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 struct epoch_tracker et; 1016 int error; 1017 1018 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1019 /* 1020 * Bypass source address selection and use the primary jail IP 1021 * if requested. 1022 */ 1023 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1024 return (0); 1025 1026 error = 0; 1027 bzero(&sro, sizeof(sro)); 1028 1029 sin = (struct sockaddr_in *)&sro.ro_dst; 1030 sin->sin_family = AF_INET; 1031 sin->sin_len = sizeof(struct sockaddr_in); 1032 sin->sin_addr.s_addr = faddr->s_addr; 1033 1034 /* 1035 * If route is known our src addr is taken from the i/f, 1036 * else punt. 1037 * 1038 * Find out route to destination. 1039 */ 1040 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1041 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 1042 1043 /* 1044 * If we found a route, use the address corresponding to 1045 * the outgoing interface. 1046 * 1047 * Otherwise assume faddr is reachable on a directly connected 1048 * network and try to find a corresponding interface to take 1049 * the source address from. 1050 */ 1051 NET_EPOCH_ENTER(et); 1052 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 1053 struct in_ifaddr *ia; 1054 struct ifnet *ifp; 1055 1056 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1057 inp->inp_socket->so_fibnum)); 1058 if (ia == NULL) { 1059 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1060 inp->inp_socket->so_fibnum)); 1061 1062 } 1063 if (ia == NULL) { 1064 error = ENETUNREACH; 1065 goto done; 1066 } 1067 1068 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1069 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1070 goto done; 1071 } 1072 1073 ifp = ia->ia_ifp; 1074 ia = NULL; 1075 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1076 1077 sa = ifa->ifa_addr; 1078 if (sa->sa_family != AF_INET) 1079 continue; 1080 sin = (struct sockaddr_in *)sa; 1081 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1082 ia = (struct in_ifaddr *)ifa; 1083 break; 1084 } 1085 } 1086 if (ia != NULL) { 1087 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1088 goto done; 1089 } 1090 1091 /* 3. As a last resort return the 'default' jail address. */ 1092 error = prison_get_ip4(cred, laddr); 1093 goto done; 1094 } 1095 1096 /* 1097 * If the outgoing interface on the route found is not 1098 * a loopback interface, use the address from that interface. 1099 * In case of jails do those three steps: 1100 * 1. check if the interface address belongs to the jail. If so use it. 1101 * 2. check if we have any address on the outgoing interface 1102 * belonging to this jail. If so use it. 1103 * 3. as a last resort return the 'default' jail address. 1104 */ 1105 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 1106 struct in_ifaddr *ia; 1107 struct ifnet *ifp; 1108 1109 /* If not jailed, use the default returned. */ 1110 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1111 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1112 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1113 goto done; 1114 } 1115 1116 /* Jailed. */ 1117 /* 1. Check if the iface address belongs to the jail. */ 1118 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 1119 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1120 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1121 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1122 goto done; 1123 } 1124 1125 /* 1126 * 2. Check if we have any address on the outgoing interface 1127 * belonging to this jail. 1128 */ 1129 ia = NULL; 1130 ifp = sro.ro_rt->rt_ifp; 1131 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1132 sa = ifa->ifa_addr; 1133 if (sa->sa_family != AF_INET) 1134 continue; 1135 sin = (struct sockaddr_in *)sa; 1136 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1137 ia = (struct in_ifaddr *)ifa; 1138 break; 1139 } 1140 } 1141 if (ia != NULL) { 1142 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1143 goto done; 1144 } 1145 1146 /* 3. As a last resort return the 'default' jail address. */ 1147 error = prison_get_ip4(cred, laddr); 1148 goto done; 1149 } 1150 1151 /* 1152 * The outgoing interface is marked with 'loopback net', so a route 1153 * to ourselves is here. 1154 * Try to find the interface of the destination address and then 1155 * take the address from there. That interface is not necessarily 1156 * a loopback interface. 1157 * In case of jails, check that it is an address of the jail 1158 * and if we cannot find, fall back to the 'default' jail address. 1159 */ 1160 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 1161 struct sockaddr_in sain; 1162 struct in_ifaddr *ia; 1163 1164 bzero(&sain, sizeof(struct sockaddr_in)); 1165 sain.sin_family = AF_INET; 1166 sain.sin_len = sizeof(struct sockaddr_in); 1167 sain.sin_addr.s_addr = faddr->s_addr; 1168 1169 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), 1170 inp->inp_socket->so_fibnum)); 1171 if (ia == NULL) 1172 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, 1173 inp->inp_socket->so_fibnum)); 1174 if (ia == NULL) 1175 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 1176 1177 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1178 if (ia == NULL) { 1179 error = ENETUNREACH; 1180 goto done; 1181 } 1182 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1183 goto done; 1184 } 1185 1186 /* Jailed. */ 1187 if (ia != NULL) { 1188 struct ifnet *ifp; 1189 1190 ifp = ia->ia_ifp; 1191 ia = NULL; 1192 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1193 sa = ifa->ifa_addr; 1194 if (sa->sa_family != AF_INET) 1195 continue; 1196 sin = (struct sockaddr_in *)sa; 1197 if (prison_check_ip4(cred, 1198 &sin->sin_addr) == 0) { 1199 ia = (struct in_ifaddr *)ifa; 1200 break; 1201 } 1202 } 1203 if (ia != NULL) { 1204 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1205 goto done; 1206 } 1207 } 1208 1209 /* 3. As a last resort return the 'default' jail address. */ 1210 error = prison_get_ip4(cred, laddr); 1211 goto done; 1212 } 1213 1214 done: 1215 NET_EPOCH_EXIT(et); 1216 if (sro.ro_rt != NULL) 1217 RTFREE(sro.ro_rt); 1218 return (error); 1219 } 1220 1221 /* 1222 * Set up for a connect from a socket to the specified address. 1223 * On entry, *laddrp and *lportp should contain the current local 1224 * address and port for the PCB; these are updated to the values 1225 * that should be placed in inp_laddr and inp_lport to complete 1226 * the connect. 1227 * 1228 * On success, *faddrp and *fportp will be set to the remote address 1229 * and port. These are not updated in the error case. 1230 * 1231 * If the operation fails because the connection already exists, 1232 * *oinpp will be set to the PCB of that connection so that the 1233 * caller can decide to override it. In all other cases, *oinpp 1234 * is set to NULL. 1235 */ 1236 int 1237 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1238 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1239 struct inpcb **oinpp, struct ucred *cred) 1240 { 1241 struct rm_priotracker in_ifa_tracker; 1242 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1243 struct in_ifaddr *ia; 1244 struct inpcb *oinp; 1245 struct in_addr laddr, faddr; 1246 u_short lport, fport; 1247 int error; 1248 1249 /* 1250 * Because a global state change doesn't actually occur here, a read 1251 * lock is sufficient. 1252 */ 1253 INP_LOCK_ASSERT(inp); 1254 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1255 1256 if (oinpp != NULL) 1257 *oinpp = NULL; 1258 if (nam->sa_len != sizeof (*sin)) 1259 return (EINVAL); 1260 if (sin->sin_family != AF_INET) 1261 return (EAFNOSUPPORT); 1262 if (sin->sin_port == 0) 1263 return (EADDRNOTAVAIL); 1264 laddr.s_addr = *laddrp; 1265 lport = *lportp; 1266 faddr = sin->sin_addr; 1267 fport = sin->sin_port; 1268 1269 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1270 /* 1271 * If the destination address is INADDR_ANY, 1272 * use the primary local address. 1273 * If the supplied address is INADDR_BROADCAST, 1274 * and the primary interface supports broadcast, 1275 * choose the broadcast address for that interface. 