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 #include <netinet/ip_var.h> 90 #include <netinet/tcp_var.h> 91 #ifdef TCPHPTS 92 #include <netinet/tcp_hpts.h> 93 #endif 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 #endif 97 #ifdef INET 98 #include <netinet/in_var.h> 99 #endif 100 #ifdef INET6 101 #include <netinet/ip6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #endif /* INET6 */ 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 #endif /* INET */ 214 215 /* 216 * in_pcb.c: manage the Protocol Control Blocks. 217 * 218 * NOTE: It is assumed that most of these functions will be called with 219 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 220 * functions often modify hash chains or addresses in pcbs. 221 */ 222 223 static struct inpcblbgroup * 224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag, 225 uint16_t port, const union in_dependaddr *addr, int size) 226 { 227 struct inpcblbgroup *grp; 228 size_t bytes; 229 230 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 231 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 232 if (!grp) 233 return (NULL); 234 grp->il_vflag = vflag; 235 grp->il_lport = port; 236 grp->il_dependladdr = *addr; 237 grp->il_inpsiz = size; 238 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 239 return (grp); 240 } 241 242 static void 243 in_pcblbgroup_free_deferred(epoch_context_t ctx) 244 { 245 struct inpcblbgroup *grp; 246 247 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 248 free(grp, M_PCB); 249 } 250 251 static void 252 in_pcblbgroup_free(struct inpcblbgroup *grp) 253 { 254 255 CK_LIST_REMOVE(grp, il_list); 256 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx, 257 in_pcblbgroup_free_deferred); 258 } 259 260 static struct inpcblbgroup * 261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 262 struct inpcblbgroup *old_grp, int size) 263 { 264 struct inpcblbgroup *grp; 265 int i; 266 267 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, 268 old_grp->il_lport, &old_grp->il_dependladdr, size); 269 if (grp == NULL) 270 return (NULL); 271 272 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 273 ("invalid new local group size %d and old local group count %d", 274 grp->il_inpsiz, old_grp->il_inpcnt)); 275 276 for (i = 0; i < old_grp->il_inpcnt; ++i) 277 grp->il_inp[i] = old_grp->il_inp[i]; 278 grp->il_inpcnt = old_grp->il_inpcnt; 279 in_pcblbgroup_free(old_grp); 280 return (grp); 281 } 282 283 /* 284 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 285 * and shrink group if possible. 286 */ 287 static void 288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 289 int i) 290 { 291 struct inpcblbgroup *grp, *new_grp; 292 293 grp = *grpp; 294 for (; i + 1 < grp->il_inpcnt; ++i) 295 grp->il_inp[i] = grp->il_inp[i + 1]; 296 grp->il_inpcnt--; 297 298 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 299 grp->il_inpcnt <= grp->il_inpsiz / 4) { 300 /* Shrink this group. */ 301 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 302 if (new_grp != NULL) 303 *grpp = new_grp; 304 } 305 } 306 307 /* 308 * Add PCB to load balance group for SO_REUSEPORT_LB option. 309 */ 310 static int 311 in_pcbinslbgrouphash(struct inpcb *inp) 312 { 313 const static struct timeval interval = { 60, 0 }; 314 static struct timeval lastprint; 315 struct inpcbinfo *pcbinfo; 316 struct inpcblbgrouphead *hdr; 317 struct inpcblbgroup *grp; 318 uint32_t idx; 319 320 pcbinfo = inp->inp_pcbinfo; 321 322 INP_WLOCK_ASSERT(inp); 323 INP_HASH_WLOCK_ASSERT(pcbinfo); 324 325 /* 326 * Don't allow jailed socket to join local group. 327 */ 328 if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred)) 329 return (0); 330 331 #ifdef INET6 332 /* 333 * Don't allow IPv4 mapped INET6 wild socket. 334 */ 335 if ((inp->inp_vflag & INP_IPV4) && 336 inp->inp_laddr.s_addr == INADDR_ANY && 337 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 338 return (0); 339 } 340 #endif 341 342 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); 343 hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; 344 CK_LIST_FOREACH(grp, hdr, il_list) { 345 if (grp->il_vflag == inp->inp_vflag && 346 grp->il_lport == inp->inp_lport && 347 memcmp(&grp->il_dependladdr, 348 &inp->inp_inc.inc_ie.ie_dependladdr, 349 sizeof(grp->il_dependladdr)) == 0) 350 break; 351 } 352 if (grp == NULL) { 353 /* Create new load balance group. */ 354 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, 355 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 356 INPCBLBGROUP_SIZMIN); 357 if (grp == NULL) 358 return (ENOBUFS); 359 } else if (grp->il_inpcnt == grp->il_inpsiz) { 360 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 361 if (ratecheck(&lastprint, &interval)) 362 printf("lb group port %d, limit reached\n", 363 ntohs(grp->il_lport)); 364 return (0); 365 } 366 367 /* Expand this local group. */ 368 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 369 if (grp == NULL) 370 return (ENOBUFS); 371 } 372 373 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 374 ("invalid local group size %d and count %d", grp->il_inpsiz, 375 grp->il_inpcnt)); 376 377 grp->il_inp[grp->il_inpcnt] = inp; 378 grp->il_inpcnt++; 379 return (0); 380 } 381 382 /* 383 * Remove PCB from load balance group. 384 */ 385 static void 386 in_pcbremlbgrouphash(struct inpcb *inp) 387 { 388 struct inpcbinfo *pcbinfo; 389 struct inpcblbgrouphead *hdr; 390 struct inpcblbgroup *grp; 391 int i; 392 393 pcbinfo = inp->inp_pcbinfo; 394 395 INP_WLOCK_ASSERT(inp); 396 INP_HASH_WLOCK_ASSERT(pcbinfo); 397 398 hdr = &pcbinfo->ipi_lbgrouphashbase[ 399 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; 400 CK_LIST_FOREACH(grp, hdr, il_list) { 401 for (i = 0; i < grp->il_inpcnt; ++i) { 402 if (grp->il_inp[i] != inp) 403 continue; 404 405 if (grp->il_inpcnt == 1) { 406 /* We are the last, free this local group. */ 407 in_pcblbgroup_free(grp); 408 } else { 409 /* Pull up inpcbs, shrink group if possible. */ 410 in_pcblbgroup_reorder(hdr, &grp, i); 411 } 412 return; 413 } 414 } 415 } 416 417 /* 418 * Different protocols initialize their inpcbs differently - giving 419 * different name to the lock. But they all are disposed the same. 420 */ 421 static void 422 inpcb_fini(void *mem, int size) 423 { 424 struct inpcb *inp = mem; 425 426 INP_LOCK_DESTROY(inp); 427 } 428 429 /* 430 * Initialize an inpcbinfo -- we should be able to reduce the number of 431 * arguments in time. 432 */ 433 void 434 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 435 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 436 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) 437 { 438 439 porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 440 441 INP_INFO_LOCK_INIT(pcbinfo, name); 442 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 443 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 444 #ifdef VIMAGE 445 pcbinfo->ipi_vnet = curvnet; 446 #endif 447 pcbinfo->ipi_listhead = listhead; 448 CK_LIST_INIT(pcbinfo->ipi_listhead); 449 pcbinfo->ipi_count = 0; 450 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 451 &pcbinfo->ipi_hashmask); 452 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 453 &pcbinfo->ipi_porthashmask); 454 pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB, 455 &pcbinfo->ipi_lbgrouphashmask); 456 #ifdef PCBGROUP 457 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 458 #endif 459 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 460 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); 461 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 462 uma_zone_set_warning(pcbinfo->ipi_zone, 463 "kern.ipc.maxsockets limit reached"); 464 } 465 466 /* 467 * Destroy an inpcbinfo. 468 */ 469 void 470 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 471 { 472 473 KASSERT(pcbinfo->ipi_count == 0, 474 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 475 476 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 477 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 478 pcbinfo->ipi_porthashmask); 479 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 480 pcbinfo->ipi_lbgrouphashmask); 481 #ifdef PCBGROUP 482 in_pcbgroup_destroy(pcbinfo); 483 #endif 484 uma_zdestroy(pcbinfo->ipi_zone); 485 INP_LIST_LOCK_DESTROY(pcbinfo); 486 INP_HASH_LOCK_DESTROY(pcbinfo); 487 INP_INFO_LOCK_DESTROY(pcbinfo); 488 } 489 490 /* 491 * Allocate a PCB and associate it with the socket. 492 * On success return with the PCB locked. 493 */ 494 int 495 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 496 { 497 struct inpcb *inp; 498 int error; 499 500 #ifdef INVARIANTS 501 if (pcbinfo == &V_tcbinfo) { 502 INP_INFO_RLOCK_ASSERT(pcbinfo); 503 } else { 504 INP_INFO_WLOCK_ASSERT(pcbinfo); 505 } 506 #endif 507 508 error = 0; 509 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 510 if (inp == NULL) 511 return (ENOBUFS); 512 bzero(&inp->inp_start_zero, inp_zero_size); 513 inp->inp_pcbinfo = pcbinfo; 514 inp->inp_socket = so; 515 inp->inp_cred = crhold(so->so_cred); 516 inp->inp_inc.inc_fibnum = so->so_fibnum; 517 #ifdef MAC 518 error = mac_inpcb_init(inp, M_NOWAIT); 519 if (error != 0) 520 goto out; 521 mac_inpcb_create(so, inp); 522 #endif 523 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 524 error = ipsec_init_pcbpolicy(inp); 525 if (error != 0) { 526 #ifdef MAC 527 mac_inpcb_destroy(inp); 528 #endif 529 goto out; 530 } 531 #endif /*IPSEC*/ 532 #ifdef INET6 533 if (INP_SOCKAF(so) == AF_INET6) { 534 inp->inp_vflag |= INP_IPV6PROTO; 535 if (V_ip6_v6only) 536 inp->inp_flags |= IN6P_IPV6_V6ONLY; 537 } 538 #endif 539 INP_WLOCK(inp); 540 INP_LIST_WLOCK(pcbinfo); 541 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 542 pcbinfo->ipi_count++; 543 so->so_pcb = (caddr_t)inp; 544 #ifdef INET6 545 if (V_ip6_auto_flowlabel) 546 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 547 #endif 548 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 549 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 550 551 /* 552 * Routes in inpcb's can cache L2 as well; they are guaranteed 553 * to be cleaned up. 554 */ 555 inp->inp_route.ro_flags = RT_LLE_CACHE; 556 INP_LIST_WUNLOCK(pcbinfo); 557 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 558 out: 559 if (error != 0) { 560 crfree(inp->inp_cred); 561 uma_zfree(pcbinfo->ipi_zone, inp); 562 } 563 #endif 564 return (error); 565 } 566 567 #ifdef INET 568 int 569 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 570 { 571 int anonport, error; 572 573 INP_WLOCK_ASSERT(inp); 574 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 575 576 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 577 return (EINVAL); 578 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 579 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 580 &inp->inp_lport, cred); 581 if (error) 582 return (error); 583 if (in_pcbinshash(inp) != 0) { 584 inp->inp_laddr.s_addr = INADDR_ANY; 585 inp->inp_lport = 0; 586 return (EAGAIN); 587 } 588 if (anonport) 589 inp->inp_flags |= INP_ANONPORT; 590 return (0); 591 } 592 #endif 593 594 /* 595 * Select a local port (number) to use. 596 */ 597 #if defined(INET) || defined(INET6) 598 int 599 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 600 struct ucred *cred, int lookupflags) 601 { 602 struct inpcbinfo *pcbinfo; 603 struct inpcb *tmpinp; 604 unsigned short *lastport; 605 int count, dorandom, error; 606 u_short aux, first, last, lport; 607 #ifdef INET 608 struct in_addr laddr; 609 #endif 610 611 pcbinfo = inp->inp_pcbinfo; 612 613 /* 614 * Because no actual state changes occur here, a global write lock on 615 * the pcbinfo isn't required. 