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 LIST_INSERT_HEAD(hdr, grp, il_list); 239 return (grp); 240 } 241 242 static void 243 in_pcblbgroup_free(struct inpcblbgroup *grp) 244 { 245 246 LIST_REMOVE(grp, il_list); 247 free(grp, M_TEMP); 248 } 249 250 static struct inpcblbgroup * 251 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 252 struct inpcblbgroup *old_grp, int size) 253 { 254 struct inpcblbgroup *grp; 255 int i; 256 257 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, 258 old_grp->il_lport, &old_grp->il_dependladdr, size); 259 if (!grp) 260 return (NULL); 261 262 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 263 ("invalid new local group size %d and old local group count %d", 264 grp->il_inpsiz, old_grp->il_inpcnt)); 265 266 for (i = 0; i < old_grp->il_inpcnt; ++i) 267 grp->il_inp[i] = old_grp->il_inp[i]; 268 grp->il_inpcnt = old_grp->il_inpcnt; 269 in_pcblbgroup_free(old_grp); 270 return (grp); 271 } 272 273 /* 274 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 275 * and shrink group if possible. 276 */ 277 static void 278 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 279 int i) 280 { 281 struct inpcblbgroup *grp = *grpp; 282 283 for (; i + 1 < grp->il_inpcnt; ++i) 284 grp->il_inp[i] = grp->il_inp[i + 1]; 285 grp->il_inpcnt--; 286 287 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 288 grp->il_inpcnt <= (grp->il_inpsiz / 4)) { 289 /* Shrink this group. */ 290 struct inpcblbgroup *new_grp = 291 in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 292 if (new_grp) 293 *grpp = new_grp; 294 } 295 return; 296 } 297 298 /* 299 * Add PCB to load balance group for SO_REUSEPORT_LB option. 300 */ 301 static int 302 in_pcbinslbgrouphash(struct inpcb *inp) 303 { 304 struct inpcbinfo *pcbinfo; 305 struct inpcblbgrouphead *hdr; 306 struct inpcblbgroup *grp; 307 uint16_t hashmask, lport; 308 uint32_t group_index; 309 struct ucred *cred; 310 static int limit_logged = 0; 311 312 pcbinfo = inp->inp_pcbinfo; 313 314 INP_WLOCK_ASSERT(inp); 315 INP_HASH_WLOCK_ASSERT(pcbinfo); 316 317 if (pcbinfo->ipi_lbgrouphashbase == NULL) 318 return (0); 319 320 hashmask = pcbinfo->ipi_lbgrouphashmask; 321 lport = inp->inp_lport; 322 group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask); 323 hdr = &pcbinfo->ipi_lbgrouphashbase[group_index]; 324 325 /* 326 * Don't allow jailed socket to join local group. 327 */ 328 if (inp->inp_socket != NULL) 329 cred = inp->inp_socket->so_cred; 330 else 331 cred = NULL; 332 if (cred != NULL && jailed(cred)) 333 return (0); 334 335 #ifdef INET6 336 /* 337 * Don't allow IPv4 mapped INET6 wild socket. 338 */ 339 if ((inp->inp_vflag & INP_IPV4) && 340 inp->inp_laddr.s_addr == INADDR_ANY && 341 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 342 return (0); 343 } 344 #endif 345 346 hdr = &pcbinfo->ipi_lbgrouphashbase[ 347 INP_PCBLBGROUP_PORTHASH(inp->inp_lport, 348 pcbinfo->ipi_lbgrouphashmask)]; 349 LIST_FOREACH(grp, hdr, il_list) { 350 if (grp->il_vflag == inp->inp_vflag && 351 grp->il_lport == inp->inp_lport && 352 memcmp(&grp->il_dependladdr, 353 &inp->inp_inc.inc_ie.ie_dependladdr, 354 sizeof(grp->il_dependladdr)) == 0) { 355 break; 356 } 357 } 358 if (grp == NULL) { 359 /* Create new load balance group. */ 360 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, 361 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 362 INPCBLBGROUP_SIZMIN); 363 if (!grp) 364 return (ENOBUFS); 365 } else if (grp->il_inpcnt == grp->il_inpsiz) { 366 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 367 if (!limit_logged) { 368 limit_logged = 1; 369 printf("lb group port %d, limit reached\n", 370 ntohs(grp->il_lport)); 371 } 372 return (0); 373 } 374 375 /* Expand this local group. */ 376 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 377 if (!grp) 378 return (ENOBUFS); 379 } 380 381 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 382 ("invalid local group size %d and count %d", 383 grp->il_inpsiz, grp->il_inpcnt)); 384 385 grp->il_inp[grp->il_inpcnt] = inp; 386 grp->il_inpcnt++; 387 return (0); 388 } 389 390 /* 391 * Remove PCB from load balance group. 392 */ 393 static void 394 in_pcbremlbgrouphash(struct inpcb *inp) 395 { 396 struct inpcbinfo *pcbinfo; 397 struct inpcblbgrouphead *hdr; 398 struct inpcblbgroup *grp; 399 int i; 400 401 pcbinfo = inp->inp_pcbinfo; 402 403 INP_WLOCK_ASSERT(inp); 404 INP_HASH_WLOCK_ASSERT(pcbinfo); 405 406 if (pcbinfo->ipi_lbgrouphashbase == NULL) 407 return; 408 409 hdr = &pcbinfo->ipi_lbgrouphashbase[ 410 INP_PCBLBGROUP_PORTHASH(inp->inp_lport, 411 pcbinfo->ipi_lbgrouphashmask)]; 412 413 LIST_FOREACH(grp, hdr, il_list) { 414 for (i = 0; i < grp->il_inpcnt; ++i) { 415 if (grp->il_inp[i] != inp) 416 continue; 417 418 if (grp->il_inpcnt == 1) { 419 /* We are the last, free this local group. */ 420 in_pcblbgroup_free(grp); 421 } else { 422 /* Pull up inpcbs, shrink group if possible. */ 423 in_pcblbgroup_reorder(hdr, &grp, i); 424 } 425 return; 426 } 427 } 428 } 429 430 /* 431 * Different protocols initialize their inpcbs differently - giving 432 * different name to the lock. But they all are disposed the same. 433 */ 434 static void 435 inpcb_fini(void *mem, int size) 436 { 437 struct inpcb *inp = mem; 438 439 INP_LOCK_DESTROY(inp); 440 } 441 442 /* 443 * Initialize an inpcbinfo -- we should be able to reduce the number of 444 * arguments in time. 445 */ 446 void 447 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 448 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 449 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) 450 { 451 452 INP_INFO_LOCK_INIT(pcbinfo, name); 453 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 454 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 455 #ifdef VIMAGE 456 pcbinfo->ipi_vnet = curvnet; 457 #endif 458 pcbinfo->ipi_listhead = listhead; 459 CK_LIST_INIT(pcbinfo->ipi_listhead); 460 pcbinfo->ipi_count = 0; 461 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 462 &pcbinfo->ipi_hashmask); 463 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 464 &pcbinfo->ipi_porthashmask); 465 pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB, 466 &pcbinfo->ipi_lbgrouphashmask); 467 #ifdef PCBGROUP 468 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 469 #endif 470 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 471 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); 472 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 473 uma_zone_set_warning(pcbinfo->ipi_zone, 474 "kern.ipc.maxsockets limit reached"); 475 } 476 477 /* 478 * Destroy an inpcbinfo. 479 */ 480 void 481 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 482 { 483 484 KASSERT(pcbinfo->ipi_count == 0, 485 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 486 487 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 488 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 489 pcbinfo->ipi_porthashmask); 490 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 491 pcbinfo->ipi_lbgrouphashmask); 492 #ifdef PCBGROUP 493 in_pcbgroup_destroy(pcbinfo); 494 #endif 495 uma_zdestroy(pcbinfo->ipi_zone); 496 INP_LIST_LOCK_DESTROY(pcbinfo); 497 INP_HASH_LOCK_DESTROY(pcbinfo); 498 INP_INFO_LOCK_DESTROY(pcbinfo); 499 } 500 501 /* 502 * Allocate a PCB and associate it with the socket. 503 * On success return with the PCB locked. 504 */ 505 int 506 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 507 { 508 struct inpcb *inp; 509 int error; 510 511 #ifdef INVARIANTS 512 if (pcbinfo == &V_tcbinfo) { 513 INP_INFO_RLOCK_ASSERT(pcbinfo); 514 } else { 515 INP_INFO_WLOCK_ASSERT(pcbinfo); 516 } 517 #endif 518 519 error = 0; 520 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 521 if (inp == NULL) 522 return (ENOBUFS); 523 bzero(&inp->inp_start_zero, inp_zero_size); 524 inp->inp_pcbinfo = pcbinfo; 525 inp->inp_socket = so; 526 inp->inp_cred = crhold(so->so_cred); 527 inp->inp_inc.inc_fibnum = so->so_fibnum; 528 #ifdef MAC 529 error = mac_inpcb_init(inp, M_NOWAIT); 530 if (error != 0) 531 goto out; 532 mac_inpcb_create(so, inp); 533 #endif 534 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 535 error = ipsec_init_pcbpolicy(inp); 536 if (error != 0) { 537 #ifdef MAC 538 mac_inpcb_destroy(inp); 539 #endif 540 goto out; 541 } 542 #endif /*IPSEC*/ 543 #ifdef INET6 544 if (INP_SOCKAF(so) == AF_INET6) { 545 inp->inp_vflag |= INP_IPV6PROTO; 546 if (V_ip6_v6only) 547 inp->inp_flags |= IN6P_IPV6_V6ONLY; 548 } 549 #endif 550 INP_WLOCK(inp); 551 INP_LIST_WLOCK(pcbinfo); 552 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 553 pcbinfo->ipi_count++; 554 so->so_pcb = (caddr_t)inp; 555 #ifdef INET6 556 if (V_ip6_auto_flowlabel) 557 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 558 #endif 559 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 560 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 561 562 /* 563 * Routes in inpcb's can cache L2 as well; they are guaranteed 564 * to be cleaned up. 565 */ 566 inp->inp_route.ro_flags = RT_LLE_CACHE; 567 INP_LIST_WUNLOCK(pcbinfo); 568 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 569 out: 570 if (error != 0) { 571 crfree(inp->inp_cred); 572 uma_zfree(pcbinfo->ipi_zone, inp); 573 } 574 #endif 575 return (error); 576 } 577 578 #ifdef INET 579 int 580 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 581 { 582 int anonport, error; 583 584 INP_WLOCK_ASSERT(inp); 585 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 586 587 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 588 return (EINVAL); 589 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 590 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 591 &inp->inp_lport, cred); 592 if (error) 593 return (error); 594 if (in_pcbinshash(inp) != 0) { 595 inp->inp_laddr.s_addr = INADDR_ANY; 596 inp->inp_lport = 0; 597 return (EAGAIN); 598 } 599 if (anonport) 600 inp->inp_flags |= INP_ANONPORT; 601 return (0); 602 } 603 #endif 604 605 /* 606 * Select a local port (number) to use. 607 */ 608 #if defined(INET) || defined(INET6) 609 int 610 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 611 struct ucred *cred, int lookupflags) 612 { 613 struct inpcbinfo *pcbinfo; 614 struct inpcb *tmpinp; 615 unsigned short *lastport; 616 int count, dorandom, error; 617 u_short aux, first, last, lport; 618 #ifdef INET 619 struct in_addr laddr; 620 #endif 621 622 pcbinfo = inp->inp_pcbinfo; 623 624 /* 625 * Because no actual state changes occur here, a global write lock on 626 * the pcbinfo isn't required. 627 */ 628 INP_LOCK_ASSERT(inp); 629 INP_HASH_LOCK_ASSERT(pcbinfo); 630 631 if (inp->inp_flags & INP_HIGHPORT) { 632 first = V_ipport_hifirstauto; /* sysctl */ 633 last = V_ipport_hilastauto; 634 lastport = &pcbinfo->ipi_lasthi; 635 } else if (inp->inp_flags & INP_LOWPORT) { 636 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0); 637 if (error) 638 return (error); 639 first = V_ipport_lowfirstauto; /* 1023 */ 640 last = V_ipport_lowlastauto; /* 600 */ 641 lastport = &pcbinfo->ipi_lastlow; 642 } else { 643 first = V_ipport_firstauto; /* sysctl */ 644 last = V_ipport_lastauto; 645 lastport = &pcbinfo->ipi_lastport; 646 } 647 /* 648 * For UDP(-Lite), use random port allocation as long as the user 649 * allows it. For TCP (and as of yet unknown) connections, 650 * use random port allocation only if the user allows it AND 651 * ipport_tick() allows it. 652 */ 653 if (V_ipport_randomized && 654 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 655 pcbinfo == &V_ulitecbinfo)) 656 dorandom = 1; 657 else 658 dorandom = 0; 659 /* 660 * It makes no sense to do random port allocation if 661 * we have the only port available. 