1276 */ 1277 if (faddr.s_addr == INADDR_ANY) { 1278 IN_IFADDR_RLOCK(&in_ifa_tracker); 1279 faddr = 1280 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1281 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1282 if (cred != NULL && 1283 (error = prison_get_ip4(cred, &faddr)) != 0) 1284 return (error); 1285 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1286 IN_IFADDR_RLOCK(&in_ifa_tracker); 1287 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1288 IFF_BROADCAST) 1289 faddr = satosin(&CK_STAILQ_FIRST( 1290 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1291 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1292 } 1293 } 1294 if (laddr.s_addr == INADDR_ANY) { 1295 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1296 /* 1297 * If the destination address is multicast and an outgoing 1298 * interface has been set as a multicast option, prefer the 1299 * address of that interface as our source address. 1300 */ 1301 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1302 inp->inp_moptions != NULL) { 1303 struct ip_moptions *imo; 1304 struct ifnet *ifp; 1305 1306 imo = inp->inp_moptions; 1307 if (imo->imo_multicast_ifp != NULL) { 1308 ifp = imo->imo_multicast_ifp; 1309 IN_IFADDR_RLOCK(&in_ifa_tracker); 1310 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1311 if ((ia->ia_ifp == ifp) && 1312 (cred == NULL || 1313 prison_check_ip4(cred, 1314 &ia->ia_addr.sin_addr) == 0)) 1315 break; 1316 } 1317 if (ia == NULL) 1318 error = EADDRNOTAVAIL; 1319 else { 1320 laddr = ia->ia_addr.sin_addr; 1321 error = 0; 1322 } 1323 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1324 } 1325 } 1326 if (error) 1327 return (error); 1328 } 1329 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1330 laddr, lport, 0, NULL); 1331 if (oinp != NULL) { 1332 if (oinpp != NULL) 1333 *oinpp = oinp; 1334 return (EADDRINUSE); 1335 } 1336 if (lport == 0) { 1337 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1338 cred); 1339 if (error) 1340 return (error); 1341 } 1342 *laddrp = laddr.s_addr; 1343 *lportp = lport; 1344 *faddrp = faddr.s_addr; 1345 *fportp = fport; 1346 return (0); 1347 } 1348 1349 void 1350 in_pcbdisconnect(struct inpcb *inp) 1351 { 1352 1353 INP_WLOCK_ASSERT(inp); 1354 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1355 1356 inp->inp_faddr.s_addr = INADDR_ANY; 1357 inp->inp_fport = 0; 1358 in_pcbrehash(inp); 1359 } 1360 #endif /* INET */ 1361 1362 /* 1363 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1364 * For most protocols, this will be invoked immediately prior to calling 1365 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1366 * socket, in which case in_pcbfree() is deferred. 1367 */ 1368 void 1369 in_pcbdetach(struct inpcb *inp) 1370 { 1371 1372 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1373 1374 #ifdef RATELIMIT 1375 if (inp->inp_snd_tag != NULL) 1376 in_pcbdetach_txrtlmt(inp); 1377 #endif 1378 inp->inp_socket->so_pcb = NULL; 1379 inp->inp_socket = NULL; 1380 } 1381 1382 /* 1383 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1384 * stability of an inpcb pointer despite the inpcb lock being released. This 1385 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1386 * but where the inpcb lock may already held, or when acquiring a reference 1387 * via a pcbgroup. 1388 * 1389 * in_pcbref() should be used only to provide brief memory stability, and 1390 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1391 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1392 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1393 * lock and rele are the *only* safe operations that may be performed on the 1394 * inpcb. 1395 * 1396 * While the inpcb will not be freed, releasing the inpcb lock means that the 1397 * connection's state may change, so the caller should be careful to 1398 * revalidate any cached state on reacquiring the lock. Drop the reference 1399 * using in_pcbrele(). 1400 */ 1401 void 1402 in_pcbref(struct inpcb *inp) 1403 { 1404 1405 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1406 1407 refcount_acquire(&inp->inp_refcount); 1408 } 1409 1410 /* 1411 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1412 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1413 * return a flag indicating whether or not the inpcb remains valid. If it is 1414 * valid, we return with the inpcb lock held. 1415 * 1416 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1417 * reference on an inpcb. Historically more work was done here (actually, in 1418 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1419 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1420 * about memory stability (and continued use of the write lock). 1421 */ 1422 int 1423 in_pcbrele_rlocked(struct inpcb *inp) 1424 { 1425 struct inpcbinfo *pcbinfo; 1426 1427 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1428 1429 INP_RLOCK_ASSERT(inp); 1430 1431 if (refcount_release(&inp->inp_refcount) == 0) { 1432 /* 1433 * If the inpcb has been freed, let the caller know, even if 1434 * this isn't the last reference. 1435 */ 1436 if (inp->inp_flags2 & INP_FREED) { 1437 INP_RUNLOCK(inp); 1438 return (1); 1439 } 1440 return (0); 1441 } 1442 1443 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1444 #ifdef TCPHPTS 1445 if (inp->inp_in_hpts || inp->inp_in_input) { 1446 struct tcp_hpts_entry *hpts; 1447 /* 1448 * We should not be on the hpts at 1449 * this point in any form. we must 1450 * get the lock to be sure. 1451 */ 1452 hpts = tcp_hpts_lock(inp); 1453 if (inp->inp_in_hpts) 1454 panic("Hpts:%p inp:%p at free still on hpts", 1455 hpts, inp); 1456 mtx_unlock(&hpts->p_mtx); 1457 hpts = tcp_input_lock(inp); 1458 if (inp->inp_in_input) 1459 panic("Hpts:%p inp:%p at free still on input hpts", 1460 hpts, inp); 1461 mtx_unlock(&hpts->p_mtx); 1462 } 1463 #endif 1464 INP_RUNLOCK(inp); 1465 pcbinfo = inp->inp_pcbinfo; 1466 uma_zfree(pcbinfo->ipi_zone, inp); 1467 return (1); 1468 } 1469 1470 int 1471 in_pcbrele_wlocked(struct inpcb *inp) 1472 { 1473 struct inpcbinfo *pcbinfo; 1474 1475 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1476 1477 INP_WLOCK_ASSERT(inp); 1478 1479 if (refcount_release(&inp->inp_refcount) == 0) { 1480 /* 1481 * If the inpcb has been freed, let the caller know, even if 1482 * this isn't the last reference. 1483 */ 1484 if (inp->inp_flags2 & INP_FREED) { 1485 INP_WUNLOCK(inp); 1486 return (1); 1487 } 1488 return (0); 1489 } 1490 1491 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1492 #ifdef TCPHPTS 1493 if (inp->inp_in_hpts || inp->inp_in_input) { 1494 struct tcp_hpts_entry *hpts; 1495 /* 1496 * We should not be on the hpts at 1497 * this point in any form. we must 1498 * get the lock to be sure. 1499 */ 1500 hpts = tcp_hpts_lock(inp); 1501 if (inp->inp_in_hpts) 1502 panic("Hpts:%p inp:%p at free still on hpts", 1503 hpts, inp); 1504 mtx_unlock(&hpts->p_mtx); 1505 hpts = tcp_input_lock(inp); 1506 if (inp->inp_in_input) 1507 panic("Hpts:%p inp:%p at free still on input hpts", 1508 hpts, inp); 1509 mtx_unlock(&hpts->p_mtx); 1510 } 1511 #endif 1512 INP_WUNLOCK(inp); 1513 pcbinfo = inp->inp_pcbinfo; 1514 uma_zfree(pcbinfo->ipi_zone, inp); 1515 return (1); 1516 } 1517 1518 /* 1519 * Temporary wrapper. 1520 */ 1521 int 1522 in_pcbrele(struct inpcb *inp) 1523 { 1524 1525 return (in_pcbrele_wlocked(inp)); 1526 } 1527 1528 void 1529 in_pcblist_rele_rlocked(epoch_context_t ctx) 1530 { 1531 struct in_pcblist *il; 1532 struct inpcb *inp; 1533 struct inpcbinfo *pcbinfo; 1534 int i, n; 1535 1536 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1537 pcbinfo = il->il_pcbinfo; 1538 n = il->il_count; 1539 INP_INFO_WLOCK(pcbinfo); 1540 for (i = 0; i < n; i++) { 1541 inp = il->il_inp_list[i]; 1542 INP_RLOCK(inp); 1543 if (!in_pcbrele_rlocked(inp)) 1544 INP_RUNLOCK(inp); 1545 } 1546 INP_INFO_WUNLOCK(pcbinfo); 1547 free(il, M_TEMP); 1548 } 1549 1550 static void 1551 inpcbport_free(epoch_context_t ctx) 1552 { 1553 struct inpcbport *phd; 1554 1555 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); 1556 free(phd, M_PCB); 1557 } 1558 1559 static void 1560 in_pcbfree_deferred(epoch_context_t ctx) 1561 { 1562 struct inpcb *inp; 1563 int released __unused; 1564 1565 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); 1566 1567 INP_WLOCK(inp); 1568 CURVNET_SET(inp->inp_vnet); 1569 #ifdef INET 1570 struct ip_moptions *imo = inp->inp_moptions; 1571 inp->inp_moptions = NULL; 1572 #endif 1573 /* XXXRW: Do as much as possible here. */ 1574 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1575 if (inp->inp_sp != NULL) 1576 ipsec_delete_pcbpolicy(inp); 1577 #endif 1578 #ifdef INET6 1579 struct ip6_moptions *im6o = NULL; 1580 if (inp->inp_vflag & INP_IPV6PROTO) { 1581 ip6_freepcbopts(inp->in6p_outputopts); 