2046 } 2047 } 2048 if (tmpinp != NULL) { 2049 inp = tmpinp; 2050 goto found; 2051 } 2052 2053 #ifdef RSS 2054 /* 2055 * For incoming connections, we may wish to do a wildcard 2056 * match for an RSS-local socket. 2057 */ 2058 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2059 struct inpcb *local_wild = NULL, *local_exact = NULL; 2060 #ifdef INET6 2061 struct inpcb *local_wild_mapped = NULL; 2062 #endif 2063 struct inpcb *jail_wild = NULL; 2064 struct inpcbhead *head; 2065 int injail; 2066 2067 /* 2068 * Order of socket selection - we always prefer jails. 2069 * 1. jailed, non-wild. 2070 * 2. jailed, wild. 2071 * 3. non-jailed, non-wild. 2072 * 4. non-jailed, wild. 2073 */ 2074 2075 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 2076 lport, 0, pcbgroup->ipg_hashmask)]; 2077 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2078 #ifdef INET6 2079 /* XXX inp locking */ 2080 if ((inp->inp_vflag & INP_IPV4) == 0) 2081 continue; 2082 #endif 2083 if (inp->inp_faddr.s_addr != INADDR_ANY || 2084 inp->inp_lport != lport) 2085 continue; 2086 2087 injail = prison_flag(inp->inp_cred, PR_IP4); 2088 if (injail) { 2089 if (prison_check_ip4(inp->inp_cred, 2090 &laddr) != 0) 2091 continue; 2092 } else { 2093 if (local_exact != NULL) 2094 continue; 2095 } 2096 2097 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2098 if (injail) 2099 goto found; 2100 else 2101 local_exact = inp; 2102 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2103 #ifdef INET6 2104 /* XXX inp locking, NULL check */ 2105 if (inp->inp_vflag & INP_IPV6PROTO) 2106 local_wild_mapped = inp; 2107 else 2108 #endif 2109 if (injail) 2110 jail_wild = inp; 2111 else 2112 local_wild = inp; 2113 } 2114 } /* LIST_FOREACH */ 2115 2116 inp = jail_wild; 2117 if (inp == NULL) 2118 inp = local_exact; 2119 if (inp == NULL) 2120 inp = local_wild; 2121 #ifdef INET6 2122 if (inp == NULL) 2123 inp = local_wild_mapped; 2124 #endif 2125 if (inp != NULL) 2126 goto found; 2127 } 2128 #endif 2129 2130 /* 2131 * Then look for a wildcard match, if requested. 2132 */ 2133 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2134 struct inpcb *local_wild = NULL, *local_exact = NULL; 2135 #ifdef INET6 2136 struct inpcb *local_wild_mapped = NULL; 2137 #endif 2138 struct inpcb *jail_wild = NULL; 2139 struct inpcbhead *head; 2140 int injail; 2141 2142 /* 2143 * Order of socket selection - we always prefer jails. 2144 * 1. jailed, non-wild. 2145 * 2. jailed, wild. 2146 * 3. non-jailed, non-wild. 2147 * 4. non-jailed, wild. 2148 */ 2149 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 2150 0, pcbinfo->ipi_wildmask)]; 2151 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 2152 #ifdef INET6 2153 /* XXX inp locking */ 2154 if ((inp->inp_vflag & INP_IPV4) == 0) 2155 continue; 2156 #endif 2157 if (inp->inp_faddr.s_addr != INADDR_ANY || 2158 inp->inp_lport != lport) 2159 continue; 2160 2161 injail = prison_flag(inp->inp_cred, PR_IP4); 2162 if (injail) { 2163 if (prison_check_ip4(inp->inp_cred, 2164 &laddr) != 0) 2165 continue; 2166 } else { 2167 if (local_exact != NULL) 2168 continue; 2169 } 2170 2171 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2172 if (injail) 2173 goto found; 2174 else 2175 local_exact = inp; 2176 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2177 #ifdef INET6 2178 /* XXX inp locking, NULL check */ 2179 if (inp->inp_vflag & INP_IPV6PROTO) 2180 local_wild_mapped = inp; 2181 else 2182 #endif 2183 if (injail) 2184 jail_wild = inp; 2185 else 2186 local_wild = inp; 2187 } 2188 } /* LIST_FOREACH */ 2189 inp = jail_wild; 2190 if (inp == NULL) 2191 inp = local_exact; 2192 if (inp == NULL) 2193 inp = local_wild; 2194 #ifdef INET6 2195 if (inp == NULL) 2196 inp = local_wild_mapped; 2197 #endif 2198 if (inp != NULL) 2199 goto found; 2200 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 2201 INP_GROUP_UNLOCK(pcbgroup); 2202 return (NULL); 2203 2204 found: 2205 if (lookupflags & INPLOOKUP_WLOCKPCB) 2206 locked = INP_TRY_WLOCK(inp); 2207 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2208 locked = INP_TRY_RLOCK(inp); 2209 else 2210 panic("%s: locking bug", __func__); 2211 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { 2212 if (lookupflags & INPLOOKUP_WLOCKPCB) 2213 INP_WUNLOCK(inp); 2214 else 2215 INP_RUNLOCK(inp); 2216 return (NULL); 2217 } else if (!locked) 2218 in_pcbref(inp); 2219 INP_GROUP_UNLOCK(pcbgroup); 2220 if (!locked) { 2221 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2222 INP_WLOCK(inp); 2223 if (in_pcbrele_wlocked(inp)) 2224 return (NULL); 2225 } else { 2226 INP_RLOCK(inp); 2227 if (in_pcbrele_rlocked(inp)) 2228 return (NULL); 2229 } 2230 } 2231 #ifdef INVARIANTS 2232 if (lookupflags & INPLOOKUP_WLOCKPCB) 2233 INP_WLOCK_ASSERT(inp); 2234 else 2235 INP_RLOCK_ASSERT(inp); 2236 #endif 2237 return (inp); 2238 } 2239 #endif /* PCBGROUP */ 2240 2241 /* 2242 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2243 * that the caller has locked the hash list, and will not perform any further 2244 * locking or reference operations on either the hash list or the connection. 