616 */ 617 INP_LOCK_ASSERT(inp); 618 INP_HASH_LOCK_ASSERT(pcbinfo); 619 620 if (inp->inp_flags & INP_HIGHPORT) { 621 first = V_ipport_hifirstauto; /* sysctl */ 622 last = V_ipport_hilastauto; 623 lastport = &pcbinfo->ipi_lasthi; 624 } else if (inp->inp_flags & INP_LOWPORT) { 625 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT); 626 if (error) 627 return (error); 628 first = V_ipport_lowfirstauto; /* 1023 */ 629 last = V_ipport_lowlastauto; /* 600 */ 630 lastport = &pcbinfo->ipi_lastlow; 631 } else { 632 first = V_ipport_firstauto; /* sysctl */ 633 last = V_ipport_lastauto; 634 lastport = &pcbinfo->ipi_lastport; 635 } 636 /* 637 * For UDP(-Lite), use random port allocation as long as the user 638 * allows it. For TCP (and as of yet unknown) connections, 639 * use random port allocation only if the user allows it AND 640 * ipport_tick() allows it. 641 */ 642 if (V_ipport_randomized && 643 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 644 pcbinfo == &V_ulitecbinfo)) 645 dorandom = 1; 646 else 647 dorandom = 0; 648 /* 649 * It makes no sense to do random port allocation if 650 * we have the only port available. 651 */ 652 if (first == last) 653 dorandom = 0; 654 /* Make sure to not include UDP(-Lite) packets in the count. */ 655 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 656 V_ipport_tcpallocs++; 657 /* 658 * Instead of having two loops further down counting up or down 659 * make sure that first is always <= last and go with only one 660 * code path implementing all logic. 661 */ 662 if (first > last) { 663 aux = first; 664 first = last; 665 last = aux; 666 } 667 668 #ifdef INET 669 /* Make the compiler happy. */ 670 laddr.s_addr = 0; 671 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 672 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", 673 __func__, inp)); 674 laddr = *laddrp; 675 } 676 #endif 677 tmpinp = NULL; /* Make compiler happy. */ 678 lport = *lportp; 679 680 if (dorandom) 681 *lastport = first + (arc4random() % (last - first)); 682 683 count = last - first; 684 685 do { 686 if (count-- < 0) /* completely used? */ 687 return (EADDRNOTAVAIL); 688 ++*lastport; 689 if (*lastport < first || *lastport > last) 690 *lastport = first; 691 lport = htons(*lastport); 692 693 #ifdef INET6 694 if ((inp->inp_vflag & INP_IPV6) != 0) 695 tmpinp = in6_pcblookup_local(pcbinfo, 696 &inp->in6p_laddr, lport, lookupflags, cred); 697 #endif 698 #if defined(INET) && defined(INET6) 699 else 700 #endif 701 #ifdef INET 702 tmpinp = in_pcblookup_local(pcbinfo, laddr, 703 lport, lookupflags, cred); 704 #endif 705 } while (tmpinp != NULL); 706 707 #ifdef INET 708 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) 709 laddrp->s_addr = laddr.s_addr; 710 #endif 711 *lportp = lport; 712 713 return (0); 714 } 715 716 /* 717 * Return cached socket options. 718 */ 719 int 720 inp_so_options(const struct inpcb *inp) 721 { 722 int so_options; 723 724 so_options = 0; 725 726 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 727 so_options |= SO_REUSEPORT_LB; 728 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 729 so_options |= SO_REUSEPORT; 730 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 731 so_options |= SO_REUSEADDR; 732 return (so_options); 733 } 734 #endif /* INET || INET6 */ 735 736 /* 737 * Check if a new BINDMULTI socket is allowed to be created. 738 * 739 * ni points to the new inp. 740 * oi points to the exisitng inp. 741 * 742 * This checks whether the existing inp also has BINDMULTI and 743 * whether the credentials match. 744 */ 745 int 746 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 747 { 748 /* Check permissions match */ 749 if ((ni->inp_flags2 & INP_BINDMULTI) && 750 (ni->inp_cred->cr_uid != 751 oi->inp_cred->cr_uid)) 752 return (0); 753 754 /* Check the existing inp has BINDMULTI set */ 755 if ((ni->inp_flags2 & INP_BINDMULTI) && 756 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 757 return (0); 758 759 /* 760 * We're okay - either INP_BINDMULTI isn't set on ni, or 761 * it is and it matches the checks. 762 */ 763 return (1); 764 } 765 766 #ifdef INET 767 /* 768 * Set up a bind operation on a PCB, performing port allocation 769 * as required, but do not actually modify the PCB. Callers can 770 * either complete the bind by setting inp_laddr/inp_lport and 771 * calling in_pcbinshash(), or they can just use the resulting 772 * port and address to authorise the sending of a once-off packet. 773 * 774 * On error, the values of *laddrp and *lportp are not changed. 775 */ 776 int 777 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 778 u_short *lportp, struct ucred *cred) 779 { 780 struct socket *so = inp->inp_socket; 781 struct sockaddr_in *sin; 782 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 783 struct in_addr laddr; 784 u_short lport = 0; 785 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 786 int error; 787 788 /* 789 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 790 * so that we don't have to add to the (already messy) code below. 791 */ 792 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 793 794 /* 795 * No state changes, so read locks are sufficient here. 796 */ 797 INP_LOCK_ASSERT(inp); 798 INP_HASH_LOCK_ASSERT(pcbinfo); 799 800 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 801 return (EADDRNOTAVAIL); 802 laddr.s_addr = *laddrp; 803 if (nam != NULL && laddr.s_addr != INADDR_ANY) 804 return (EINVAL); 805 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 806 lookupflags = INPLOOKUP_WILDCARD; 807 if (nam == NULL) { 808 if ((error = prison_local_ip4(cred, &laddr)) != 0) 809 return (error); 810 } else { 811 sin = (struct sockaddr_in *)nam; 812 if (nam->sa_len != sizeof (*sin)) 813 return (EINVAL); 814 #ifdef notdef 815 /* 816 * We should check the family, but old programs 817 * incorrectly fail to initialize it. 818 */ 819 if (sin->sin_family != AF_INET) 820 return (EAFNOSUPPORT); 821 #endif 822 error = prison_local_ip4(cred, &sin->sin_addr); 823 if (error) 824 return (error); 825 if (sin->sin_port != *lportp) { 826 /* Don't allow the port to change. */ 827 if (*lportp != 0) 828 return (EINVAL); 829 lport = sin->sin_port; 830 } 831 /* NB: lport is left as 0 if the port isn't being changed. */ 832 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 833 /* 834 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 835 * allow complete duplication of binding if 836 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 837 * and a multicast address is bound on both 838 * new and duplicated sockets. 839 */ 840 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 841 reuseport = SO_REUSEADDR|SO_REUSEPORT; 842 /* 843 * XXX: How to deal with SO_REUSEPORT_LB here? 844 * Treat same as SO_REUSEPORT for now. 845 */ 846 if ((so->so_options & 847 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 848 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 849 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 850 sin->sin_port = 0; /* yech... */ 851 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 852 /* 853 * Is the address a local IP address? 854 * If INP_BINDANY is set, then the socket may be bound 855 * to any endpoint address, local or not. 856 */ 857 if ((inp->inp_flags & INP_BINDANY) == 0 && 858 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 859 return (EADDRNOTAVAIL); 860 } 861 laddr = sin->sin_addr; 862 if (lport) { 863 struct inpcb *t; 864 struct tcptw *tw; 865 866 /* GROSS */ 867 if (ntohs(lport) <= V_ipport_reservedhigh && 868 ntohs(lport) >= V_ipport_reservedlow && 869 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT)) 870 return (EACCES); 871 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 872 priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) { 873 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 874 lport, INPLOOKUP_WILDCARD, cred); 875 /* 876 * XXX 877 * This entire block sorely needs a rewrite. 878 */ 879 if (t && 880 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 881 ((t->inp_flags & INP_TIMEWAIT) == 0) && 882 (so->so_type != SOCK_STREAM || 883 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 884 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 885 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 886 (t->inp_flags2 & INP_REUSEPORT) || 887 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 888 (inp->inp_cred->cr_uid != 889 t->inp_cred->cr_uid)) 890 return (EADDRINUSE); 891 892 /* 893 * If the socket is a BINDMULTI socket, then 894 * the credentials need to match and the 895 * original socket also has to have been bound 896 * with BINDMULTI. 897 */ 898 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 899 return (EADDRINUSE); 900 } 901 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 902 lport, lookupflags, cred); 903 if (t && (t->inp_flags & INP_TIMEWAIT)) { 904 /* 905 * XXXRW: If an incpb has had its timewait 906 * state recycled, we treat the address as 907 * being in use (for now). This is better 908 * than a panic, but not desirable. 909 */ 910 tw = intotw(t); 911 if (tw == NULL || 912 ((reuseport & tw->tw_so_options) == 0 && 913 (reuseport_lb & 914 tw->tw_so_options) == 0)) { 915 return (EADDRINUSE); 916 } 917 } else if (t && 918 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 919 (reuseport & inp_so_options(t)) == 0 && 920 (reuseport_lb & inp_so_options(t)) == 0) { 921 #ifdef INET6 922 if (ntohl(sin->sin_addr.s_addr) != 923 INADDR_ANY || 924 ntohl(t->inp_laddr.s_addr) != 925 INADDR_ANY || 926 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 927 (t->inp_vflag & INP_IPV6PROTO) == 0) 928 #endif 929 return (EADDRINUSE); 930 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 931 return (EADDRINUSE); 932 } 933 } 934 } 935 if (*lportp != 0) 936 lport = *lportp; 937 if (lport == 0) { 938 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 939 if (error != 0) 940 return (error); 941 942 } 943 *laddrp = laddr.s_addr; 944 *lportp = lport; 945 return (0); 946 } 947 948 /* 949 * Connect from a socket to a specified address. 950 * Both address and port must be specified in argument sin. 951 * If don't have a local address for this socket yet, 952 * then pick one. 953 */ 954 int 955 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 956 struct ucred *cred, struct mbuf *m) 957 { 958 u_short lport, fport; 959 in_addr_t laddr, faddr; 960 int anonport, error; 961 962 INP_WLOCK_ASSERT(inp); 963 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 964 965 lport = inp->inp_lport; 966 laddr = inp->inp_laddr.s_addr; 967 anonport = (lport == 0); 968 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 969 NULL, cred); 970 if (error) 971 return (error); 972 973 /* Do the initial binding of the local address if required. */ 974 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 975 inp->inp_lport = lport; 976 inp->inp_laddr.s_addr = laddr; 977 if (in_pcbinshash(inp) != 0) { 978 inp->inp_laddr.s_addr = INADDR_ANY; 979 inp->inp_lport = 0; 980 return (EAGAIN); 981 } 982 } 983 984 /* Commit the remaining changes. */ 985 inp->inp_lport = lport; 986 inp->inp_laddr.s_addr = laddr; 987 inp->inp_faddr.s_addr = faddr; 988 inp->inp_fport = fport; 989 in_pcbrehash_mbuf(inp, m); 990 991 if (anonport) 992 inp->inp_flags |= INP_ANONPORT; 993 return (0); 994 } 995 996 int 997 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 998 { 999 1000 return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); 1001 } 1002 1003 /* 1004 * Do proper source address selection on an unbound socket in case 1005 * of connect. Take jails into account as well. 1006 */ 1007 int 1008 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1009 struct ucred *cred) 1010 { 1011 struct ifaddr *ifa; 1012 struct sockaddr *sa; 1013 struct sockaddr_in *sin; 1014 struct route sro; 1015 int error; 1016 1017 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1018 /* 1019 * Bypass source address selection and use the primary jail IP 1020 * if requested. 