1582 im6o = inp->in6p_moptions; 1583 inp->in6p_moptions = NULL; 1584 } 1585 #endif 1586 if (inp->inp_options) 1587 (void)m_free(inp->inp_options); 1588 inp->inp_vflag = 0; 1589 crfree(inp->inp_cred); 1590 #ifdef MAC 1591 mac_inpcb_destroy(inp); 1592 #endif 1593 released = in_pcbrele_wlocked(inp); 1594 MPASS(released); 1595 #ifdef INET6 1596 ip6_freemoptions(im6o); 1597 #endif 1598 #ifdef INET 1599 inp_freemoptions(imo); 1600 #endif 1601 CURVNET_RESTORE(); 1602 } 1603 1604 /* 1605 * Unconditionally schedule an inpcb to be freed by decrementing its 1606 * reference count, which should occur only after the inpcb has been detached 1607 * from its socket. If another thread holds a temporary reference (acquired 1608 * using in_pcbref()) then the free is deferred until that reference is 1609 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1610 * work, including removal from global lists, is done in this context, where 1611 * the pcbinfo lock is held. 1612 */ 1613 void 1614 in_pcbfree(struct inpcb *inp) 1615 { 1616 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1617 1618 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1619 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1620 ("%s: called twice for pcb %p", __func__, inp)); 1621 if (inp->inp_flags2 & INP_FREED) { 1622 INP_WUNLOCK(inp); 1623 return; 1624 } 1625 1626 #ifdef INVARIANTS 1627 if (pcbinfo == &V_tcbinfo) { 1628 INP_INFO_LOCK_ASSERT(pcbinfo); 1629 } else { 1630 INP_INFO_WLOCK_ASSERT(pcbinfo); 1631 } 1632 #endif 1633 INP_WLOCK_ASSERT(inp); 1634 INP_LIST_WLOCK(pcbinfo); 1635 in_pcbremlists(inp); 1636 INP_LIST_WUNLOCK(pcbinfo); 1637 RO_INVALIDATE_CACHE(&inp->inp_route); 1638 /* mark as destruction in progress */ 1639 inp->inp_flags2 |= INP_FREED; 1640 INP_WUNLOCK(inp); 1641 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred); 1642 } 1643 1644 /* 1645 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1646 * port reservation, and preventing it from being returned by inpcb lookups. 1647 * 1648 * It is used by TCP to mark an inpcb as unused and avoid future packet 1649 * delivery or event notification when a socket remains open but TCP has 1650 * closed. This might occur as a result of a shutdown()-initiated TCP close 1651 * or a RST on the wire, and allows the port binding to be reused while still 1652 * maintaining the invariant that so_pcb always points to a valid inpcb until 1653 * in_pcbdetach(). 1654 * 1655 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1656 * in_pcbnotifyall() and in_pcbpurgeif0()? 1657 */ 1658 void 1659 in_pcbdrop(struct inpcb *inp) 1660 { 1661 1662 INP_WLOCK_ASSERT(inp); 1663 #ifdef INVARIANTS 1664 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1665 MPASS(inp->inp_refcount > 1); 1666 #endif 1667 1668 /* 1669 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1670 * the hash lock...? 1671 */ 1672 inp->inp_flags |= INP_DROPPED; 1673 if (inp->inp_flags & INP_INHASHLIST) { 1674 struct inpcbport *phd = inp->inp_phd; 1675 1676 INP_HASH_WLOCK(inp->inp_pcbinfo); 1677 in_pcbremlbgrouphash(inp); 1678 CK_LIST_REMOVE(inp, inp_hash); 1679 CK_LIST_REMOVE(inp, inp_portlist); 1680 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 1681 CK_LIST_REMOVE(phd, phd_hash); 1682 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 1683 } 1684 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1685 inp->inp_flags &= ~INP_INHASHLIST; 1686 #ifdef PCBGROUP 1687 in_pcbgroup_remove(inp); 1688 #endif 1689 } 1690 } 1691 1692 #ifdef INET 1693 /* 1694 * Common routines to return the socket addresses associated with inpcbs. 1695 */ 1696 struct sockaddr * 1697 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1698 { 1699 struct sockaddr_in *sin; 1700 1701 sin = malloc(sizeof *sin, M_SONAME, 1702 M_WAITOK | M_ZERO); 1703 sin->sin_family = AF_INET; 1704 sin->sin_len = sizeof(*sin); 1705 sin->sin_addr = *addr_p; 1706 sin->sin_port = port; 1707 1708 return (struct sockaddr *)sin; 1709 } 1710 1711 int 1712 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1713 { 1714 struct inpcb *inp; 1715 struct in_addr addr; 1716 in_port_t port; 1717 1718 inp = sotoinpcb(so); 1719 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1720 1721 INP_RLOCK(inp); 1722 port = inp->inp_lport; 1723 addr = inp->inp_laddr; 1724 INP_RUNLOCK(inp); 1725 1726 *nam = in_sockaddr(port, &addr); 1727 return 0; 1728 } 1729 1730 int 1731 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1732 { 1733 struct inpcb *inp; 1734 struct in_addr addr; 1735 in_port_t port; 1736 1737 inp = sotoinpcb(so); 1738 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1739 1740 INP_RLOCK(inp); 1741 port = inp->inp_fport; 1742 addr = inp->inp_faddr; 1743 INP_RUNLOCK(inp); 1744 1745 *nam = in_sockaddr(port, &addr); 1746 return 0; 1747 } 1748 1749 void 1750 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1751 struct inpcb *(*notify)(struct inpcb *, int)) 1752 { 1753 struct inpcb *inp, *inp_temp; 1754 1755 INP_INFO_WLOCK(pcbinfo); 1756 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1757 INP_WLOCK(inp); 1758 #ifdef INET6 1759 if ((inp->inp_vflag & INP_IPV4) == 0) { 1760 INP_WUNLOCK(inp); 1761 continue; 1762 } 1763 #endif 1764 if (inp->inp_faddr.s_addr != faddr.s_addr || 1765 inp->inp_socket == NULL) { 1766 INP_WUNLOCK(inp); 1767 continue; 1768 } 1769 if ((*notify)(inp, errno)) 1770 INP_WUNLOCK(inp); 1771 } 1772 INP_INFO_WUNLOCK(pcbinfo); 1773 } 1774 1775 void 1776 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1777 { 1778 struct inpcb *inp; 1779 struct ip_moptions *imo; 1780 int i, gap; 1781 1782 INP_INFO_WLOCK(pcbinfo); 1783 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1784 INP_WLOCK(inp); 1785 imo = inp->inp_moptions; 1786 if ((inp->inp_vflag & INP_IPV4) && 1787 imo != NULL) { 1788 /* 1789 * Unselect the outgoing interface if it is being 1790 * detached. 1791 */ 1792 if (imo->imo_multicast_ifp == ifp) 1793 imo->imo_multicast_ifp = NULL; 1794 1795 /* 1796 * Drop multicast group membership if we joined 1797 * through the interface being detached. 1798 * 1799 * XXX This can all be deferred to an epoch_call 1800 */ 1801 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1802 i++) { 1803 if (imo->imo_membership[i]->inm_ifp == ifp) { 1804 IN_MULTI_LOCK_ASSERT(); 1805 in_leavegroup_locked(imo->imo_membership[i], NULL); 1806 gap++; 1807 } else if (gap != 0) 1808 imo->imo_membership[i - gap] = 1809 imo->imo_membership[i]; 1810 } 1811 imo->imo_num_memberships -= gap; 1812 } 1813 INP_WUNLOCK(inp); 1814 } 1815 INP_INFO_WUNLOCK(pcbinfo); 1816 } 1817 1818 /* 1819 * Lookup a PCB based on the local address and port. Caller must hold the 1820 * hash lock. No inpcb locks or references are acquired. 1821 */ 1822 #define INP_LOOKUP_MAPPED_PCB_COST 3 1823 struct inpcb * 1824 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1825 u_short lport, int lookupflags, struct ucred *cred) 1826 { 1827 struct inpcb *inp; 1828 #ifdef INET6 1829 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1830 #else 1831 int matchwild = 3; 1832 #endif 1833 int wildcard; 1834 1835 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1836 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1837 1838 INP_HASH_LOCK_ASSERT(pcbinfo); 1839 1840 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1841 struct inpcbhead *head; 1842 /* 1843 * Look for an unconnected (wildcard foreign addr) PCB that 1844 * matches the local address and port we're looking for. 1845 */ 1846 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1847 0, pcbinfo->ipi_hashmask)]; 1848 CK_LIST_FOREACH(inp, head, inp_hash) { 1849 #ifdef INET6 1850 /* XXX inp locking */ 1851 if ((inp->inp_vflag & INP_IPV4) == 0) 1852 continue; 1853 #endif 1854 if (inp->inp_faddr.s_addr == INADDR_ANY && 1855 inp->inp_laddr.s_addr == laddr.s_addr && 1856 inp->inp_lport == lport) { 1857 /* 1858 * Found? 1859 */ 1860 if (cred == NULL || 1861 prison_equal_ip4(cred->cr_prison, 1862 inp->inp_cred->cr_prison)) 1863 return (inp); 1864 } 1865 } 1866 /* 1867 * Not found. 1868 */ 1869 return (NULL); 1870 } else { 1871 struct inpcbporthead *porthash; 1872 struct inpcbport *phd; 1873 struct inpcb *match = NULL; 1874 /* 1875 * Best fit PCB lookup. 