2245 */ 2246 static struct inpcb * 2247 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2248 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2249 struct ifnet *ifp) 2250 { 2251 struct inpcbhead *head; 2252 struct inpcb *inp, *tmpinp; 2253 u_short fport = fport_arg, lport = lport_arg; 2254 2255 #ifdef INVARIANTS 2256 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2257 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2258 if (!mtx_owned(&pcbinfo->ipi_hash_lock)) 2259 MPASS(in_epoch_verbose(net_epoch_preempt, 1)); 2260 #endif 2261 /* 2262 * First look for an exact match. 2263 */ 2264 tmpinp = NULL; 2265 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2266 pcbinfo->ipi_hashmask)]; 2267 CK_LIST_FOREACH(inp, head, inp_hash) { 2268 #ifdef INET6 2269 /* XXX inp locking */ 2270 if ((inp->inp_vflag & INP_IPV4) == 0) 2271 continue; 2272 #endif 2273 if (inp->inp_faddr.s_addr == faddr.s_addr && 2274 inp->inp_laddr.s_addr == laddr.s_addr && 2275 inp->inp_fport == fport && 2276 inp->inp_lport == lport) { 2277 /* 2278 * XXX We should be able to directly return 2279 * the inp here, without any checks. 2280 * Well unless both bound with SO_REUSEPORT? 2281 */ 2282 if (prison_flag(inp->inp_cred, PR_IP4)) 2283 return (inp); 2284 if (tmpinp == NULL) 2285 tmpinp = inp; 2286 } 2287 } 2288 if (tmpinp != NULL) 2289 return (tmpinp); 2290 2291 /* 2292 * Then look in lb group (for wildcard match). 2293 */ 2294 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2295 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, 2296 fport, lookupflags); 2297 if (inp != NULL) 2298 return (inp); 2299 } 2300 2301 /* 2302 * Then look for a wildcard match, if requested. 2303 */ 2304 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2305 struct inpcb *local_wild = NULL, *local_exact = NULL; 2306 #ifdef INET6 2307 struct inpcb *local_wild_mapped = NULL; 2308 #endif 2309 struct inpcb *jail_wild = NULL; 2310 int injail; 2311 2312 /* 2313 * Order of socket selection - we always prefer jails. 2314 * 1. jailed, non-wild. 2315 * 2. jailed, wild. 2316 * 3. non-jailed, non-wild. 2317 * 4. non-jailed, wild. 2318 */ 2319 2320 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2321 0, pcbinfo->ipi_hashmask)]; 2322 CK_LIST_FOREACH(inp, head, inp_hash) { 2323 #ifdef INET6 2324 /* XXX inp locking */ 2325 if ((inp->inp_vflag & INP_IPV4) == 0) 2326 continue; 2327 #endif 2328 if (inp->inp_faddr.s_addr != INADDR_ANY || 2329 inp->inp_lport != lport) 2330 continue; 2331 2332 injail = prison_flag(inp->inp_cred, PR_IP4); 2333 if (injail) { 2334 if (prison_check_ip4(inp->inp_cred, 2335 &laddr) != 0) 2336 continue; 2337 } else { 2338 if (local_exact != NULL) 2339 continue; 2340 } 2341 2342 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2343 if (injail) 2344 return (inp); 2345 else 2346 local_exact = inp; 2347 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2348 #ifdef INET6 2349 /* XXX inp locking, NULL check */ 2350 if (inp->inp_vflag & INP_IPV6PROTO) 2351 local_wild_mapped = inp; 2352 else 2353 #endif 2354 if (injail) 2355 jail_wild = inp; 2356 else 2357 local_wild = inp; 2358 } 2359 } /* LIST_FOREACH */ 2360 if (jail_wild != NULL) 2361 return (jail_wild); 2362 if (local_exact != NULL) 2363 return (local_exact); 2364 if (local_wild != NULL) 2365 return (local_wild); 2366 #ifdef INET6 2367 if (local_wild_mapped != NULL) 2368 return (local_wild_mapped); 2369 #endif 2370 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2371 2372 return (NULL); 2373 } 2374 2375 /* 2376 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2377 * hash list lock, and will return the inpcb locked (i.e., requires 2378 * INPLOOKUP_LOCKPCB). 2379 */ 2380 static struct inpcb * 2381 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2382 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2383 struct ifnet *ifp) 2384 { 2385 struct inpcb *inp; 2386 2387 INP_HASH_RLOCK(pcbinfo); 2388 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2389 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 2390 if (inp != NULL) { 2391 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2392 INP_WLOCK(inp); 2393 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2394 INP_WUNLOCK(inp); 2395 inp = NULL; 2396 } 2397 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2398 INP_RLOCK(inp); 2399 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2400 INP_RUNLOCK(inp); 2401 inp = NULL; 2402 } 2403 } else 2404 panic("%s: locking bug", __func__); 2405 #ifdef INVARIANTS 2406 if (inp != NULL) { 2407 if (lookupflags & INPLOOKUP_WLOCKPCB) 2408 INP_WLOCK_ASSERT(inp); 2409 else 2410 INP_RLOCK_ASSERT(inp); 2411 } 2412 #endif 2413 } 2414 INP_HASH_RUNLOCK(pcbinfo); 2415 return (inp); 2416 } 2417 2418 /* 2419 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2420 * from which a pre-calculated hash value may be extracted. 