662 */ 663 if (first == last) 664 dorandom = 0; 665 /* Make sure to not include UDP(-Lite) packets in the count. */ 666 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 667 V_ipport_tcpallocs++; 668 /* 669 * Instead of having two loops further down counting up or down 670 * make sure that first is always <= last and go with only one 671 * code path implementing all logic. 672 */ 673 if (first > last) { 674 aux = first; 675 first = last; 676 last = aux; 677 } 678 679 #ifdef INET 680 /* Make the compiler happy. */ 681 laddr.s_addr = 0; 682 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 683 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", 684 __func__, inp)); 685 laddr = *laddrp; 686 } 687 #endif 688 tmpinp = NULL; /* Make compiler happy. */ 689 lport = *lportp; 690 691 if (dorandom) 692 *lastport = first + (arc4random() % (last - first)); 693 694 count = last - first; 695 696 do { 697 if (count-- < 0) /* completely used? */ 698 return (EADDRNOTAVAIL); 699 ++*lastport; 700 if (*lastport < first || *lastport > last) 701 *lastport = first; 702 lport = htons(*lastport); 703 704 #ifdef INET6 705 if ((inp->inp_vflag & INP_IPV6) != 0) 706 tmpinp = in6_pcblookup_local(pcbinfo, 707 &inp->in6p_laddr, lport, lookupflags, cred); 708 #endif 709 #if defined(INET) && defined(INET6) 710 else 711 #endif 712 #ifdef INET 713 tmpinp = in_pcblookup_local(pcbinfo, laddr, 714 lport, lookupflags, cred); 715 #endif 716 } while (tmpinp != NULL); 717 718 #ifdef INET 719 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) 720 laddrp->s_addr = laddr.s_addr; 721 #endif 722 *lportp = lport; 723 724 return (0); 725 } 726 727 /* 728 * Return cached socket options. 729 */ 730 int 731 inp_so_options(const struct inpcb *inp) 732 { 733 int so_options; 734 735 so_options = 0; 736 737 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 738 so_options |= SO_REUSEPORT_LB; 739 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 740 so_options |= SO_REUSEPORT; 741 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 742 so_options |= SO_REUSEADDR; 743 return (so_options); 744 } 745 #endif /* INET || INET6 */ 746 747 /* 748 * Check if a new BINDMULTI socket is allowed to be created. 749 * 750 * ni points to the new inp. 751 * oi points to the exisitng inp. 752 * 753 * This checks whether the existing inp also has BINDMULTI and 754 * whether the credentials match. 755 */ 756 int 757 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 758 { 759 /* Check permissions match */ 760 if ((ni->inp_flags2 & INP_BINDMULTI) && 761 (ni->inp_cred->cr_uid != 762 oi->inp_cred->cr_uid)) 763 return (0); 764 765 /* Check the existing inp has BINDMULTI set */ 766 if ((ni->inp_flags2 & INP_BINDMULTI) && 767 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 768 return (0); 769 770 /* 771 * We're okay - either INP_BINDMULTI isn't set on ni, or 772 * it is and it matches the checks. 773 */ 774 return (1); 775 } 776 777 #ifdef INET 778 /* 779 * Set up a bind operation on a PCB, performing port allocation 780 * as required, but do not actually modify the PCB. Callers can 781 * either complete the bind by setting inp_laddr/inp_lport and 782 * calling in_pcbinshash(), or they can just use the resulting 783 * port and address to authorise the sending of a once-off packet. 784 * 785 * On error, the values of *laddrp and *lportp are not changed. 786 */ 787 int 788 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 789 u_short *lportp, struct ucred *cred) 790 { 791 struct socket *so = inp->inp_socket; 792 struct sockaddr_in *sin; 793 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 794 struct in_addr laddr; 795 u_short lport = 0; 796 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 797 int error; 798 799 /* 800 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 801 * so that we don't have to add to the (already messy) code below. 802 */ 803 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 804 805 /* 806 * No state changes, so read locks are sufficient here. 807 */ 808 INP_LOCK_ASSERT(inp); 809 INP_HASH_LOCK_ASSERT(pcbinfo); 810 811 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 812 return (EADDRNOTAVAIL); 813 laddr.s_addr = *laddrp; 814 if (nam != NULL && laddr.s_addr != INADDR_ANY) 815 return (EINVAL); 816 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 817 lookupflags = INPLOOKUP_WILDCARD; 818 if (nam == NULL) { 819 if ((error = prison_local_ip4(cred, &laddr)) != 0) 820 return (error); 821 } else { 822 sin = (struct sockaddr_in *)nam; 823 if (nam->sa_len != sizeof (*sin)) 824 return (EINVAL); 825 #ifdef notdef 826 /* 827 * We should check the family, but old programs 828 * incorrectly fail to initialize it. 829 */ 830 if (sin->sin_family != AF_INET) 831 return (EAFNOSUPPORT); 832 #endif 833 error = prison_local_ip4(cred, &sin->sin_addr); 834 if (error) 835 return (error); 836 if (sin->sin_port != *lportp) { 837 /* Don't allow the port to change. */ 838 if (*lportp != 0) 839 return (EINVAL); 840 lport = sin->sin_port; 841 } 842 /* NB: lport is left as 0 if the port isn't being changed. */ 843 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 844 /* 845 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 846 * allow complete duplication of binding if 847 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 848 * and a multicast address is bound on both 849 * new and duplicated sockets. 850 */ 851 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 852 reuseport = SO_REUSEADDR|SO_REUSEPORT; 853 /* 854 * XXX: How to deal with SO_REUSEPORT_LB here? 855 * Treat same as SO_REUSEPORT for now. 856 */ 857 if ((so->so_options & 858 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 859 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 860 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 861 sin->sin_port = 0; /* yech... */ 862 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 863 /* 864 * Is the address a local IP address? 865 * If INP_BINDANY is set, then the socket may be bound 866 * to any endpoint address, local or not. 867 */ 868 if ((inp->inp_flags & INP_BINDANY) == 0 && 869 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 870 return (EADDRNOTAVAIL); 871 } 872 laddr = sin->sin_addr; 873 if (lport) { 874 struct inpcb *t; 875 struct tcptw *tw; 876 877 /* GROSS */ 878 if (ntohs(lport) <= V_ipport_reservedhigh && 879 ntohs(lport) >= V_ipport_reservedlow && 880 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 881 0)) 882 return (EACCES); 883 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 884 priv_check_cred(inp->inp_cred, 885 PRIV_NETINET_REUSEPORT, 0) != 0) { 886 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 887 lport, INPLOOKUP_WILDCARD, cred); 888 /* 889 * XXX 890 * This entire block sorely needs a rewrite. 891 */ 892 if (t && 893 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 894 ((t->inp_flags & INP_TIMEWAIT) == 0) && 895 (so->so_type != SOCK_STREAM || 896 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 897 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 898 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 899 (t->inp_flags2 & INP_REUSEPORT) || 900 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 901 (inp->inp_cred->cr_uid != 902 t->inp_cred->cr_uid)) 903 return (EADDRINUSE); 904 905 /* 906 * If the socket is a BINDMULTI socket, then 907 * the credentials need to match and the 908 * original socket also has to have been bound 909 * with BINDMULTI. 910 */ 911 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 912 return (EADDRINUSE); 913 } 914 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 915 lport, lookupflags, cred); 916 if (t && (t->inp_flags & INP_TIMEWAIT)) { 917 /* 918 * XXXRW: If an incpb has had its timewait 919 * state recycled, we treat the address as 920 * being in use (for now). This is better 921 * than a panic, but not desirable. 922 */ 923 tw = intotw(t); 924 if (tw == NULL || 925 ((reuseport & tw->tw_so_options) == 0 && 926 (reuseport_lb & 927 tw->tw_so_options) == 0)) { 928 return (EADDRINUSE); 929 } 930 } else if (t && 931 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 932 (reuseport & inp_so_options(t)) == 0 && 933 (reuseport_lb & inp_so_options(t)) == 0) { 934 #ifdef INET6 935 if (ntohl(sin->sin_addr.s_addr) != 936 INADDR_ANY || 937 ntohl(t->inp_laddr.s_addr) != 938 INADDR_ANY || 939 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 940 (t->inp_vflag & INP_IPV6PROTO) == 0) 941 #endif 942 return (EADDRINUSE); 943 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 944 return (EADDRINUSE); 945 } 946 } 947 } 948 if (*lportp != 0) 949 lport = *lportp; 950 if (lport == 0) { 951 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 952 if (error != 0) 953 return (error); 954 955 } 956 *laddrp = laddr.s_addr; 957 *lportp = lport; 958 return (0); 959 } 960 961 /* 962 * Connect from a socket to a specified address. 963 * Both address and port must be specified in argument sin. 964 * If don't have a local address for this socket yet, 965 * then pick one. 966 */ 967 int 968 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 969 struct ucred *cred, struct mbuf *m) 970 { 971 u_short lport, fport; 972 in_addr_t laddr, faddr; 973 int anonport, error; 974 975 INP_WLOCK_ASSERT(inp); 976 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 977 978 lport = inp->inp_lport; 979 laddr = inp->inp_laddr.s_addr; 980 anonport = (lport == 0); 981 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 982 NULL, cred); 983 if (error) 984 return (error); 985 986 /* Do the initial binding of the local address if required. */ 987 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 988 inp->inp_lport = lport; 989 inp->inp_laddr.s_addr = laddr; 990 if (in_pcbinshash(inp) != 0) { 991 inp->inp_laddr.s_addr = INADDR_ANY; 992 inp->inp_lport = 0; 993 return (EAGAIN); 994 } 995 } 996 997 /* Commit the remaining changes. */ 998 inp->inp_lport = lport; 999 inp->inp_laddr.s_addr = laddr; 1000 inp->inp_faddr.s_addr = faddr; 1001 inp->inp_fport = fport; 1002 in_pcbrehash_mbuf(inp, m); 1003 1004 if (anonport) 1005 inp->inp_flags |= INP_ANONPORT; 1006 return (0); 1007 } 1008 1009 int 1010 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 1011 { 1012 1013 return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); 1014 } 1015 1016 /* 1017 * Do proper source address selection on an unbound socket in case 1018 * of connect. Take jails into account as well. 1019 */ 1020 int 1021 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1022 struct ucred *cred) 1023 { 1024 struct ifaddr *ifa; 1025 struct sockaddr *sa; 1026 struct sockaddr_in *sin; 1027 struct route sro; 1028 int error; 1029 1030 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1031 /* 1032 * Bypass source address selection and use the primary jail IP 1033 * if requested. 