1876 * 1877 * First see if this local port is in use by looking on the 1878 * port hash list. 1879 */ 1880 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1881 pcbinfo->ipi_porthashmask)]; 1882 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1883 if (phd->phd_port == lport) 1884 break; 1885 } 1886 if (phd != NULL) { 1887 /* 1888 * Port is in use by one or more PCBs. Look for best 1889 * fit. 1890 */ 1891 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1892 wildcard = 0; 1893 if (cred != NULL && 1894 !prison_equal_ip4(inp->inp_cred->cr_prison, 1895 cred->cr_prison)) 1896 continue; 1897 #ifdef INET6 1898 /* XXX inp locking */ 1899 if ((inp->inp_vflag & INP_IPV4) == 0) 1900 continue; 1901 /* 1902 * We never select the PCB that has 1903 * INP_IPV6 flag and is bound to :: if 1904 * we have another PCB which is bound 1905 * to 0.0.0.0. If a PCB has the 1906 * INP_IPV6 flag, then we set its cost 1907 * higher than IPv4 only PCBs. 1908 * 1909 * Note that the case only happens 1910 * when a socket is bound to ::, under 1911 * the condition that the use of the 1912 * mapped address is allowed. 1913 */ 1914 if ((inp->inp_vflag & INP_IPV6) != 0) 1915 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1916 #endif 1917 if (inp->inp_faddr.s_addr != INADDR_ANY) 1918 wildcard++; 1919 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1920 if (laddr.s_addr == INADDR_ANY) 1921 wildcard++; 1922 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1923 continue; 1924 } else { 1925 if (laddr.s_addr != INADDR_ANY) 1926 wildcard++; 1927 } 1928 if (wildcard < matchwild) { 1929 match = inp; 1930 matchwild = wildcard; 1931 if (matchwild == 0) 1932 break; 1933 } 1934 } 1935 } 1936 return (match); 1937 } 1938 } 1939 #undef INP_LOOKUP_MAPPED_PCB_COST 1940 1941 static struct inpcb * 1942 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 1943 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, 1944 uint16_t fport, int lookupflags) 1945 { 1946 struct inpcb *local_wild; 1947 const struct inpcblbgrouphead *hdr; 1948 struct inpcblbgroup *grp; 1949 uint32_t idx; 1950 1951 INP_HASH_LOCK_ASSERT(pcbinfo); 1952 1953 hdr = &pcbinfo->ipi_lbgrouphashbase[ 1954 INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 1955 1956 /* 1957 * Order of socket selection: 1958 * 1. non-wild. 1959 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). 1960 * 1961 * NOTE: 1962 * - Load balanced group does not contain jailed sockets 1963 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets 1964 */ 1965 local_wild = NULL; 1966 CK_LIST_FOREACH(grp, hdr, il_list) { 1967 #ifdef INET6 1968 if (!(grp->il_vflag & INP_IPV4)) 1969 continue; 1970 #endif 1971 if (grp->il_lport != lport) 1972 continue; 1973 1974 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % 1975 grp->il_inpcnt; 1976 if (grp->il_laddr.s_addr == laddr->s_addr) 1977 return (grp->il_inp[idx]); 1978 if (grp->il_laddr.s_addr == INADDR_ANY && 1979 (lookupflags & INPLOOKUP_WILDCARD) != 0) 1980 local_wild = grp->il_inp[idx]; 1981 } 1982 return (local_wild); 1983 } 1984 1985 #ifdef PCBGROUP 1986 /* 1987 * Lookup PCB in hash list, using pcbgroup tables. 1988 */ 1989 static struct inpcb * 1990 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 1991 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 1992 u_int lport_arg, int lookupflags, struct ifnet *ifp) 1993 { 1994 struct inpcbhead *head; 1995 struct inpcb *inp, *tmpinp; 1996 u_short fport = fport_arg, lport = lport_arg; 1997 bool locked; 1998 1999 /* 2000 * First look for an exact match. 2001 */ 2002 tmpinp = NULL; 2003 INP_GROUP_LOCK(pcbgroup); 2004 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2005 pcbgroup->ipg_hashmask)]; 2006 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2007 #ifdef INET6 2008 /* XXX inp locking */ 2009 if ((inp->inp_vflag & INP_IPV4) == 0) 2010 continue; 2011 #endif 2012 if (inp->inp_faddr.s_addr == faddr.s_addr && 2013 inp->inp_laddr.s_addr == laddr.s_addr && 2014 inp->inp_fport == fport && 2015 inp->inp_lport == lport) { 2016 /* 2017 * XXX We should be able to directly return 2018 * the inp here, without any checks. 2019 * Well unless both bound with SO_REUSEPORT? 2020 */ 2021 if (prison_flag(inp->inp_cred, PR_IP4)) 2022 goto found; 2023 if (tmpinp == NULL) 2024 tmpinp = inp; 2025 } 2026 } 2027 if (tmpinp != NULL) { 2028 inp = tmpinp; 2029 goto found; 2030 } 2031 2032 #ifdef RSS 2033 /* 2034 * For incoming connections, we may wish to do a wildcard 2035 * match for an RSS-local socket. 2036 */ 2037 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2038 struct inpcb *local_wild = NULL, *local_exact = NULL; 2039 #ifdef INET6 2040 struct inpcb *local_wild_mapped = NULL; 2041 #endif 2042 struct inpcb *jail_wild = NULL; 2043 struct inpcbhead *head; 2044 int injail; 2045 2046 /* 2047 * Order of socket selection - we always prefer jails. 2048 * 1. jailed, non-wild. 2049 * 2. jailed, wild. 2050 * 3. non-jailed, non-wild. 2051 * 4. non-jailed, wild. 2052 */ 2053 2054 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 2055 lport, 0, pcbgroup->ipg_hashmask)]; 2056 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2057 #ifdef INET6 2058 /* XXX inp locking */ 2059 if ((inp->inp_vflag & INP_IPV4) == 0) 2060 continue; 2061 #endif 2062 if (inp->inp_faddr.s_addr != INADDR_ANY || 2063 inp->inp_lport != lport) 2064 continue; 2065 2066 injail = prison_flag(inp->inp_cred, PR_IP4); 2067 if (injail) { 2068 if (prison_check_ip4(inp->inp_cred, 2069 &laddr) != 0) 2070 continue; 2071 } else { 2072 if (local_exact != NULL) 2073 continue; 2074 } 2075 2076 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2077 if (injail) 2078 goto found; 2079 else 2080 local_exact = inp; 2081 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2082 #ifdef INET6 2083 /* XXX inp locking, NULL check */ 2084 if (inp->inp_vflag & INP_IPV6PROTO) 2085 local_wild_mapped = inp; 2086 else 2087 #endif 2088 if (injail) 2089 jail_wild = inp; 2090 else 2091 local_wild = inp; 2092 } 2093 } /* LIST_FOREACH */ 2094 2095 inp = jail_wild; 2096 if (inp == NULL) 2097 inp = local_exact; 2098 if (inp == NULL) 2099 inp = local_wild; 2100 #ifdef INET6 2101 if (inp == NULL) 2102 inp = local_wild_mapped; 2103 #endif 2104 if (inp != NULL) 2105 goto found; 2106 } 2107 #endif 2108 2109 /* 2110 * Then look for a wildcard match, if requested. 2111 */ 2112 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2113 struct inpcb *local_wild = NULL, *local_exact = NULL; 2114 #ifdef INET6 2115 struct inpcb *local_wild_mapped = NULL; 2116 #endif 2117 struct inpcb *jail_wild = NULL; 2118 struct inpcbhead *head; 2119 int injail; 2120 2121 /* 2122 * Order of socket selection - we always prefer jails. 2123 * 1. jailed, non-wild. 2124 * 2. jailed, wild. 2125 * 3. non-jailed, non-wild. 2126 * 4. non-jailed, wild. 2127 */ 2128 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 2129 0, pcbinfo->ipi_wildmask)]; 2130 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 2131 #ifdef INET6 2132 /* XXX inp locking */ 2133 if ((inp->inp_vflag & INP_IPV4) == 0) 2134 continue; 2135 #endif 2136 if (inp->inp_faddr.s_addr != INADDR_ANY || 2137 inp->inp_lport != lport) 2138 continue; 2139 2140 injail = prison_flag(inp->inp_cred, PR_IP4); 2141 if (injail) { 2142 if (prison_check_ip4(inp->inp_cred, 2143 &laddr) != 0) 2144 continue; 2145 } else { 2146 if (local_exact != NULL) 2147 continue; 2148 } 2149 2150 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2151 if (injail) 2152 goto found; 2153 else 2154 local_exact = inp; 2155 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2156 #ifdef INET6 2157 /* XXX inp locking, NULL check */ 2158 if (inp->inp_vflag & INP_IPV6PROTO) 2159 local_wild_mapped = inp; 2160 else 2161 #endif 2162 if (injail) 2163 jail_wild = inp; 2164 else 2165 local_wild = inp; 2166 } 2167 } /* LIST_FOREACH */ 2168 inp = jail_wild; 2169 if (inp == NULL) 2170 inp = local_exact; 2171 if (inp == NULL) 2172 inp = local_wild; 2173 #ifdef INET6 2174 if (inp == NULL) 2175 inp = local_wild_mapped; 2176 #endif 2177 if (inp != NULL) 2178 goto found; 2179 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 2180 INP_GROUP_UNLOCK(pcbgroup); 2181 return (NULL); 2182 2183 found: 2184 if (lookupflags & INPLOOKUP_WLOCKPCB) 2185 locked = INP_TRY_WLOCK(inp); 2186 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2187 locked = INP_TRY_RLOCK(inp); 2188 else 2189 panic("%s: locking bug", __func__); 2190 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { 2191 if (lookupflags & INPLOOKUP_WLOCKPCB) 2192 INP_WUNLOCK(inp); 2193 else 2194 INP_RUNLOCK(inp); 2195 return (NULL); 2196 } else if (!locked) 2197 in_pcbref(inp); 2198 INP_GROUP_UNLOCK(pcbgroup); 2199 if (!locked) { 2200 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2201 INP_WLOCK(inp); 2202 if (in_pcbrele_wlocked(inp)) 2203 return (NULL); 2204 } else { 2205 INP_RLOCK(inp); 2206 if (in_pcbrele_rlocked(inp)) 2207 return (NULL); 2208 } 2209 } 2210 #ifdef INVARIANTS 2211 if (lookupflags & INPLOOKUP_WLOCKPCB) 2212 INP_WLOCK_ASSERT(inp); 2213 else 2214 INP_RLOCK_ASSERT(inp); 2215 #endif 2216 return (inp); 2217 } 2218 #endif /* PCBGROUP */ 2219 2220 /* 2221 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2222 * that the caller has locked the hash list, and will not perform any further 2223 * locking or reference operations on either the hash list or the connection. 