2421 * 2422 * Possibly more of this logic should be in in_pcbgroup.c. 2423 */ 2424 struct inpcb * 2425 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2426 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2427 { 2428 #if defined(PCBGROUP) && !defined(RSS) 2429 struct inpcbgroup *pcbgroup; 2430 #endif 2431 2432 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2433 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2434 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2435 ("%s: LOCKPCB not set", __func__)); 2436 2437 /* 2438 * When not using RSS, use connection groups in preference to the 2439 * reservation table when looking up 4-tuples. When using RSS, just 2440 * use the reservation table, due to the cost of the Toeplitz hash 2441 * in software. 2442 * 2443 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2444 * we could be doing RSS with a non-Toeplitz hash that is affordable 2445 * in software. 2446 */ 2447 #if defined(PCBGROUP) && !defined(RSS) 2448 if (in_pcbgroup_enabled(pcbinfo)) { 2449 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2450 fport); 2451 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2452 laddr, lport, lookupflags, ifp)); 2453 } 2454 #endif 2455 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2456 lookupflags, ifp)); 2457 } 2458 2459 struct inpcb * 2460 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2461 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2462 struct ifnet *ifp, struct mbuf *m) 2463 { 2464 #ifdef PCBGROUP 2465 struct inpcbgroup *pcbgroup; 2466 #endif 2467 2468 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2469 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2470 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2471 ("%s: LOCKPCB not set", __func__)); 2472 2473 #ifdef PCBGROUP 2474 /* 2475 * If we can use a hardware-generated hash to look up the connection 2476 * group, use that connection group to find the inpcb. Otherwise 2477 * fall back on a software hash -- or the reservation table if we're 2478 * using RSS. 2479 * 2480 * XXXRW: As above, that policy belongs in the pcbgroup code. 2481 */ 2482 if (in_pcbgroup_enabled(pcbinfo) && 2483 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2484 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2485 m->m_pkthdr.flowid); 2486 if (pcbgroup != NULL) 2487 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2488 fport, laddr, lport, lookupflags, ifp)); 2489 #ifndef RSS 2490 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2491 fport); 2492 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2493 laddr, lport, lookupflags, ifp)); 2494 #endif 2495 } 2496 #endif 2497 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2498 lookupflags, ifp)); 2499 } 2500 #endif /* INET */ 2501 2502 /* 2503 * Insert PCB onto various hash lists. 2504 */ 2505 static int 2506 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 2507 { 2508 struct inpcbhead *pcbhash; 2509 struct inpcbporthead *pcbporthash; 2510 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2511 struct inpcbport *phd; 2512 u_int32_t hashkey_faddr; 2513 int so_options; 2514 2515 INP_WLOCK_ASSERT(inp); 2516 INP_HASH_WLOCK_ASSERT(pcbinfo); 2517 2518 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2519 ("in_pcbinshash: INP_INHASHLIST")); 2520 2521 #ifdef INET6 2522 if (inp->inp_vflag & INP_IPV6) 2523 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2524 else 2525 #endif 2526 hashkey_faddr = inp->inp_faddr.s_addr; 2527 2528 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2529 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2530 2531 pcbporthash = &pcbinfo->ipi_porthashbase[ 2532 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2533 2534 /* 2535 * Add entry to load balance group. 2536 * Only do this if SO_REUSEPORT_LB is set. 2537 */ 2538 so_options = inp_so_options(inp); 2539 if (so_options & SO_REUSEPORT_LB) { 2540 int ret = in_pcbinslbgrouphash(inp); 2541 if (ret) { 2542 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2543 return (ret); 2544 } 2545 } 2546 2547 /* 2548 * Go through port list and look for a head for this lport. 2549 */ 2550 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2551 if (phd->phd_port == inp->inp_lport) 2552 break; 2553 } 2554 /* 2555 * If none exists, malloc one and tack it on. 2556 */ 2557 if (phd == NULL) { 2558 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2559 if (phd == NULL) { 2560 return (ENOBUFS); /* XXX */ 2561 } 2562 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2563 phd->phd_port = inp->inp_lport; 2564 CK_LIST_INIT(&phd->phd_pcblist); 2565 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2566 } 2567 inp->inp_phd = phd; 2568 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2569 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2570 inp->inp_flags |= INP_INHASHLIST; 2571 #ifdef PCBGROUP 2572 if (do_pcbgroup_update) 2573 in_pcbgroup_update(inp); 2574 #endif 2575 return (0); 2576 } 2577 2578 /* 2579 * For now, there are two public interfaces to insert an inpcb into the hash 2580 * lists -- one that does update pcbgroups, and one that doesn't. The latter 2581 * is used only in the TCP syncache, where in_pcbinshash is called before the 2582 * full 4-tuple is set for the inpcb, and we don't want to install in the 2583 * pcbgroup until later. 