1021 */ 1022 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1023 return (0); 1024 1025 error = 0; 1026 bzero(&sro, sizeof(sro)); 1027 1028 sin = (struct sockaddr_in *)&sro.ro_dst; 1029 sin->sin_family = AF_INET; 1030 sin->sin_len = sizeof(struct sockaddr_in); 1031 sin->sin_addr.s_addr = faddr->s_addr; 1032 1033 /* 1034 * If route is known our src addr is taken from the i/f, 1035 * else punt. 1036 * 1037 * Find out route to destination. 1038 */ 1039 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1040 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 1041 1042 /* 1043 * If we found a route, use the address corresponding to 1044 * the outgoing interface. 1045 * 1046 * Otherwise assume faddr is reachable on a directly connected 1047 * network and try to find a corresponding interface to take 1048 * the source address from. 1049 */ 1050 NET_EPOCH_ENTER(); 1051 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 1052 struct in_ifaddr *ia; 1053 struct ifnet *ifp; 1054 1055 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1056 inp->inp_socket->so_fibnum)); 1057 if (ia == NULL) { 1058 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1059 inp->inp_socket->so_fibnum)); 1060 1061 } 1062 if (ia == NULL) { 1063 error = ENETUNREACH; 1064 goto done; 1065 } 1066 1067 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1068 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1069 goto done; 1070 } 1071 1072 ifp = ia->ia_ifp; 1073 ia = NULL; 1074 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1075 1076 sa = ifa->ifa_addr; 1077 if (sa->sa_family != AF_INET) 1078 continue; 1079 sin = (struct sockaddr_in *)sa; 1080 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1081 ia = (struct in_ifaddr *)ifa; 1082 break; 1083 } 1084 } 1085 if (ia != NULL) { 1086 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1087 goto done; 1088 } 1089 1090 /* 3. As a last resort return the 'default' jail address. */ 1091 error = prison_get_ip4(cred, laddr); 1092 goto done; 1093 } 1094 1095 /* 1096 * If the outgoing interface on the route found is not 1097 * a loopback interface, use the address from that interface. 1098 * In case of jails do those three steps: 1099 * 1. check if the interface address belongs to the jail. If so use it. 1100 * 2. check if we have any address on the outgoing interface 1101 * belonging to this jail. If so use it. 1102 * 3. as a last resort return the 'default' jail address. 1103 */ 1104 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 1105 struct in_ifaddr *ia; 1106 struct ifnet *ifp; 1107 1108 /* If not jailed, use the default returned. */ 1109 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1110 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1111 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1112 goto done; 1113 } 1114 1115 /* Jailed. */ 1116 /* 1. Check if the iface address belongs to the jail. */ 1117 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 1118 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1119 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1120 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1121 goto done; 1122 } 1123 1124 /* 1125 * 2. Check if we have any address on the outgoing interface 1126 * belonging to this jail. 1127 */ 1128 ia = NULL; 1129 ifp = sro.ro_rt->rt_ifp; 1130 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1131 sa = ifa->ifa_addr; 1132 if (sa->sa_family != AF_INET) 1133 continue; 1134 sin = (struct sockaddr_in *)sa; 1135 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1136 ia = (struct in_ifaddr *)ifa; 1137 break; 1138 } 1139 } 1140 if (ia != NULL) { 1141 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1142 goto done; 1143 } 1144 1145 /* 3. As a last resort return the 'default' jail address. */ 1146 error = prison_get_ip4(cred, laddr); 1147 goto done; 1148 } 1149 1150 /* 1151 * The outgoing interface is marked with 'loopback net', so a route 1152 * to ourselves is here. 1153 * Try to find the interface of the destination address and then 1154 * take the address from there. That interface is not necessarily 1155 * a loopback interface. 1156 * In case of jails, check that it is an address of the jail 1157 * and if we cannot find, fall back to the 'default' jail address. 1158 */ 1159 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 1160 struct sockaddr_in sain; 1161 struct in_ifaddr *ia; 1162 1163 bzero(&sain, sizeof(struct sockaddr_in)); 1164 sain.sin_family = AF_INET; 1165 sain.sin_len = sizeof(struct sockaddr_in); 1166 sain.sin_addr.s_addr = faddr->s_addr; 1167 1168 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), 1169 inp->inp_socket->so_fibnum)); 1170 if (ia == NULL) 1171 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, 1172 inp->inp_socket->so_fibnum)); 1173 if (ia == NULL) 1174 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 1175 1176 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1177 if (ia == NULL) { 1178 error = ENETUNREACH; 1179 goto done; 1180 } 1181 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1182 goto done; 1183 } 1184 1185 /* Jailed. */ 1186 if (ia != NULL) { 1187 struct ifnet *ifp; 1188 1189 ifp = ia->ia_ifp; 1190 ia = NULL; 1191 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1192 sa = ifa->ifa_addr; 1193 if (sa->sa_family != AF_INET) 1194 continue; 1195 sin = (struct sockaddr_in *)sa; 1196 if (prison_check_ip4(cred, 1197 &sin->sin_addr) == 0) { 1198 ia = (struct in_ifaddr *)ifa; 1199 break; 1200 } 1201 } 1202 if (ia != NULL) { 1203 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1204 goto done; 1205 } 1206 } 1207 1208 /* 3. As a last resort return the 'default' jail address. */ 1209 error = prison_get_ip4(cred, laddr); 1210 goto done; 1211 } 1212 1213 done: 1214 NET_EPOCH_EXIT(); 1215 if (sro.ro_rt != NULL) 1216 RTFREE(sro.ro_rt); 1217 return (error); 1218 } 1219 1220 /* 1221 * Set up for a connect from a socket to the specified address. 1222 * On entry, *laddrp and *lportp should contain the current local 1223 * address and port for the PCB; these are updated to the values 1224 * that should be placed in inp_laddr and inp_lport to complete 1225 * the connect. 1226 * 1227 * On success, *faddrp and *fportp will be set to the remote address 1228 * and port. These are not updated in the error case. 1229 * 1230 * If the operation fails because the connection already exists, 1231 * *oinpp will be set to the PCB of that connection so that the 1232 * caller can decide to override it. In all other cases, *oinpp 1233 * is set to NULL. 1234 */ 1235 int 1236 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1237 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1238 struct inpcb **oinpp, struct ucred *cred) 1239 { 1240 struct rm_priotracker in_ifa_tracker; 1241 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1242 struct in_ifaddr *ia; 1243 struct inpcb *oinp; 1244 struct in_addr laddr, faddr; 1245 u_short lport, fport; 1246 int error; 1247 1248 /* 1249 * Because a global state change doesn't actually occur here, a read 1250 * lock is sufficient. 1251 */ 1252 INP_LOCK_ASSERT(inp); 1253 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1254 1255 if (oinpp != NULL) 1256 *oinpp = NULL; 1257 if (nam->sa_len != sizeof (*sin)) 1258 return (EINVAL); 1259 if (sin->sin_family != AF_INET) 1260 return (EAFNOSUPPORT); 1261 if (sin->sin_port == 0) 1262 return (EADDRNOTAVAIL); 1263 laddr.s_addr = *laddrp; 1264 lport = *lportp; 1265 faddr = sin->sin_addr; 1266 fport = sin->sin_port; 1267 1268 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1269 /* 1270 * If the destination address is INADDR_ANY, 1271 * use the primary local address. 1272 * If the supplied address is INADDR_BROADCAST, 1273 * and the primary interface supports broadcast, 1274 * choose the broadcast address for that interface. 1275 */ 1276 if (faddr.s_addr == INADDR_ANY) { 1277 IN_IFADDR_RLOCK(&in_ifa_tracker); 1278 faddr = 1279 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1280 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1281 if (cred != NULL && 1282 (error = prison_get_ip4(cred, &faddr)) != 0) 1283 return (error); 1284 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1285 IN_IFADDR_RLOCK(&in_ifa_tracker); 1286 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1287 IFF_BROADCAST) 1288 faddr = satosin(&CK_STAILQ_FIRST( 1289 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1290 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1291 } 1292 } 1293 if (laddr.s_addr == INADDR_ANY) { 1294 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1295 /* 1296 * If the destination address is multicast and an outgoing 1297 * interface has been set as a multicast option, prefer the 1298 * address of that interface as our source address. 1299 */ 1300 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1301 inp->inp_moptions != NULL) { 1302 struct ip_moptions *imo; 1303 struct ifnet *ifp; 1304 1305 imo = inp->inp_moptions; 1306 if (imo->imo_multicast_ifp != NULL) { 1307 ifp = imo->imo_multicast_ifp; 1308 IN_IFADDR_RLOCK(&in_ifa_tracker); 1309 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1310 if ((ia->ia_ifp == ifp) && 1311 (cred == NULL || 1312 prison_check_ip4(cred, 1313 &ia->ia_addr.sin_addr) == 0)) 1314 break; 1315 } 1316 if (ia == NULL) 1317 error = EADDRNOTAVAIL; 1318 else { 1319 laddr = ia->ia_addr.sin_addr; 1320 error = 0; 1321 } 1322 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1323 } 1324 } 1325 if (error) 1326 return (error); 1327 } 1328 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1329 laddr, lport, 0, NULL); 1330 if (oinp != NULL) { 1331 if (oinpp != NULL) 1332 *oinpp = oinp; 1333 return (EADDRINUSE); 1334 } 1335 if (lport == 0) { 1336 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1337 cred); 1338 if (error) 1339 return (error); 1340 } 1341 *laddrp = laddr.s_addr; 1342 *lportp = lport; 1343 *faddrp = faddr.s_addr; 1344 *fportp = fport; 1345 return (0); 1346 } 1347 1348 void 1349 in_pcbdisconnect(struct inpcb *inp) 1350 { 1351 1352 INP_WLOCK_ASSERT(inp); 1353 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1354 1355 inp->inp_faddr.s_addr = INADDR_ANY; 1356 inp->inp_fport = 0; 1357 in_pcbrehash(inp); 1358 } 1359 #endif /* INET */ 1360 1361 /* 1362 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1363 * For most protocols, this will be invoked immediately prior to calling 1364 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1365 * socket, in which case in_pcbfree() is deferred. 