2224 */ 2225 static struct inpcb * 2226 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2227 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2228 struct ifnet *ifp) 2229 { 2230 struct inpcbhead *head; 2231 struct inpcb *inp, *tmpinp; 2232 u_short fport = fport_arg, lport = lport_arg; 2233 2234 #ifdef INVARIANTS 2235 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2236 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2237 if (!mtx_owned(&pcbinfo->ipi_hash_lock)) 2238 MPASS(in_epoch_verbose(net_epoch_preempt, 1)); 2239 #endif 2240 /* 2241 * First look for an exact match. 2242 */ 2243 tmpinp = NULL; 2244 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2245 pcbinfo->ipi_hashmask)]; 2246 CK_LIST_FOREACH(inp, head, inp_hash) { 2247 #ifdef INET6 2248 /* XXX inp locking */ 2249 if ((inp->inp_vflag & INP_IPV4) == 0) 2250 continue; 2251 #endif 2252 if (inp->inp_faddr.s_addr == faddr.s_addr && 2253 inp->inp_laddr.s_addr == laddr.s_addr && 2254 inp->inp_fport == fport && 2255 inp->inp_lport == lport) { 2256 /* 2257 * XXX We should be able to directly return 2258 * the inp here, without any checks. 2259 * Well unless both bound with SO_REUSEPORT? 2260 */ 2261 if (prison_flag(inp->inp_cred, PR_IP4)) 2262 return (inp); 2263 if (tmpinp == NULL) 2264 tmpinp = inp; 2265 } 2266 } 2267 if (tmpinp != NULL) 2268 return (tmpinp); 2269 2270 /* 2271 * Then look in lb group (for wildcard match). 2272 */ 2273 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2274 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, 2275 fport, lookupflags); 2276 if (inp != NULL) 2277 return (inp); 2278 } 2279 2280 /* 2281 * Then look for a wildcard match, if requested. 2282 */ 2283 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2284 struct inpcb *local_wild = NULL, *local_exact = NULL; 2285 #ifdef INET6 2286 struct inpcb *local_wild_mapped = NULL; 2287 #endif 2288 struct inpcb *jail_wild = NULL; 2289 int injail; 2290 2291 /* 2292 * Order of socket selection - we always prefer jails. 2293 * 1. jailed, non-wild. 2294 * 2. jailed, wild. 2295 * 3. non-jailed, non-wild. 2296 * 4. non-jailed, wild. 2297 */ 2298 2299 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2300 0, pcbinfo->ipi_hashmask)]; 2301 CK_LIST_FOREACH(inp, head, inp_hash) { 2302 #ifdef INET6 2303 /* XXX inp locking */ 2304 if ((inp->inp_vflag & INP_IPV4) == 0) 2305 continue; 2306 #endif 2307 if (inp->inp_faddr.s_addr != INADDR_ANY || 2308 inp->inp_lport != lport) 2309 continue; 2310 2311 injail = prison_flag(inp->inp_cred, PR_IP4); 2312 if (injail) { 2313 if (prison_check_ip4(inp->inp_cred, 2314 &laddr) != 0) 2315 continue; 2316 } else { 2317 if (local_exact != NULL) 2318 continue; 2319 } 2320 2321 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2322 if (injail) 2323 return (inp); 2324 else 2325 local_exact = inp; 2326 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2327 #ifdef INET6 2328 /* XXX inp locking, NULL check */ 2329 if (inp->inp_vflag & INP_IPV6PROTO) 2330 local_wild_mapped = inp; 2331 else 2332 #endif 2333 if (injail) 2334 jail_wild = inp; 2335 else 2336 local_wild = inp; 2337 } 2338 } /* LIST_FOREACH */ 2339 if (jail_wild != NULL) 2340 return (jail_wild); 2341 if (local_exact != NULL) 2342 return (local_exact); 2343 if (local_wild != NULL) 2344 return (local_wild); 2345 #ifdef INET6 2346 if (local_wild_mapped != NULL) 2347 return (local_wild_mapped); 2348 #endif 2349 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2350 2351 return (NULL); 2352 } 2353 2354 /* 2355 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2356 * hash list lock, and will return the inpcb locked (i.e., requires 2357 * INPLOOKUP_LOCKPCB). 2358 */ 2359 static struct inpcb * 2360 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2361 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2362 struct ifnet *ifp) 2363 { 2364 struct inpcb *inp; 2365 2366 INP_HASH_RLOCK(pcbinfo); 2367 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2368 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 2369 if (inp != NULL) { 2370 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2371 INP_WLOCK(inp); 2372 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2373 INP_WUNLOCK(inp); 2374 inp = NULL; 2375 } 2376 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2377 INP_RLOCK(inp); 2378 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2379 INP_RUNLOCK(inp); 2380 inp = NULL; 2381 } 2382 } else 2383 panic("%s: locking bug", __func__); 2384 #ifdef INVARIANTS 2385 if (inp != NULL) { 2386 if (lookupflags & INPLOOKUP_WLOCKPCB) 2387 INP_WLOCK_ASSERT(inp); 2388 else 2389 INP_RLOCK_ASSERT(inp); 2390 } 2391 #endif 2392 } 2393 INP_HASH_RUNLOCK(pcbinfo); 2394 return (inp); 2395 } 2396 2397 /* 2398 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2399 * from which a pre-calculated hash value may be extracted. 2400 * 2401 * Possibly more of this logic should be in in_pcbgroup.c. 2402 */ 2403 struct inpcb * 2404 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2405 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2406 { 2407 #if defined(PCBGROUP) && !defined(RSS) 2408 struct inpcbgroup *pcbgroup; 2409 #endif 2410 2411 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2412 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2413 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2414 ("%s: LOCKPCB not set", __func__)); 2415 2416 /* 2417 * When not using RSS, use connection groups in preference to the 2418 * reservation table when looking up 4-tuples. When using RSS, just 2419 * use the reservation table, due to the cost of the Toeplitz hash 2420 * in software. 2421 * 2422 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2423 * we could be doing RSS with a non-Toeplitz hash that is affordable 2424 * in software. 2425 */ 2426 #if defined(PCBGROUP) && !defined(RSS) 2427 if (in_pcbgroup_enabled(pcbinfo)) { 2428 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2429 fport); 2430 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2431 laddr, lport, lookupflags, ifp)); 2432 } 2433 #endif 2434 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2435 lookupflags, ifp)); 2436 } 2437 2438 struct inpcb * 2439 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2440 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2441 struct ifnet *ifp, struct mbuf *m) 2442 { 2443 #ifdef PCBGROUP 2444 struct inpcbgroup *pcbgroup; 2445 #endif 2446 2447 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2448 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2449 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2450 ("%s: LOCKPCB not set", __func__)); 2451 2452 #ifdef PCBGROUP 2453 /* 2454 * If we can use a hardware-generated hash to look up the connection 2455 * group, use that connection group to find the inpcb. Otherwise 2456 * fall back on a software hash -- or the reservation table if we're 2457 * using RSS. 2458 * 2459 * XXXRW: As above, that policy belongs in the pcbgroup code. 2460 */ 2461 if (in_pcbgroup_enabled(pcbinfo) && 2462 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2463 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2464 m->m_pkthdr.flowid); 2465 if (pcbgroup != NULL) 2466 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2467 fport, laddr, lport, lookupflags, ifp)); 2468 #ifndef RSS 2469 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2470 fport); 2471 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2472 laddr, lport, lookupflags, ifp)); 2473 #endif 2474 } 2475 #endif 2476 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2477 lookupflags, ifp)); 2478 } 2479 #endif /* INET */ 2480 2481 /* 2482 * Insert PCB onto various hash lists. 