1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1991, 1993, 1995 5 * The Regents of the University of California. 6 * Copyright (c) 2007-2009 Robert N. M. Watson 7 * Copyright (c) 2010-2011 Juniper Networks, Inc. 8 * All rights reserved. 9 * 10 * Portions of this software were developed by Robert N. M. Watson under 11 * contract to Juniper Networks, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_ddb.h" 44 #include "opt_ipsec.h" 45 #include "opt_inet.h" 46 #include "opt_inet6.h" 47 #include "opt_ratelimit.h" 48 #include "opt_pcbgroup.h" 49 #include "opt_rss.h" 50 51 #include <sys/param.h> 52 #include <sys/systm.h> 53 #include <sys/lock.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/callout.h> 57 #include <sys/eventhandler.h> 58 #include <sys/domain.h> 59 #include <sys/protosw.h> 60 #include <sys/rmlock.h> 61 #include <sys/smp.h> 62 #include <sys/socket.h> 63 #include <sys/socketvar.h> 64 #include <sys/sockio.h> 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/refcount.h> 68 #include <sys/jail.h> 69 #include <sys/kernel.h> 70 #include <sys/sysctl.h> 71 72 #ifdef DDB 73 #include <ddb/ddb.h> 74 #endif 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/if_types.h> 81 #include <net/if_llatbl.h> 82 #include <net/route.h> 83 #include <net/rss_config.h> 84 #include <net/vnet.h> 85 86 #if defined(INET) || defined(INET6) 87 #include <netinet/in.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/ip_var.h> 90 #include <netinet/tcp_var.h> 91 #ifdef TCPHPTS 92 #include <netinet/tcp_hpts.h> 93 #endif 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 #endif 97 #ifdef INET 98 #include <netinet/in_var.h> 99 #endif 100 #ifdef INET6 101 #include <netinet/ip6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #endif /* INET6 */ 106 107 #include <netipsec/ipsec_support.h> 108 109 #include <security/mac/mac_framework.h> 110 111 #define INPCBLBGROUP_SIZMIN 8 112 #define INPCBLBGROUP_SIZMAX 256 113 114 static struct callout ipport_tick_callout; 115 116 /* 117 * These configure the range of local port addresses assigned to 118 * "unspecified" outgoing connections/packets/whatever. 119 */ 120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ 121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ 122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ 123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ 124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ 125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ 126 127 /* 128 * Reserved ports accessible only to root. There are significant 129 * security considerations that must be accounted for when changing these, 130 * but the security benefits can be great. Please be careful. 131 */ 132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ 133 VNET_DEFINE(int, ipport_reservedlow); 134 135 /* Variables dealing with random ephemeral port allocation. */ 136 VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ 137 VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ 138 VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ 139 VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ 140 VNET_DEFINE(int, ipport_tcpallocs); 141 VNET_DEFINE_STATIC(int, ipport_tcplastcount); 142 143 #define V_ipport_tcplastcount VNET(ipport_tcplastcount) 144 145 static void in_pcbremlists(struct inpcb *inp); 146 #ifdef INET 147 static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, 148 struct in_addr faddr, u_int fport_arg, 149 struct in_addr laddr, u_int lport_arg, 150 int lookupflags, struct ifnet *ifp); 151 152 #define RANGECHK(var, min, max) \ 153 if ((var) < (min)) { (var) = (min); } \ 154 else if ((var) > (max)) { (var) = (max); } 155 156 static int 157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) 158 { 159 int error; 160 161 error = sysctl_handle_int(oidp, arg1, arg2, req); 162 if (error == 0) { 163 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); 164 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); 165 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); 166 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); 167 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); 168 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); 169 } 170 return (error); 171 } 172 173 #undef RANGECHK 174 175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, 176 "IP Ports"); 177 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, 179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 180 &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", ""); 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, 182 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 183 &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", ""); 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, 185 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 186 &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", ""); 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, 188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 189 &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", ""); 190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, 191 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 192 &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", ""); 193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, 194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 195 &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", ""); 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, 197 CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, 198 &VNET_NAME(ipport_reservedhigh), 0, ""); 199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, 200 CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, 202 CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); 204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, 205 CTLFLAG_VNET | CTLFLAG_RW, 206 &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " 207 "allocations before switching to a sequental one"); 208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, 209 CTLFLAG_VNET | CTLFLAG_RW, 210 &VNET_NAME(ipport_randomtime), 0, 211 "Minimum time to keep sequental port " 212 "allocation before switching to a random one"); 213 #endif /* INET */ 214 215 /* 216 * in_pcb.c: manage the Protocol Control Blocks. 