1034 */ 1035 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1036 return (0); 1037 1038 error = 0; 1039 bzero(&sro, sizeof(sro)); 1040 1041 sin = (struct sockaddr_in *)&sro.ro_dst; 1042 sin->sin_family = AF_INET; 1043 sin->sin_len = sizeof(struct sockaddr_in); 1044 sin->sin_addr.s_addr = faddr->s_addr; 1045 1046 /* 1047 * If route is known our src addr is taken from the i/f, 1048 * else punt. 1049 * 1050 * Find out route to destination. 1051 */ 1052 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1053 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 1054 1055 /* 1056 * If we found a route, use the address corresponding to 1057 * the outgoing interface. 1058 * 1059 * Otherwise assume faddr is reachable on a directly connected 1060 * network and try to find a corresponding interface to take 1061 * the source address from. 1062 */ 1063 NET_EPOCH_ENTER(); 1064 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 1065 struct in_ifaddr *ia; 1066 struct ifnet *ifp; 1067 1068 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1069 inp->inp_socket->so_fibnum)); 1070 if (ia == NULL) { 1071 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1072 inp->inp_socket->so_fibnum)); 1073 1074 } 1075 if (ia == NULL) { 1076 error = ENETUNREACH; 1077 goto done; 1078 } 1079 1080 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1081 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1082 goto done; 1083 } 1084 1085 ifp = ia->ia_ifp; 1086 ia = NULL; 1087 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1088 1089 sa = ifa->ifa_addr; 1090 if (sa->sa_family != AF_INET) 1091 continue; 1092 sin = (struct sockaddr_in *)sa; 1093 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1094 ia = (struct in_ifaddr *)ifa; 1095 break; 1096 } 1097 } 1098 if (ia != NULL) { 1099 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1100 goto done; 1101 } 1102 1103 /* 3. As a last resort return the 'default' jail address. */ 1104 error = prison_get_ip4(cred, laddr); 1105 goto done; 1106 } 1107 1108 /* 1109 * If the outgoing interface on the route found is not 1110 * a loopback interface, use the address from that interface. 1111 * In case of jails do those three steps: 1112 * 1. check if the interface address belongs to the jail. If so use it. 1113 * 2. check if we have any address on the outgoing interface 1114 * belonging to this jail. If so use it. 1115 * 3. as a last resort return the 'default' jail address. 1116 */ 1117 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 1118 struct in_ifaddr *ia; 1119 struct ifnet *ifp; 1120 1121 /* If not jailed, use the default returned. */ 1122 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1123 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1124 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1125 goto done; 1126 } 1127 1128 /* Jailed. */ 1129 /* 1. Check if the iface address belongs to the jail. */ 1130 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 1131 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1132 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1133 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1134 goto done; 1135 } 1136 1137 /* 1138 * 2. Check if we have any address on the outgoing interface 1139 * belonging to this jail. 1140 */ 1141 ia = NULL; 1142 ifp = sro.ro_rt->rt_ifp; 1143 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1144 sa = ifa->ifa_addr; 1145 if (sa->sa_family != AF_INET) 1146 continue; 1147 sin = (struct sockaddr_in *)sa; 1148 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1149 ia = (struct in_ifaddr *)ifa; 1150 break; 1151 } 1152 } 1153 if (ia != NULL) { 1154 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1155 goto done; 1156 } 1157 1158 /* 3. As a last resort return the 'default' jail address. */ 1159 error = prison_get_ip4(cred, laddr); 1160 goto done; 1161 } 1162 1163 /* 1164 * The outgoing interface is marked with 'loopback net', so a route 1165 * to ourselves is here. 1166 * Try to find the interface of the destination address and then 1167 * take the address from there. That interface is not necessarily 1168 * a loopback interface. 1169 * In case of jails, check that it is an address of the jail 1170 * and if we cannot find, fall back to the 'default' jail address. 1171 */ 1172 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 1173 struct sockaddr_in sain; 1174 struct in_ifaddr *ia; 1175 1176 bzero(&sain, sizeof(struct sockaddr_in)); 1177 sain.sin_family = AF_INET; 1178 sain.sin_len = sizeof(struct sockaddr_in); 1179 sain.sin_addr.s_addr = faddr->s_addr; 1180 1181 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), 1182 inp->inp_socket->so_fibnum)); 1183 if (ia == NULL) 1184 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, 1185 inp->inp_socket->so_fibnum)); 1186 if (ia == NULL) 1187 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 1188 1189 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1190 if (ia == NULL) { 1191 error = ENETUNREACH; 1192 goto done; 1193 } 1194 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1195 goto done; 1196 } 1197 1198 /* Jailed. */ 1199 if (ia != NULL) { 1200 struct ifnet *ifp; 1201 1202 ifp = ia->ia_ifp; 1203 ia = NULL; 1204 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1205 sa = ifa->ifa_addr; 1206 if (sa->sa_family != AF_INET) 1207 continue; 1208 sin = (struct sockaddr_in *)sa; 1209 if (prison_check_ip4(cred, 1210 &sin->sin_addr) == 0) { 1211 ia = (struct in_ifaddr *)ifa; 1212 break; 1213 } 1214 } 1215 if (ia != NULL) { 1216 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1217 goto done; 1218 } 1219 } 1220 1221 /* 3. As a last resort return the 'default' jail address. */ 1222 error = prison_get_ip4(cred, laddr); 1223 goto done; 1224 } 1225 1226 done: 1227 NET_EPOCH_EXIT(); 1228 if (sro.ro_rt != NULL) 1229 RTFREE(sro.ro_rt); 1230 return (error); 1231 } 1232 1233 /* 1234 * Set up for a connect from a socket to the specified address. 1235 * On entry, *laddrp and *lportp should contain the current local 1236 * address and port for the PCB; these are updated to the values 1237 * that should be placed in inp_laddr and inp_lport to complete 1238 * the connect. 1239 * 1240 * On success, *faddrp and *fportp will be set to the remote address 1241 * and port. These are not updated in the error case. 1242 * 1243 * If the operation fails because the connection already exists, 1244 * *oinpp will be set to the PCB of that connection so that the 1245 * caller can decide to override it. In all other cases, *oinpp 1246 * is set to NULL. 1247 */ 1248 int 1249 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1250 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1251 struct inpcb **oinpp, struct ucred *cred) 1252 { 1253 struct rm_priotracker in_ifa_tracker; 1254 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1255 struct in_ifaddr *ia; 1256 struct inpcb *oinp; 1257 struct in_addr laddr, faddr; 1258 u_short lport, fport; 1259 int error; 1260 1261 /* 1262 * Because a global state change doesn't actually occur here, a read 1263 * lock is sufficient. 1264 */ 1265 INP_LOCK_ASSERT(inp); 1266 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1267 1268 if (oinpp != NULL) 1269 *oinpp = NULL; 1270 if (nam->sa_len != sizeof (*sin)) 1271 return (EINVAL); 1272 if (sin->sin_family != AF_INET) 1273 return (EAFNOSUPPORT); 1274 if (sin->sin_port == 0) 1275 return (EADDRNOTAVAIL); 1276 laddr.s_addr = *laddrp; 1277 lport = *lportp; 1278 faddr = sin->sin_addr; 1279 fport = sin->sin_port; 1280 1281 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1282 /* 1283 * If the destination address is INADDR_ANY, 1284 * use the primary local address. 1285 * If the supplied address is INADDR_BROADCAST, 1286 * and the primary interface supports broadcast, 1287 * choose the broadcast address for that interface. 1288 */ 1289 if (faddr.s_addr == INADDR_ANY) { 1290 IN_IFADDR_RLOCK(&in_ifa_tracker); 1291 faddr = 1292 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1293 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1294 if (cred != NULL && 1295 (error = prison_get_ip4(cred, &faddr)) != 0) 1296 return (error); 1297 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1298 IN_IFADDR_RLOCK(&in_ifa_tracker); 1299 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1300 IFF_BROADCAST) 1301 faddr = satosin(&CK_STAILQ_FIRST( 1302 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1303 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1304 } 1305 } 1306 if (laddr.s_addr == INADDR_ANY) { 1307 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1308 /* 1309 * If the destination address is multicast and an outgoing 1310 * interface has been set as a multicast option, prefer the 1311 * address of that interface as our source address. 1312 */ 1313 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1314 inp->inp_moptions != NULL) { 1315 struct ip_moptions *imo; 1316 struct ifnet *ifp; 1317 1318 imo = inp->inp_moptions; 1319 if (imo->imo_multicast_ifp != NULL) { 1320 ifp = imo->imo_multicast_ifp; 1321 IN_IFADDR_RLOCK(&in_ifa_tracker); 1322 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1323 if ((ia->ia_ifp == ifp) && 1324 (cred == NULL || 1325 prison_check_ip4(cred, 1326 &ia->ia_addr.sin_addr) == 0)) 1327 break; 1328 } 1329 if (ia == NULL) 1330 error = EADDRNOTAVAIL; 1331 else { 1332 laddr = ia->ia_addr.sin_addr; 1333 error = 0; 1334 } 1335 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1336 } 1337 } 1338 if (error) 1339 return (error); 1340 } 1341 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1342 laddr, lport, 0, NULL); 1343 if (oinp != NULL) { 1344 if (oinpp != NULL) 1345 *oinpp = oinp; 1346 return (EADDRINUSE); 1347 } 1348 if (lport == 0) { 1349 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1350 cred); 1351 if (error) 1352 return (error); 1353 } 1354 *laddrp = laddr.s_addr; 1355 *lportp = lport; 1356 *faddrp = faddr.s_addr; 1357 *fportp = fport; 1358 return (0); 1359 } 1360 1361 void 1362 in_pcbdisconnect(struct inpcb *inp) 1363 { 1364 1365 INP_WLOCK_ASSERT(inp); 1366 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1367 1368 inp->inp_faddr.s_addr = INADDR_ANY; 1369 inp->inp_fport = 0; 1370 in_pcbrehash(inp); 1371 } 1372 #endif /* INET */ 1373 1374 /* 1375 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1376 * For most protocols, this will be invoked immediately prior to calling 1377 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1378 * socket, in which case in_pcbfree() is deferred. 1379 */ 1380 void 1381 in_pcbdetach(struct inpcb *inp) 1382 { 1383 1384 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1385 1386 #ifdef RATELIMIT 1387 if (inp->inp_snd_tag != NULL) 1388 in_pcbdetach_txrtlmt(inp); 1389 #endif 1390 inp->inp_socket->so_pcb = NULL; 1391 inp->inp_socket = NULL; 1392 } 1393 1394 /* 1395 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1396 * stability of an inpcb pointer despite the inpcb lock being released. This 1397 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1398 * but where the inpcb lock may already held, or when acquiring a reference 1399 * via a pcbgroup. 