2483 */ 2484 static int 2485 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 2486 { 2487 struct inpcbhead *pcbhash; 2488 struct inpcbporthead *pcbporthash; 2489 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2490 struct inpcbport *phd; 2491 u_int32_t hashkey_faddr; 2492 int so_options; 2493 2494 INP_WLOCK_ASSERT(inp); 2495 INP_HASH_WLOCK_ASSERT(pcbinfo); 2496 2497 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2498 ("in_pcbinshash: INP_INHASHLIST")); 2499 2500 #ifdef INET6 2501 if (inp->inp_vflag & INP_IPV6) 2502 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2503 else 2504 #endif 2505 hashkey_faddr = inp->inp_faddr.s_addr; 2506 2507 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2508 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2509 2510 pcbporthash = &pcbinfo->ipi_porthashbase[ 2511 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2512 2513 /* 2514 * Add entry to load balance group. 2515 * Only do this if SO_REUSEPORT_LB is set. 2516 */ 2517 so_options = inp_so_options(inp); 2518 if (so_options & SO_REUSEPORT_LB) { 2519 int ret = in_pcbinslbgrouphash(inp); 2520 if (ret) { 2521 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2522 return (ret); 2523 } 2524 } 2525 2526 /* 2527 * Go through port list and look for a head for this lport. 2528 */ 2529 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2530 if (phd->phd_port == inp->inp_lport) 2531 break; 2532 } 2533 /* 2534 * If none exists, malloc one and tack it on. 2535 */ 2536 if (phd == NULL) { 2537 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2538 if (phd == NULL) { 2539 return (ENOBUFS); /* XXX */ 2540 } 2541 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2542 phd->phd_port = inp->inp_lport; 2543 CK_LIST_INIT(&phd->phd_pcblist); 2544 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2545 } 2546 inp->inp_phd = phd; 2547 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2548 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2549 inp->inp_flags |= INP_INHASHLIST; 2550 #ifdef PCBGROUP 2551 if (do_pcbgroup_update) 2552 in_pcbgroup_update(inp); 2553 #endif 2554 return (0); 2555 } 2556 2557 /* 2558 * For now, there are two public interfaces to insert an inpcb into the hash 2559 * lists -- one that does update pcbgroups, and one that doesn't. The latter 2560 * is used only in the TCP syncache, where in_pcbinshash is called before the 2561 * full 4-tuple is set for the inpcb, and we don't want to install in the 2562 * pcbgroup until later. 2563 * 2564 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 2565 * connection groups, and partially initialised inpcbs should not be exposed 2566 * to either reservation hash tables or pcbgroups. 2567 */ 2568 int 2569 in_pcbinshash(struct inpcb *inp) 2570 { 2571 2572 return (in_pcbinshash_internal(inp, 1)); 2573 } 2574 2575 int 2576 in_pcbinshash_nopcbgroup(struct inpcb *inp) 2577 { 2578 2579 return (in_pcbinshash_internal(inp, 0)); 2580 } 2581 2582 /* 2583 * Move PCB to the proper hash bucket when { faddr, fport } have been 2584 * changed. NOTE: This does not handle the case of the lport changing (the 2585 * hashed port list would have to be updated as well), so the lport must 2586 * not change after in_pcbinshash() has been called. 2587 */ 2588 void 2589 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2590 { 2591 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2592 struct inpcbhead *head; 2593 u_int32_t hashkey_faddr; 2594 2595 INP_WLOCK_ASSERT(inp); 2596 INP_HASH_WLOCK_ASSERT(pcbinfo); 2597 2598 KASSERT(inp->inp_flags & INP_INHASHLIST, 2599 ("in_pcbrehash: !INP_INHASHLIST")); 2600 2601 #ifdef INET6 2602 if (inp->inp_vflag & INP_IPV6) 2603 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2604 else 2605 #endif 2606 hashkey_faddr = inp->inp_faddr.s_addr; 2607 2608 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2609 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2610 2611 CK_LIST_REMOVE(inp, inp_hash); 2612 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2613 2614 #ifdef PCBGROUP 2615 if (m != NULL) 2616 in_pcbgroup_update_mbuf(inp, m); 2617 else 2618 in_pcbgroup_update(inp); 2619 #endif 2620 } 2621 2622 void 2623 in_pcbrehash(struct inpcb *inp) 2624 { 2625 2626 in_pcbrehash_mbuf(inp, NULL); 2627 } 2628 2629 /* 2630 * Remove PCB from various lists. 2631 */ 2632 static void 2633 in_pcbremlists(struct inpcb *inp) 2634 { 2635 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2636 2637 #ifdef INVARIANTS 2638 if (pcbinfo == &V_tcbinfo) { 2639 INP_INFO_RLOCK_ASSERT(pcbinfo); 2640 } else { 2641 INP_INFO_WLOCK_ASSERT(pcbinfo); 2642 } 2643 #endif 2644 2645 INP_WLOCK_ASSERT(inp); 2646 INP_LIST_WLOCK_ASSERT(pcbinfo); 2647 2648 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2649 if (inp->inp_flags & INP_INHASHLIST) { 2650 struct inpcbport *phd = inp->inp_phd; 2651 2652 INP_HASH_WLOCK(pcbinfo); 2653 2654 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2655 in_pcbremlbgrouphash(inp); 2656 2657 CK_LIST_REMOVE(inp, inp_hash); 2658 CK_LIST_REMOVE(inp, inp_portlist); 2659 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2660 CK_LIST_REMOVE(phd, phd_hash); 2661 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 2662 } 2663 INP_HASH_WUNLOCK(pcbinfo); 2664 inp->inp_flags &= ~INP_INHASHLIST; 2665 } 2666 CK_LIST_REMOVE(inp, inp_list); 2667 pcbinfo->ipi_count--; 2668 #ifdef PCBGROUP 2669 in_pcbgroup_remove(inp); 2670 #endif 2671 } 2672 2673 /* 2674 * Check for alternatives when higher level complains 2675 * about service problems. For now, invalidate cached 2676 * routing information. If the route was created dynamically 2677 * (by a redirect), time to try a default gateway again. 2678 */ 2679 void 2680 in_losing(struct inpcb *inp) 2681 { 2682 2683 RO_INVALIDATE_CACHE(&inp->inp_route); 2684 return; 2685 } 2686 2687 /* 2688 * A set label operation has occurred at the socket layer, propagate the 2689 * label change into the in_pcb for the socket. 2690 */ 2691 void 2692 in_pcbsosetlabel(struct socket *so) 2693 { 2694 #ifdef MAC 2695 struct inpcb *inp; 2696 2697 inp = sotoinpcb(so); 2698 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2699 2700 INP_WLOCK(inp); 2701 SOCK_LOCK(so); 2702 mac_inpcb_sosetlabel(so, inp); 2703 SOCK_UNLOCK(so); 2704 INP_WUNLOCK(inp); 2705 #endif 2706 } 2707 2708 /* 2709 * ipport_tick runs once per second, determining if random port allocation 2710 * should be continued. If more than ipport_randomcps ports have been 2711 * allocated in the last second, then we return to sequential port 2712 * allocation. We return to random allocation only once we drop below 2713 * ipport_randomcps for at least ipport_randomtime seconds. 2714 */ 2715 static void 2716 ipport_tick(void *xtp) 2717 { 2718 VNET_ITERATOR_DECL(vnet_iter); 2719 2720 VNET_LIST_RLOCK_NOSLEEP(); 2721 VNET_FOREACH(vnet_iter) { 2722 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2723 if (V_ipport_tcpallocs <= 2724 V_ipport_tcplastcount + V_ipport_randomcps) { 2725 if (V_ipport_stoprandom > 0) 2726 V_ipport_stoprandom--; 2727 } else 2728 V_ipport_stoprandom = V_ipport_randomtime; 2729 V_ipport_tcplastcount = V_ipport_tcpallocs; 2730 CURVNET_RESTORE(); 2731 } 2732 VNET_LIST_RUNLOCK_NOSLEEP(); 2733 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2734 } 2735 2736 static void 2737 ip_fini(void *xtp) 2738 { 2739 2740 callout_stop(&ipport_tick_callout); 2741 } 2742 2743 /* 2744 * The ipport_callout should start running at about the time we attach the 2745 * inet or inet6 domains. 