217 * 218 * NOTE: It is assumed that most of these functions will be called with 219 * the pcbinfo lock held, and often, the inpcb lock held, as these utility 220 * functions often modify hash chains or addresses in pcbs. 221 */ 222 223 static struct inpcblbgroup * 224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag, 225 uint16_t port, const union in_dependaddr *addr, int size) 226 { 227 struct inpcblbgroup *grp; 228 size_t bytes; 229 230 bytes = __offsetof(struct inpcblbgroup, il_inp[size]); 231 grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); 232 if (!grp) 233 return (NULL); 234 grp->il_vflag = vflag; 235 grp->il_lport = port; 236 grp->il_dependladdr = *addr; 237 grp->il_inpsiz = size; 238 CK_LIST_INSERT_HEAD(hdr, grp, il_list); 239 return (grp); 240 } 241 242 static void 243 in_pcblbgroup_free_deferred(epoch_context_t ctx) 244 { 245 struct inpcblbgroup *grp; 246 247 grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); 248 free(grp, M_PCB); 249 } 250 251 static void 252 in_pcblbgroup_free(struct inpcblbgroup *grp) 253 { 254 255 CK_LIST_REMOVE(grp, il_list); 256 epoch_call(net_epoch_preempt, &grp->il_epoch_ctx, 257 in_pcblbgroup_free_deferred); 258 } 259 260 static struct inpcblbgroup * 261 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, 262 struct inpcblbgroup *old_grp, int size) 263 { 264 struct inpcblbgroup *grp; 265 int i; 266 267 grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, 268 old_grp->il_lport, &old_grp->il_dependladdr, size); 269 if (!grp) 270 return (NULL); 271 272 KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, 273 ("invalid new local group size %d and old local group count %d", 274 grp->il_inpsiz, old_grp->il_inpcnt)); 275 276 for (i = 0; i < old_grp->il_inpcnt; ++i) 277 grp->il_inp[i] = old_grp->il_inp[i]; 278 grp->il_inpcnt = old_grp->il_inpcnt; 279 in_pcblbgroup_free(old_grp); 280 return (grp); 281 } 282 283 /* 284 * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] 285 * and shrink group if possible. 286 */ 287 static void 288 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, 289 int i) 290 { 291 struct inpcblbgroup *grp = *grpp; 292 293 for (; i + 1 < grp->il_inpcnt; ++i) 294 grp->il_inp[i] = grp->il_inp[i + 1]; 295 grp->il_inpcnt--; 296 297 if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && 298 grp->il_inpcnt <= (grp->il_inpsiz / 4)) { 299 /* Shrink this group. */ 300 struct inpcblbgroup *new_grp = 301 in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); 302 if (new_grp) 303 *grpp = new_grp; 304 } 305 return; 306 } 307 308 /* 309 * Add PCB to load balance group for SO_REUSEPORT_LB option. 310 */ 311 static int 312 in_pcbinslbgrouphash(struct inpcb *inp) 313 { 314 const static struct timeval interval = { 60, 0 }; 315 static struct timeval lastprint; 316 struct inpcbinfo *pcbinfo; 317 struct inpcblbgrouphead *hdr; 318 struct inpcblbgroup *grp; 319 uint16_t hashmask, lport; 320 uint32_t group_index; 321 struct ucred *cred; 322 323 pcbinfo = inp->inp_pcbinfo; 324 325 INP_WLOCK_ASSERT(inp); 326 INP_HASH_WLOCK_ASSERT(pcbinfo); 327 328 if (pcbinfo->ipi_lbgrouphashbase == NULL) 329 return (0); 330 331 hashmask = pcbinfo->ipi_lbgrouphashmask; 332 lport = inp->inp_lport; 333 group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask); 334 hdr = &pcbinfo->ipi_lbgrouphashbase[group_index]; 335 336 /* 337 * Don't allow jailed socket to join local group. 338 */ 339 if (inp->inp_socket != NULL) 340 cred = inp->inp_socket->so_cred; 341 else 342 cred = NULL; 343 if (cred != NULL && jailed(cred)) 344 return (0); 345 346 #ifdef INET6 347 /* 348 * Don't allow IPv4 mapped INET6 wild socket. 349 */ 350 if ((inp->inp_vflag & INP_IPV4) && 351 inp->inp_laddr.s_addr == INADDR_ANY && 352 INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { 353 return (0); 354 } 355 #endif 356 357 hdr = &pcbinfo->ipi_lbgrouphashbase[ 358 INP_PCBLBGROUP_PORTHASH(inp->inp_lport, 359 pcbinfo->ipi_lbgrouphashmask)]; 360 CK_LIST_FOREACH(grp, hdr, il_list) { 361 if (grp->il_vflag == inp->inp_vflag && 362 grp->il_lport == inp->inp_lport && 363 memcmp(&grp->il_dependladdr, 364 &inp->inp_inc.inc_ie.ie_dependladdr, 365 sizeof(grp->il_dependladdr)) == 0) { 366 break; 367 } 368 } 369 if (grp == NULL) { 370 /* Create new load balance group. */ 371 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, 372 inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, 373 INPCBLBGROUP_SIZMIN); 374 if (!grp) 375 return (ENOBUFS); 376 } else if (grp->il_inpcnt == grp->il_inpsiz) { 377 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { 378 if (ratecheck(&lastprint, &interval)) 379 printf("lb group port %d, limit reached\n", 380 ntohs(grp->il_lport)); 381 return (0); 382 } 383 384 /* Expand this local group. */ 385 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); 386 if (!grp) 387 return (ENOBUFS); 388 } 389 390 KASSERT(grp->il_inpcnt < grp->il_inpsiz, 391 ("invalid local group size %d and count %d", 392 grp->il_inpsiz, grp->il_inpcnt)); 393 394 grp->il_inp[grp->il_inpcnt] = inp; 395 grp->il_inpcnt++; 396 return (0); 397 } 398 399 /* 400 * Remove PCB from load balance group. 401 */ 402 static void 403 in_pcbremlbgrouphash(struct inpcb *inp) 404 { 405 struct inpcbinfo *pcbinfo; 406 struct inpcblbgrouphead *hdr; 407 struct inpcblbgroup *grp; 408 int i; 409 410 pcbinfo = inp->inp_pcbinfo; 411 412 INP_WLOCK_ASSERT(inp); 413 INP_HASH_WLOCK_ASSERT(pcbinfo); 414 415 if (pcbinfo->ipi_lbgrouphashbase == NULL) 416 return; 417 418 hdr = &pcbinfo->ipi_lbgrouphashbase[ 419 INP_PCBLBGROUP_PORTHASH(inp->inp_lport, 420 pcbinfo->ipi_lbgrouphashmask)]; 421 422 CK_LIST_FOREACH(grp, hdr, il_list) { 423 for (i = 0; i < grp->il_inpcnt; ++i) { 424 if (grp->il_inp[i] != inp) 425 continue; 426 427 if (grp->il_inpcnt == 1) { 428 /* We are the last, free this local group. */ 429 in_pcblbgroup_free(grp); 430 } else { 431 /* Pull up inpcbs, shrink group if possible. */ 432 in_pcblbgroup_reorder(hdr, &grp, i); 433 } 434 return; 435 } 436 } 437 } 438 439 /* 440 * Different protocols initialize their inpcbs differently - giving 441 * different name to the lock. But they all are disposed the same. 442 */ 443 static void 444 inpcb_fini(void *mem, int size) 445 { 446 struct inpcb *inp = mem; 447 448 INP_LOCK_DESTROY(inp); 449 } 450 451 /* 452 * Initialize an inpcbinfo -- we should be able to reduce the number of 453 * arguments in time. 454 */ 455 void 456 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, 457 struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, 458 char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) 459 { 460 461 INP_INFO_LOCK_INIT(pcbinfo, name); 462 INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ 463 INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); 464 #ifdef VIMAGE 465 pcbinfo->ipi_vnet = curvnet; 466 #endif 467 pcbinfo->ipi_listhead = listhead; 468 CK_LIST_INIT(pcbinfo->ipi_listhead); 469 pcbinfo->ipi_count = 0; 470 pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, 471 &pcbinfo->ipi_hashmask); 472 pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, 473 &pcbinfo->ipi_porthashmask); 474 pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB, 475 &pcbinfo->ipi_lbgrouphashmask); 476 #ifdef PCBGROUP 477 in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); 478 #endif 479 pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), 480 NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); 481 uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); 482 uma_zone_set_warning(pcbinfo->ipi_zone, 483 "kern.ipc.maxsockets limit reached"); 484 } 485 486 /* 487 * Destroy an inpcbinfo. 