1400 * 1401 * in_pcbref() should be used only to provide brief memory stability, and 1402 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1403 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1404 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1405 * lock and rele are the *only* safe operations that may be performed on the 1406 * inpcb. 1407 * 1408 * While the inpcb will not be freed, releasing the inpcb lock means that the 1409 * connection's state may change, so the caller should be careful to 1410 * revalidate any cached state on reacquiring the lock. Drop the reference 1411 * using in_pcbrele(). 1412 */ 1413 void 1414 in_pcbref(struct inpcb *inp) 1415 { 1416 1417 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1418 1419 refcount_acquire(&inp->inp_refcount); 1420 } 1421 1422 /* 1423 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1424 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1425 * return a flag indicating whether or not the inpcb remains valid. If it is 1426 * valid, we return with the inpcb lock held. 1427 * 1428 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1429 * reference on an inpcb. Historically more work was done here (actually, in 1430 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1431 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1432 * about memory stability (and continued use of the write lock). 1433 */ 1434 int 1435 in_pcbrele_rlocked(struct inpcb *inp) 1436 { 1437 struct inpcbinfo *pcbinfo; 1438 1439 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1440 1441 INP_RLOCK_ASSERT(inp); 1442 1443 if (refcount_release(&inp->inp_refcount) == 0) { 1444 /* 1445 * If the inpcb has been freed, let the caller know, even if 1446 * this isn't the last reference. 1447 */ 1448 if (inp->inp_flags2 & INP_FREED) { 1449 INP_RUNLOCK(inp); 1450 return (1); 1451 } 1452 return (0); 1453 } 1454 1455 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1456 #ifdef TCPHPTS 1457 if (inp->inp_in_hpts || inp->inp_in_input) { 1458 struct tcp_hpts_entry *hpts; 1459 /* 1460 * We should not be on the hpts at 1461 * this point in any form. we must 1462 * get the lock to be sure. 1463 */ 1464 hpts = tcp_hpts_lock(inp); 1465 if (inp->inp_in_hpts) 1466 panic("Hpts:%p inp:%p at free still on hpts", 1467 hpts, inp); 1468 mtx_unlock(&hpts->p_mtx); 1469 hpts = tcp_input_lock(inp); 1470 if (inp->inp_in_input) 1471 panic("Hpts:%p inp:%p at free still on input hpts", 1472 hpts, inp); 1473 mtx_unlock(&hpts->p_mtx); 1474 } 1475 #endif 1476 INP_RUNLOCK(inp); 1477 pcbinfo = inp->inp_pcbinfo; 1478 uma_zfree(pcbinfo->ipi_zone, inp); 1479 return (1); 1480 } 1481 1482 int 1483 in_pcbrele_wlocked(struct inpcb *inp) 1484 { 1485 struct inpcbinfo *pcbinfo; 1486 1487 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1488 1489 INP_WLOCK_ASSERT(inp); 1490 1491 if (refcount_release(&inp->inp_refcount) == 0) { 1492 /* 1493 * If the inpcb has been freed, let the caller know, even if 1494 * this isn't the last reference. 1495 */ 1496 if (inp->inp_flags2 & INP_FREED) { 1497 INP_WUNLOCK(inp); 1498 return (1); 1499 } 1500 return (0); 1501 } 1502 1503 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1504 #ifdef TCPHPTS 1505 if (inp->inp_in_hpts || inp->inp_in_input) { 1506 struct tcp_hpts_entry *hpts; 1507 /* 1508 * We should not be on the hpts at 1509 * this point in any form. we must 1510 * get the lock to be sure. 1511 */ 1512 hpts = tcp_hpts_lock(inp); 1513 if (inp->inp_in_hpts) 1514 panic("Hpts:%p inp:%p at free still on hpts", 1515 hpts, inp); 1516 mtx_unlock(&hpts->p_mtx); 1517 hpts = tcp_input_lock(inp); 1518 if (inp->inp_in_input) 1519 panic("Hpts:%p inp:%p at free still on input hpts", 1520 hpts, inp); 1521 mtx_unlock(&hpts->p_mtx); 1522 } 1523 #endif 1524 INP_WUNLOCK(inp); 1525 pcbinfo = inp->inp_pcbinfo; 1526 uma_zfree(pcbinfo->ipi_zone, inp); 1527 return (1); 1528 } 1529 1530 /* 1531 * Temporary wrapper. 1532 */ 1533 int 1534 in_pcbrele(struct inpcb *inp) 1535 { 1536 1537 return (in_pcbrele_wlocked(inp)); 1538 } 1539 1540 void 1541 in_pcblist_rele_rlocked(epoch_context_t ctx) 1542 { 1543 struct in_pcblist *il; 1544 struct inpcb *inp; 1545 struct inpcbinfo *pcbinfo; 1546 int i, n; 1547 1548 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1549 pcbinfo = il->il_pcbinfo; 1550 n = il->il_count; 1551 INP_INFO_WLOCK(pcbinfo); 1552 for (i = 0; i < n; i++) { 1553 inp = il->il_inp_list[i]; 1554 INP_RLOCK(inp); 1555 if (!in_pcbrele_rlocked(inp)) 1556 INP_RUNLOCK(inp); 1557 } 1558 INP_INFO_WUNLOCK(pcbinfo); 1559 free(il, M_TEMP); 1560 } 1561 1562 static void 1563 inpcbport_free(epoch_context_t ctx) 1564 { 1565 struct inpcbport *phd; 1566 1567 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); 1568 free(phd, M_PCB); 1569 } 1570 1571 static void 1572 in_pcbfree_deferred(epoch_context_t ctx) 1573 { 1574 struct inpcb *inp; 1575 int released __unused; 1576 1577 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); 1578 1579 INP_WLOCK(inp); 1580 #ifdef INET 1581 inp_freemoptions(inp->inp_moptions); 1582 inp->inp_moptions = NULL; 1583 #endif 1584 /* XXXRW: Do as much as possible here. */ 1585 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1586 if (inp->inp_sp != NULL) 1587 ipsec_delete_pcbpolicy(inp); 1588 #endif 1589 #ifdef INET6 1590 if (inp->inp_vflag & INP_IPV6PROTO) { 1591 ip6_freepcbopts(inp->in6p_outputopts); 1592 ip6_freemoptions(inp->in6p_moptions); 1593 inp->in6p_moptions = NULL; 1594 } 1595 #endif 1596 if (inp->inp_options) 1597 (void)m_free(inp->inp_options); 1598 inp->inp_vflag = 0; 1599 crfree(inp->inp_cred); 1600 #ifdef MAC 1601 mac_inpcb_destroy(inp); 1602 #endif 1603 released = in_pcbrele_wlocked(inp); 1604 MPASS(released); 1605 } 1606 1607 /* 1608 * Unconditionally schedule an inpcb to be freed by decrementing its 1609 * reference count, which should occur only after the inpcb has been detached 1610 * from its socket. If another thread holds a temporary reference (acquired 1611 * using in_pcbref()) then the free is deferred until that reference is 1612 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1613 * work, including removal from global lists, is done in this context, where 1614 * the pcbinfo lock is held. 1615 */ 1616 void 1617 in_pcbfree(struct inpcb *inp) 1618 { 1619 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1620 1621 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1622 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1623 ("%s: called twice for pcb %p", __func__, inp)); 1624 if (inp->inp_flags2 & INP_FREED) { 1625 INP_WUNLOCK(inp); 1626 return; 1627 } 1628 1629 #ifdef INVARIANTS 1630 if (pcbinfo == &V_tcbinfo) { 1631 INP_INFO_LOCK_ASSERT(pcbinfo); 1632 } else { 1633 INP_INFO_WLOCK_ASSERT(pcbinfo); 1634 } 1635 #endif 1636 INP_WLOCK_ASSERT(inp); 1637 INP_LIST_WLOCK(pcbinfo); 1638 in_pcbremlists(inp); 1639 INP_LIST_WUNLOCK(pcbinfo); 1640 RO_INVALIDATE_CACHE(&inp->inp_route); 1641 /* mark as destruction in progress */ 1642 inp->inp_flags2 |= INP_FREED; 1643 INP_WUNLOCK(inp); 1644 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred); 1645 } 1646 1647 /* 1648 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1649 * port reservation, and preventing it from being returned by inpcb lookups. 1650 * 1651 * It is used by TCP to mark an inpcb as unused and avoid future packet 1652 * delivery or event notification when a socket remains open but TCP has 1653 * closed. This might occur as a result of a shutdown()-initiated TCP close 1654 * or a RST on the wire, and allows the port binding to be reused while still 1655 * maintaining the invariant that so_pcb always points to a valid inpcb until 1656 * in_pcbdetach(). 1657 * 1658 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1659 * in_pcbnotifyall() and in_pcbpurgeif0()? 1660 */ 1661 void 1662 in_pcbdrop(struct inpcb *inp) 1663 { 1664 1665 INP_WLOCK_ASSERT(inp); 1666 #ifdef INVARIANTS 1667 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1668 MPASS(inp->inp_refcount > 1); 1669 #endif 1670 1671 /* 1672 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1673 * the hash lock...? 1674 */ 1675 inp->inp_flags |= INP_DROPPED; 1676 if (inp->inp_flags & INP_INHASHLIST) { 1677 struct inpcbport *phd = inp->inp_phd; 1678 1679 INP_HASH_WLOCK(inp->inp_pcbinfo); 1680 in_pcbremlbgrouphash(inp); 1681 CK_LIST_REMOVE(inp, inp_hash); 1682 CK_LIST_REMOVE(inp, inp_portlist); 1683 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 1684 CK_LIST_REMOVE(phd, phd_hash); 1685 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 1686 } 1687 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1688 inp->inp_flags &= ~INP_INHASHLIST; 1689 #ifdef PCBGROUP 1690 in_pcbgroup_remove(inp); 1691 #endif 1692 } 1693 } 1694 1695 #ifdef INET 1696 /* 1697 * Common routines to return the socket addresses associated with inpcbs. 1698 */ 1699 struct sockaddr * 1700 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1701 { 1702 struct sockaddr_in *sin; 1703 1704 sin = malloc(sizeof *sin, M_SONAME, 1705 M_WAITOK | M_ZERO); 1706 sin->sin_family = AF_INET; 1707 sin->sin_len = sizeof(*sin); 1708 sin->sin_addr = *addr_p; 1709 sin->sin_port = port; 1710 1711 return (struct sockaddr *)sin; 1712 } 1713 1714 int 1715 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1716 { 1717 struct inpcb *inp; 1718 struct in_addr addr; 1719 in_port_t port; 1720 1721 inp = sotoinpcb(so); 1722 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1723 1724 INP_RLOCK(inp); 1725 port = inp->inp_lport; 1726 addr = inp->inp_laddr; 1727 INP_RUNLOCK(inp); 1728 1729 *nam = in_sockaddr(port, &addr); 1730 return 0; 1731 } 1732 1733 int 1734 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1735 { 1736 struct inpcb *inp; 1737 struct in_addr addr; 1738 in_port_t port; 1739 1740 inp = sotoinpcb(so); 1741 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1742 1743 INP_RLOCK(inp); 1744 port = inp->inp_fport; 1745 addr = inp->inp_faddr; 1746 INP_RUNLOCK(inp); 1747 1748 *nam = in_sockaddr(port, &addr); 1749 return 0; 1750 } 1751 1752 void 1753 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1754 struct inpcb *(*notify)(struct inpcb *, int)) 1755 { 1756 struct inpcb *inp, *inp_temp; 1757 1758 INP_INFO_WLOCK(pcbinfo); 1759 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1760 INP_WLOCK(inp); 1761 #ifdef INET6 1762 if ((inp->inp_vflag & INP_IPV4) == 0) { 1763 INP_WUNLOCK(inp); 1764 continue; 1765 } 1766 #endif 1767 if (inp->inp_faddr.s_addr != faddr.s_addr || 1768 inp->inp_socket == NULL) { 1769 INP_WUNLOCK(inp); 1770 continue; 1771 } 1772 if ((*notify)(inp, errno)) 1773 INP_WUNLOCK(inp); 1774 } 1775 INP_INFO_WUNLOCK(pcbinfo); 1776 } 1777 1778 void 1779 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1780 { 1781 struct inpcb *inp; 1782 struct ip_moptions *imo; 1783 int i, gap; 1784 1785 INP_INFO_WLOCK(pcbinfo); 1786 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1787 INP_WLOCK(inp); 1788 imo = inp->inp_moptions; 1789 if ((inp->inp_vflag & INP_IPV4) && 1790 imo != NULL) { 1791 /* 1792 * Unselect the outgoing interface if it is being 1793 * detached. 1794 */ 1795 if (imo->imo_multicast_ifp == ifp) 1796 imo->imo_multicast_ifp = NULL; 1797 1798 /* 1799 * Drop multicast group membership if we joined 1800 * through the interface being detached. 