2584 * 2585 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 2586 * connection groups, and partially initialised inpcbs should not be exposed 2587 * to either reservation hash tables or pcbgroups. 2588 */ 2589 int 2590 in_pcbinshash(struct inpcb *inp) 2591 { 2592 2593 return (in_pcbinshash_internal(inp, 1)); 2594 } 2595 2596 int 2597 in_pcbinshash_nopcbgroup(struct inpcb *inp) 2598 { 2599 2600 return (in_pcbinshash_internal(inp, 0)); 2601 } 2602 2603 /* 2604 * Move PCB to the proper hash bucket when { faddr, fport } have been 2605 * changed. NOTE: This does not handle the case of the lport changing (the 2606 * hashed port list would have to be updated as well), so the lport must 2607 * not change after in_pcbinshash() has been called. 2608 */ 2609 void 2610 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2611 { 2612 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2613 struct inpcbhead *head; 2614 u_int32_t hashkey_faddr; 2615 2616 INP_WLOCK_ASSERT(inp); 2617 INP_HASH_WLOCK_ASSERT(pcbinfo); 2618 2619 KASSERT(inp->inp_flags & INP_INHASHLIST, 2620 ("in_pcbrehash: !INP_INHASHLIST")); 2621 2622 #ifdef INET6 2623 if (inp->inp_vflag & INP_IPV6) 2624 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2625 else 2626 #endif 2627 hashkey_faddr = inp->inp_faddr.s_addr; 2628 2629 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2630 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2631 2632 CK_LIST_REMOVE(inp, inp_hash); 2633 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2634 2635 #ifdef PCBGROUP 2636 if (m != NULL) 2637 in_pcbgroup_update_mbuf(inp, m); 2638 else 2639 in_pcbgroup_update(inp); 2640 #endif 2641 } 2642 2643 void 2644 in_pcbrehash(struct inpcb *inp) 2645 { 2646 2647 in_pcbrehash_mbuf(inp, NULL); 2648 } 2649 2650 /* 2651 * Remove PCB from various lists. 2652 */ 2653 static void 2654 in_pcbremlists(struct inpcb *inp) 2655 { 2656 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2657 2658 #ifdef INVARIANTS 2659 if (pcbinfo == &V_tcbinfo) { 2660 INP_INFO_RLOCK_ASSERT(pcbinfo); 2661 } else { 2662 INP_INFO_WLOCK_ASSERT(pcbinfo); 2663 } 2664 #endif 2665 2666 INP_WLOCK_ASSERT(inp); 2667 INP_LIST_WLOCK_ASSERT(pcbinfo); 2668 2669 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2670 if (inp->inp_flags & INP_INHASHLIST) { 2671 struct inpcbport *phd = inp->inp_phd; 2672 2673 INP_HASH_WLOCK(pcbinfo); 2674 2675 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2676 in_pcbremlbgrouphash(inp); 2677 2678 CK_LIST_REMOVE(inp, inp_hash); 2679 CK_LIST_REMOVE(inp, inp_portlist); 2680 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2681 CK_LIST_REMOVE(phd, phd_hash); 2682 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 2683 } 2684 INP_HASH_WUNLOCK(pcbinfo); 2685 inp->inp_flags &= ~INP_INHASHLIST; 2686 } 2687 CK_LIST_REMOVE(inp, inp_list); 2688 pcbinfo->ipi_count--; 2689 #ifdef PCBGROUP 2690 in_pcbgroup_remove(inp); 2691 #endif 2692 } 2693 2694 /* 2695 * Check for alternatives when higher level complains 2696 * about service problems. For now, invalidate cached 2697 * routing information. If the route was created dynamically 2698 * (by a redirect), time to try a default gateway again. 2699 */ 2700 void 2701 in_losing(struct inpcb *inp) 2702 { 2703 2704 RO_INVALIDATE_CACHE(&inp->inp_route); 2705 return; 2706 } 2707 2708 /* 2709 * A set label operation has occurred at the socket layer, propagate the 2710 * label change into the in_pcb for the socket. 2711 */ 2712 void 2713 in_pcbsosetlabel(struct socket *so) 2714 { 2715 #ifdef MAC 2716 struct inpcb *inp; 2717 2718 inp = sotoinpcb(so); 2719 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2720 2721 INP_WLOCK(inp); 2722 SOCK_LOCK(so); 2723 mac_inpcb_sosetlabel(so, inp); 2724 SOCK_UNLOCK(so); 2725 INP_WUNLOCK(inp); 2726 #endif 2727 } 2728 2729 /* 2730 * ipport_tick runs once per second, determining if random port allocation 2731 * should be continued. If more than ipport_randomcps ports have been 2732 * allocated in the last second, then we return to sequential port 2733 * allocation. We return to random allocation only once we drop below 2734 * ipport_randomcps for at least ipport_randomtime seconds. 