488 */ 489 void 490 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) 491 { 492 493 KASSERT(pcbinfo->ipi_count == 0, 494 ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); 495 496 hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); 497 hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, 498 pcbinfo->ipi_porthashmask); 499 hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, 500 pcbinfo->ipi_lbgrouphashmask); 501 #ifdef PCBGROUP 502 in_pcbgroup_destroy(pcbinfo); 503 #endif 504 uma_zdestroy(pcbinfo->ipi_zone); 505 INP_LIST_LOCK_DESTROY(pcbinfo); 506 INP_HASH_LOCK_DESTROY(pcbinfo); 507 INP_INFO_LOCK_DESTROY(pcbinfo); 508 } 509 510 /* 511 * Allocate a PCB and associate it with the socket. 512 * On success return with the PCB locked. 513 */ 514 int 515 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) 516 { 517 struct inpcb *inp; 518 int error; 519 520 #ifdef INVARIANTS 521 if (pcbinfo == &V_tcbinfo) { 522 INP_INFO_RLOCK_ASSERT(pcbinfo); 523 } else { 524 INP_INFO_WLOCK_ASSERT(pcbinfo); 525 } 526 #endif 527 528 error = 0; 529 inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); 530 if (inp == NULL) 531 return (ENOBUFS); 532 bzero(&inp->inp_start_zero, inp_zero_size); 533 inp->inp_pcbinfo = pcbinfo; 534 inp->inp_socket = so; 535 inp->inp_cred = crhold(so->so_cred); 536 inp->inp_inc.inc_fibnum = so->so_fibnum; 537 #ifdef MAC 538 error = mac_inpcb_init(inp, M_NOWAIT); 539 if (error != 0) 540 goto out; 541 mac_inpcb_create(so, inp); 542 #endif 543 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 544 error = ipsec_init_pcbpolicy(inp); 545 if (error != 0) { 546 #ifdef MAC 547 mac_inpcb_destroy(inp); 548 #endif 549 goto out; 550 } 551 #endif /*IPSEC*/ 552 #ifdef INET6 553 if (INP_SOCKAF(so) == AF_INET6) { 554 inp->inp_vflag |= INP_IPV6PROTO; 555 if (V_ip6_v6only) 556 inp->inp_flags |= IN6P_IPV6_V6ONLY; 557 } 558 #endif 559 INP_WLOCK(inp); 560 INP_LIST_WLOCK(pcbinfo); 561 CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); 562 pcbinfo->ipi_count++; 563 so->so_pcb = (caddr_t)inp; 564 #ifdef INET6 565 if (V_ip6_auto_flowlabel) 566 inp->inp_flags |= IN6P_AUTOFLOWLABEL; 567 #endif 568 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 569 refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ 570 571 /* 572 * Routes in inpcb's can cache L2 as well; they are guaranteed 573 * to be cleaned up. 574 */ 575 inp->inp_route.ro_flags = RT_LLE_CACHE; 576 INP_LIST_WUNLOCK(pcbinfo); 577 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) 578 out: 579 if (error != 0) { 580 crfree(inp->inp_cred); 581 uma_zfree(pcbinfo->ipi_zone, inp); 582 } 583 #endif 584 return (error); 585 } 586 587 #ifdef INET 588 int 589 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 590 { 591 int anonport, error; 592 593 INP_WLOCK_ASSERT(inp); 594 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 595 596 if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) 597 return (EINVAL); 598 anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; 599 error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, 600 &inp->inp_lport, cred); 601 if (error) 602 return (error); 603 if (in_pcbinshash(inp) != 0) { 604 inp->inp_laddr.s_addr = INADDR_ANY; 605 inp->inp_lport = 0; 606 return (EAGAIN); 607 } 608 if (anonport) 609 inp->inp_flags |= INP_ANONPORT; 610 return (0); 611 } 612 #endif 613 614 /* 615 * Select a local port (number) to use. 616 */ 617 #if defined(INET) || defined(INET6) 618 int 619 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, 620 struct ucred *cred, int lookupflags) 621 { 622 struct inpcbinfo *pcbinfo; 623 struct inpcb *tmpinp; 624 unsigned short *lastport; 625 int count, dorandom, error; 626 u_short aux, first, last, lport; 627 #ifdef INET 628 struct in_addr laddr; 629 #endif 630 631 pcbinfo = inp->inp_pcbinfo; 632 633 /* 634 * Because no actual state changes occur here, a global write lock on 635 * the pcbinfo isn't required. 636 */ 637 INP_LOCK_ASSERT(inp); 638 INP_HASH_LOCK_ASSERT(pcbinfo); 639 640 if (inp->inp_flags & INP_HIGHPORT) { 641 first = V_ipport_hifirstauto; /* sysctl */ 642 last = V_ipport_hilastauto; 643 lastport = &pcbinfo->ipi_lasthi; 644 } else if (inp->inp_flags & INP_LOWPORT) { 645 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0); 646 if (error) 647 return (error); 648 first = V_ipport_lowfirstauto; /* 1023 */ 649 last = V_ipport_lowlastauto; /* 600 */ 650 lastport = &pcbinfo->ipi_lastlow; 651 } else { 652 first = V_ipport_firstauto; /* sysctl */ 653 last = V_ipport_lastauto; 654 lastport = &pcbinfo->ipi_lastport; 655 } 656 /* 657 * For UDP(-Lite), use random port allocation as long as the user 658 * allows it. For TCP (and as of yet unknown) connections, 659 * use random port allocation only if the user allows it AND 660 * ipport_tick() allows it. 661 */ 662 if (V_ipport_randomized && 663 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || 664 pcbinfo == &V_ulitecbinfo)) 665 dorandom = 1; 666 else 667 dorandom = 0; 668 /* 669 * It makes no sense to do random port allocation if 670 * we have the only port available. 671 */ 672 if (first == last) 673 dorandom = 0; 674 /* Make sure to not include UDP(-Lite) packets in the count. */ 675 if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) 676 V_ipport_tcpallocs++; 677 /* 678 * Instead of having two loops further down counting up or down 679 * make sure that first is always <= last and go with only one 680 * code path implementing all logic. 681 */ 682 if (first > last) { 683 aux = first; 684 first = last; 685 last = aux; 686 } 687 688 #ifdef INET 689 /* Make the compiler happy. */ 690 laddr.s_addr = 0; 691 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { 692 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", 693 __func__, inp)); 694 laddr = *laddrp; 695 } 696 #endif 697 tmpinp = NULL; /* Make compiler happy. */ 698 lport = *lportp; 699 700 if (dorandom) 701 *lastport = first + (arc4random() % (last - first)); 702 703 count = last - first; 704 705 do { 706 if (count-- < 0) /* completely used? */ 707 return (EADDRNOTAVAIL); 708 ++*lastport; 709 if (*lastport < first || *lastport > last) 710 *lastport = first; 711 lport = htons(*lastport); 712 713 #ifdef INET6 714 if ((inp->inp_vflag & INP_IPV6) != 0) 715 tmpinp = in6_pcblookup_local(pcbinfo, 716 &inp->in6p_laddr, lport, lookupflags, cred); 717 #endif 718 #if defined(INET) && defined(INET6) 719 else 720 #endif 721 #ifdef INET 722 tmpinp = in_pcblookup_local(pcbinfo, laddr, 723 lport, lookupflags, cred); 724 #endif 725 } while (tmpinp != NULL); 726 727 #ifdef INET 728 if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) 729 laddrp->s_addr = laddr.s_addr; 730 #endif 731 *lportp = lport; 732 733 return (0); 734 } 735 736 /* 737 * Return cached socket options. 