2746 */ 2747 static void 2748 ipport_tick_init(const void *unused __unused) 2749 { 2750 2751 /* Start ipport_tick. */ 2752 callout_init(&ipport_tick_callout, 1); 2753 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2754 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2755 SHUTDOWN_PRI_DEFAULT); 2756 } 2757 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2758 ipport_tick_init, NULL); 2759 2760 void 2761 inp_wlock(struct inpcb *inp) 2762 { 2763 2764 INP_WLOCK(inp); 2765 } 2766 2767 void 2768 inp_wunlock(struct inpcb *inp) 2769 { 2770 2771 INP_WUNLOCK(inp); 2772 } 2773 2774 void 2775 inp_rlock(struct inpcb *inp) 2776 { 2777 2778 INP_RLOCK(inp); 2779 } 2780 2781 void 2782 inp_runlock(struct inpcb *inp) 2783 { 2784 2785 INP_RUNLOCK(inp); 2786 } 2787 2788 #ifdef INVARIANT_SUPPORT 2789 void 2790 inp_lock_assert(struct inpcb *inp) 2791 { 2792 2793 INP_WLOCK_ASSERT(inp); 2794 } 2795 2796 void 2797 inp_unlock_assert(struct inpcb *inp) 2798 { 2799 2800 INP_UNLOCK_ASSERT(inp); 2801 } 2802 #endif 2803 2804 void 2805 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2806 { 2807 struct inpcb *inp; 2808 2809 INP_INFO_WLOCK(&V_tcbinfo); 2810 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2811 INP_WLOCK(inp); 2812 func(inp, arg); 2813 INP_WUNLOCK(inp); 2814 } 2815 INP_INFO_WUNLOCK(&V_tcbinfo); 2816 } 2817 2818 struct socket * 2819 inp_inpcbtosocket(struct inpcb *inp) 2820 { 2821 2822 INP_WLOCK_ASSERT(inp); 2823 return (inp->inp_socket); 2824 } 2825 2826 struct tcpcb * 2827 inp_inpcbtotcpcb(struct inpcb *inp) 2828 { 2829 2830 INP_WLOCK_ASSERT(inp); 2831 return ((struct tcpcb *)inp->inp_ppcb); 2832 } 2833 2834 int 2835 inp_ip_tos_get(const struct inpcb *inp) 2836 { 2837 2838 return (inp->inp_ip_tos); 2839 } 2840 2841 void 2842 inp_ip_tos_set(struct inpcb *inp, int val) 2843 { 2844 2845 inp->inp_ip_tos = val; 2846 } 2847 2848 void 2849 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2850 uint32_t *faddr, uint16_t *fp) 2851 { 2852 2853 INP_LOCK_ASSERT(inp); 2854 *laddr = inp->inp_laddr.s_addr; 2855 *faddr = inp->inp_faddr.s_addr; 2856 *lp = inp->inp_lport; 2857 *fp = inp->inp_fport; 2858 } 2859 2860 struct inpcb * 2861 so_sotoinpcb(struct socket *so) 2862 { 2863 2864 return (sotoinpcb(so)); 2865 } 2866 2867 struct tcpcb * 2868 so_sototcpcb(struct socket *so) 2869 { 2870 2871 return (sototcpcb(so)); 2872 } 2873 2874 /* 2875 * Create an external-format (``xinpcb'') structure using the information in 2876 * the kernel-format in_pcb structure pointed to by inp. This is done to 2877 * reduce the spew of irrelevant information over this interface, to isolate 2878 * user code from changes in the kernel structure, and potentially to provide 2879 * information-hiding if we decide that some of this information should be 2880 * hidden from users. 2881 */ 2882 void 2883 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2884 { 2885 2886 bzero(xi, sizeof(*xi)); 2887 xi->xi_len = sizeof(struct xinpcb); 2888 if (inp->inp_socket) 2889 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2890 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2891 xi->inp_gencnt = inp->inp_gencnt; 2892 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2893 xi->inp_flow = inp->inp_flow; 2894 xi->inp_flowid = inp->inp_flowid; 2895 xi->inp_flowtype = inp->inp_flowtype; 2896 xi->inp_flags = inp->inp_flags; 2897 xi->inp_flags2 = inp->inp_flags2; 2898 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2899 xi->in6p_cksum = inp->in6p_cksum; 2900 xi->in6p_hops = inp->in6p_hops; 2901 xi->inp_ip_tos = inp->inp_ip_tos; 2902 xi->inp_vflag = inp->inp_vflag; 2903 xi->inp_ip_ttl = inp->inp_ip_ttl; 2904 xi->inp_ip_p = inp->inp_ip_p; 2905 xi->inp_ip_minttl = inp->inp_ip_minttl; 2906 } 2907 2908 #ifdef DDB 2909 static void 2910 db_print_indent(int indent) 2911 { 2912 int i; 2913 2914 for (i = 0; i < indent; i++) 2915 db_printf(" "); 2916 } 2917 2918 static void 2919 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2920 { 2921 char faddr_str[48], laddr_str[48]; 2922 2923 db_print_indent(indent); 2924 db_printf("%s at %p\n", name, inc); 2925 2926 indent += 2; 2927 2928 #ifdef INET6 2929 if (inc->inc_flags & INC_ISIPV6) { 2930 /* IPv6. */ 2931 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2932 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2933 } else 2934 #endif 2935 { 2936 /* IPv4. */ 2937 inet_ntoa_r(inc->inc_laddr, laddr_str); 2938 inet_ntoa_r(inc->inc_faddr, faddr_str); 2939 } 2940 db_print_indent(indent); 2941 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2942 ntohs(inc->inc_lport)); 2943 db_print_indent(indent); 2944 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2945 ntohs(inc->inc_fport)); 2946 } 2947 2948 static void 2949 db_print_inpflags(int inp_flags) 2950 { 2951 int comma; 2952 2953 comma = 0; 2954 if (inp_flags & INP_RECVOPTS) { 2955 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2956 comma = 1; 2957 } 2958 if (inp_flags & INP_RECVRETOPTS) { 2959 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2960 comma = 1; 2961 } 2962 if (inp_flags & INP_RECVDSTADDR) { 2963 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2964 comma = 1; 2965 } 2966 if (inp_flags & INP_ORIGDSTADDR) { 2967 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2968 comma = 1; 2969 } 2970 if (inp_flags & INP_HDRINCL) { 2971 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2972 comma = 1; 2973 } 2974 if (inp_flags & INP_HIGHPORT) { 2975 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2976 comma = 1; 2977 } 2978 if (inp_flags & INP_LOWPORT) { 2979 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 2980 comma = 1; 2981 } 2982 if (inp_flags & INP_ANONPORT) { 2983 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 2984 comma = 1; 2985 } 2986 if (inp_flags & INP_RECVIF) { 2987 db_printf("%sINP_RECVIF", comma ? ", " : ""); 2988 comma = 1; 2989 } 2990 if (inp_flags & INP_MTUDISC) { 2991 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 2992 comma = 1; 2993 } 2994 if (inp_flags & INP_RECVTTL) { 2995 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 2996 comma = 1; 2997 } 2998 if (inp_flags & INP_DONTFRAG) { 2999 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3000 comma = 1; 3001 } 3002 if (inp_flags & INP_RECVTOS) { 3003 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3004 comma = 1; 3005 } 3006 if (inp_flags & IN6P_IPV6_V6ONLY) { 3007 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3008 comma = 1; 3009 } 3010 if (inp_flags & IN6P_PKTINFO) { 3011 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3012 comma = 1; 3013 } 3014 if (inp_flags & IN6P_HOPLIMIT) { 3015 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3016 comma = 1; 3017 } 3018 if (inp_flags & IN6P_HOPOPTS) { 3019 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3020 comma = 1; 3021 } 3022 if (inp_flags & IN6P_DSTOPTS) { 3023 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3024 comma = 1; 3025 } 3026 if (inp_flags & IN6P_RTHDR) { 3027 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3028 comma = 1; 3029 } 3030 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3031 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3032 comma = 1; 3033 } 3034 if (inp_flags & IN6P_TCLASS) { 3035 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3036 comma = 1; 3037 } 3038 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3039 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3040 comma = 1; 3041 } 3042 if (inp_flags & INP_TIMEWAIT) { 3043 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3044 comma = 1; 3045 } 3046 if (inp_flags & INP_ONESBCAST) { 3047 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3048 comma = 1; 3049 } 3050 if (inp_flags & INP_DROPPED) { 3051 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3052 comma = 1; 3053 } 3054 if (inp_flags & INP_SOCKREF) { 3055 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3056 comma = 1; 3057 } 3058 if (inp_flags & IN6P_RFC2292) { 3059 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3060 comma = 1; 3061 } 3062 if (inp_flags & IN6P_MTU) { 3063 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3064 comma = 1; 3065 } 3066 } 3067 3068 static void 3069 db_print_inpvflag(u_char inp_vflag) 3070 { 3071 int comma; 3072 3073 comma = 0; 3074 if (inp_vflag & INP_IPV4) { 3075 db_printf("%sINP_IPV4", comma ? ", " : ""); 3076 comma = 1; 3077 } 3078 if (inp_vflag & INP_IPV6) { 3079 db_printf("%sINP_IPV6", comma ? ", " : ""); 3080 comma = 1; 3081 } 3082 if (inp_vflag & INP_IPV6PROTO) { 3083 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3084 comma = 1; 3085 } 3086 } 3087 3088 static void 3089 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3090 { 3091 3092 db_print_indent(indent); 3093 db_printf("%s at %p\n", name, inp); 3094 3095 indent += 2; 3096 3097 db_print_indent(indent); 3098 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3099 3100 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3101 3102 db_print_indent(indent); 3103 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3104 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3105 3106 db_print_indent(indent); 3107 db_printf("inp_label: %p inp_flags: 0x%x (", 3108 inp->inp_label, inp->inp_flags); 3109 db_print_inpflags(inp->inp_flags); 3110 db_printf(")\n"); 