1366 */ 1367 void 1368 in_pcbdetach(struct inpcb *inp) 1369 { 1370 1371 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1372 1373 #ifdef RATELIMIT 1374 if (inp->inp_snd_tag != NULL) 1375 in_pcbdetach_txrtlmt(inp); 1376 #endif 1377 inp->inp_socket->so_pcb = NULL; 1378 inp->inp_socket = NULL; 1379 } 1380 1381 /* 1382 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1383 * stability of an inpcb pointer despite the inpcb lock being released. This 1384 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1385 * but where the inpcb lock may already held, or when acquiring a reference 1386 * via a pcbgroup. 1387 * 1388 * in_pcbref() should be used only to provide brief memory stability, and 1389 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1390 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1391 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1392 * lock and rele are the *only* safe operations that may be performed on the 1393 * inpcb. 1394 * 1395 * While the inpcb will not be freed, releasing the inpcb lock means that the 1396 * connection's state may change, so the caller should be careful to 1397 * revalidate any cached state on reacquiring the lock. Drop the reference 1398 * using in_pcbrele(). 1399 */ 1400 void 1401 in_pcbref(struct inpcb *inp) 1402 { 1403 1404 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1405 1406 refcount_acquire(&inp->inp_refcount); 1407 } 1408 1409 /* 1410 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1411 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1412 * return a flag indicating whether or not the inpcb remains valid. If it is 1413 * valid, we return with the inpcb lock held. 1414 * 1415 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1416 * reference on an inpcb. Historically more work was done here (actually, in 1417 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1418 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1419 * about memory stability (and continued use of the write lock). 1420 */ 1421 int 1422 in_pcbrele_rlocked(struct inpcb *inp) 1423 { 1424 struct inpcbinfo *pcbinfo; 1425 1426 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1427 1428 INP_RLOCK_ASSERT(inp); 1429 1430 if (refcount_release(&inp->inp_refcount) == 0) { 1431 /* 1432 * If the inpcb has been freed, let the caller know, even if 1433 * this isn't the last reference. 1434 */ 1435 if (inp->inp_flags2 & INP_FREED) { 1436 INP_RUNLOCK(inp); 1437 return (1); 1438 } 1439 return (0); 1440 } 1441 1442 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1443 #ifdef TCPHPTS 1444 if (inp->inp_in_hpts || inp->inp_in_input) { 1445 struct tcp_hpts_entry *hpts; 1446 /* 1447 * We should not be on the hpts at 1448 * this point in any form. we must 1449 * get the lock to be sure. 1450 */ 1451 hpts = tcp_hpts_lock(inp); 1452 if (inp->inp_in_hpts) 1453 panic("Hpts:%p inp:%p at free still on hpts", 1454 hpts, inp); 1455 mtx_unlock(&hpts->p_mtx); 1456 hpts = tcp_input_lock(inp); 1457 if (inp->inp_in_input) 1458 panic("Hpts:%p inp:%p at free still on input hpts", 1459 hpts, inp); 1460 mtx_unlock(&hpts->p_mtx); 1461 } 1462 #endif 1463 INP_RUNLOCK(inp); 1464 pcbinfo = inp->inp_pcbinfo; 1465 uma_zfree(pcbinfo->ipi_zone, inp); 1466 return (1); 1467 } 1468 1469 int 1470 in_pcbrele_wlocked(struct inpcb *inp) 1471 { 1472 struct inpcbinfo *pcbinfo; 1473 1474 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1475 1476 INP_WLOCK_ASSERT(inp); 1477 1478 if (refcount_release(&inp->inp_refcount) == 0) { 1479 /* 1480 * If the inpcb has been freed, let the caller know, even if 1481 * this isn't the last reference. 1482 */ 1483 if (inp->inp_flags2 & INP_FREED) { 1484 INP_WUNLOCK(inp); 1485 return (1); 1486 } 1487 return (0); 1488 } 1489 1490 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1491 #ifdef TCPHPTS 1492 if (inp->inp_in_hpts || inp->inp_in_input) { 1493 struct tcp_hpts_entry *hpts; 1494 /* 1495 * We should not be on the hpts at 1496 * this point in any form. we must 1497 * get the lock to be sure. 1498 */ 1499 hpts = tcp_hpts_lock(inp); 1500 if (inp->inp_in_hpts) 1501 panic("Hpts:%p inp:%p at free still on hpts", 1502 hpts, inp); 1503 mtx_unlock(&hpts->p_mtx); 1504 hpts = tcp_input_lock(inp); 1505 if (inp->inp_in_input) 1506 panic("Hpts:%p inp:%p at free still on input hpts", 1507 hpts, inp); 1508 mtx_unlock(&hpts->p_mtx); 1509 } 1510 #endif 1511 INP_WUNLOCK(inp); 1512 pcbinfo = inp->inp_pcbinfo; 1513 uma_zfree(pcbinfo->ipi_zone, inp); 1514 return (1); 1515 } 1516 1517 /* 1518 * Temporary wrapper. 1519 */ 1520 int 1521 in_pcbrele(struct inpcb *inp) 1522 { 1523 1524 return (in_pcbrele_wlocked(inp)); 1525 } 1526 1527 void 1528 in_pcblist_rele_rlocked(epoch_context_t ctx) 1529 { 1530 struct in_pcblist *il; 1531 struct inpcb *inp; 1532 struct inpcbinfo *pcbinfo; 1533 int i, n; 1534 1535 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1536 pcbinfo = il->il_pcbinfo; 1537 n = il->il_count; 1538 INP_INFO_WLOCK(pcbinfo); 1539 for (i = 0; i < n; i++) { 1540 inp = il->il_inp_list[i]; 1541 INP_RLOCK(inp); 1542 if (!in_pcbrele_rlocked(inp)) 1543 INP_RUNLOCK(inp); 1544 } 1545 INP_INFO_WUNLOCK(pcbinfo); 1546 free(il, M_TEMP); 1547 } 1548 1549 static void 1550 inpcbport_free(epoch_context_t ctx) 1551 { 1552 struct inpcbport *phd; 1553 1554 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); 1555 free(phd, M_PCB); 1556 } 1557 1558 static void 1559 in_pcbfree_deferred(epoch_context_t ctx) 1560 { 1561 struct inpcb *inp; 1562 int released __unused; 1563 1564 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); 1565 1566 INP_WLOCK(inp); 1567 #ifdef INET 1568 struct ip_moptions *imo = inp->inp_moptions; 1569 inp->inp_moptions = NULL; 1570 #endif 1571 /* XXXRW: Do as much as possible here. */ 1572 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1573 if (inp->inp_sp != NULL) 1574 ipsec_delete_pcbpolicy(inp); 1575 #endif 1576 #ifdef INET6 1577 struct ip6_moptions *im6o = NULL; 1578 if (inp->inp_vflag & INP_IPV6PROTO) { 1579 ip6_freepcbopts(inp->in6p_outputopts); 1580 im6o = inp->in6p_moptions; 1581 inp->in6p_moptions = NULL; 1582 } 1583 #endif 1584 if (inp->inp_options) 1585 (void)m_free(inp->inp_options); 1586 inp->inp_vflag = 0; 1587 crfree(inp->inp_cred); 1588 #ifdef MAC 1589 mac_inpcb_destroy(inp); 1590 #endif 1591 released = in_pcbrele_wlocked(inp); 1592 MPASS(released); 1593 #ifdef INET6 1594 ip6_freemoptions(im6o); 1595 #endif 1596 #ifdef INET 1597 inp_freemoptions(imo); 1598 #endif 1599 } 1600 1601 /* 1602 * Unconditionally schedule an inpcb to be freed by decrementing its 1603 * reference count, which should occur only after the inpcb has been detached 1604 * from its socket. If another thread holds a temporary reference (acquired 1605 * using in_pcbref()) then the free is deferred until that reference is 1606 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1607 * work, including removal from global lists, is done in this context, where 1608 * the pcbinfo lock is held. 1609 */ 1610 void 1611 in_pcbfree(struct inpcb *inp) 1612 { 1613 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1614 1615 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1616 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1617 ("%s: called twice for pcb %p", __func__, inp)); 1618 if (inp->inp_flags2 & INP_FREED) { 1619 INP_WUNLOCK(inp); 1620 return; 1621 } 1622 1623 #ifdef INVARIANTS 1624 if (pcbinfo == &V_tcbinfo) { 1625 INP_INFO_LOCK_ASSERT(pcbinfo); 1626 } else { 1627 INP_INFO_WLOCK_ASSERT(pcbinfo); 1628 } 1629 #endif 1630 INP_WLOCK_ASSERT(inp); 1631 INP_LIST_WLOCK(pcbinfo); 1632 in_pcbremlists(inp); 1633 INP_LIST_WUNLOCK(pcbinfo); 1634 RO_INVALIDATE_CACHE(&inp->inp_route); 1635 /* mark as destruction in progress */ 1636 inp->inp_flags2 |= INP_FREED; 1637 INP_WUNLOCK(inp); 1638 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred); 1639 } 1640 1641 /* 1642 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1643 * port reservation, and preventing it from being returned by inpcb lookups. 1644 * 1645 * It is used by TCP to mark an inpcb as unused and avoid future packet 1646 * delivery or event notification when a socket remains open but TCP has 1647 * closed. This might occur as a result of a shutdown()-initiated TCP close 1648 * or a RST on the wire, and allows the port binding to be reused while still 1649 * maintaining the invariant that so_pcb always points to a valid inpcb until 1650 * in_pcbdetach(). 1651 * 1652 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1653 * in_pcbnotifyall() and in_pcbpurgeif0()? 1654 */ 1655 void 1656 in_pcbdrop(struct inpcb *inp) 1657 { 1658 1659 INP_WLOCK_ASSERT(inp); 1660 #ifdef INVARIANTS 1661 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1662 MPASS(inp->inp_refcount > 1); 1663 #endif 1664 1665 /* 1666 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1667 * the hash lock...? 1668 */ 1669 inp->inp_flags |= INP_DROPPED; 1670 if (inp->inp_flags & INP_INHASHLIST) { 1671 struct inpcbport *phd = inp->inp_phd; 1672 1673 INP_HASH_WLOCK(inp->inp_pcbinfo); 1674 in_pcbremlbgrouphash(inp); 1675 CK_LIST_REMOVE(inp, inp_hash); 1676 CK_LIST_REMOVE(inp, inp_portlist); 1677 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 1678 CK_LIST_REMOVE(phd, phd_hash); 1679 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 1680 } 1681 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1682 inp->inp_flags &= ~INP_INHASHLIST; 1683 #ifdef PCBGROUP 1684 in_pcbgroup_remove(inp); 1685 #endif 1686 } 1687 } 1688 1689 #ifdef INET 1690 /* 1691 * Common routines to return the socket addresses associated with inpcbs. 1692 */ 1693 struct sockaddr * 1694 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1695 { 1696 struct sockaddr_in *sin; 1697 1698 sin = malloc(sizeof *sin, M_SONAME, 1699 M_WAITOK | M_ZERO); 1700 sin->sin_family = AF_INET; 1701 sin->sin_len = sizeof(*sin); 1702 sin->sin_addr = *addr_p; 1703 sin->sin_port = port; 1704 1705 return (struct sockaddr *)sin; 1706 } 1707 1708 int 1709 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1710 { 1711 struct inpcb *inp; 1712 struct in_addr addr; 1713 in_port_t port; 1714 1715 inp = sotoinpcb(so); 1716 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1717 1718 INP_RLOCK(inp); 1719 port = inp->inp_lport; 1720 addr = inp->inp_laddr; 1721 INP_RUNLOCK(inp); 1722 1723 *nam = in_sockaddr(port, &addr); 1724 return 0; 1725 } 1726 1727 int 1728 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1729 { 1730 struct inpcb *inp; 1731 struct in_addr addr; 1732 in_port_t port; 1733 1734 inp = sotoinpcb(so); 1735 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1736 1737 INP_RLOCK(inp); 1738 port = inp->inp_fport; 1739 addr = inp->inp_faddr; 1740 INP_RUNLOCK(inp); 1741 1742 *nam = in_sockaddr(port, &addr); 1743 return 0; 1744 } 1745 1746 void 1747 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1748 struct inpcb *(*notify)(struct inpcb *, int)) 1749 { 1750 struct inpcb *inp, *inp_temp; 1751 1752 INP_INFO_WLOCK(pcbinfo); 1753 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1754 INP_WLOCK(inp); 1755 #ifdef INET6 1756 if ((inp->inp_vflag & INP_IPV4) == 0) { 1757 INP_WUNLOCK(inp); 1758 continue; 1759 } 1760 #endif 1761 if (inp->inp_faddr.s_addr != faddr.s_addr || 1762 inp->inp_socket == NULL) { 1763 INP_WUNLOCK(inp); 1764 continue; 1765 } 1766 if ((*notify)(inp, errno)) 1767 INP_WUNLOCK(inp); 1768 } 1769 INP_INFO_WUNLOCK(pcbinfo); 1770 } 1771 1772 void 1773 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1774 { 1775 struct inpcb *inp; 1776 struct ip_moptions *imo; 1777 int i, gap; 1778 1779 INP_INFO_WLOCK(pcbinfo); 1780 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1781 INP_WLOCK(inp); 1782 imo = inp->inp_moptions; 1783 if ((inp->inp_vflag & INP_IPV4) && 1784 imo != NULL) { 1785 /* 1786 * Unselect the outgoing interface if it is being 1787 * detached. 