738 */ 739 int 740 inp_so_options(const struct inpcb *inp) 741 { 742 int so_options; 743 744 so_options = 0; 745 746 if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) 747 so_options |= SO_REUSEPORT_LB; 748 if ((inp->inp_flags2 & INP_REUSEPORT) != 0) 749 so_options |= SO_REUSEPORT; 750 if ((inp->inp_flags2 & INP_REUSEADDR) != 0) 751 so_options |= SO_REUSEADDR; 752 return (so_options); 753 } 754 #endif /* INET || INET6 */ 755 756 /* 757 * Check if a new BINDMULTI socket is allowed to be created. 758 * 759 * ni points to the new inp. 760 * oi points to the exisitng inp. 761 * 762 * This checks whether the existing inp also has BINDMULTI and 763 * whether the credentials match. 764 */ 765 int 766 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) 767 { 768 /* Check permissions match */ 769 if ((ni->inp_flags2 & INP_BINDMULTI) && 770 (ni->inp_cred->cr_uid != 771 oi->inp_cred->cr_uid)) 772 return (0); 773 774 /* Check the existing inp has BINDMULTI set */ 775 if ((ni->inp_flags2 & INP_BINDMULTI) && 776 ((oi->inp_flags2 & INP_BINDMULTI) == 0)) 777 return (0); 778 779 /* 780 * We're okay - either INP_BINDMULTI isn't set on ni, or 781 * it is and it matches the checks. 782 */ 783 return (1); 784 } 785 786 #ifdef INET 787 /* 788 * Set up a bind operation on a PCB, performing port allocation 789 * as required, but do not actually modify the PCB. Callers can 790 * either complete the bind by setting inp_laddr/inp_lport and 791 * calling in_pcbinshash(), or they can just use the resulting 792 * port and address to authorise the sending of a once-off packet. 793 * 794 * On error, the values of *laddrp and *lportp are not changed. 795 */ 796 int 797 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, 798 u_short *lportp, struct ucred *cred) 799 { 800 struct socket *so = inp->inp_socket; 801 struct sockaddr_in *sin; 802 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 803 struct in_addr laddr; 804 u_short lport = 0; 805 int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); 806 int error; 807 808 /* 809 * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here 810 * so that we don't have to add to the (already messy) code below. 811 */ 812 int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); 813 814 /* 815 * No state changes, so read locks are sufficient here. 816 */ 817 INP_LOCK_ASSERT(inp); 818 INP_HASH_LOCK_ASSERT(pcbinfo); 819 820 if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ 821 return (EADDRNOTAVAIL); 822 laddr.s_addr = *laddrp; 823 if (nam != NULL && laddr.s_addr != INADDR_ANY) 824 return (EINVAL); 825 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) 826 lookupflags = INPLOOKUP_WILDCARD; 827 if (nam == NULL) { 828 if ((error = prison_local_ip4(cred, &laddr)) != 0) 829 return (error); 830 } else { 831 sin = (struct sockaddr_in *)nam; 832 if (nam->sa_len != sizeof (*sin)) 833 return (EINVAL); 834 #ifdef notdef 835 /* 836 * We should check the family, but old programs 837 * incorrectly fail to initialize it. 838 */ 839 if (sin->sin_family != AF_INET) 840 return (EAFNOSUPPORT); 841 #endif 842 error = prison_local_ip4(cred, &sin->sin_addr); 843 if (error) 844 return (error); 845 if (sin->sin_port != *lportp) { 846 /* Don't allow the port to change. */ 847 if (*lportp != 0) 848 return (EINVAL); 849 lport = sin->sin_port; 850 } 851 /* NB: lport is left as 0 if the port isn't being changed. */ 852 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { 853 /* 854 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; 855 * allow complete duplication of binding if 856 * SO_REUSEPORT is set, or if SO_REUSEADDR is set 857 * and a multicast address is bound on both 858 * new and duplicated sockets. 859 */ 860 if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) 861 reuseport = SO_REUSEADDR|SO_REUSEPORT; 862 /* 863 * XXX: How to deal with SO_REUSEPORT_LB here? 864 * Treat same as SO_REUSEPORT for now. 865 */ 866 if ((so->so_options & 867 (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) 868 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; 869 } else if (sin->sin_addr.s_addr != INADDR_ANY) { 870 sin->sin_port = 0; /* yech... */ 871 bzero(&sin->sin_zero, sizeof(sin->sin_zero)); 872 /* 873 * Is the address a local IP address? 874 * If INP_BINDANY is set, then the socket may be bound 875 * to any endpoint address, local or not. 876 */ 877 if ((inp->inp_flags & INP_BINDANY) == 0 && 878 ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) 879 return (EADDRNOTAVAIL); 880 } 881 laddr = sin->sin_addr; 882 if (lport) { 883 struct inpcb *t; 884 struct tcptw *tw; 885 886 /* GROSS */ 887 if (ntohs(lport) <= V_ipport_reservedhigh && 888 ntohs(lport) >= V_ipport_reservedlow && 889 priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 890 0)) 891 return (EACCES); 892 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && 893 priv_check_cred(inp->inp_cred, 894 PRIV_NETINET_REUSEPORT, 0) != 0) { 895 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 896 lport, INPLOOKUP_WILDCARD, cred); 897 /* 898 * XXX 899 * This entire block sorely needs a rewrite. 900 */ 901 if (t && 902 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 903 ((t->inp_flags & INP_TIMEWAIT) == 0) && 904 (so->so_type != SOCK_STREAM || 905 ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && 906 (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || 907 ntohl(t->inp_laddr.s_addr) != INADDR_ANY || 908 (t->inp_flags2 & INP_REUSEPORT) || 909 (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && 910 (inp->inp_cred->cr_uid != 911 t->inp_cred->cr_uid)) 912 return (EADDRINUSE); 913 914 /* 915 * If the socket is a BINDMULTI socket, then 916 * the credentials need to match and the 917 * original socket also has to have been bound 918 * with BINDMULTI. 919 */ 920 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 921 return (EADDRINUSE); 922 } 923 t = in_pcblookup_local(pcbinfo, sin->sin_addr, 924 lport, lookupflags, cred); 925 if (t && (t->inp_flags & INP_TIMEWAIT)) { 926 /* 927 * XXXRW: If an incpb has had its timewait 928 * state recycled, we treat the address as 929 * being in use (for now). This is better 930 * than a panic, but not desirable. 