1801 * 1802 * XXX This can all be deferred to an epoch_call 1803 */ 1804 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1805 i++) { 1806 if (imo->imo_membership[i]->inm_ifp == ifp) { 1807 IN_MULTI_LOCK_ASSERT(); 1808 in_leavegroup_locked(imo->imo_membership[i], NULL); 1809 gap++; 1810 } else if (gap != 0) 1811 imo->imo_membership[i - gap] = 1812 imo->imo_membership[i]; 1813 } 1814 imo->imo_num_memberships -= gap; 1815 } 1816 INP_WUNLOCK(inp); 1817 } 1818 INP_INFO_WUNLOCK(pcbinfo); 1819 } 1820 1821 /* 1822 * Lookup a PCB based on the local address and port. Caller must hold the 1823 * hash lock. No inpcb locks or references are acquired. 1824 */ 1825 #define INP_LOOKUP_MAPPED_PCB_COST 3 1826 struct inpcb * 1827 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1828 u_short lport, int lookupflags, struct ucred *cred) 1829 { 1830 struct inpcb *inp; 1831 #ifdef INET6 1832 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1833 #else 1834 int matchwild = 3; 1835 #endif 1836 int wildcard; 1837 1838 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1839 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1840 1841 INP_HASH_LOCK_ASSERT(pcbinfo); 1842 1843 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1844 struct inpcbhead *head; 1845 /* 1846 * Look for an unconnected (wildcard foreign addr) PCB that 1847 * matches the local address and port we're looking for. 1848 */ 1849 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1850 0, pcbinfo->ipi_hashmask)]; 1851 CK_LIST_FOREACH(inp, head, inp_hash) { 1852 #ifdef INET6 1853 /* XXX inp locking */ 1854 if ((inp->inp_vflag & INP_IPV4) == 0) 1855 continue; 1856 #endif 1857 if (inp->inp_faddr.s_addr == INADDR_ANY && 1858 inp->inp_laddr.s_addr == laddr.s_addr && 1859 inp->inp_lport == lport) { 1860 /* 1861 * Found? 1862 */ 1863 if (cred == NULL || 1864 prison_equal_ip4(cred->cr_prison, 1865 inp->inp_cred->cr_prison)) 1866 return (inp); 1867 } 1868 } 1869 /* 1870 * Not found. 1871 */ 1872 return (NULL); 1873 } else { 1874 struct inpcbporthead *porthash; 1875 struct inpcbport *phd; 1876 struct inpcb *match = NULL; 1877 /* 1878 * Best fit PCB lookup. 1879 * 1880 * First see if this local port is in use by looking on the 1881 * port hash list. 1882 */ 1883 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1884 pcbinfo->ipi_porthashmask)]; 1885 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1886 if (phd->phd_port == lport) 1887 break; 1888 } 1889 if (phd != NULL) { 1890 /* 1891 * Port is in use by one or more PCBs. Look for best 1892 * fit. 1893 */ 1894 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1895 wildcard = 0; 1896 if (cred != NULL && 1897 !prison_equal_ip4(inp->inp_cred->cr_prison, 1898 cred->cr_prison)) 1899 continue; 1900 #ifdef INET6 1901 /* XXX inp locking */ 1902 if ((inp->inp_vflag & INP_IPV4) == 0) 1903 continue; 1904 /* 1905 * We never select the PCB that has 1906 * INP_IPV6 flag and is bound to :: if 1907 * we have another PCB which is bound 1908 * to 0.0.0.0. If a PCB has the 1909 * INP_IPV6 flag, then we set its cost 1910 * higher than IPv4 only PCBs. 1911 * 1912 * Note that the case only happens 1913 * when a socket is bound to ::, under 1914 * the condition that the use of the 1915 * mapped address is allowed. 1916 */ 1917 if ((inp->inp_vflag & INP_IPV6) != 0) 1918 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1919 #endif 1920 if (inp->inp_faddr.s_addr != INADDR_ANY) 1921 wildcard++; 1922 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1923 if (laddr.s_addr == INADDR_ANY) 1924 wildcard++; 1925 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1926 continue; 1927 } else { 1928 if (laddr.s_addr != INADDR_ANY) 1929 wildcard++; 1930 } 1931 if (wildcard < matchwild) { 1932 match = inp; 1933 matchwild = wildcard; 1934 if (matchwild == 0) 1935 break; 1936 } 1937 } 1938 } 1939 return (match); 1940 } 1941 } 1942 #undef INP_LOOKUP_MAPPED_PCB_COST 1943 1944 static struct inpcb * 1945 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 1946 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, 1947 uint16_t fport, int lookupflags) 1948 { 1949 struct inpcb *local_wild = NULL; 1950 const struct inpcblbgrouphead *hdr; 1951 struct inpcblbgroup *grp; 1952 struct inpcblbgroup *grp_local_wild; 1953 1954 INP_HASH_LOCK_ASSERT(pcbinfo); 1955 1956 hdr = &pcbinfo->ipi_lbgrouphashbase[ 1957 INP_PCBLBGROUP_PORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; 1958 1959 /* 1960 * Order of socket selection: 1961 * 1. non-wild. 1962 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). 1963 * 1964 * NOTE: 1965 * - Load balanced group does not contain jailed sockets 1966 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets 1967 */ 1968 LIST_FOREACH(grp, hdr, il_list) { 1969 #ifdef INET6 1970 if (!(grp->il_vflag & INP_IPV4)) 1971 continue; 1972 #endif 1973 1974 if (grp->il_lport == lport) { 1975 1976 uint32_t idx = 0; 1977 int pkt_hash = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, 1978 lport, fport); 1979 1980 idx = pkt_hash % grp->il_inpcnt; 1981 1982 if (grp->il_laddr.s_addr == laddr->s_addr) { 1983 return (grp->il_inp[idx]); 1984 } else { 1985 if (grp->il_laddr.s_addr == INADDR_ANY && 1986 (lookupflags & INPLOOKUP_WILDCARD)) { 1987 local_wild = grp->il_inp[idx]; 1988 grp_local_wild = grp; 1989 } 1990 } 1991 } 1992 } 1993 if (local_wild != NULL) { 1994 return (local_wild); 1995 } 1996 return (NULL); 1997 } 1998 1999 #ifdef PCBGROUP 2000 /* 2001 * Lookup PCB in hash list, using pcbgroup tables. 2002 */ 2003 static struct inpcb * 2004 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 2005 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 2006 u_int lport_arg, int lookupflags, struct ifnet *ifp) 2007 { 2008 struct inpcbhead *head; 2009 struct inpcb *inp, *tmpinp; 2010 u_short fport = fport_arg, lport = lport_arg; 2011 bool locked; 2012 2013 /* 2014 * First look for an exact match. 2015 */ 2016 tmpinp = NULL; 2017 INP_GROUP_LOCK(pcbgroup); 2018 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2019 pcbgroup->ipg_hashmask)]; 2020 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2021 #ifdef INET6 2022 /* XXX inp locking */ 2023 if ((inp->inp_vflag & INP_IPV4) == 0) 2024 continue; 2025 #endif 2026 if (inp->inp_faddr.s_addr == faddr.s_addr && 2027 inp->inp_laddr.s_addr == laddr.s_addr && 2028 inp->inp_fport == fport && 2029 inp->inp_lport == lport) { 2030 /* 2031 * XXX We should be able to directly return 2032 * the inp here, without any checks. 2033 * Well unless both bound with SO_REUSEPORT? 2034 */ 2035 if (prison_flag(inp->inp_cred, PR_IP4)) 2036 goto found; 2037 if (tmpinp == NULL) 2038 tmpinp = inp; 2039 } 2040 } 2041 if (tmpinp != NULL) { 2042 inp = tmpinp; 2043 goto found; 2044 } 2045 2046 #ifdef RSS 2047 /* 2048 * For incoming connections, we may wish to do a wildcard 2049 * match for an RSS-local socket. 2050 */ 2051 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2052 struct inpcb *local_wild = NULL, *local_exact = NULL; 2053 #ifdef INET6 2054 struct inpcb *local_wild_mapped = NULL; 2055 #endif 2056 struct inpcb *jail_wild = NULL; 2057 struct inpcbhead *head; 2058 int injail; 2059 2060 /* 2061 * Order of socket selection - we always prefer jails. 2062 * 1. jailed, non-wild. 2063 * 2. jailed, wild. 2064 * 3. non-jailed, non-wild. 2065 * 4. non-jailed, wild. 2066 */ 2067 2068 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 2069 lport, 0, pcbgroup->ipg_hashmask)]; 2070 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2071 #ifdef INET6 2072 /* XXX inp locking */ 2073 if ((inp->inp_vflag & INP_IPV4) == 0) 2074 continue; 2075 #endif 2076 if (inp->inp_faddr.s_addr != INADDR_ANY || 2077 inp->inp_lport != lport) 2078 continue; 2079 2080 injail = prison_flag(inp->inp_cred, PR_IP4); 2081 if (injail) { 2082 if (prison_check_ip4(inp->inp_cred, 2083 &laddr) != 0) 2084 continue; 2085 } else { 2086 if (local_exact != NULL) 2087 continue; 2088 } 2089 2090 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2091 if (injail) 2092 goto found; 2093 else 2094 local_exact = inp; 2095 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2096 #ifdef INET6 2097 /* XXX inp locking, NULL check */ 2098 if (inp->inp_vflag & INP_IPV6PROTO) 2099 local_wild_mapped = inp; 2100 else 2101 #endif 2102 if (injail) 2103 jail_wild = inp; 2104 else 2105 local_wild = inp; 2106 } 2107 } /* LIST_FOREACH */ 2108 2109 inp = jail_wild; 2110 if (inp == NULL) 2111 inp = local_exact; 2112 if (inp == NULL) 2113 inp = local_wild; 2114 #ifdef INET6 2115 if (inp == NULL) 2116 inp = local_wild_mapped; 2117 #endif 2118 if (inp != NULL) 2119 goto found; 2120 } 2121 #endif 2122 2123 /* 2124 * Then look for a wildcard match, if requested. 2125 */ 2126 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2127 struct inpcb *local_wild = NULL, *local_exact = NULL; 2128 #ifdef INET6 2129 struct inpcb *local_wild_mapped = NULL; 2130 #endif 2131 struct inpcb *jail_wild = NULL; 2132 struct inpcbhead *head; 2133 int injail; 2134 2135 /* 2136 * Order of socket selection - we always prefer jails. 2137 * 1. jailed, non-wild. 2138 * 2. jailed, wild. 2139 * 3. non-jailed, non-wild. 2140 * 4. non-jailed, wild. 2141 */ 2142 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 2143 0, pcbinfo->ipi_wildmask)]; 2144 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 2145 #ifdef INET6 2146 /* XXX inp locking */ 2147 if ((inp->inp_vflag & INP_IPV4) == 0) 2148 continue; 2149 #endif 2150 if (inp->inp_faddr.s_addr != INADDR_ANY || 2151 inp->inp_lport != lport) 2152 continue; 2153 2154 injail = prison_flag(inp->inp_cred, PR_IP4); 2155 if (injail) { 2156 if (prison_check_ip4(inp->inp_cred, 2157 &laddr) != 0) 2158 continue; 2159 } else { 2160 if (local_exact != NULL) 2161 continue; 2162 } 2163 2164 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2165 if (injail) 2166 goto found; 2167 else 2168 local_exact = inp; 2169 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2170 #ifdef INET6 2171 /* XXX inp locking, NULL check */ 2172 if (inp->inp_vflag & INP_IPV6PROTO) 2173 local_wild_mapped = inp; 2174 else 2175 #endif 2176 if (injail) 2177 jail_wild = inp; 2178 else 2179 local_wild = inp; 2180 } 2181 } /* LIST_FOREACH */ 2182 inp = jail_wild; 2183 if (inp == NULL) 2184 inp = local_exact; 2185 if (inp == NULL) 2186 inp = local_wild; 2187 #ifdef INET6 2188 if (inp == NULL) 2189 inp = local_wild_mapped; 2190 #endif 2191 if (inp != NULL) 2192 goto found; 2193 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 2194 INP_GROUP_UNLOCK(pcbgroup); 2195 return (NULL); 2196 2197 found: 2198 if (lookupflags & INPLOOKUP_WLOCKPCB) 2199 locked = INP_TRY_WLOCK(inp); 2200 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2201 locked = INP_TRY_RLOCK(inp); 2202 else 2203 panic("%s: locking bug", __func__); 2204 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { 2205 if (lookupflags & INPLOOKUP_WLOCKPCB) 2206 INP_WUNLOCK(inp); 2207 else 2208 INP_RUNLOCK(inp); 2209 return (NULL); 2210 } else if (!locked) 2211 in_pcbref(inp); 2212 INP_GROUP_UNLOCK(pcbgroup); 2213 if (!locked) { 2214 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2215 INP_WLOCK(inp); 2216 if (in_pcbrele_wlocked(inp)) 2217 return (NULL); 2218 } else { 2219 INP_RLOCK(inp); 2220 if (in_pcbrele_rlocked(inp)) 2221 return (NULL); 2222 } 2223 } 2224 #ifdef INVARIANTS 2225 if (lookupflags & INPLOOKUP_WLOCKPCB) 2226 INP_WLOCK_ASSERT(inp); 2227 else 2228 INP_RLOCK_ASSERT(inp); 2229 #endif 2230 return (inp); 2231 } 2232 #endif /* PCBGROUP */ 2233 2234 /* 2235 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2236 * that the caller has locked the hash list, and will not perform any further 2237 * locking or reference operations on either the hash list or the connection. 