1788 */ 1789 if (imo->imo_multicast_ifp == ifp) 1790 imo->imo_multicast_ifp = NULL; 1791 1792 /* 1793 * Drop multicast group membership if we joined 1794 * through the interface being detached. 1795 * 1796 * XXX This can all be deferred to an epoch_call 1797 */ 1798 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1799 i++) { 1800 if (imo->imo_membership[i]->inm_ifp == ifp) { 1801 IN_MULTI_LOCK_ASSERT(); 1802 in_leavegroup_locked(imo->imo_membership[i], NULL); 1803 gap++; 1804 } else if (gap != 0) 1805 imo->imo_membership[i - gap] = 1806 imo->imo_membership[i]; 1807 } 1808 imo->imo_num_memberships -= gap; 1809 } 1810 INP_WUNLOCK(inp); 1811 } 1812 INP_INFO_WUNLOCK(pcbinfo); 1813 } 1814 1815 /* 1816 * Lookup a PCB based on the local address and port. Caller must hold the 1817 * hash lock. No inpcb locks or references are acquired. 1818 */ 1819 #define INP_LOOKUP_MAPPED_PCB_COST 3 1820 struct inpcb * 1821 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1822 u_short lport, int lookupflags, struct ucred *cred) 1823 { 1824 struct inpcb *inp; 1825 #ifdef INET6 1826 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1827 #else 1828 int matchwild = 3; 1829 #endif 1830 int wildcard; 1831 1832 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1833 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1834 1835 INP_HASH_LOCK_ASSERT(pcbinfo); 1836 1837 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1838 struct inpcbhead *head; 1839 /* 1840 * Look for an unconnected (wildcard foreign addr) PCB that 1841 * matches the local address and port we're looking for. 1842 */ 1843 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1844 0, pcbinfo->ipi_hashmask)]; 1845 CK_LIST_FOREACH(inp, head, inp_hash) { 1846 #ifdef INET6 1847 /* XXX inp locking */ 1848 if ((inp->inp_vflag & INP_IPV4) == 0) 1849 continue; 1850 #endif 1851 if (inp->inp_faddr.s_addr == INADDR_ANY && 1852 inp->inp_laddr.s_addr == laddr.s_addr && 1853 inp->inp_lport == lport) { 1854 /* 1855 * Found? 1856 */ 1857 if (cred == NULL || 1858 prison_equal_ip4(cred->cr_prison, 1859 inp->inp_cred->cr_prison)) 1860 return (inp); 1861 } 1862 } 1863 /* 1864 * Not found. 1865 */ 1866 return (NULL); 1867 } else { 1868 struct inpcbporthead *porthash; 1869 struct inpcbport *phd; 1870 struct inpcb *match = NULL; 1871 /* 1872 * Best fit PCB lookup. 1873 * 1874 * First see if this local port is in use by looking on the 1875 * port hash list. 1876 */ 1877 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1878 pcbinfo->ipi_porthashmask)]; 1879 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1880 if (phd->phd_port == lport) 1881 break; 1882 } 1883 if (phd != NULL) { 1884 /* 1885 * Port is in use by one or more PCBs. Look for best 1886 * fit. 1887 */ 1888 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1889 wildcard = 0; 1890 if (cred != NULL && 1891 !prison_equal_ip4(inp->inp_cred->cr_prison, 1892 cred->cr_prison)) 1893 continue; 1894 #ifdef INET6 1895 /* XXX inp locking */ 1896 if ((inp->inp_vflag & INP_IPV4) == 0) 1897 continue; 1898 /* 1899 * We never select the PCB that has 1900 * INP_IPV6 flag and is bound to :: if 1901 * we have another PCB which is bound 1902 * to 0.0.0.0. If a PCB has the 1903 * INP_IPV6 flag, then we set its cost 1904 * higher than IPv4 only PCBs. 1905 * 1906 * Note that the case only happens 1907 * when a socket is bound to ::, under 1908 * the condition that the use of the 1909 * mapped address is allowed. 1910 */ 1911 if ((inp->inp_vflag & INP_IPV6) != 0) 1912 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1913 #endif 1914 if (inp->inp_faddr.s_addr != INADDR_ANY) 1915 wildcard++; 1916 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1917 if (laddr.s_addr == INADDR_ANY) 1918 wildcard++; 1919 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1920 continue; 1921 } else { 1922 if (laddr.s_addr != INADDR_ANY) 1923 wildcard++; 1924 } 1925 if (wildcard < matchwild) { 1926 match = inp; 1927 matchwild = wildcard; 1928 if (matchwild == 0) 1929 break; 1930 } 1931 } 1932 } 1933 return (match); 1934 } 1935 } 1936 #undef INP_LOOKUP_MAPPED_PCB_COST 1937 1938 static struct inpcb * 1939 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 1940 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, 1941 uint16_t fport, int lookupflags) 1942 { 1943 struct inpcb *local_wild; 1944 const struct inpcblbgrouphead *hdr; 1945 struct inpcblbgroup *grp; 1946 uint32_t idx; 1947 1948 INP_HASH_LOCK_ASSERT(pcbinfo); 1949 1950 hdr = &pcbinfo->ipi_lbgrouphashbase[ 1951 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 1952 1953 /* 1954 * Order of socket selection: 1955 * 1. non-wild. 1956 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). 1957 * 1958 * NOTE: 1959 * - Load balanced group does not contain jailed sockets 1960 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets 1961 */ 1962 local_wild = NULL; 1963 CK_LIST_FOREACH(grp, hdr, il_list) { 1964 #ifdef INET6 1965 if (!(grp->il_vflag & INP_IPV4)) 1966 continue; 1967 #endif 1968 if (grp->il_lport != lport) 1969 continue; 1970 1971 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % 1972 grp->il_inpcnt; 1973 if (grp->il_laddr.s_addr == laddr->s_addr) 1974 return (grp->il_inp[idx]); 1975 if (grp->il_laddr.s_addr == INADDR_ANY && 1976 (lookupflags & INPLOOKUP_WILDCARD) != 0) 1977 local_wild = grp->il_inp[idx]; 1978 } 1979 return (local_wild); 1980 } 1981 1982 #ifdef PCBGROUP 1983 /* 1984 * Lookup PCB in hash list, using pcbgroup tables. 1985 */ 1986 static struct inpcb * 1987 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 1988 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 1989 u_int lport_arg, int lookupflags, struct ifnet *ifp) 1990 { 1991 struct inpcbhead *head; 1992 struct inpcb *inp, *tmpinp; 1993 u_short fport = fport_arg, lport = lport_arg; 1994 bool locked; 1995 1996 /* 1997 * First look for an exact match. 1998 */ 1999 tmpinp = NULL; 2000 INP_GROUP_LOCK(pcbgroup); 2001 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2002 pcbgroup->ipg_hashmask)]; 2003 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2004 #ifdef INET6 2005 /* XXX inp locking */ 2006 if ((inp->inp_vflag & INP_IPV4) == 0) 2007 continue; 2008 #endif 2009 if (inp->inp_faddr.s_addr == faddr.s_addr && 2010 inp->inp_laddr.s_addr == laddr.s_addr && 2011 inp->inp_fport == fport && 2012 inp->inp_lport == lport) { 2013 /* 2014 * XXX We should be able to directly return 2015 * the inp here, without any checks. 2016 * Well unless both bound with SO_REUSEPORT? 2017 */ 2018 if (prison_flag(inp->inp_cred, PR_IP4)) 2019 goto found; 2020 if (tmpinp == NULL) 2021 tmpinp = inp; 2022 } 2023 } 2024 if (tmpinp != NULL) { 2025 inp = tmpinp; 2026 goto found; 2027 } 2028 2029 #ifdef RSS 2030 /* 2031 * For incoming connections, we may wish to do a wildcard 2032 * match for an RSS-local socket. 2033 */ 2034 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2035 struct inpcb *local_wild = NULL, *local_exact = NULL; 2036 #ifdef INET6 2037 struct inpcb *local_wild_mapped = NULL; 2038 #endif 2039 struct inpcb *jail_wild = NULL; 2040 struct inpcbhead *head; 2041 int injail; 2042 2043 /* 2044 * Order of socket selection - we always prefer jails. 2045 * 1. jailed, non-wild. 2046 * 2. jailed, wild. 2047 * 3. non-jailed, non-wild. 2048 * 4. non-jailed, wild. 2049 */ 2050 2051 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 2052 lport, 0, pcbgroup->ipg_hashmask)]; 2053 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2054 #ifdef INET6 2055 /* XXX inp locking */ 2056 if ((inp->inp_vflag & INP_IPV4) == 0) 2057 continue; 2058 #endif 2059 if (inp->inp_faddr.s_addr != INADDR_ANY || 2060 inp->inp_lport != lport) 2061 continue; 2062 2063 injail = prison_flag(inp->inp_cred, PR_IP4); 2064 if (injail) { 2065 if (prison_check_ip4(inp->inp_cred, 2066 &laddr) != 0) 2067 continue; 2068 } else { 2069 if (local_exact != NULL) 2070 continue; 2071 } 2072 2073 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2074 if (injail) 2075 goto found; 2076 else 2077 local_exact = inp; 2078 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2079 #ifdef INET6 2080 /* XXX inp locking, NULL check */ 2081 if (inp->inp_vflag & INP_IPV6PROTO) 2082 local_wild_mapped = inp; 2083 else 2084 #endif 2085 if (injail) 2086 jail_wild = inp; 2087 else 2088 local_wild = inp; 2089 } 2090 } /* LIST_FOREACH */ 2091 2092 inp = jail_wild; 2093 if (inp == NULL) 2094 inp = local_exact; 2095 if (inp == NULL) 2096 inp = local_wild; 2097 #ifdef INET6 2098 if (inp == NULL) 2099 inp = local_wild_mapped; 2100 #endif 2101 if (inp != NULL) 2102 goto found; 2103 } 2104 #endif 2105 2106 /* 2107 * Then look for a wildcard match, if requested. 2108 */ 2109 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2110 struct inpcb *local_wild = NULL, *local_exact = NULL; 2111 #ifdef INET6 2112 struct inpcb *local_wild_mapped = NULL; 2113 #endif 2114 struct inpcb *jail_wild = NULL; 2115 struct inpcbhead *head; 2116 int injail; 2117 2118 /* 2119 * Order of socket selection - we always prefer jails. 2120 * 1. jailed, non-wild. 2121 * 2. jailed, wild. 2122 * 3. non-jailed, non-wild. 2123 * 4. non-jailed, wild. 2124 */ 2125 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 2126 0, pcbinfo->ipi_wildmask)]; 2127 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 2128 #ifdef INET6 2129 /* XXX inp locking */ 2130 if ((inp->inp_vflag & INP_IPV4) == 0) 2131 continue; 2132 #endif 2133 if (inp->inp_faddr.s_addr != INADDR_ANY || 2134 inp->inp_lport != lport) 2135 continue; 2136 2137 injail = prison_flag(inp->inp_cred, PR_IP4); 2138 if (injail) { 2139 if (prison_check_ip4(inp->inp_cred, 2140 &laddr) != 0) 2141 continue; 2142 } else { 2143 if (local_exact != NULL) 2144 continue; 2145 } 2146 2147 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2148 if (injail) 2149 goto found; 2150 else 2151 local_exact = inp; 2152 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2153 #ifdef INET6 2154 /* XXX inp locking, NULL check */ 2155 if (inp->inp_vflag & INP_IPV6PROTO) 2156 local_wild_mapped = inp; 2157 else 2158 #endif 2159 if (injail) 2160 jail_wild = inp; 2161 else 2162 local_wild = inp; 2163 } 2164 } /* LIST_FOREACH */ 2165 inp = jail_wild; 2166 if (inp == NULL) 2167 inp = local_exact; 2168 if (inp == NULL) 2169 inp = local_wild; 2170 #ifdef INET6 2171 if (inp == NULL) 2172 inp = local_wild_mapped; 2173 #endif 2174 if (inp != NULL) 2175 goto found; 2176 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 2177 INP_GROUP_UNLOCK(pcbgroup); 2178 return (NULL); 2179 2180 found: 2181 if (lookupflags & INPLOOKUP_WLOCKPCB) 2182 locked = INP_TRY_WLOCK(inp); 2183 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2184 locked = INP_TRY_RLOCK(inp); 2185 else 2186 panic("%s: locking bug", __func__); 2187 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { 2188 if (lookupflags & INPLOOKUP_WLOCKPCB) 2189 INP_WUNLOCK(inp); 2190 else 2191 INP_RUNLOCK(inp); 2192 return (NULL); 2193 } else if (!locked) 2194 in_pcbref(inp); 2195 INP_GROUP_UNLOCK(pcbgroup); 2196 if (!locked) { 2197 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2198 INP_WLOCK(inp); 2199 if (in_pcbrele_wlocked(inp)) 2200 return (NULL); 2201 } else { 2202 INP_RLOCK(inp); 2203 if (in_pcbrele_rlocked(inp)) 2204 return (NULL); 2205 } 2206 } 2207 #ifdef INVARIANTS 2208 if (lookupflags & INPLOOKUP_WLOCKPCB) 2209 INP_WLOCK_ASSERT(inp); 2210 else 2211 INP_RLOCK_ASSERT(inp); 2212 #endif 2213 return (inp); 2214 } 2215 #endif /* PCBGROUP */ 2216 2217 /* 2218 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2219 * that the caller has locked the hash list, and will not perform any further 2220 * locking or reference operations on either the hash list or the connection. 