931 */ 932 tw = intotw(t); 933 if (tw == NULL || 934 ((reuseport & tw->tw_so_options) == 0 && 935 (reuseport_lb & 936 tw->tw_so_options) == 0)) { 937 return (EADDRINUSE); 938 } 939 } else if (t && 940 ((inp->inp_flags2 & INP_BINDMULTI) == 0) && 941 (reuseport & inp_so_options(t)) == 0 && 942 (reuseport_lb & inp_so_options(t)) == 0) { 943 #ifdef INET6 944 if (ntohl(sin->sin_addr.s_addr) != 945 INADDR_ANY || 946 ntohl(t->inp_laddr.s_addr) != 947 INADDR_ANY || 948 (inp->inp_vflag & INP_IPV6PROTO) == 0 || 949 (t->inp_vflag & INP_IPV6PROTO) == 0) 950 #endif 951 return (EADDRINUSE); 952 if (t && (! in_pcbbind_check_bindmulti(inp, t))) 953 return (EADDRINUSE); 954 } 955 } 956 } 957 if (*lportp != 0) 958 lport = *lportp; 959 if (lport == 0) { 960 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); 961 if (error != 0) 962 return (error); 963 964 } 965 *laddrp = laddr.s_addr; 966 *lportp = lport; 967 return (0); 968 } 969 970 /* 971 * Connect from a socket to a specified address. 972 * Both address and port must be specified in argument sin. 973 * If don't have a local address for this socket yet, 974 * then pick one. 975 */ 976 int 977 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, 978 struct ucred *cred, struct mbuf *m) 979 { 980 u_short lport, fport; 981 in_addr_t laddr, faddr; 982 int anonport, error; 983 984 INP_WLOCK_ASSERT(inp); 985 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 986 987 lport = inp->inp_lport; 988 laddr = inp->inp_laddr.s_addr; 989 anonport = (lport == 0); 990 error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, 991 NULL, cred); 992 if (error) 993 return (error); 994 995 /* Do the initial binding of the local address if required. */ 996 if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { 997 inp->inp_lport = lport; 998 inp->inp_laddr.s_addr = laddr; 999 if (in_pcbinshash(inp) != 0) { 1000 inp->inp_laddr.s_addr = INADDR_ANY; 1001 inp->inp_lport = 0; 1002 return (EAGAIN); 1003 } 1004 } 1005 1006 /* Commit the remaining changes. */ 1007 inp->inp_lport = lport; 1008 inp->inp_laddr.s_addr = laddr; 1009 inp->inp_faddr.s_addr = faddr; 1010 inp->inp_fport = fport; 1011 in_pcbrehash_mbuf(inp, m); 1012 1013 if (anonport) 1014 inp->inp_flags |= INP_ANONPORT; 1015 return (0); 1016 } 1017 1018 int 1019 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) 1020 { 1021 1022 return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); 1023 } 1024 1025 /* 1026 * Do proper source address selection on an unbound socket in case 1027 * of connect. Take jails into account as well. 1028 */ 1029 int 1030 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, 1031 struct ucred *cred) 1032 { 1033 struct ifaddr *ifa; 1034 struct sockaddr *sa; 1035 struct sockaddr_in *sin; 1036 struct route sro; 1037 int error; 1038 1039 KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); 1040 /* 1041 * Bypass source address selection and use the primary jail IP 1042 * if requested. 1043 */ 1044 if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) 1045 return (0); 1046 1047 error = 0; 1048 bzero(&sro, sizeof(sro)); 1049 1050 sin = (struct sockaddr_in *)&sro.ro_dst; 1051 sin->sin_family = AF_INET; 1052 sin->sin_len = sizeof(struct sockaddr_in); 1053 sin->sin_addr.s_addr = faddr->s_addr; 1054 1055 /* 1056 * If route is known our src addr is taken from the i/f, 1057 * else punt. 1058 * 1059 * Find out route to destination. 1060 */ 1061 if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) 1062 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); 1063 1064 /* 1065 * If we found a route, use the address corresponding to 1066 * the outgoing interface. 1067 * 1068 * Otherwise assume faddr is reachable on a directly connected 1069 * network and try to find a corresponding interface to take 1070 * the source address from. 1071 */ 1072 NET_EPOCH_ENTER(); 1073 if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { 1074 struct in_ifaddr *ia; 1075 struct ifnet *ifp; 1076 1077 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, 1078 inp->inp_socket->so_fibnum)); 1079 if (ia == NULL) { 1080 ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, 1081 inp->inp_socket->so_fibnum)); 1082 1083 } 1084 if (ia == NULL) { 1085 error = ENETUNREACH; 1086 goto done; 1087 } 1088 1089 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1090 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1091 goto done; 1092 } 1093 1094 ifp = ia->ia_ifp; 1095 ia = NULL; 1096 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1097 1098 sa = ifa->ifa_addr; 1099 if (sa->sa_family != AF_INET) 1100 continue; 1101 sin = (struct sockaddr_in *)sa; 1102 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1103 ia = (struct in_ifaddr *)ifa; 1104 break; 1105 } 1106 } 1107 if (ia != NULL) { 1108 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1109 goto done; 1110 } 1111 1112 /* 3. As a last resort return the 'default' jail address. */ 1113 error = prison_get_ip4(cred, laddr); 1114 goto done; 1115 } 1116 1117 /* 1118 * If the outgoing interface on the route found is not 1119 * a loopback interface, use the address from that interface. 1120 * In case of jails do those three steps: 1121 * 1. check if the interface address belongs to the jail. If so use it. 1122 * 2. check if we have any address on the outgoing interface 1123 * belonging to this jail. If so use it. 1124 * 3. as a last resort return the 'default' jail address. 1125 */ 1126 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { 1127 struct in_ifaddr *ia; 1128 struct ifnet *ifp; 1129 1130 /* If not jailed, use the default returned. */ 1131 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 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 /* Jailed. */ 1138 /* 1. Check if the iface address belongs to the jail. */ 1139 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; 1140 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1141 ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; 1142 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1143 goto done; 1144 } 1145 1146 /* 1147 * 2. Check if we have any address on the outgoing interface 1148 * belonging to this jail. 1149 */ 1150 ia = NULL; 1151 ifp = sro.ro_rt->rt_ifp; 1152 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1153 sa = ifa->ifa_addr; 1154 if (sa->sa_family != AF_INET) 1155 continue; 1156 sin = (struct sockaddr_in *)sa; 1157 if (prison_check_ip4(cred, &sin->sin_addr) == 0) { 1158 ia = (struct in_ifaddr *)ifa; 1159 break; 1160 } 1161 } 1162 if (ia != NULL) { 1163 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1164 goto done; 1165 } 1166 1167 /* 3. As a last resort return the 'default' jail address. */ 1168 error = prison_get_ip4(cred, laddr); 1169 goto done; 1170 } 1171 1172 /* 1173 * The outgoing interface is marked with 'loopback net', so a route 1174 * to ourselves is here. 1175 * Try to find the interface of the destination address and then 1176 * take the address from there. That interface is not necessarily 1177 * a loopback interface. 