2238 */ 2239 static struct inpcb * 2240 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2241 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2242 struct ifnet *ifp) 2243 { 2244 struct inpcbhead *head; 2245 struct inpcb *inp, *tmpinp; 2246 u_short fport = fport_arg, lport = lport_arg; 2247 2248 #ifdef INVARIANTS 2249 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2250 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2251 if (!mtx_owned(&pcbinfo->ipi_hash_lock)) 2252 MPASS(in_epoch_verbose(net_epoch_preempt, 1)); 2253 #endif 2254 /* 2255 * First look for an exact match. 2256 */ 2257 tmpinp = NULL; 2258 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2259 pcbinfo->ipi_hashmask)]; 2260 CK_LIST_FOREACH(inp, head, inp_hash) { 2261 #ifdef INET6 2262 /* XXX inp locking */ 2263 if ((inp->inp_vflag & INP_IPV4) == 0) 2264 continue; 2265 #endif 2266 if (inp->inp_faddr.s_addr == faddr.s_addr && 2267 inp->inp_laddr.s_addr == laddr.s_addr && 2268 inp->inp_fport == fport && 2269 inp->inp_lport == lport) { 2270 /* 2271 * XXX We should be able to directly return 2272 * the inp here, without any checks. 2273 * Well unless both bound with SO_REUSEPORT? 2274 */ 2275 if (prison_flag(inp->inp_cred, PR_IP4)) 2276 return (inp); 2277 if (tmpinp == NULL) 2278 tmpinp = inp; 2279 } 2280 } 2281 if (tmpinp != NULL) 2282 return (tmpinp); 2283 2284 /* 2285 * Then look in lb group (for wildcard match). 2286 */ 2287 if (pcbinfo->ipi_lbgrouphashbase != NULL && 2288 (lookupflags & INPLOOKUP_WILDCARD)) { 2289 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, 2290 fport, lookupflags); 2291 if (inp != NULL) { 2292 return (inp); 2293 } 2294 } 2295 2296 /* 2297 * Then look for a wildcard match, if requested. 2298 */ 2299 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2300 struct inpcb *local_wild = NULL, *local_exact = NULL; 2301 #ifdef INET6 2302 struct inpcb *local_wild_mapped = NULL; 2303 #endif 2304 struct inpcb *jail_wild = NULL; 2305 int injail; 2306 2307 /* 2308 * Order of socket selection - we always prefer jails. 2309 * 1. jailed, non-wild. 2310 * 2. jailed, wild. 2311 * 3. non-jailed, non-wild. 2312 * 4. non-jailed, wild. 2313 */ 2314 2315 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2316 0, pcbinfo->ipi_hashmask)]; 2317 CK_LIST_FOREACH(inp, head, inp_hash) { 2318 #ifdef INET6 2319 /* XXX inp locking */ 2320 if ((inp->inp_vflag & INP_IPV4) == 0) 2321 continue; 2322 #endif 2323 if (inp->inp_faddr.s_addr != INADDR_ANY || 2324 inp->inp_lport != lport) 2325 continue; 2326 2327 injail = prison_flag(inp->inp_cred, PR_IP4); 2328 if (injail) { 2329 if (prison_check_ip4(inp->inp_cred, 2330 &laddr) != 0) 2331 continue; 2332 } else { 2333 if (local_exact != NULL) 2334 continue; 2335 } 2336 2337 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2338 if (injail) 2339 return (inp); 2340 else 2341 local_exact = inp; 2342 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2343 #ifdef INET6 2344 /* XXX inp locking, NULL check */ 2345 if (inp->inp_vflag & INP_IPV6PROTO) 2346 local_wild_mapped = inp; 2347 else 2348 #endif 2349 if (injail) 2350 jail_wild = inp; 2351 else 2352 local_wild = inp; 2353 } 2354 } /* LIST_FOREACH */ 2355 if (jail_wild != NULL) 2356 return (jail_wild); 2357 if (local_exact != NULL) 2358 return (local_exact); 2359 if (local_wild != NULL) 2360 return (local_wild); 2361 #ifdef INET6 2362 if (local_wild_mapped != NULL) 2363 return (local_wild_mapped); 2364 #endif 2365 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2366 2367 return (NULL); 2368 } 2369 2370 /* 2371 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2372 * hash list lock, and will return the inpcb locked (i.e., requires 2373 * INPLOOKUP_LOCKPCB). 2374 */ 2375 static struct inpcb * 2376 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2377 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2378 struct ifnet *ifp) 2379 { 2380 struct inpcb *inp; 2381 2382 INP_HASH_RLOCK(pcbinfo); 2383 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2384 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 2385 if (inp != NULL) { 2386 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2387 INP_WLOCK(inp); 2388 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2389 INP_WUNLOCK(inp); 2390 inp = NULL; 2391 } 2392 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2393 INP_RLOCK(inp); 2394 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2395 INP_RUNLOCK(inp); 2396 inp = NULL; 2397 } 2398 } else 2399 panic("%s: locking bug", __func__); 2400 #ifdef INVARIANTS 2401 if (inp != NULL) { 2402 if (lookupflags & INPLOOKUP_WLOCKPCB) 2403 INP_WLOCK_ASSERT(inp); 2404 else 2405 INP_RLOCK_ASSERT(inp); 2406 } 2407 #endif 2408 } 2409 INP_HASH_RUNLOCK(pcbinfo); 2410 return (inp); 2411 } 2412 2413 /* 2414 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2415 * from which a pre-calculated hash value may be extracted. 2416 * 2417 * Possibly more of this logic should be in in_pcbgroup.c. 2418 */ 2419 struct inpcb * 2420 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2421 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2422 { 2423 #if defined(PCBGROUP) && !defined(RSS) 2424 struct inpcbgroup *pcbgroup; 2425 #endif 2426 2427 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2428 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2429 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2430 ("%s: LOCKPCB not set", __func__)); 2431 2432 /* 2433 * When not using RSS, use connection groups in preference to the 2434 * reservation table when looking up 4-tuples. When using RSS, just 2435 * use the reservation table, due to the cost of the Toeplitz hash 2436 * in software. 2437 * 2438 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2439 * we could be doing RSS with a non-Toeplitz hash that is affordable 2440 * in software. 2441 */ 2442 #if defined(PCBGROUP) && !defined(RSS) 2443 if (in_pcbgroup_enabled(pcbinfo)) { 2444 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2445 fport); 2446 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2447 laddr, lport, lookupflags, ifp)); 2448 } 2449 #endif 2450 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2451 lookupflags, ifp)); 2452 } 2453 2454 struct inpcb * 2455 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2456 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2457 struct ifnet *ifp, struct mbuf *m) 2458 { 2459 #ifdef PCBGROUP 2460 struct inpcbgroup *pcbgroup; 2461 #endif 2462 2463 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2464 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2465 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2466 ("%s: LOCKPCB not set", __func__)); 2467 2468 #ifdef PCBGROUP 2469 /* 2470 * If we can use a hardware-generated hash to look up the connection 2471 * group, use that connection group to find the inpcb. Otherwise 2472 * fall back on a software hash -- or the reservation table if we're 2473 * using RSS. 2474 * 2475 * XXXRW: As above, that policy belongs in the pcbgroup code. 2476 */ 2477 if (in_pcbgroup_enabled(pcbinfo) && 2478 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2479 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2480 m->m_pkthdr.flowid); 2481 if (pcbgroup != NULL) 2482 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2483 fport, laddr, lport, lookupflags, ifp)); 2484 #ifndef RSS 2485 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2486 fport); 2487 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2488 laddr, lport, lookupflags, ifp)); 2489 #endif 2490 } 2491 #endif 2492 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2493 lookupflags, ifp)); 2494 } 2495 #endif /* INET */ 2496 2497 /* 2498 * Insert PCB onto various hash lists. 2499 */ 2500 static int 2501 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 2502 { 2503 struct inpcbhead *pcbhash; 2504 struct inpcbporthead *pcbporthash; 2505 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2506 struct inpcbport *phd; 2507 u_int32_t hashkey_faddr; 2508 int so_options; 2509 2510 INP_WLOCK_ASSERT(inp); 2511 INP_HASH_WLOCK_ASSERT(pcbinfo); 2512 2513 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2514 ("in_pcbinshash: INP_INHASHLIST")); 2515 2516 #ifdef INET6 2517 if (inp->inp_vflag & INP_IPV6) 2518 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2519 else 2520 #endif 2521 hashkey_faddr = inp->inp_faddr.s_addr; 2522 2523 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2524 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2525 2526 pcbporthash = &pcbinfo->ipi_porthashbase[ 2527 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2528 2529 /* 2530 * Add entry to load balance group. 2531 * Only do this if SO_REUSEPORT_LB is set. 2532 */ 2533 so_options = inp_so_options(inp); 2534 if (so_options & SO_REUSEPORT_LB) { 2535 int ret = in_pcbinslbgrouphash(inp); 2536 if (ret) { 2537 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2538 return (ret); 2539 } 2540 } 2541 2542 /* 2543 * Go through port list and look for a head for this lport. 2544 */ 2545 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2546 if (phd->phd_port == inp->inp_lport) 2547 break; 2548 } 2549 /* 2550 * If none exists, malloc one and tack it on. 2551 */ 2552 if (phd == NULL) { 2553 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2554 if (phd == NULL) { 2555 return (ENOBUFS); /* XXX */ 2556 } 2557 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2558 phd->phd_port = inp->inp_lport; 2559 CK_LIST_INIT(&phd->phd_pcblist); 2560 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2561 } 2562 inp->inp_phd = phd; 2563 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2564 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2565 inp->inp_flags |= INP_INHASHLIST; 2566 #ifdef PCBGROUP 2567 if (do_pcbgroup_update) 2568 in_pcbgroup_update(inp); 2569 #endif 2570 return (0); 2571 } 2572 2573 /* 2574 * For now, there are two public interfaces to insert an inpcb into the hash 2575 * lists -- one that does update pcbgroups, and one that doesn't. The latter 2576 * is used only in the TCP syncache, where in_pcbinshash is called before the 2577 * full 4-tuple is set for the inpcb, and we don't want to install in the 2578 * pcbgroup until later. 2579 * 2580 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 2581 * connection groups, and partially initialised inpcbs should not be exposed 2582 * to either reservation hash tables or pcbgroups. 