2221 */ 2222 static struct inpcb * 2223 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2224 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2225 struct ifnet *ifp) 2226 { 2227 struct inpcbhead *head; 2228 struct inpcb *inp, *tmpinp; 2229 u_short fport = fport_arg, lport = lport_arg; 2230 2231 #ifdef INVARIANTS 2232 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2233 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2234 if (!mtx_owned(&pcbinfo->ipi_hash_lock)) 2235 MPASS(in_epoch_verbose(net_epoch_preempt, 1)); 2236 #endif 2237 /* 2238 * First look for an exact match. 2239 */ 2240 tmpinp = NULL; 2241 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2242 pcbinfo->ipi_hashmask)]; 2243 CK_LIST_FOREACH(inp, head, inp_hash) { 2244 #ifdef INET6 2245 /* XXX inp locking */ 2246 if ((inp->inp_vflag & INP_IPV4) == 0) 2247 continue; 2248 #endif 2249 if (inp->inp_faddr.s_addr == faddr.s_addr && 2250 inp->inp_laddr.s_addr == laddr.s_addr && 2251 inp->inp_fport == fport && 2252 inp->inp_lport == lport) { 2253 /* 2254 * XXX We should be able to directly return 2255 * the inp here, without any checks. 2256 * Well unless both bound with SO_REUSEPORT? 2257 */ 2258 if (prison_flag(inp->inp_cred, PR_IP4)) 2259 return (inp); 2260 if (tmpinp == NULL) 2261 tmpinp = inp; 2262 } 2263 } 2264 if (tmpinp != NULL) 2265 return (tmpinp); 2266 2267 /* 2268 * Then look in lb group (for wildcard match). 2269 */ 2270 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2271 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, 2272 fport, lookupflags); 2273 if (inp != NULL) 2274 return (inp); 2275 } 2276 2277 /* 2278 * Then look for a wildcard match, if requested. 2279 */ 2280 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2281 struct inpcb *local_wild = NULL, *local_exact = NULL; 2282 #ifdef INET6 2283 struct inpcb *local_wild_mapped = NULL; 2284 #endif 2285 struct inpcb *jail_wild = NULL; 2286 int injail; 2287 2288 /* 2289 * Order of socket selection - we always prefer jails. 2290 * 1. jailed, non-wild. 2291 * 2. jailed, wild. 2292 * 3. non-jailed, non-wild. 2293 * 4. non-jailed, wild. 2294 */ 2295 2296 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2297 0, pcbinfo->ipi_hashmask)]; 2298 CK_LIST_FOREACH(inp, head, inp_hash) { 2299 #ifdef INET6 2300 /* XXX inp locking */ 2301 if ((inp->inp_vflag & INP_IPV4) == 0) 2302 continue; 2303 #endif 2304 if (inp->inp_faddr.s_addr != INADDR_ANY || 2305 inp->inp_lport != lport) 2306 continue; 2307 2308 injail = prison_flag(inp->inp_cred, PR_IP4); 2309 if (injail) { 2310 if (prison_check_ip4(inp->inp_cred, 2311 &laddr) != 0) 2312 continue; 2313 } else { 2314 if (local_exact != NULL) 2315 continue; 2316 } 2317 2318 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2319 if (injail) 2320 return (inp); 2321 else 2322 local_exact = inp; 2323 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2324 #ifdef INET6 2325 /* XXX inp locking, NULL check */ 2326 if (inp->inp_vflag & INP_IPV6PROTO) 2327 local_wild_mapped = inp; 2328 else 2329 #endif 2330 if (injail) 2331 jail_wild = inp; 2332 else 2333 local_wild = inp; 2334 } 2335 } /* LIST_FOREACH */ 2336 if (jail_wild != NULL) 2337 return (jail_wild); 2338 if (local_exact != NULL) 2339 return (local_exact); 2340 if (local_wild != NULL) 2341 return (local_wild); 2342 #ifdef INET6 2343 if (local_wild_mapped != NULL) 2344 return (local_wild_mapped); 2345 #endif 2346 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2347 2348 return (NULL); 2349 } 2350 2351 /* 2352 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2353 * hash list lock, and will return the inpcb locked (i.e., requires 2354 * INPLOOKUP_LOCKPCB). 2355 */ 2356 static struct inpcb * 2357 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2358 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2359 struct ifnet *ifp) 2360 { 2361 struct inpcb *inp; 2362 2363 INP_HASH_RLOCK(pcbinfo); 2364 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2365 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 2366 if (inp != NULL) { 2367 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2368 INP_WLOCK(inp); 2369 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2370 INP_WUNLOCK(inp); 2371 inp = NULL; 2372 } 2373 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2374 INP_RLOCK(inp); 2375 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2376 INP_RUNLOCK(inp); 2377 inp = NULL; 2378 } 2379 } else 2380 panic("%s: locking bug", __func__); 2381 #ifdef INVARIANTS 2382 if (inp != NULL) { 2383 if (lookupflags & INPLOOKUP_WLOCKPCB) 2384 INP_WLOCK_ASSERT(inp); 2385 else 2386 INP_RLOCK_ASSERT(inp); 2387 } 2388 #endif 2389 } 2390 INP_HASH_RUNLOCK(pcbinfo); 2391 return (inp); 2392 } 2393 2394 /* 2395 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2396 * from which a pre-calculated hash value may be extracted. 2397 * 2398 * Possibly more of this logic should be in in_pcbgroup.c. 2399 */ 2400 struct inpcb * 2401 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2402 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2403 { 2404 #if defined(PCBGROUP) && !defined(RSS) 2405 struct inpcbgroup *pcbgroup; 2406 #endif 2407 2408 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2409 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2410 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2411 ("%s: LOCKPCB not set", __func__)); 2412 2413 /* 2414 * When not using RSS, use connection groups in preference to the 2415 * reservation table when looking up 4-tuples. When using RSS, just 2416 * use the reservation table, due to the cost of the Toeplitz hash 2417 * in software. 2418 * 2419 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2420 * we could be doing RSS with a non-Toeplitz hash that is affordable 2421 * in software. 2422 */ 2423 #if defined(PCBGROUP) && !defined(RSS) 2424 if (in_pcbgroup_enabled(pcbinfo)) { 2425 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2426 fport); 2427 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2428 laddr, lport, lookupflags, ifp)); 2429 } 2430 #endif 2431 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2432 lookupflags, ifp)); 2433 } 2434 2435 struct inpcb * 2436 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2437 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2438 struct ifnet *ifp, struct mbuf *m) 2439 { 2440 #ifdef PCBGROUP 2441 struct inpcbgroup *pcbgroup; 2442 #endif 2443 2444 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2445 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2446 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2447 ("%s: LOCKPCB not set", __func__)); 2448 2449 #ifdef PCBGROUP 2450 /* 2451 * If we can use a hardware-generated hash to look up the connection 2452 * group, use that connection group to find the inpcb. Otherwise 2453 * fall back on a software hash -- or the reservation table if we're 2454 * using RSS. 2455 * 2456 * XXXRW: As above, that policy belongs in the pcbgroup code. 2457 */ 2458 if (in_pcbgroup_enabled(pcbinfo) && 2459 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2460 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2461 m->m_pkthdr.flowid); 2462 if (pcbgroup != NULL) 2463 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2464 fport, laddr, lport, lookupflags, ifp)); 2465 #ifndef RSS 2466 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2467 fport); 2468 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2469 laddr, lport, lookupflags, ifp)); 2470 #endif 2471 } 2472 #endif 2473 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2474 lookupflags, ifp)); 2475 } 2476 #endif /* INET */ 2477 2478 /* 2479 * Insert PCB onto various hash lists. 2480 */ 2481 static int 2482 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 2483 { 2484 struct inpcbhead *pcbhash; 2485 struct inpcbporthead *pcbporthash; 2486 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2487 struct inpcbport *phd; 2488 u_int32_t hashkey_faddr; 2489 int so_options; 2490 2491 INP_WLOCK_ASSERT(inp); 2492 INP_HASH_WLOCK_ASSERT(pcbinfo); 2493 2494 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2495 ("in_pcbinshash: INP_INHASHLIST")); 2496 2497 #ifdef INET6 2498 if (inp->inp_vflag & INP_IPV6) 2499 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2500 else 2501 #endif 2502 hashkey_faddr = inp->inp_faddr.s_addr; 2503 2504 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2505 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2506 2507 pcbporthash = &pcbinfo->ipi_porthashbase[ 2508 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2509 2510 /* 2511 * Add entry to load balance group. 2512 * Only do this if SO_REUSEPORT_LB is set. 2513 */ 2514 so_options = inp_so_options(inp); 2515 if (so_options & SO_REUSEPORT_LB) { 2516 int ret = in_pcbinslbgrouphash(inp); 2517 if (ret) { 2518 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2519 return (ret); 2520 } 2521 } 2522 2523 /* 2524 * Go through port list and look for a head for this lport. 2525 */ 2526 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2527 if (phd->phd_port == inp->inp_lport) 2528 break; 2529 } 2530 /* 2531 * If none exists, malloc one and tack it on. 2532 */ 2533 if (phd == NULL) { 2534 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2535 if (phd == NULL) { 2536 return (ENOBUFS); /* XXX */ 2537 } 2538 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2539 phd->phd_port = inp->inp_lport; 2540 CK_LIST_INIT(&phd->phd_pcblist); 2541 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2542 } 2543 inp->inp_phd = phd; 2544 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2545 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2546 inp->inp_flags |= INP_INHASHLIST; 2547 #ifdef PCBGROUP 2548 if (do_pcbgroup_update) 2549 in_pcbgroup_update(inp); 2550 #endif 2551 return (0); 2552 } 2553 2554 /* 2555 * For now, there are two public interfaces to insert an inpcb into the hash 2556 * lists -- one that does update pcbgroups, and one that doesn't. The latter 2557 * is used only in the TCP syncache, where in_pcbinshash is called before the 2558 * full 4-tuple is set for the inpcb, and we don't want to install in the 2559 * pcbgroup until later. 2560 * 2561 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 2562 * connection groups, and partially initialised inpcbs should not be exposed 2563 * to either reservation hash tables or pcbgroups. 