1178 * In case of jails, check that it is an address of the jail 1179 * and if we cannot find, fall back to the 'default' jail address. 1180 */ 1181 if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { 1182 struct sockaddr_in sain; 1183 struct in_ifaddr *ia; 1184 1185 bzero(&sain, sizeof(struct sockaddr_in)); 1186 sain.sin_family = AF_INET; 1187 sain.sin_len = sizeof(struct sockaddr_in); 1188 sain.sin_addr.s_addr = faddr->s_addr; 1189 1190 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), 1191 inp->inp_socket->so_fibnum)); 1192 if (ia == NULL) 1193 ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, 1194 inp->inp_socket->so_fibnum)); 1195 if (ia == NULL) 1196 ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); 1197 1198 if (cred == NULL || !prison_flag(cred, PR_IP4)) { 1199 if (ia == NULL) { 1200 error = ENETUNREACH; 1201 goto done; 1202 } 1203 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1204 goto done; 1205 } 1206 1207 /* Jailed. */ 1208 if (ia != NULL) { 1209 struct ifnet *ifp; 1210 1211 ifp = ia->ia_ifp; 1212 ia = NULL; 1213 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1214 sa = ifa->ifa_addr; 1215 if (sa->sa_family != AF_INET) 1216 continue; 1217 sin = (struct sockaddr_in *)sa; 1218 if (prison_check_ip4(cred, 1219 &sin->sin_addr) == 0) { 1220 ia = (struct in_ifaddr *)ifa; 1221 break; 1222 } 1223 } 1224 if (ia != NULL) { 1225 laddr->s_addr = ia->ia_addr.sin_addr.s_addr; 1226 goto done; 1227 } 1228 } 1229 1230 /* 3. As a last resort return the 'default' jail address. */ 1231 error = prison_get_ip4(cred, laddr); 1232 goto done; 1233 } 1234 1235 done: 1236 NET_EPOCH_EXIT(); 1237 if (sro.ro_rt != NULL) 1238 RTFREE(sro.ro_rt); 1239 return (error); 1240 } 1241 1242 /* 1243 * Set up for a connect from a socket to the specified address. 1244 * On entry, *laddrp and *lportp should contain the current local 1245 * address and port for the PCB; these are updated to the values 1246 * that should be placed in inp_laddr and inp_lport to complete 1247 * the connect. 1248 * 1249 * On success, *faddrp and *fportp will be set to the remote address 1250 * and port. These are not updated in the error case. 1251 * 1252 * If the operation fails because the connection already exists, 1253 * *oinpp will be set to the PCB of that connection so that the 1254 * caller can decide to override it. In all other cases, *oinpp 1255 * is set to NULL. 1256 */ 1257 int 1258 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, 1259 in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, 1260 struct inpcb **oinpp, struct ucred *cred) 1261 { 1262 struct rm_priotracker in_ifa_tracker; 1263 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1264 struct in_ifaddr *ia; 1265 struct inpcb *oinp; 1266 struct in_addr laddr, faddr; 1267 u_short lport, fport; 1268 int error; 1269 1270 /* 1271 * Because a global state change doesn't actually occur here, a read 1272 * lock is sufficient. 1273 */ 1274 INP_LOCK_ASSERT(inp); 1275 INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); 1276 1277 if (oinpp != NULL) 1278 *oinpp = NULL; 1279 if (nam->sa_len != sizeof (*sin)) 1280 return (EINVAL); 1281 if (sin->sin_family != AF_INET) 1282 return (EAFNOSUPPORT); 1283 if (sin->sin_port == 0) 1284 return (EADDRNOTAVAIL); 1285 laddr.s_addr = *laddrp; 1286 lport = *lportp; 1287 faddr = sin->sin_addr; 1288 fport = sin->sin_port; 1289 1290 if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { 1291 /* 1292 * If the destination address is INADDR_ANY, 1293 * use the primary local address. 1294 * If the supplied address is INADDR_BROADCAST, 1295 * and the primary interface supports broadcast, 1296 * choose the broadcast address for that interface. 1297 */ 1298 if (faddr.s_addr == INADDR_ANY) { 1299 IN_IFADDR_RLOCK(&in_ifa_tracker); 1300 faddr = 1301 IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; 1302 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1303 if (cred != NULL && 1304 (error = prison_get_ip4(cred, &faddr)) != 0) 1305 return (error); 1306 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { 1307 IN_IFADDR_RLOCK(&in_ifa_tracker); 1308 if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & 1309 IFF_BROADCAST) 1310 faddr = satosin(&CK_STAILQ_FIRST( 1311 &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; 1312 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1313 } 1314 } 1315 if (laddr.s_addr == INADDR_ANY) { 1316 error = in_pcbladdr(inp, &faddr, &laddr, cred); 1317 /* 1318 * If the destination address is multicast and an outgoing 1319 * interface has been set as a multicast option, prefer the 1320 * address of that interface as our source address. 1321 */ 1322 if (IN_MULTICAST(ntohl(faddr.s_addr)) && 1323 inp->inp_moptions != NULL) { 1324 struct ip_moptions *imo; 1325 struct ifnet *ifp; 1326 1327 imo = inp->inp_moptions; 1328 if (imo->imo_multicast_ifp != NULL) { 1329 ifp = imo->imo_multicast_ifp; 1330 IN_IFADDR_RLOCK(&in_ifa_tracker); 1331 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1332 if ((ia->ia_ifp == ifp) && 1333 (cred == NULL || 1334 prison_check_ip4(cred, 1335 &ia->ia_addr.sin_addr) == 0)) 1336 break; 1337 } 1338 if (ia == NULL) 1339 error = EADDRNOTAVAIL; 1340 else { 1341 laddr = ia->ia_addr.sin_addr; 1342 error = 0; 1343 } 1344 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1345 } 1346 } 1347 if (error) 1348 return (error); 1349 } 1350 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, 1351 laddr, lport, 0, NULL); 1352 if (oinp != NULL) { 1353 if (oinpp != NULL) 1354 *oinpp = oinp; 1355 return (EADDRINUSE); 1356 } 1357 if (lport == 0) { 1358 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, 1359 cred); 1360 if (error) 1361 return (error); 1362 } 1363 *laddrp = laddr.s_addr; 1364 *lportp = lport; 1365 *faddrp = faddr.s_addr; 1366 *fportp = fport; 1367 return (0); 1368 } 1369 1370 void 1371 in_pcbdisconnect(struct inpcb *inp) 1372 { 1373 1374 INP_WLOCK_ASSERT(inp); 1375 INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); 1376 1377 inp->inp_faddr.s_addr = INADDR_ANY; 1378 inp->inp_fport = 0; 1379 in_pcbrehash(inp); 1380 } 1381 #endif /* INET */ 1382 1383 /* 1384 * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. 1385 * For most protocols, this will be invoked immediately prior to calling 1386 * in_pcbfree(). However, with TCP the inpcb may significantly outlive the 1387 * socket, in which case in_pcbfree() is deferred. 1388 */ 1389 void 1390 in_pcbdetach(struct inpcb *inp) 1391 { 1392 1393 KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); 1394 1395 #ifdef RATELIMIT 1396 if (inp->inp_snd_tag != NULL) 1397 in_pcbdetach_txrtlmt(inp); 1398 #endif 1399 inp->inp_socket->so_pcb = NULL; 1400 inp->inp_socket = NULL; 1401 } 1402 1403 /* 1404 * in_pcbref() bumps the reference count on an inpcb in order to maintain 1405 * stability of an inpcb pointer despite the inpcb lock being released. This 1406 * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, 1407 * but where the inpcb lock may already held, or when acquiring a reference 1408 * via a pcbgroup. 