2735 */ 2736 static void 2737 ipport_tick(void *xtp) 2738 { 2739 VNET_ITERATOR_DECL(vnet_iter); 2740 2741 VNET_LIST_RLOCK_NOSLEEP(); 2742 VNET_FOREACH(vnet_iter) { 2743 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2744 if (V_ipport_tcpallocs <= 2745 V_ipport_tcplastcount + V_ipport_randomcps) { 2746 if (V_ipport_stoprandom > 0) 2747 V_ipport_stoprandom--; 2748 } else 2749 V_ipport_stoprandom = V_ipport_randomtime; 2750 V_ipport_tcplastcount = V_ipport_tcpallocs; 2751 CURVNET_RESTORE(); 2752 } 2753 VNET_LIST_RUNLOCK_NOSLEEP(); 2754 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2755 } 2756 2757 static void 2758 ip_fini(void *xtp) 2759 { 2760 2761 callout_stop(&ipport_tick_callout); 2762 } 2763 2764 /* 2765 * The ipport_callout should start running at about the time we attach the 2766 * inet or inet6 domains. 2767 */ 2768 static void 2769 ipport_tick_init(const void *unused __unused) 2770 { 2771 2772 /* Start ipport_tick. */ 2773 callout_init(&ipport_tick_callout, 1); 2774 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2775 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2776 SHUTDOWN_PRI_DEFAULT); 2777 } 2778 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2779 ipport_tick_init, NULL); 2780 2781 void 2782 inp_wlock(struct inpcb *inp) 2783 { 2784 2785 INP_WLOCK(inp); 2786 } 2787 2788 void 2789 inp_wunlock(struct inpcb *inp) 2790 { 2791 2792 INP_WUNLOCK(inp); 2793 } 2794 2795 void 2796 inp_rlock(struct inpcb *inp) 2797 { 2798 2799 INP_RLOCK(inp); 2800 } 2801 2802 void 2803 inp_runlock(struct inpcb *inp) 2804 { 2805 2806 INP_RUNLOCK(inp); 2807 } 2808 2809 #ifdef INVARIANT_SUPPORT 2810 void 2811 inp_lock_assert(struct inpcb *inp) 2812 { 2813 2814 INP_WLOCK_ASSERT(inp); 2815 } 2816 2817 void 2818 inp_unlock_assert(struct inpcb *inp) 2819 { 2820 2821 INP_UNLOCK_ASSERT(inp); 2822 } 2823 #endif 2824 2825 void 2826 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2827 { 2828 struct inpcb *inp; 2829 2830 INP_INFO_WLOCK(&V_tcbinfo); 2831 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2832 INP_WLOCK(inp); 2833 func(inp, arg); 2834 INP_WUNLOCK(inp); 2835 } 2836 INP_INFO_WUNLOCK(&V_tcbinfo); 2837 } 2838 2839 struct socket * 2840 inp_inpcbtosocket(struct inpcb *inp) 2841 { 2842 2843 INP_WLOCK_ASSERT(inp); 2844 return (inp->inp_socket); 2845 } 2846 2847 struct tcpcb * 2848 inp_inpcbtotcpcb(struct inpcb *inp) 2849 { 2850 2851 INP_WLOCK_ASSERT(inp); 2852 return ((struct tcpcb *)inp->inp_ppcb); 2853 } 2854 2855 int 2856 inp_ip_tos_get(const struct inpcb *inp) 2857 { 2858 2859 return (inp->inp_ip_tos); 2860 } 2861 2862 void 2863 inp_ip_tos_set(struct inpcb *inp, int val) 2864 { 2865 2866 inp->inp_ip_tos = val; 2867 } 2868 2869 void 2870 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2871 uint32_t *faddr, uint16_t *fp) 2872 { 2873 2874 INP_LOCK_ASSERT(inp); 2875 *laddr = inp->inp_laddr.s_addr; 2876 *faddr = inp->inp_faddr.s_addr; 2877 *lp = inp->inp_lport; 2878 *fp = inp->inp_fport; 2879 } 2880 2881 struct inpcb * 2882 so_sotoinpcb(struct socket *so) 2883 { 2884 2885 return (sotoinpcb(so)); 2886 } 2887 2888 struct tcpcb * 2889 so_sototcpcb(struct socket *so) 2890 { 2891 2892 return (sototcpcb(so)); 2893 } 2894 2895 /* 2896 * Create an external-format (``xinpcb'') structure using the information in 2897 * the kernel-format in_pcb structure pointed to by inp. This is done to 2898 * reduce the spew of irrelevant information over this interface, to isolate 2899 * user code from changes in the kernel structure, and potentially to provide 2900 * information-hiding if we decide that some of this information should be 2901 * hidden from users. 2902 */ 2903 void 2904 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2905 { 2906 2907 bzero(xi, sizeof(*xi)); 2908 xi->xi_len = sizeof(struct xinpcb); 2909 if (inp->inp_socket) 2910 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2911 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2912 xi->inp_gencnt = inp->inp_gencnt; 2913 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2914 xi->inp_flow = inp->inp_flow; 2915 xi->inp_flowid = inp->inp_flowid; 2916 xi->inp_flowtype = inp->inp_flowtype; 2917 xi->inp_flags = inp->inp_flags; 2918 xi->inp_flags2 = inp->inp_flags2; 2919 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2920 xi->in6p_cksum = inp->in6p_cksum; 2921 xi->in6p_hops = inp->in6p_hops; 2922 xi->inp_ip_tos = inp->inp_ip_tos; 2923 xi->inp_vflag = inp->inp_vflag; 2924 xi->inp_ip_ttl = inp->inp_ip_ttl; 2925 xi->inp_ip_p = inp->inp_ip_p; 2926 xi->inp_ip_minttl = inp->inp_ip_minttl; 2927 } 2928 2929 #ifdef DDB 2930 static void 2931 db_print_indent(int indent) 2932 { 2933 int i; 2934 2935 for (i = 0; i < indent; i++) 2936 db_printf(" "); 2937 } 2938 2939 static void 2940 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2941 { 2942 char faddr_str[48], laddr_str[48]; 2943 2944 db_print_indent(indent); 2945 db_printf("%s at %p\n", name, inc); 2946 2947 indent += 2; 2948 2949 #ifdef INET6 2950 if (inc->inc_flags & INC_ISIPV6) { 2951 /* IPv6. */ 2952 