2564 */ 2565 int 2566 in_pcbinshash(struct inpcb *inp) 2567 { 2568 2569 return (in_pcbinshash_internal(inp, 1)); 2570 } 2571 2572 int 2573 in_pcbinshash_nopcbgroup(struct inpcb *inp) 2574 { 2575 2576 return (in_pcbinshash_internal(inp, 0)); 2577 } 2578 2579 /* 2580 * Move PCB to the proper hash bucket when { faddr, fport } have been 2581 * changed. NOTE: This does not handle the case of the lport changing (the 2582 * hashed port list would have to be updated as well), so the lport must 2583 * not change after in_pcbinshash() has been called. 2584 */ 2585 void 2586 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2587 { 2588 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2589 struct inpcbhead *head; 2590 u_int32_t hashkey_faddr; 2591 2592 INP_WLOCK_ASSERT(inp); 2593 INP_HASH_WLOCK_ASSERT(pcbinfo); 2594 2595 KASSERT(inp->inp_flags & INP_INHASHLIST, 2596 ("in_pcbrehash: !INP_INHASHLIST")); 2597 2598 #ifdef INET6 2599 if (inp->inp_vflag & INP_IPV6) 2600 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2601 else 2602 #endif 2603 hashkey_faddr = inp->inp_faddr.s_addr; 2604 2605 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2606 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2607 2608 CK_LIST_REMOVE(inp, inp_hash); 2609 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2610 2611 #ifdef PCBGROUP 2612 if (m != NULL) 2613 in_pcbgroup_update_mbuf(inp, m); 2614 else 2615 in_pcbgroup_update(inp); 2616 #endif 2617 } 2618 2619 void 2620 in_pcbrehash(struct inpcb *inp) 2621 { 2622 2623 in_pcbrehash_mbuf(inp, NULL); 2624 } 2625 2626 /* 2627 * Remove PCB from various lists. 2628 */ 2629 static void 2630 in_pcbremlists(struct inpcb *inp) 2631 { 2632 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2633 2634 #ifdef INVARIANTS 2635 if (pcbinfo == &V_tcbinfo) { 2636 INP_INFO_RLOCK_ASSERT(pcbinfo); 2637 } else { 2638 INP_INFO_WLOCK_ASSERT(pcbinfo); 2639 } 2640 #endif 2641 2642 INP_WLOCK_ASSERT(inp); 2643 INP_LIST_WLOCK_ASSERT(pcbinfo); 2644 2645 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2646 if (inp->inp_flags & INP_INHASHLIST) { 2647 struct inpcbport *phd = inp->inp_phd; 2648 2649 INP_HASH_WLOCK(pcbinfo); 2650 2651 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2652 in_pcbremlbgrouphash(inp); 2653 2654 CK_LIST_REMOVE(inp, inp_hash); 2655 CK_LIST_REMOVE(inp, inp_portlist); 2656 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2657 CK_LIST_REMOVE(phd, phd_hash); 2658 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 2659 } 2660 INP_HASH_WUNLOCK(pcbinfo); 2661 inp->inp_flags &= ~INP_INHASHLIST; 2662 } 2663 CK_LIST_REMOVE(inp, inp_list); 2664 pcbinfo->ipi_count--; 2665 #ifdef PCBGROUP 2666 in_pcbgroup_remove(inp); 2667 #endif 2668 } 2669 2670 /* 2671 * Check for alternatives when higher level complains 2672 * about service problems. For now, invalidate cached 2673 * routing information. If the route was created dynamically 2674 * (by a redirect), time to try a default gateway again. 2675 */ 2676 void 2677 in_losing(struct inpcb *inp) 2678 { 2679 2680 RO_INVALIDATE_CACHE(&inp->inp_route); 2681 return; 2682 } 2683 2684 /* 2685 * A set label operation has occurred at the socket layer, propagate the 2686 * label change into the in_pcb for the socket. 2687 */ 2688 void 2689 in_pcbsosetlabel(struct socket *so) 2690 { 2691 #ifdef MAC 2692 struct inpcb *inp; 2693 2694 inp = sotoinpcb(so); 2695 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2696 2697 INP_WLOCK(inp); 2698 SOCK_LOCK(so); 2699 mac_inpcb_sosetlabel(so, inp); 2700 SOCK_UNLOCK(so); 2701 INP_WUNLOCK(inp); 2702 #endif 2703 } 2704 2705 /* 2706 * ipport_tick runs once per second, determining if random port allocation 2707 * should be continued. If more than ipport_randomcps ports have been 2708 * allocated in the last second, then we return to sequential port 2709 * allocation. We return to random allocation only once we drop below 2710 * ipport_randomcps for at least ipport_randomtime seconds. 2711 */ 2712 static void 2713 ipport_tick(void *xtp) 2714 { 2715 VNET_ITERATOR_DECL(vnet_iter); 2716 2717 VNET_LIST_RLOCK_NOSLEEP(); 2718 VNET_FOREACH(vnet_iter) { 2719 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2720 if (V_ipport_tcpallocs <= 2721 V_ipport_tcplastcount + V_ipport_randomcps) { 2722 if (V_ipport_stoprandom > 0) 2723 V_ipport_stoprandom--; 2724 } else 2725 V_ipport_stoprandom = V_ipport_randomtime; 2726 V_ipport_tcplastcount = V_ipport_tcpallocs; 2727 CURVNET_RESTORE(); 2728 } 2729 VNET_LIST_RUNLOCK_NOSLEEP(); 2730 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2731 } 2732 2733 static void 2734 ip_fini(void *xtp) 2735 { 2736 2737 callout_stop(&ipport_tick_callout); 2738 } 2739 2740 /* 2741 * The ipport_callout should start running at about the time we attach the 2742 * inet or inet6 domains. 2743 */ 2744 static void 2745 ipport_tick_init(const void *unused __unused) 2746 { 2747 2748 /* Start ipport_tick. */ 2749 callout_init(&ipport_tick_callout, 1); 2750 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2751 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2752 SHUTDOWN_PRI_DEFAULT); 2753 } 2754 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2755 ipport_tick_init, NULL); 2756 2757 void 2758 inp_wlock(struct inpcb *inp) 2759 { 2760 2761 INP_WLOCK(inp); 2762 } 2763 2764 void 2765 inp_wunlock(struct inpcb *inp) 2766 { 2767 2768 INP_WUNLOCK(inp); 2769 } 2770 2771 void 2772 inp_rlock(struct inpcb *inp) 2773 { 2774 2775 INP_RLOCK(inp); 2776 } 2777 2778 void 2779 inp_runlock(struct inpcb *inp) 2780 { 2781 2782 INP_RUNLOCK(inp); 2783 } 2784 2785 #ifdef INVARIANT_SUPPORT 2786 void 2787 inp_lock_assert(struct inpcb *inp) 2788 { 2789 2790 INP_WLOCK_ASSERT(inp); 2791 } 2792 2793 void 2794 inp_unlock_assert(struct inpcb *inp) 2795 { 2796 2797 INP_UNLOCK_ASSERT(inp); 2798 } 2799 #endif 2800 2801 void 2802 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2803 { 2804 struct inpcb *inp; 2805 2806 INP_INFO_WLOCK(&V_tcbinfo); 2807 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2808 INP_WLOCK(inp); 2809 func(inp, arg); 2810 INP_WUNLOCK(inp); 2811 } 2812 INP_INFO_WUNLOCK(&V_tcbinfo); 2813 } 2814 2815 struct socket * 2816 inp_inpcbtosocket(struct inpcb *inp) 2817 { 2818 2819 INP_WLOCK_ASSERT(inp); 2820 return (inp->inp_socket); 2821 } 2822 2823 struct tcpcb * 2824 inp_inpcbtotcpcb(struct inpcb *inp) 2825 { 2826 2827 INP_WLOCK_ASSERT(inp); 2828 return ((struct tcpcb *)inp->inp_ppcb); 2829 } 2830 2831 int 2832 inp_ip_tos_get(const struct inpcb *inp) 2833 { 2834 2835 return (inp->inp_ip_tos); 2836 } 2837 2838 void 2839 inp_ip_tos_set(struct inpcb *inp, int val) 2840 { 2841 2842 inp->inp_ip_tos = val; 2843 } 2844 2845 void 2846 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2847 uint32_t *faddr, uint16_t *fp) 2848 { 2849 2850 INP_LOCK_ASSERT(inp); 2851 *laddr = inp->inp_laddr.s_addr; 2852 *faddr = inp->inp_faddr.s_addr; 2853 *lp = inp->inp_lport; 2854 *fp = inp->inp_fport; 2855 } 2856 2857 struct inpcb * 2858 so_sotoinpcb(struct socket *so) 2859 { 2860 2861 return (sotoinpcb(so)); 2862 } 2863 2864 struct tcpcb * 2865 so_sototcpcb(struct socket *so) 2866 { 2867 2868 return (sototcpcb(so)); 2869 } 2870 2871 /* 2872 * Create an external-format (``xinpcb'') structure using the information in 2873 * the kernel-format in_pcb structure pointed to by inp. This is done to 2874 * reduce the spew of irrelevant information over this interface, to isolate 2875 * user code from changes in the kernel structure, and potentially to provide 2876 * information-hiding if we decide that some of this information should be 2877 * hidden from users. 2878 */ 2879 void 2880 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2881 { 2882 2883 bzero(xi, sizeof(*xi)); 2884 xi->xi_len = sizeof(struct xinpcb); 2885 if (inp->inp_socket) 2886 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2887 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2888 xi->inp_gencnt = inp->inp_gencnt; 2889 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2890 xi->inp_flow = inp->inp_flow; 2891 xi->inp_flowid = inp->inp_flowid; 2892 xi->inp_flowtype = inp->inp_flowtype; 2893 xi->inp_flags = inp->inp_flags; 2894 xi->inp_flags2 = inp->inp_flags2; 2895 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2896 xi->in6p_cksum = inp->in6p_cksum; 2897 xi->in6p_hops = inp->in6p_hops; 2898 xi->inp_ip_tos = inp->inp_ip_tos; 2899 xi->inp_vflag = inp->inp_vflag; 2900 xi->inp_ip_ttl = inp->inp_ip_ttl; 2901 xi->inp_ip_p = inp->inp_ip_p; 2902 xi->inp_ip_minttl = inp->inp_ip_minttl; 2903 } 2904 2905 #ifdef DDB 2906 static void 2907 db_print_indent(int indent) 2908 { 2909 int i; 2910 2911 for (i = 0; i < indent; i++) 2912 db_printf(" "); 2913 } 2914 2915 static void 2916 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2917 { 2918 char faddr_str[48], laddr_str[48]; 2919 2920 db_print_indent(indent); 2921 db_printf("%s at %p\n", name, inc); 2922 2923 indent += 2; 2924 2925 #ifdef INET6 2926 if (inc->inc_flags & INC_ISIPV6) { 2927 /* IPv6. */ 2928 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2929 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2930 } else 2931 #endif 2932 { 2933 /* IPv4. */ 2934 inet_ntoa_r(inc->inc_laddr, laddr_str); 2935 inet_ntoa_r(inc->inc_faddr, faddr_str); 2936 } 2937 db_print_indent(indent); 2938 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2939 ntohs(inc->inc_lport)); 2940 db_print_indent(indent); 2941 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2942 ntohs(inc->inc_fport)); 2943 } 2944 2945 static void 2946 db_print_inpflags(int inp_flags) 2947 { 2948 int comma; 2949 2950 comma = 0; 2951 if (inp_flags & INP_RECVOPTS) { 2952 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2953 comma = 1; 2954 } 2955 if (inp_flags & INP_RECVRETOPTS) { 2956 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2957 comma = 1; 2958 } 2959 if (inp_flags & INP_RECVDSTADDR) { 2960 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2961 comma = 1; 2962 } 2963 if (inp_flags & INP_ORIGDSTADDR) { 2964 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2965 comma = 1; 2966 } 2967 if (inp_flags & INP_HDRINCL) { 2968 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2969 comma = 1; 2970 } 2971 if (inp_flags & INP_HIGHPORT) { 2972 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2973 comma = 1; 2974 } 2975 if (inp_flags & INP_LOWPORT) { 2976 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 2977 comma = 1; 2978 } 2979 if (inp_flags & INP_ANONPORT) { 2980 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 2981 comma = 1; 2982 } 2983 if (inp_flags & INP_RECVIF) { 2984 db_printf("%sINP_RECVIF", comma ? ", " : ""); 2985 comma = 1; 2986 } 2987 if (inp_flags & INP_MTUDISC) { 2988 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 2989 comma = 1; 2990 } 2991 if (inp_flags & INP_RECVTTL) { 2992 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 2993 comma = 1; 2994 } 2995 if (inp_flags & INP_DONTFRAG) { 2996 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 2997 comma = 1; 2998 } 2999 if (inp_flags & INP_RECVTOS) { 3000 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3001 comma = 1; 3002 } 3003 if (inp_flags & IN6P_IPV6_V6ONLY) { 3004 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3005 comma = 1; 3006 } 3007 