1409 * 1410 * in_pcbref() should be used only to provide brief memory stability, and 1411 * must always be followed by a call to INP_WLOCK() and in_pcbrele() to 1412 * garbage collect the inpcb if it has been in_pcbfree()'d from another 1413 * context. Until in_pcbrele() has returned that the inpcb is still valid, 1414 * lock and rele are the *only* safe operations that may be performed on the 1415 * inpcb. 1416 * 1417 * While the inpcb will not be freed, releasing the inpcb lock means that the 1418 * connection's state may change, so the caller should be careful to 1419 * revalidate any cached state on reacquiring the lock. Drop the reference 1420 * using in_pcbrele(). 1421 */ 1422 void 1423 in_pcbref(struct inpcb *inp) 1424 { 1425 1426 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1427 1428 refcount_acquire(&inp->inp_refcount); 1429 } 1430 1431 /* 1432 * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to 1433 * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we 1434 * return a flag indicating whether or not the inpcb remains valid. If it is 1435 * valid, we return with the inpcb lock held. 1436 * 1437 * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a 1438 * reference on an inpcb. Historically more work was done here (actually, in 1439 * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the 1440 * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely 1441 * about memory stability (and continued use of the write lock). 1442 */ 1443 int 1444 in_pcbrele_rlocked(struct inpcb *inp) 1445 { 1446 struct inpcbinfo *pcbinfo; 1447 1448 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1449 1450 INP_RLOCK_ASSERT(inp); 1451 1452 if (refcount_release(&inp->inp_refcount) == 0) { 1453 /* 1454 * If the inpcb has been freed, let the caller know, even if 1455 * this isn't the last reference. 1456 */ 1457 if (inp->inp_flags2 & INP_FREED) { 1458 INP_RUNLOCK(inp); 1459 return (1); 1460 } 1461 return (0); 1462 } 1463 1464 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1465 #ifdef TCPHPTS 1466 if (inp->inp_in_hpts || inp->inp_in_input) { 1467 struct tcp_hpts_entry *hpts; 1468 /* 1469 * We should not be on the hpts at 1470 * this point in any form. we must 1471 * get the lock to be sure. 1472 */ 1473 hpts = tcp_hpts_lock(inp); 1474 if (inp->inp_in_hpts) 1475 panic("Hpts:%p inp:%p at free still on hpts", 1476 hpts, inp); 1477 mtx_unlock(&hpts->p_mtx); 1478 hpts = tcp_input_lock(inp); 1479 if (inp->inp_in_input) 1480 panic("Hpts:%p inp:%p at free still on input hpts", 1481 hpts, inp); 1482 mtx_unlock(&hpts->p_mtx); 1483 } 1484 #endif 1485 INP_RUNLOCK(inp); 1486 pcbinfo = inp->inp_pcbinfo; 1487 uma_zfree(pcbinfo->ipi_zone, inp); 1488 return (1); 1489 } 1490 1491 int 1492 in_pcbrele_wlocked(struct inpcb *inp) 1493 { 1494 struct inpcbinfo *pcbinfo; 1495 1496 KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); 1497 1498 INP_WLOCK_ASSERT(inp); 1499 1500 if (refcount_release(&inp->inp_refcount) == 0) { 1501 /* 1502 * If the inpcb has been freed, let the caller know, even if 1503 * this isn't the last reference. 1504 */ 1505 if (inp->inp_flags2 & INP_FREED) { 1506 INP_WUNLOCK(inp); 1507 return (1); 1508 } 1509 return (0); 1510 } 1511 1512 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1513 #ifdef TCPHPTS 1514 if (inp->inp_in_hpts || inp->inp_in_input) { 1515 struct tcp_hpts_entry *hpts; 1516 /* 1517 * We should not be on the hpts at 1518 * this point in any form. we must 1519 * get the lock to be sure. 1520 */ 1521 hpts = tcp_hpts_lock(inp); 1522 if (inp->inp_in_hpts) 1523 panic("Hpts:%p inp:%p at free still on hpts", 1524 hpts, inp); 1525 mtx_unlock(&hpts->p_mtx); 1526 hpts = tcp_input_lock(inp); 1527 if (inp->inp_in_input) 1528 panic("Hpts:%p inp:%p at free still on input hpts", 1529 hpts, inp); 1530 mtx_unlock(&hpts->p_mtx); 1531 } 1532 #endif 1533 INP_WUNLOCK(inp); 1534 pcbinfo = inp->inp_pcbinfo; 1535 uma_zfree(pcbinfo->ipi_zone, inp); 1536 return (1); 1537 } 1538 1539 /* 1540 * Temporary wrapper. 1541 */ 1542 int 1543 in_pcbrele(struct inpcb *inp) 1544 { 1545 1546 return (in_pcbrele_wlocked(inp)); 1547 } 1548 1549 void 1550 in_pcblist_rele_rlocked(epoch_context_t ctx) 1551 { 1552 struct in_pcblist *il; 1553 struct inpcb *inp; 1554 struct inpcbinfo *pcbinfo; 1555 int i, n; 1556 1557 il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); 1558 pcbinfo = il->il_pcbinfo; 1559 n = il->il_count; 1560 INP_INFO_WLOCK(pcbinfo); 1561 for (i = 0; i < n; i++) { 1562 inp = il->il_inp_list[i]; 1563 INP_RLOCK(inp); 1564 if (!in_pcbrele_rlocked(inp)) 1565 INP_RUNLOCK(inp); 1566 } 1567 INP_INFO_WUNLOCK(pcbinfo); 1568 free(il, M_TEMP); 1569 } 1570 1571 static void 1572 inpcbport_free(epoch_context_t ctx) 1573 { 1574 struct inpcbport *phd; 1575 1576 phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); 1577 free(phd, M_PCB); 1578 } 1579 1580 static void 1581 in_pcbfree_deferred(epoch_context_t ctx) 1582 { 1583 struct inpcb *inp; 1584 int released __unused; 1585 1586 inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); 1587 1588 INP_WLOCK(inp); 1589 #ifdef INET 1590 struct ip_moptions *imo = inp->inp_moptions; 1591 inp->inp_moptions = NULL; 1592 #endif 1593 /* XXXRW: Do as much as possible here. */ 1594 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 1595 if (inp->inp_sp != NULL) 1596 ipsec_delete_pcbpolicy(inp); 1597 #endif 1598 #ifdef INET6 1599 struct ip6_moptions *im6o = NULL; 1600 if (inp->inp_vflag & INP_IPV6PROTO) { 1601 ip6_freepcbopts(inp->in6p_outputopts); 1602 im6o = inp->in6p_moptions; 1603 inp->in6p_moptions = NULL; 1604 } 1605 #endif 1606 if (inp->inp_options) 1607 (void)m_free(inp->inp_options); 1608 inp->inp_vflag = 0; 1609 crfree(inp->inp_cred); 1610 #ifdef MAC 1611 mac_inpcb_destroy(inp); 1612 #endif 1613 released = in_pcbrele_wlocked(inp); 1614 MPASS(released); 1615 #ifdef INET6 1616 ip6_freemoptions(im6o); 1617 #endif 1618 #ifdef INET 1619 inp_freemoptions(imo); 1620 #endif 1621 } 1622 1623 /* 1624 * Unconditionally schedule an inpcb to be freed by decrementing its 1625 * reference count, which should occur only after the inpcb has been detached 1626 * from its socket. If another thread holds a temporary reference (acquired 1627 * using in_pcbref()) then the free is deferred until that reference is 1628 * released using in_pcbrele(), but the inpcb is still unlocked. Almost all 1629 * work, including removal from global lists, is done in this context, where 1630 * the pcbinfo lock is held. 1631 */ 1632 void 1633 in_pcbfree(struct inpcb *inp) 1634 { 1635 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 1636 1637 KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); 1638 KASSERT((inp->inp_flags2 & INP_FREED) == 0, 1639 ("%s: called twice for pcb %p", __func__, inp)); 1640 if (inp->inp_flags2 & INP_FREED) { 1641 INP_WUNLOCK(inp); 1642 return; 1643 } 1644 1645 #ifdef INVARIANTS 1646 if (pcbinfo == &V_tcbinfo) { 1647 INP_INFO_LOCK_ASSERT(pcbinfo); 1648 } else { 1649 INP_INFO_WLOCK_ASSERT(pcbinfo); 1650 } 1651 #endif 1652 INP_WLOCK_ASSERT(inp); 1653 INP_LIST_WLOCK(pcbinfo); 1654 in_pcbremlists(inp); 1655 INP_LIST_WUNLOCK(pcbinfo); 1656 RO_INVALIDATE_CACHE(&inp->inp_route); 1657 /* mark as destruction in progress */ 1658 inp->inp_flags2 |= INP_FREED; 1659 INP_WUNLOCK(inp); 1660 epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred); 1661 } 1662 1663 /* 1664 * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and 1665 * port reservation, and preventing it from being returned by inpcb lookups. 