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2953 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2954 } else 2955 #endif 2956 { 2957 /* IPv4. */ 2958 inet_ntoa_r(inc->inc_laddr, laddr_str); 2959 inet_ntoa_r(inc->inc_faddr, faddr_str); 2960 } 2961 db_print_indent(indent); 2962 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2963 ntohs(inc->inc_lport)); 2964 db_print_indent(indent); 2965 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2966 ntohs(inc->inc_fport)); 2967 } 2968 2969 static void 2970 db_print_inpflags(int inp_flags) 2971 { 2972 int comma; 2973 2974 comma = 0; 2975 if (inp_flags & INP_RECVOPTS) { 2976 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2977 comma = 1; 2978 } 2979 if (inp_flags & INP_RECVRETOPTS) { 2980 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2981 comma = 1; 2982 } 2983 if (inp_flags & INP_RECVDSTADDR) { 2984 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2985 comma = 1; 2986 } 2987 if (inp_flags & INP_ORIGDSTADDR) { 2988 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2989 comma = 1; 2990 } 2991 if (inp_flags & INP_HDRINCL) { 2992 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2993 comma = 1; 2994 } 2995 if (inp_flags & INP_HIGHPORT) { 2996 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2997 comma = 1; 2998 } 2999 if (inp_flags & INP_LOWPORT) { 3000 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3001 comma = 1; 3002 } 3003 if (inp_flags & INP_ANONPORT) { 3004 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3005 comma = 1; 3006 } 3007 if (inp_flags & INP_RECVIF) { 3008 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3009 comma = 1; 3010 } 3011 if (inp_flags & INP_MTUDISC) { 3012 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3013 comma = 1; 3014 } 3015 if (inp_flags & INP_RECVTTL) { 3016 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3017 comma = 1; 3018 } 3019 if (inp_flags & INP_DONTFRAG) { 3020 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3021 comma = 1; 3022 } 3023 if (inp_flags & INP_RECVTOS) { 3024 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3025 comma = 1; 3026 } 3027 if (inp_flags & IN6P_IPV6_V6ONLY) { 3028 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3029 comma = 1; 3030 } 3031 if (inp_flags & IN6P_PKTINFO) { 3032 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3033 comma = 1; 3034 } 3035 if (inp_flags & IN6P_HOPLIMIT) { 3036 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3037 comma = 1; 3038 } 3039 if (inp_flags & IN6P_HOPOPTS) { 3040 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3041 comma = 1; 3042 } 3043 if (inp_flags & IN6P_DSTOPTS) { 3044 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3045 comma = 1; 3046 } 3047 if (inp_flags & IN6P_RTHDR) { 3048 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3049 comma = 1; 3050 } 3051 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3052 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3053 comma = 1; 3054 } 3055 if (inp_flags & IN6P_TCLASS) { 3056 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3057 comma = 1; 3058 } 3059 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3060 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3061 comma = 1; 3062 } 3063 if (inp_flags & INP_TIMEWAIT) { 3064 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3065 comma = 1; 3066 } 3067 if (inp_flags & INP_ONESBCAST) { 3068 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3069 comma = 1; 3070 } 3071 if (inp_flags & INP_DROPPED) { 3072 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3073 comma = 1; 3074 } 3075 if (inp_flags & INP_SOCKREF) { 3076 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3077 comma = 1; 3078 } 3079 if (inp_flags & IN6P_RFC2292) { 3080 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3081 comma = 1; 3082 } 3083 if (inp_flags & IN6P_MTU) { 3084 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3085 comma = 1; 3086 } 3087 } 3088 3089 static void 3090 db_print_inpvflag(u_char inp_vflag) 3091 { 3092 int comma; 3093 3094 comma = 0; 3095 if (inp_vflag & INP_IPV4) { 3096 db_printf("%sINP_IPV4", comma ? ", " : ""); 3097 comma = 1; 3098 } 3099 if (inp_vflag & INP_IPV6) { 3100 db_printf("%sINP_IPV6", comma ? ", " : ""); 3101 comma = 1; 3102 } 3103 if (inp_vflag & INP_IPV6PROTO) { 3104 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3105 comma = 1; 3106 } 3107 } 3108 3109 static void 3110 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3111 { 3112 3113 db_print_indent(indent); 3114 db_printf("%s at %p\n", name, inp); 3115 3116 indent += 2; 3117 3118 db_print_indent(indent); 3119 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3120 3121 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3122 3123 db_print_indent(indent); 3124 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3125 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3126 3127 