3111 3112 db_print_indent(indent); 3113 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3114 inp->inp_vflag); 3115 db_print_inpvflag(inp->inp_vflag); 3116 db_printf(")\n"); 3117 3118 db_print_indent(indent); 3119 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3120 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3121 3122 db_print_indent(indent); 3123 #ifdef INET6 3124 if (inp->inp_vflag & INP_IPV6) { 3125 db_printf("in6p_options: %p in6p_outputopts: %p " 3126 "in6p_moptions: %p\n", inp->in6p_options, 3127 inp->in6p_outputopts, inp->in6p_moptions); 3128 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3129 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3130 inp->in6p_hops); 3131 } else 3132 #endif 3133 { 3134 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3135 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3136 inp->inp_options, inp->inp_moptions); 3137 } 3138 3139 db_print_indent(indent); 3140 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3141 (uintmax_t)inp->inp_gencnt); 3142 } 3143 3144 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3145 { 3146 struct inpcb *inp; 3147 3148 if (!have_addr) { 3149 db_printf("usage: show inpcb <addr>\n"); 3150 return; 3151 } 3152 inp = (struct inpcb *)addr; 3153 3154 db_print_inpcb(inp, "inpcb", 0); 3155 } 3156 #endif /* DDB */ 3157 3158 #ifdef RATELIMIT 3159 /* 3160 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3161 * if any. 3162 */ 3163 int 3164 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3165 { 3166 union if_snd_tag_modify_params params = { 3167 .rate_limit.max_rate = max_pacing_rate, 3168 }; 3169 struct m_snd_tag *mst; 3170 struct ifnet *ifp; 3171 int error; 3172 3173 mst = inp->inp_snd_tag; 3174 if (mst == NULL) 3175 return (EINVAL); 3176 3177 ifp = mst->ifp; 3178 if (ifp == NULL) 3179 return (EINVAL); 3180 3181 if (ifp->if_snd_tag_modify == NULL) { 3182 error = EOPNOTSUPP; 3183 } else { 3184 error = ifp->if_snd_tag_modify(mst, ¶ms); 3185 } 3186 return (error); 3187 } 3188 3189 /* 3190 * Query existing TX rate limit based on the existing 3191 * "inp->inp_snd_tag", if any. 3192 */ 3193 int 3194 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3195 { 3196 union if_snd_tag_query_params params = { }; 3197 struct m_snd_tag *mst; 3198 struct ifnet *ifp; 3199 int error; 3200 3201 mst = inp->inp_snd_tag; 3202 if (mst == NULL) 3203 return (EINVAL); 3204 3205 ifp = mst->ifp; 3206 if (ifp == NULL) 3207 return (EINVAL); 3208 3209 if (ifp->if_snd_tag_query == NULL) { 3210 error = EOPNOTSUPP; 3211 } else { 3212 error = ifp->if_snd_tag_query(mst, ¶ms); 3213 if (error == 0 && p_max_pacing_rate != NULL) 3214 *p_max_pacing_rate = params.rate_limit.max_rate; 3215 } 3216 return (error); 3217 } 3218 3219 /* 3220 * Query existing TX queue level based on the existing 3221 * "inp->inp_snd_tag", if any. 3222 */ 3223 int 3224 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3225 { 3226 union if_snd_tag_query_params params = { }; 3227 struct m_snd_tag *mst; 3228 struct ifnet *ifp; 3229 int error; 3230 3231 mst = inp->inp_snd_tag; 3232 if (mst == NULL) 3233 return (EINVAL); 3234 3235 ifp = mst->ifp; 3236 if (ifp == NULL) 3237 return (EINVAL); 3238 3239 if (ifp->if_snd_tag_query == NULL) 3240 return (EOPNOTSUPP); 3241 3242 error = ifp->if_snd_tag_query(mst, ¶ms); 3243 if (error == 0 && p_txqueue_level != NULL) 3244 *p_txqueue_level = params.rate_limit.queue_level; 3245 return (error); 3246 } 3247 3248 /* 3249 * Allocate a new TX rate limit send tag from the network interface 3250 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3251 */ 3252 int 3253 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3254 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate) 3255 { 3256 union if_snd_tag_alloc_params params = { 3257 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3258 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3259 .rate_limit.hdr.flowid = flowid, 3260 .rate_limit.hdr.flowtype = flowtype, 3261 .rate_limit.max_rate = max_pacing_rate, 3262 }; 3263 int error; 3264 3265 INP_WLOCK_ASSERT(inp); 3266 3267 if (inp->inp_snd_tag != NULL) 3268 return (EINVAL); 3269 3270 if (ifp->if_snd_tag_alloc == NULL) { 3271 error = EOPNOTSUPP; 3272 } else { 3273 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); 3274 3275 /* 3276 * At success increment the refcount on 3277 * the send tag's network interface: 3278 */ 3279 if (error == 0) 3280 if_ref(inp->inp_snd_tag->ifp); 3281 } 3282 return (error); 3283 } 3284 3285 /* 3286 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3287 * if any: 3288 */ 3289 void 3290 in_pcbdetach_txrtlmt(struct inpcb *inp) 3291 { 3292 struct m_snd_tag *mst; 3293 struct ifnet *ifp; 3294 3295 INP_WLOCK_ASSERT(inp); 3296 3297 mst = inp->inp_snd_tag; 3298 inp->inp_snd_tag = NULL; 3299 3300 if (mst == NULL) 3301 return; 3302 3303 ifp = mst->ifp; 3304 if (ifp == NULL) 3305 return; 3306 3307 /* 3308 * If the device was detached while we still had reference(s) 3309 * on the ifp, we assume if_snd_tag_free() was replaced with 3310 * stubs. 3311 */ 3312 ifp->if_snd_tag_free(mst); 3313 3314 /* release reference count on network interface */ 3315 if_rele(ifp); 3316 } 3317 3318 /* 3319 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3320 * is set in the fast path and will attach/detach/modify the TX rate 3321 * limit send tag based on the socket's so_max_pacing_rate value. 3322 */ 3323 void 3324 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3325 { 3326 struct socket *socket; 3327 uint32_t max_pacing_rate; 3328 bool did_upgrade; 3329 int error; 3330 3331 if (inp == NULL) 3332 return; 3333 3334 socket = inp->inp_socket; 3335 if (socket == NULL) 3336 return; 3337 3338 if (!INP_WLOCKED(inp)) { 3339 /* 3340 * NOTE: If the write locking fails, we need to bail 3341 * out and use the non-ratelimited ring for the 3342 * transmit until there is a new chance to get the 3343 * write lock. 3344 */ 3345 if (!INP_TRY_UPGRADE(inp)) 3346 return; 3347 did_upgrade = 1; 3348 } else { 3349 did_upgrade = 0; 3350 } 3351 3352 /* 3353 * NOTE: The so_max_pacing_rate value is read unlocked, 3354 * because atomic updates are not required since the variable 3355 * is checked at every mbuf we send. It is assumed that the 3356 * variable read itself will be atomic. 3357 */ 3358 max_pacing_rate = socket->so_max_pacing_rate; 3359 3360 /* 3361 * NOTE: When attaching to a network interface a reference is 3362 * made to ensure the network interface doesn't go away until 3363 * all ratelimit connections are gone. The network interface 3364 * pointers compared below represent valid network interfaces, 3365 * except when comparing towards NULL. 3366 */ 3367 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3368 error = 0; 3369 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3370 if (inp->inp_snd_tag != NULL) 3371 in_pcbdetach_txrtlmt(inp); 3372 error = 0; 3373 } else if (inp->inp_snd_tag == NULL) { 3374 /* 3375 * In order to utilize packet pacing with RSS, we need 3376 * to wait until there is a valid RSS hash before we 3377 * can proceed: 3378 */ 3379 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3380 error = EAGAIN; 3381 } else { 3382 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3383 mb->m_pkthdr.flowid, max_pacing_rate); 3384 } 3385 } else { 3386 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3387 } 3388 if (error == 0 || error == EOPNOTSUPP) 3389 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3390 if (did_upgrade) 3391 INP_DOWNGRADE(inp); 3392 } 3393 3394 /* 3395 * Track route changes for TX rate limiting. 3396 */ 3397 void 3398 in_pcboutput_eagain(struct inpcb *inp) 3399 { 3400 bool did_upgrade; 3401 3402 if (inp == NULL) 3403 return; 3404 3405 if (inp->inp_snd_tag == NULL) 3406 return; 3407 3408 if (!INP_WLOCKED(inp)) { 3409 /* 3410 * NOTE: If the write locking fails, we need to bail 3411 * out and use the non-ratelimited ring for the 3412 * transmit until there is a new chance to get the 3413 * write lock. 3414 */ 3415 if (!INP_TRY_UPGRADE(inp)) 3416 return; 3417 did_upgrade = 1; 3418 } else { 3419 did_upgrade = 0; 3420 } 3421 3422 /* detach rate limiting */ 3423 in_pcbdetach_txrtlmt(inp); 3424 3425 /* make sure new mbuf send tag allocation is made */ 3426 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3427 3428 if (did_upgrade) 3429 INP_DOWNGRADE(inp); 3430 } 3431 #endif /* RATELIMIT */ 3432