if (inp_flags & IN6P_PKTINFO) { 3008 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3009 comma = 1; 3010 } 3011 if (inp_flags & IN6P_HOPLIMIT) { 3012 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3013 comma = 1; 3014 } 3015 if (inp_flags & IN6P_HOPOPTS) { 3016 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3017 comma = 1; 3018 } 3019 if (inp_flags & IN6P_DSTOPTS) { 3020 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3021 comma = 1; 3022 } 3023 if (inp_flags & IN6P_RTHDR) { 3024 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3025 comma = 1; 3026 } 3027 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3028 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3029 comma = 1; 3030 } 3031 if (inp_flags & IN6P_TCLASS) { 3032 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3033 comma = 1; 3034 } 3035 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3036 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3037 comma = 1; 3038 } 3039 if (inp_flags & INP_TIMEWAIT) { 3040 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3041 comma = 1; 3042 } 3043 if (inp_flags & INP_ONESBCAST) { 3044 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3045 comma = 1; 3046 } 3047 if (inp_flags & INP_DROPPED) { 3048 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3049 comma = 1; 3050 } 3051 if (inp_flags & INP_SOCKREF) { 3052 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3053 comma = 1; 3054 } 3055 if (inp_flags & IN6P_RFC2292) { 3056 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3057 comma = 1; 3058 } 3059 if (inp_flags & IN6P_MTU) { 3060 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3061 comma = 1; 3062 } 3063 } 3064 3065 static void 3066 db_print_inpvflag(u_char inp_vflag) 3067 { 3068 int comma; 3069 3070 comma = 0; 3071 if (inp_vflag & INP_IPV4) { 3072 db_printf("%sINP_IPV4", comma ? ", " : ""); 3073 comma = 1; 3074 } 3075 if (inp_vflag & INP_IPV6) { 3076 db_printf("%sINP_IPV6", comma ? ", " : ""); 3077 comma = 1; 3078 } 3079 if (inp_vflag & INP_IPV6PROTO) { 3080 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3081 comma = 1; 3082 } 3083 } 3084 3085 static void 3086 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3087 { 3088 3089 db_print_indent(indent); 3090 db_printf("%s at %p\n", name, inp); 3091 3092 indent += 2; 3093 3094 db_print_indent(indent); 3095 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3096 3097 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3098 3099 db_print_indent(indent); 3100 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3101 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3102 3103 db_print_indent(indent); 3104 db_printf("inp_label: %p inp_flags: 0x%x (", 3105 inp->inp_label, inp->inp_flags); 3106 db_print_inpflags(inp->inp_flags); 3107 db_printf(")\n"); 3108 3109 db_print_indent(indent); 3110 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3111 inp->inp_vflag); 3112 db_print_inpvflag(inp->inp_vflag); 3113 db_printf(")\n"); 3114 3115 db_print_indent(indent); 3116 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3117 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3118 3119 db_print_indent(indent); 3120 #ifdef INET6 3121 if (inp->inp_vflag & INP_IPV6) { 3122 db_printf("in6p_options: %p in6p_outputopts: %p " 3123 "in6p_moptions: %p\n", inp->in6p_options, 3124 inp->in6p_outputopts, inp->in6p_moptions); 3125 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3126 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3127 inp->in6p_hops); 3128 } else 3129 #endif 3130 { 3131 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3132 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3133 inp->inp_options, inp->inp_moptions); 3134 } 3135 3136 db_print_indent(indent); 3137 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3138 (uintmax_t)inp->inp_gencnt); 3139 } 3140 3141 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3142 { 3143 struct inpcb *inp; 3144 3145 if (!have_addr) { 3146 db_printf("usage: show inpcb <addr>\n"); 3147 return; 3148 } 3149 inp = (struct inpcb *)addr; 3150 3151 db_print_inpcb(inp, "inpcb", 0); 3152 } 3153 #endif /* DDB */ 3154 3155 #ifdef RATELIMIT 3156 /* 3157 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3158 * if any. 3159 */ 3160 int 3161 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3162 { 3163 union if_snd_tag_modify_params params = { 3164 .rate_limit.max_rate = max_pacing_rate, 3165 }; 3166 struct m_snd_tag *mst; 3167 struct ifnet *ifp; 3168 int error; 3169 3170 mst = inp->inp_snd_tag; 3171 if (mst == NULL) 3172 return (EINVAL); 3173 3174 ifp = mst->ifp; 3175 if (ifp == NULL) 3176 return (EINVAL); 3177 3178 if (ifp->if_snd_tag_modify == NULL) { 3179 error = EOPNOTSUPP; 3180 } else { 3181 error = ifp->if_snd_tag_modify(mst, ¶ms); 3182 } 3183 return (error); 3184 } 3185 3186 /* 3187 * Query existing TX rate limit based on the existing 3188 * "inp->inp_snd_tag", if any. 3189 */ 3190 int 3191 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3192 { 3193 union if_snd_tag_query_params params = { }; 3194 struct m_snd_tag *mst; 3195 struct ifnet *ifp; 3196 int error; 3197 3198 mst = inp->inp_snd_tag; 3199 if (mst == NULL) 3200 return (EINVAL); 3201 3202 ifp = mst->ifp; 3203 if (ifp == NULL) 3204 return (EINVAL); 3205 3206 if (ifp->if_snd_tag_query == NULL) { 3207 error = EOPNOTSUPP; 3208 } else { 3209 error = ifp->if_snd_tag_query(mst, ¶ms); 3210 if (error == 0 && p_max_pacing_rate != NULL) 3211 *p_max_pacing_rate = params.rate_limit.max_rate; 3212 } 3213 return (error); 3214 } 3215 3216 /* 3217 * Query existing TX queue level based on the existing 3218 * "inp->inp_snd_tag", if any. 3219 */ 3220 int 3221 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3222 { 3223 union if_snd_tag_query_params params = { }; 3224 struct m_snd_tag *mst; 3225 struct ifnet *ifp; 3226 int error; 3227 3228 mst = inp->inp_snd_tag; 3229 if (mst == NULL) 3230 return (EINVAL); 3231 3232 ifp = mst->ifp; 3233 if (ifp == NULL) 3234 return (EINVAL); 3235 3236 if (ifp->if_snd_tag_query == NULL) 3237 return (EOPNOTSUPP); 3238 3239 error = ifp->if_snd_tag_query(mst, ¶ms); 3240 if (error == 0 && p_txqueue_level != NULL) 3241 *p_txqueue_level = params.rate_limit.queue_level; 3242 return (error); 3243 } 3244 3245 /* 3246 * Allocate a new TX rate limit send tag from the network interface 3247 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3248 */ 3249 int 3250 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3251 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate) 3252 { 3253 union if_snd_tag_alloc_params params = { 3254 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3255 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3256 .rate_limit.hdr.flowid = flowid, 3257 .rate_limit.hdr.flowtype = flowtype, 3258 .rate_limit.max_rate = max_pacing_rate, 3259 }; 3260 int error; 3261 3262 INP_WLOCK_ASSERT(inp); 3263 3264 if (inp->inp_snd_tag != NULL) 3265 return (EINVAL); 3266 3267 if (ifp->if_snd_tag_alloc == NULL) { 3268 error = EOPNOTSUPP; 3269 } else { 3270 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); 3271 3272 /* 3273 * At success increment the refcount on 3274 * the send tag's network interface: 3275 */ 3276 if (error == 0) 3277 if_ref(inp->inp_snd_tag->ifp); 3278 } 3279 return (error); 3280 } 3281 3282 /* 3283 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3284 * if any: 3285 */ 3286 void 3287 in_pcbdetach_txrtlmt(struct inpcb *inp) 3288 { 3289 struct m_snd_tag *mst; 3290 struct ifnet *ifp; 3291 3292 INP_WLOCK_ASSERT(inp); 3293 3294 mst = inp->inp_snd_tag; 3295 inp->inp_snd_tag = NULL; 3296 3297 if (mst == NULL) 3298 return; 3299 3300 ifp = mst->ifp; 3301 if (ifp == NULL) 3302 return; 3303 3304 /* 3305 * If the device was detached while we still had reference(s) 3306 * on the ifp, we assume if_snd_tag_free() was replaced with 3307 * stubs. 3308 */ 3309 ifp->if_snd_tag_free(mst); 3310 3311 /* release reference count on network interface */ 3312 if_rele(ifp); 3313 } 3314 3315 /* 3316 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3317 * is set in the fast path and will attach/detach/modify the TX rate 3318 * limit send tag based on the socket's so_max_pacing_rate value. 3319 */ 3320 void 3321 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3322 { 3323 struct socket *socket; 3324 uint32_t max_pacing_rate; 3325 bool did_upgrade; 3326 int error; 3327 3328 if (inp == NULL) 3329 return; 3330 3331 socket = inp->inp_socket; 3332 if (socket == NULL) 3333 return; 3334 3335 if (!INP_WLOCKED(inp)) { 3336 /* 3337 * NOTE: If the write locking fails, we need to bail 3338 * out and use the non-ratelimited ring for the 3339 * transmit until there is a new chance to get the 3340 * write lock. 3341 */ 3342 if (!INP_TRY_UPGRADE(inp)) 3343 return; 3344 did_upgrade = 1; 3345 } else { 3346 did_upgrade = 0; 3347 } 3348 3349 /* 3350 * NOTE: The so_max_pacing_rate value is read unlocked, 3351 * because atomic updates are not required since the variable 3352 * is checked at every mbuf we send. It is assumed that the 3353 * variable read itself will be atomic. 3354 */ 3355 max_pacing_rate = socket->so_max_pacing_rate; 3356 3357 /* 3358 * NOTE: When attaching to a network interface a reference is 3359 * made to ensure the network interface doesn't go away until 3360 * all ratelimit connections are gone. The network interface 3361 * pointers compared below represent valid network interfaces, 3362 * except when comparing towards NULL. 3363 */ 3364 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3365 error = 0; 3366 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3367 if (inp->inp_snd_tag != NULL) 3368 in_pcbdetach_txrtlmt(inp); 3369 error = 0; 3370 } else if (inp->inp_snd_tag == NULL) { 3371 /* 3372 * In order to utilize packet pacing with RSS, we need 3373 * to wait until there is a valid RSS hash before we 3374 * can proceed: 3375 */ 3376 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3377 error = EAGAIN; 3378 } else { 3379 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3380 mb->m_pkthdr.flowid, max_pacing_rate); 3381 } 3382 } else { 3383 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3384 } 3385 if (error == 0 || error == EOPNOTSUPP) 3386 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3387 if (did_upgrade) 3388 INP_DOWNGRADE(inp); 3389 } 3390 3391 /* 3392 * Track route changes for TX rate limiting. 3393 */ 3394 void 3395 in_pcboutput_eagain(struct inpcb *inp) 3396 { 3397 struct socket *socket; 3398 bool did_upgrade; 3399 3400 if (inp == NULL) 3401 return; 3402 3403 socket = inp->inp_socket; 3404 if (socket == NULL) 3405 return; 3406 3407 if (inp->inp_snd_tag == NULL) 3408 return; 3409 3410 if (!INP_WLOCKED(inp)) { 3411 /* 3412 * NOTE: If the write locking fails, we need to bail 3413 * out and use the non-ratelimited ring for the 3414 * transmit until there is a new chance to get the 3415 * write lock. 3416 */ 3417 if (!INP_TRY_UPGRADE(inp)) 3418 return; 3419 did_upgrade = 1; 3420 } else { 3421 did_upgrade = 0; 3422 } 3423 3424 /* detach rate limiting */ 3425 in_pcbdetach_txrtlmt(inp); 3426 3427 /* make sure new mbuf send tag allocation is made */ 3428 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3429 3430 if (did_upgrade) 3431 INP_DOWNGRADE(inp); 3432 } 3433 #endif /* RATELIMIT */ 3434