db_print_indent(indent); 3128 db_printf("inp_label: %p inp_flags: 0x%x (", 3129 inp->inp_label, inp->inp_flags); 3130 db_print_inpflags(inp->inp_flags); 3131 db_printf(")\n"); 3132 3133 db_print_indent(indent); 3134 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3135 inp->inp_vflag); 3136 db_print_inpvflag(inp->inp_vflag); 3137 db_printf(")\n"); 3138 3139 db_print_indent(indent); 3140 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3141 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3142 3143 db_print_indent(indent); 3144 #ifdef INET6 3145 if (inp->inp_vflag & INP_IPV6) { 3146 db_printf("in6p_options: %p in6p_outputopts: %p " 3147 "in6p_moptions: %p\n", inp->in6p_options, 3148 inp->in6p_outputopts, inp->in6p_moptions); 3149 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3150 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3151 inp->in6p_hops); 3152 } else 3153 #endif 3154 { 3155 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3156 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3157 inp->inp_options, inp->inp_moptions); 3158 } 3159 3160 db_print_indent(indent); 3161 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3162 (uintmax_t)inp->inp_gencnt); 3163 } 3164 3165 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3166 { 3167 struct inpcb *inp; 3168 3169 if (!have_addr) { 3170 db_printf("usage: show inpcb <addr>\n"); 3171 return; 3172 } 3173 inp = (struct inpcb *)addr; 3174 3175 db_print_inpcb(inp, "inpcb", 0); 3176 } 3177 #endif /* DDB */ 3178 3179 #ifdef RATELIMIT 3180 /* 3181 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3182 * if any. 3183 */ 3184 int 3185 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3186 { 3187 union if_snd_tag_modify_params params = { 3188 .rate_limit.max_rate = max_pacing_rate, 3189 .rate_limit.flags = M_NOWAIT, 3190 }; 3191 struct m_snd_tag *mst; 3192 struct ifnet *ifp; 3193 int error; 3194 3195 mst = inp->inp_snd_tag; 3196 if (mst == NULL) 3197 return (EINVAL); 3198 3199 ifp = mst->ifp; 3200 if (ifp == NULL) 3201 return (EINVAL); 3202 3203 if (ifp->if_snd_tag_modify == NULL) { 3204 error = EOPNOTSUPP; 3205 } else { 3206 error = ifp->if_snd_tag_modify(mst, ¶ms); 3207 } 3208 return (error); 3209 } 3210 3211 /* 3212 * Query existing TX rate limit based on the existing 3213 * "inp->inp_snd_tag", if any. 3214 */ 3215 int 3216 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3217 { 3218 union if_snd_tag_query_params params = { }; 3219 struct m_snd_tag *mst; 3220 struct ifnet *ifp; 3221 int error; 3222 3223 mst = inp->inp_snd_tag; 3224 if (mst == NULL) 3225 return (EINVAL); 3226 3227 ifp = mst->ifp; 3228 if (ifp == NULL) 3229 return (EINVAL); 3230 3231 if (ifp->if_snd_tag_query == NULL) { 3232 error = EOPNOTSUPP; 3233 } else { 3234 error = ifp->if_snd_tag_query(mst, ¶ms); 3235 if (error == 0 && p_max_pacing_rate != NULL) 3236 *p_max_pacing_rate = params.rate_limit.max_rate; 3237 } 3238 return (error); 3239 } 3240 3241 /* 3242 * Query existing TX queue level based on the existing 3243 * "inp->inp_snd_tag", if any. 3244 */ 3245 int 3246 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3247 { 3248 union if_snd_tag_query_params params = { }; 3249 struct m_snd_tag *mst; 3250 struct ifnet *ifp; 3251 int error; 3252 3253 mst = inp->inp_snd_tag; 3254 if (mst == NULL) 3255 return (EINVAL); 3256 3257 ifp = mst->ifp; 3258 if (ifp == NULL) 3259 return (EINVAL); 3260 3261 if (ifp->if_snd_tag_query == NULL) 3262 return (EOPNOTSUPP); 3263 3264 error = ifp->if_snd_tag_query(mst, ¶ms); 3265 if (error == 0 && p_txqueue_level != NULL) 3266 *p_txqueue_level = params.rate_limit.queue_level; 3267 return (error); 3268 } 3269 3270 /* 3271 * Allocate a new TX rate limit send tag from the network interface 3272 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3273 */ 3274 int 3275 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3276 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st) 3277 3278 { 3279 union if_snd_tag_alloc_params params = { 3280 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3281 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3282 .rate_limit.hdr.flowid = flowid, 3283 .rate_limit.hdr.flowtype = flowtype, 3284 .rate_limit.max_rate = max_pacing_rate, 3285 .rate_limit.flags = M_NOWAIT, 3286 }; 3287 int error; 3288 3289 INP_WLOCK_ASSERT(inp); 3290 3291 if (*st != NULL) 3292 return (EINVAL); 3293 3294 if (ifp->if_snd_tag_alloc == NULL) { 3295 error = EOPNOTSUPP; 3296 } else { 3297 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); 3298 3299 #ifdef INET 3300 if (error == 0) { 3301 counter_u64_add(rate_limit_set_ok, 1); 3302 counter_u64_add(rate_limit_active, 1); 3303 } else 3304 counter_u64_add(rate_limit_alloc_fail, 1); 3305 #endif 3306 } 3307 return (error); 3308 } 3309 3310 void 3311 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst) 3312 { 3313 if (ifp == NULL) 3314 return; 3315 3316 /* 3317 * If the device was detached while we still had reference(s) 3318 * on the ifp, we assume if_snd_tag_free() was replaced with 3319 * stubs. 