2583 */ 2584 int 2585 in_pcbinshash(struct inpcb *inp) 2586 { 2587 2588 return (in_pcbinshash_internal(inp, 1)); 2589 } 2590 2591 int 2592 in_pcbinshash_nopcbgroup(struct inpcb *inp) 2593 { 2594 2595 return (in_pcbinshash_internal(inp, 0)); 2596 } 2597 2598 /* 2599 * Move PCB to the proper hash bucket when { faddr, fport } have been 2600 * changed. NOTE: This does not handle the case of the lport changing (the 2601 * hashed port list would have to be updated as well), so the lport must 2602 * not change after in_pcbinshash() has been called. 2603 */ 2604 void 2605 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2606 { 2607 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2608 struct inpcbhead *head; 2609 u_int32_t hashkey_faddr; 2610 2611 INP_WLOCK_ASSERT(inp); 2612 INP_HASH_WLOCK_ASSERT(pcbinfo); 2613 2614 KASSERT(inp->inp_flags & INP_INHASHLIST, 2615 ("in_pcbrehash: !INP_INHASHLIST")); 2616 2617 #ifdef INET6 2618 if (inp->inp_vflag & INP_IPV6) 2619 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2620 else 2621 #endif 2622 hashkey_faddr = inp->inp_faddr.s_addr; 2623 2624 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2625 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2626 2627 CK_LIST_REMOVE(inp, inp_hash); 2628 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2629 2630 #ifdef PCBGROUP 2631 if (m != NULL) 2632 in_pcbgroup_update_mbuf(inp, m); 2633 else 2634 in_pcbgroup_update(inp); 2635 #endif 2636 } 2637 2638 void 2639 in_pcbrehash(struct inpcb *inp) 2640 { 2641 2642 in_pcbrehash_mbuf(inp, NULL); 2643 } 2644 2645 /* 2646 * Remove PCB from various lists. 2647 */ 2648 static void 2649 in_pcbremlists(struct inpcb *inp) 2650 { 2651 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2652 2653 #ifdef INVARIANTS 2654 if (pcbinfo == &V_tcbinfo) { 2655 INP_INFO_RLOCK_ASSERT(pcbinfo); 2656 } else { 2657 INP_INFO_WLOCK_ASSERT(pcbinfo); 2658 } 2659 #endif 2660 2661 INP_WLOCK_ASSERT(inp); 2662 INP_LIST_WLOCK_ASSERT(pcbinfo); 2663 2664 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2665 if (inp->inp_flags & INP_INHASHLIST) { 2666 struct inpcbport *phd = inp->inp_phd; 2667 2668 INP_HASH_WLOCK(pcbinfo); 2669 2670 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2671 in_pcbremlbgrouphash(inp); 2672 2673 CK_LIST_REMOVE(inp, inp_hash); 2674 CK_LIST_REMOVE(inp, inp_portlist); 2675 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2676 CK_LIST_REMOVE(phd, phd_hash); 2677 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 2678 } 2679 INP_HASH_WUNLOCK(pcbinfo); 2680 inp->inp_flags &= ~INP_INHASHLIST; 2681 } 2682 CK_LIST_REMOVE(inp, inp_list); 2683 pcbinfo->ipi_count--; 2684 #ifdef PCBGROUP 2685 in_pcbgroup_remove(inp); 2686 #endif 2687 } 2688 2689 /* 2690 * Check for alternatives when higher level complains 2691 * about service problems. For now, invalidate cached 2692 * routing information. If the route was created dynamically 2693 * (by a redirect), time to try a default gateway again. 2694 */ 2695 void 2696 in_losing(struct inpcb *inp) 2697 { 2698 2699 RO_INVALIDATE_CACHE(&inp->inp_route); 2700 return; 2701 } 2702 2703 /* 2704 * A set label operation has occurred at the socket layer, propagate the 2705 * label change into the in_pcb for the socket. 2706 */ 2707 void 2708 in_pcbsosetlabel(struct socket *so) 2709 { 2710 #ifdef MAC 2711 struct inpcb *inp; 2712 2713 inp = sotoinpcb(so); 2714 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2715 2716 INP_WLOCK(inp); 2717 SOCK_LOCK(so); 2718 mac_inpcb_sosetlabel(so, inp); 2719 SOCK_UNLOCK(so); 2720 INP_WUNLOCK(inp); 2721 #endif 2722 } 2723 2724 /* 2725 * ipport_tick runs once per second, determining if random port allocation 2726 * should be continued. If more than ipport_randomcps ports have been 2727 * allocated in the last second, then we return to sequential port 2728 * allocation. We return to random allocation only once we drop below 2729 * ipport_randomcps for at least ipport_randomtime seconds. 2730 */ 2731 static void 2732 ipport_tick(void *xtp) 2733 { 2734 VNET_ITERATOR_DECL(vnet_iter); 2735 2736 VNET_LIST_RLOCK_NOSLEEP(); 2737 VNET_FOREACH(vnet_iter) { 2738 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2739 if (V_ipport_tcpallocs <= 2740 V_ipport_tcplastcount + V_ipport_randomcps) { 2741 if (V_ipport_stoprandom > 0) 2742 V_ipport_stoprandom--; 2743 } else 2744 V_ipport_stoprandom = V_ipport_randomtime; 2745 V_ipport_tcplastcount = V_ipport_tcpallocs; 2746 CURVNET_RESTORE(); 2747 } 2748 VNET_LIST_RUNLOCK_NOSLEEP(); 2749 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2750 } 2751 2752 static void 2753 ip_fini(void *xtp) 2754 { 2755 2756 callout_stop(&ipport_tick_callout); 2757 } 2758 2759 /* 2760 * The ipport_callout should start running at about the time we attach the 2761 * inet or inet6 domains. 2762 */ 2763 static void 2764 ipport_tick_init(const void *unused __unused) 2765 { 2766 2767 /* Start ipport_tick. */ 2768 callout_init(&ipport_tick_callout, 1); 2769 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2770 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2771 SHUTDOWN_PRI_DEFAULT); 2772 } 2773 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2774 ipport_tick_init, NULL); 2775 2776 void 2777 inp_wlock(struct inpcb *inp) 2778 { 2779 2780 INP_WLOCK(inp); 2781 } 2782 2783 void 2784 inp_wunlock(struct inpcb *inp) 2785 { 2786 2787 INP_WUNLOCK(inp); 2788 } 2789 2790 void 2791 inp_rlock(struct inpcb *inp) 2792 { 2793 2794 INP_RLOCK(inp); 2795 } 2796 2797 void 2798 inp_runlock(struct inpcb *inp) 2799 { 2800 2801 INP_RUNLOCK(inp); 2802 } 2803 2804 #ifdef INVARIANT_SUPPORT 2805 void 2806 inp_lock_assert(struct inpcb *inp) 2807 { 2808 2809 INP_WLOCK_ASSERT(inp); 2810 } 2811 2812 void 2813 inp_unlock_assert(struct inpcb *inp) 2814 { 2815 2816 INP_UNLOCK_ASSERT(inp); 2817 } 2818 #endif 2819 2820 void 2821 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2822 { 2823 struct inpcb *inp; 2824 2825 INP_INFO_WLOCK(&V_tcbinfo); 2826 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2827 INP_WLOCK(inp); 2828 func(inp, arg); 2829 INP_WUNLOCK(inp); 2830 } 2831 INP_INFO_WUNLOCK(&V_tcbinfo); 2832 } 2833 2834 struct socket * 2835 inp_inpcbtosocket(struct inpcb *inp) 2836 { 2837 2838 INP_WLOCK_ASSERT(inp); 2839 return (inp->inp_socket); 2840 } 2841 2842 struct tcpcb * 2843 inp_inpcbtotcpcb(struct inpcb *inp) 2844 { 2845 2846 INP_WLOCK_ASSERT(inp); 2847 return ((struct tcpcb *)inp->inp_ppcb); 2848 } 2849 2850 int 2851 inp_ip_tos_get(const struct inpcb *inp) 2852 { 2853 2854 return (inp->inp_ip_tos); 2855 } 2856 2857 void 2858 inp_ip_tos_set(struct inpcb *inp, int val) 2859 { 2860 2861 inp->inp_ip_tos = val; 2862 } 2863 2864 void 2865 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2866 uint32_t *faddr, uint16_t *fp) 2867 { 2868 2869 INP_LOCK_ASSERT(inp); 2870 *laddr = inp->inp_laddr.s_addr; 2871 *faddr = inp->inp_faddr.s_addr; 2872 *lp = inp->inp_lport; 2873 *fp = inp->inp_fport; 2874 } 2875 2876 struct inpcb * 2877 so_sotoinpcb(struct socket *so) 2878 { 2879 2880 return (sotoinpcb(so)); 2881 } 2882 2883 struct tcpcb * 2884 so_sototcpcb(struct socket *so) 2885 { 2886 2887 return (sototcpcb(so)); 2888 } 2889 2890 /* 2891 * Create an external-format (``xinpcb'') structure using the information in 2892 * the kernel-format in_pcb structure pointed to by inp. This is done to 2893 * reduce the spew of irrelevant information over this interface, to isolate 2894 * user code from changes in the kernel structure, and potentially to provide 2895 * information-hiding if we decide that some of this information should be 2896 * hidden from users. 2897 */ 2898 void 2899 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2900 { 2901 2902 xi->xi_len = sizeof(struct xinpcb); 2903 if (inp->inp_socket) 2904 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2905 else 2906 bzero(&xi->xi_socket, sizeof(struct xsocket)); 2907 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2908 xi->inp_gencnt = inp->inp_gencnt; 2909 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2910 xi->inp_flow = inp->inp_flow; 2911 xi->inp_flowid = inp->inp_flowid; 2912 xi->inp_flowtype = inp->inp_flowtype; 2913 xi->inp_flags = inp->inp_flags; 2914 xi->inp_flags2 = inp->inp_flags2; 2915 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2916 xi->in6p_cksum = inp->in6p_cksum; 2917 xi->in6p_hops = inp->in6p_hops; 2918 xi->inp_ip_tos = inp->inp_ip_tos; 2919 xi->inp_vflag = inp->inp_vflag; 2920 xi->inp_ip_ttl = inp->inp_ip_ttl; 2921 xi->inp_ip_p = inp->inp_ip_p; 2922 xi->inp_ip_minttl = inp->inp_ip_minttl; 2923 } 2924 2925 #ifdef DDB 2926 static void 2927 db_print_indent(int indent) 2928 { 2929 int i; 2930 2931 for (i = 0; i < indent; i++) 2932 db_printf(" "); 2933 } 2934 2935 static void 2936 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2937 { 2938 char faddr_str[48], laddr_str[48]; 2939 2940 db_print_indent(indent); 2941 db_printf("%s at %p\n", name, inc); 2942 2943 indent += 2; 2944 2945 #ifdef INET6 2946 if (inc->inc_flags & INC_ISIPV6) { 2947 /* IPv6. */ 2948 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2949 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2950 } else 2951 #endif 2952 { 2953 /* IPv4. */ 2954 inet_ntoa_r(inc->inc_laddr, laddr_str); 2955 inet_ntoa_r(inc->inc_faddr, faddr_str); 2956 } 2957 db_print_indent(indent); 2958 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2959 ntohs(inc->inc_lport)); 2960 db_print_indent(indent); 2961 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2962 ntohs(inc->inc_fport)); 2963 } 2964 2965 static void 2966 db_print_inpflags(int inp_flags) 2967 { 2968 int comma; 2969 2970 comma = 0; 2971 if (inp_flags & INP_RECVOPTS) { 2972 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2973 comma = 1; 2974 } 2975 if (inp_flags & INP_RECVRETOPTS) { 2976 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2977 comma = 1; 2978 } 2979 if (inp_flags & INP_RECVDSTADDR) { 2980 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2981 comma = 1; 2982 } 2983 if (inp_flags & INP_ORIGDSTADDR) { 2984 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2985 comma = 1; 2986 } 2987 if (inp_flags & INP_HDRINCL) { 2988 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2989 comma = 1; 2990 } 2991 if (inp_flags & INP_HIGHPORT) { 2992 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2993 comma = 1; 2994 } 2995 if (inp_flags & INP_LOWPORT) { 2996 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 2997 comma = 1; 2998 } 2999 if (inp_flags & INP_ANONPORT) { 3000 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3001 comma = 1; 3002 } 3003 if (inp_flags & INP_RECVIF) { 3004 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3005 comma = 1; 3006 } 3007 if (inp_flags & INP_MTUDISC) { 3008 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3009 comma = 1; 3010 } 3011 if (inp_flags & INP_RECVTTL) { 3012 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3013 comma = 1; 3014 } 3015 if (inp_flags & INP_DONTFRAG) { 3016 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3017 comma = 1; 3018 } 3019 if (inp_flags & INP_RECVTOS) { 3020 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3021 comma = 1; 3022 } 3023 if (inp_flags & IN6P_IPV6_V6ONLY) { 3024 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3025 comma = 1; 3026 } 3027 if (inp_flags & IN6P_PKTINFO) { 3028 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3029 comma = 1; 3030 } 3031 if (inp_flags & IN6P_HOPLIMIT) { 3032 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3033 comma = 1; 3034 } 3035 if (inp_flags & IN6P_HOPOPTS) { 3036 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3037 comma = 1; 3038 } 3039 if (inp_flags & IN6P_DSTOPTS) { 3040 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3041 comma = 1; 3042 } 3043 if (inp_flags & IN6P_RTHDR) { 3044 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3045 comma = 1; 3046 } 3047 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3048 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3049 comma = 1; 3050 } 3051 if (inp_flags & IN6P_TCLASS) { 3052 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3053 comma = 1; 3054 } 3055 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3056 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3057 comma = 1; 3058 } 3059 if (inp_flags & INP_TIMEWAIT) { 3060 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3061 comma = 1; 3062 } 3063 if (inp_flags & INP_ONESBCAST) { 3064 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3065 comma = 1; 3066 } 3067 if (inp_flags & INP_DROPPED) { 3068 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3069 comma = 1; 3070 } 3071 if (inp_flags & INP_SOCKREF) { 3072 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3073 comma = 1; 3074 } 3075 if (inp_flags & IN6P_RFC2292) { 3076 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3077 comma = 1; 3078 } 3079 if (inp_flags & IN6P_MTU) { 3080 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3081 comma = 1; 3082 } 3083 } 3084 3085 static void 3086 db_print_inpvflag(u_char inp_vflag) 3087 { 3088 int comma; 3089 3090 comma = 0; 3091 if (inp_vflag & INP_IPV4) { 3092 db_printf("%sINP_IPV4", comma ? ", " : ""); 3093 comma = 1; 3094 } 3095 if (inp_vflag & INP_IPV6) { 3096 db_printf("%sINP_IPV6", comma ? ", " : ""); 3097 comma = 1; 3098 } 3099 if (inp_vflag & INP_IPV6PROTO) { 3100 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3101 comma = 1; 3102 } 3103 } 3104 3105 static void 3106 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3107 { 3108 3109 db_print_indent(indent); 3110 db_printf("%s at %p\n", name, inp); 3111 3112 indent += 2; 3113 3114 db_print_indent(indent); 3115 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3116 3117 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3118 3119 db_print_indent(indent); 3120 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3121 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3122 3123 db_print_indent(indent); 3124 db_printf("inp_label: %p inp_flags: 0x%x (", 3125 inp->inp_label, inp->inp_flags); 3126 db_print_inpflags(inp->inp_flags); 3127 db_printf(")\n"); 3128 3129 db_print_indent(indent); 3130 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3131 inp->inp_vflag); 3132 db_print_inpvflag(inp->inp_vflag); 3133 db_printf(")\n"); 3134 3135 db_print_indent(indent); 3136 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3137 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3138 3139 db_print_indent(indent); 3140 #ifdef INET6 3141 if (inp->inp_vflag & INP_IPV6) { 3142 db_printf("in6p_options: %p in6p_outputopts: %p " 3143 "in6p_moptions: %p\n", inp->in6p_options, 3144 inp->in6p_outputopts, inp->in6p_moptions); 3145 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3146 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3147 inp->in6p_hops); 3148 } else 3149 #endif 3150 { 3151 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3152 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3153 inp->inp_options, inp->inp_moptions); 3154 } 3155 3156 db_print_indent(indent); 3157 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3158 (uintmax_t)inp->inp_gencnt); 3159 } 3160 3161 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3162 { 3163 struct inpcb *inp; 3164 3165 if (!have_addr) { 3166 db_printf("usage: show inpcb <addr>\n"); 3167 return; 3168 } 3169 inp = (struct inpcb *)addr; 3170 3171 db_print_inpcb(inp, "inpcb", 0); 3172 } 3173 #endif /* DDB */ 3174 3175 #ifdef RATELIMIT 3176 /* 3177 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3178 * if any. 3179 */ 3180 int 3181 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3182 { 3183 union if_snd_tag_modify_params params = { 3184 .rate_limit.max_rate = max_pacing_rate, 3185 }; 3186 struct m_snd_tag *mst; 3187 struct ifnet *ifp; 3188 int error; 3189 3190 mst = inp->inp_snd_tag; 3191 if (mst == NULL) 3192 return (EINVAL); 3193 3194 ifp = mst->ifp; 3195 if (ifp == NULL) 3196 return (EINVAL); 3197 3198 if (ifp->if_snd_tag_modify == NULL) { 3199 error = EOPNOTSUPP; 3200 } else { 3201 error = ifp->if_snd_tag_modify(mst, ¶ms); 3202 } 3203 return (error); 3204 } 3205 3206 /* 3207 * Query existing TX rate limit based on the existing 3208 * "inp->inp_snd_tag", if any. 3209 */ 3210 int 3211 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3212 { 3213 union if_snd_tag_query_params params = { }; 3214 struct m_snd_tag *mst; 3215 struct ifnet *ifp; 3216 int error; 3217 3218 mst = inp->inp_snd_tag; 3219 if (mst == NULL) 3220 return (EINVAL); 3221 3222 ifp = mst->ifp; 3223 if (ifp == NULL) 3224 return (EINVAL); 3225 3226 if (ifp->if_snd_tag_query == NULL) { 3227 error = EOPNOTSUPP; 3228 } else { 3229 error = ifp->if_snd_tag_query(mst, ¶ms); 3230 if (error == 0 && p_max_pacing_rate != NULL) 3231 *p_max_pacing_rate = params.rate_limit.max_rate; 3232 } 3233 return (error); 3234 } 3235 3236 /* 3237 * Query existing TX queue level based on the existing 3238 * "inp->inp_snd_tag", if any. 3239 */ 3240 int 3241 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3242 { 3243 union if_snd_tag_query_params params = { }; 3244 struct m_snd_tag *mst; 3245 struct ifnet *ifp; 3246 int error; 3247 3248 mst = inp->inp_snd_tag; 3249 if (mst == NULL) 3250 return (EINVAL); 3251 3252 ifp = mst->ifp; 3253 if (ifp == NULL) 3254 return (EINVAL); 3255 3256 if (ifp->if_snd_tag_query == NULL) 3257 return (EOPNOTSUPP); 3258 3259 error = ifp->if_snd_tag_query(mst, ¶ms); 3260 if (error == 0 && p_txqueue_level != NULL) 3261 *p_txqueue_level = params.rate_limit.queue_level; 3262 return (error); 3263 } 3264 3265 /* 3266 * Allocate a new TX rate limit send tag from the network interface 3267 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3268 */ 3269 int 3270 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3271 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate) 3272 { 3273 union if_snd_tag_alloc_params params = { 3274 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3275 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3276 .rate_limit.hdr.flowid = flowid, 3277 .rate_limit.hdr.flowtype = flowtype, 3278 .rate_limit.max_rate = max_pacing_rate, 3279 }; 3280 int error; 3281 3282 INP_WLOCK_ASSERT(inp); 3283 3284 if (inp->inp_snd_tag != NULL) 3285 return (EINVAL); 3286 3287 if (ifp->if_snd_tag_alloc == NULL) { 3288 error = EOPNOTSUPP; 3289 } else { 3290 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); 3291 3292 /* 3293 * At success increment the refcount on 3294 * the send tag's network interface: 3295 */ 3296 if (error == 0) 3297 if_ref(inp->inp_snd_tag->ifp); 3298 } 3299 return (error); 3300 } 3301 3302 /* 3303 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3304 * if any: 3305 */ 3306 void 3307 in_pcbdetach_txrtlmt(struct inpcb *inp) 3308 { 3309 struct m_snd_tag *mst; 3310 struct ifnet *ifp; 3311 3312 INP_WLOCK_ASSERT(inp); 3313 3314 mst = inp->inp_snd_tag; 3315 inp->inp_snd_tag = NULL; 3316 3317 if (mst == NULL) 3318 return; 3319 3320 ifp = mst->ifp; 3321 if (ifp == NULL) 3322 return; 3323 3324 /* 3325 * If the device was detached while we still had reference(s) 3326 * on the ifp, we assume if_snd_tag_free() was replaced with 3327 * stubs. 3328 */ 3329 ifp->if_snd_tag_free(mst); 3330 3331 /* release reference count on network interface */ 3332 if_rele(ifp); 3333 } 3334 3335 /* 3336 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3337 * is set in the fast path and will attach/detach/modify the TX rate 3338 * limit send tag based on the socket's so_max_pacing_rate value. 3339 */ 3340 void 3341 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3342 { 3343 struct socket *socket; 3344 uint32_t max_pacing_rate; 3345 bool did_upgrade; 3346 int error; 3347 3348 if (inp == NULL) 3349 return; 3350 3351 socket = inp->inp_socket; 3352 if (socket == NULL) 3353 return; 3354 3355 if (!INP_WLOCKED(inp)) { 3356 /* 3357 * NOTE: If the write locking fails, we need to bail 3358 * out and use the non-ratelimited ring for the 3359 * transmit until there is a new chance to get the 3360 * write lock. 3361 */ 3362 if (!INP_TRY_UPGRADE(inp)) 3363 return; 3364 did_upgrade = 1; 3365 } else { 3366 did_upgrade = 0; 3367 } 3368 3369 /* 3370 * NOTE: The so_max_pacing_rate value is read unlocked, 3371 * because atomic updates are not required since the variable 3372 * is checked at every mbuf we send. It is assumed that the 3373 * variable read itself will be atomic. 3374 */ 3375 max_pacing_rate = socket->so_max_pacing_rate; 3376 3377 /* 3378 * NOTE: When attaching to a network interface a reference is 3379 * made to ensure the network interface doesn't go away until 3380 * all ratelimit connections are gone. The network interface 3381 * pointers compared below represent valid network interfaces, 3382 * except when comparing towards NULL. 3383 */ 3384 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3385 error = 0; 3386 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3387 if (inp->inp_snd_tag != NULL) 3388 in_pcbdetach_txrtlmt(inp); 3389 error = 0; 3390 } else if (inp->inp_snd_tag == NULL) { 3391 /* 3392 * In order to utilize packet pacing with RSS, we need 3393 * to wait until there is a valid RSS hash before we 3394 * can proceed: 3395 */ 3396 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3397 error = EAGAIN; 3398 } else { 3399 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3400 mb->m_pkthdr.flowid, max_pacing_rate); 3401 } 3402 } else { 3403 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3404 } 3405 if (error == 0 || error == EOPNOTSUPP) 3406 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3407 if (did_upgrade) 3408 INP_DOWNGRADE(inp); 3409 } 3410 3411 /* 3412 * Track route changes for TX rate limiting. 3413 */ 3414 void 3415 in_pcboutput_eagain(struct inpcb *inp) 3416 { 3417 struct socket *socket; 3418 bool did_upgrade; 3419 3420 if (inp == NULL) 3421 return; 3422 3423 socket = inp->inp_socket; 3424 if (socket == NULL) 3425 return; 3426 3427 if (inp->inp_snd_tag == NULL) 3428 return; 3429 3430 if (!INP_WLOCKED(inp)) { 3431 /* 3432 * NOTE: If the write locking fails, we need to bail 3433 * out and use the non-ratelimited ring for the 3434 * transmit until there is a new chance to get the 3435 * write lock. 3436 */ 3437 if (!INP_TRY_UPGRADE(inp)) 3438 return; 3439 did_upgrade = 1; 3440 } else { 3441 did_upgrade = 0; 3442 } 3443 3444 /* detach rate limiting */ 3445 in_pcbdetach_txrtlmt(inp); 3446 3447 /* make sure new mbuf send tag allocation is made */ 3448 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3449 3450 if (did_upgrade) 3451 INP_DOWNGRADE(inp); 3452 } 3453 #endif /* RATELIMIT */ 3454