3320 */ 3321 ifp->if_snd_tag_free(mst); 3322 3323 /* release reference count on network interface */ 3324 if_rele(ifp); 3325 #ifdef INET 3326 counter_u64_add(rate_limit_active, -1); 3327 #endif 3328 } 3329 3330 /* 3331 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3332 * if any: 3333 */ 3334 void 3335 in_pcbdetach_txrtlmt(struct inpcb *inp) 3336 { 3337 struct m_snd_tag *mst; 3338 3339 INP_WLOCK_ASSERT(inp); 3340 3341 mst = inp->inp_snd_tag; 3342 inp->inp_snd_tag = NULL; 3343 3344 if (mst == NULL) 3345 return; 3346 3347 m_snd_tag_rele(mst); 3348 } 3349 3350 int 3351 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) 3352 { 3353 int error; 3354 3355 /* 3356 * If the existing send tag is for the wrong interface due to 3357 * a route change, first drop the existing tag. Set the 3358 * CHANGED flag so that we will keep trying to allocate a new 3359 * tag if we fail to allocate one this time. 3360 */ 3361 if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { 3362 in_pcbdetach_txrtlmt(inp); 3363 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3364 } 3365 3366 /* 3367 * NOTE: When attaching to a network interface a reference is 3368 * made to ensure the network interface doesn't go away until 3369 * all ratelimit connections are gone. The network interface 3370 * pointers compared below represent valid network interfaces, 3371 * except when comparing towards NULL. 3372 */ 3373 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3374 error = 0; 3375 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3376 if (inp->inp_snd_tag != NULL) 3377 in_pcbdetach_txrtlmt(inp); 3378 error = 0; 3379 } else if (inp->inp_snd_tag == NULL) { 3380 /* 3381 * In order to utilize packet pacing with RSS, we need 3382 * to wait until there is a valid RSS hash before we 3383 * can proceed: 3384 */ 3385 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3386 error = EAGAIN; 3387 } else { 3388 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3389 mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag); 3390 } 3391 } else { 3392 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3393 } 3394 if (error == 0 || error == EOPNOTSUPP) 3395 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3396 3397 return (error); 3398 } 3399 3400 /* 3401 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3402 * is set in the fast path and will attach/detach/modify the TX rate 3403 * limit send tag based on the socket's so_max_pacing_rate value. 3404 */ 3405 void 3406 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3407 { 3408 struct socket *socket; 3409 uint32_t max_pacing_rate; 3410 bool did_upgrade; 3411 int error; 3412 3413 if (inp == NULL) 3414 return; 3415 3416 socket = inp->inp_socket; 3417 if (socket == NULL) 3418 return; 3419 3420 if (!INP_WLOCKED(inp)) { 3421 /* 3422 * NOTE: If the write locking fails, we need to bail 3423 * out and use the non-ratelimited ring for the 3424 * transmit until there is a new chance to get the 3425 * write lock. 3426 */ 3427 if (!INP_TRY_UPGRADE(inp)) 3428 return; 3429 did_upgrade = 1; 3430 } else { 3431 did_upgrade = 0; 3432 } 3433 3434 /* 3435 * NOTE: The so_max_pacing_rate value is read unlocked, 3436 * because atomic updates are not required since the variable 3437 * is checked at every mbuf we send. It is assumed that the 3438 * variable read itself will be atomic. 3439 */ 3440 max_pacing_rate = socket->so_max_pacing_rate; 3441 3442 error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); 3443 3444 if (did_upgrade) 3445 INP_DOWNGRADE(inp); 3446 } 3447 3448 /* 3449 * Track route changes for TX rate limiting. 3450 */ 3451 void 3452 in_pcboutput_eagain(struct inpcb *inp) 3453 { 3454 bool did_upgrade; 3455 3456 if (inp == NULL) 3457 return; 3458 3459 if (inp->inp_snd_tag == NULL) 3460 return; 3461 3462 if (!INP_WLOCKED(inp)) { 3463 /* 3464 * NOTE: If the write locking fails, we need to bail 3465 * out and use the non-ratelimited ring for the 3466 * transmit until there is a new chance to get the 3467 * write lock. 3468 */ 3469 if (!INP_TRY_UPGRADE(inp)) 3470 return; 3471 did_upgrade = 1; 3472 } else { 3473 did_upgrade = 0; 3474 } 3475 3476 /* detach rate limiting */ 3477 in_pcbdetach_txrtlmt(inp); 3478 3479 /* make sure new mbuf send tag allocation is made */ 3480 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3481 3482 if (did_upgrade) 3483 INP_DOWNGRADE(inp); 3484 } 3485 3486 #ifdef INET 3487 static void 3488 rl_init(void *st) 3489 { 3490 rate_limit_active = counter_u64_alloc(M_WAITOK); 3491 rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); 3492 rate_limit_set_ok = counter_u64_alloc(M_WAITOK); 3493 } 3494 3495 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); 3496 #endif 3497 #endif /* RATELIMIT */ 3498