1666 * 1667 * It is used by TCP to mark an inpcb as unused and avoid future packet 1668 * delivery or event notification when a socket remains open but TCP has 1669 * closed. This might occur as a result of a shutdown()-initiated TCP close 1670 * or a RST on the wire, and allows the port binding to be reused while still 1671 * maintaining the invariant that so_pcb always points to a valid inpcb until 1672 * in_pcbdetach(). 1673 * 1674 * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by 1675 * in_pcbnotifyall() and in_pcbpurgeif0()? 1676 */ 1677 void 1678 in_pcbdrop(struct inpcb *inp) 1679 { 1680 1681 INP_WLOCK_ASSERT(inp); 1682 #ifdef INVARIANTS 1683 if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) 1684 MPASS(inp->inp_refcount > 1); 1685 #endif 1686 1687 /* 1688 * XXXRW: Possibly we should protect the setting of INP_DROPPED with 1689 * the hash lock...? 1690 */ 1691 inp->inp_flags |= INP_DROPPED; 1692 if (inp->inp_flags & INP_INHASHLIST) { 1693 struct inpcbport *phd = inp->inp_phd; 1694 1695 INP_HASH_WLOCK(inp->inp_pcbinfo); 1696 in_pcbremlbgrouphash(inp); 1697 CK_LIST_REMOVE(inp, inp_hash); 1698 CK_LIST_REMOVE(inp, inp_portlist); 1699 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 1700 CK_LIST_REMOVE(phd, phd_hash); 1701 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 1702 } 1703 INP_HASH_WUNLOCK(inp->inp_pcbinfo); 1704 inp->inp_flags &= ~INP_INHASHLIST; 1705 #ifdef PCBGROUP 1706 in_pcbgroup_remove(inp); 1707 #endif 1708 } 1709 } 1710 1711 #ifdef INET 1712 /* 1713 * Common routines to return the socket addresses associated with inpcbs. 1714 */ 1715 struct sockaddr * 1716 in_sockaddr(in_port_t port, struct in_addr *addr_p) 1717 { 1718 struct sockaddr_in *sin; 1719 1720 sin = malloc(sizeof *sin, M_SONAME, 1721 M_WAITOK | M_ZERO); 1722 sin->sin_family = AF_INET; 1723 sin->sin_len = sizeof(*sin); 1724 sin->sin_addr = *addr_p; 1725 sin->sin_port = port; 1726 1727 return (struct sockaddr *)sin; 1728 } 1729 1730 int 1731 in_getsockaddr(struct socket *so, struct sockaddr **nam) 1732 { 1733 struct inpcb *inp; 1734 struct in_addr addr; 1735 in_port_t port; 1736 1737 inp = sotoinpcb(so); 1738 KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); 1739 1740 INP_RLOCK(inp); 1741 port = inp->inp_lport; 1742 addr = inp->inp_laddr; 1743 INP_RUNLOCK(inp); 1744 1745 *nam = in_sockaddr(port, &addr); 1746 return 0; 1747 } 1748 1749 int 1750 in_getpeeraddr(struct socket *so, struct sockaddr **nam) 1751 { 1752 struct inpcb *inp; 1753 struct in_addr addr; 1754 in_port_t port; 1755 1756 inp = sotoinpcb(so); 1757 KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); 1758 1759 INP_RLOCK(inp); 1760 port = inp->inp_fport; 1761 addr = inp->inp_faddr; 1762 INP_RUNLOCK(inp); 1763 1764 *nam = in_sockaddr(port, &addr); 1765 return 0; 1766 } 1767 1768 void 1769 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, 1770 struct inpcb *(*notify)(struct inpcb *, int)) 1771 { 1772 struct inpcb *inp, *inp_temp; 1773 1774 INP_INFO_WLOCK(pcbinfo); 1775 CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { 1776 INP_WLOCK(inp); 1777 #ifdef INET6 1778 if ((inp->inp_vflag & INP_IPV4) == 0) { 1779 INP_WUNLOCK(inp); 1780 continue; 1781 } 1782 #endif 1783 if (inp->inp_faddr.s_addr != faddr.s_addr || 1784 inp->inp_socket == NULL) { 1785 INP_WUNLOCK(inp); 1786 continue; 1787 } 1788 if ((*notify)(inp, errno)) 1789 INP_WUNLOCK(inp); 1790 } 1791 INP_INFO_WUNLOCK(pcbinfo); 1792 } 1793 1794 void 1795 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) 1796 { 1797 struct inpcb *inp; 1798 struct ip_moptions *imo; 1799 int i, gap; 1800 1801 INP_INFO_WLOCK(pcbinfo); 1802 CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { 1803 INP_WLOCK(inp); 1804 imo = inp->inp_moptions; 1805 if ((inp->inp_vflag & INP_IPV4) && 1806 imo != NULL) { 1807 /* 1808 * Unselect the outgoing interface if it is being 1809 * detached. 1810 */ 1811 if (imo->imo_multicast_ifp == ifp) 1812 imo->imo_multicast_ifp = NULL; 1813 1814 /* 1815 * Drop multicast group membership if we joined 1816 * through the interface being detached. 1817 * 1818 * XXX This can all be deferred to an epoch_call 1819 */ 1820 for (i = 0, gap = 0; i < imo->imo_num_memberships; 1821 i++) { 1822 if (imo->imo_membership[i]->inm_ifp == ifp) { 1823 IN_MULTI_LOCK_ASSERT(); 1824 in_leavegroup_locked(imo->imo_membership[i], NULL); 1825 gap++; 1826 } else if (gap != 0) 1827 imo->imo_membership[i - gap] = 1828 imo->imo_membership[i]; 1829 } 1830 imo->imo_num_memberships -= gap; 1831 } 1832 INP_WUNLOCK(inp); 1833 } 1834 INP_INFO_WUNLOCK(pcbinfo); 1835 } 1836 1837 /* 1838 * Lookup a PCB based on the local address and port. Caller must hold the 1839 * hash lock. No inpcb locks or references are acquired. 1840 */ 1841 #define INP_LOOKUP_MAPPED_PCB_COST 3 1842 struct inpcb * 1843 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, 1844 u_short lport, int lookupflags, struct ucred *cred) 1845 { 1846 struct inpcb *inp; 1847 #ifdef INET6 1848 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; 1849 #else 1850 int matchwild = 3; 1851 #endif 1852 int wildcard; 1853 1854 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 1855 ("%s: invalid lookup flags %d", __func__, lookupflags)); 1856 1857 INP_HASH_LOCK_ASSERT(pcbinfo); 1858 1859 if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { 1860 struct inpcbhead *head; 1861 /* 1862 * Look for an unconnected (wildcard foreign addr) PCB that 1863 * matches the local address and port we're looking for. 1864 */ 1865 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 1866 0, pcbinfo->ipi_hashmask)]; 1867 CK_LIST_FOREACH(inp, head, inp_hash) { 1868 #ifdef INET6 1869 /* XXX inp locking */ 1870 if ((inp->inp_vflag & INP_IPV4) == 0) 1871 continue; 1872 #endif 1873 if (inp->inp_faddr.s_addr == INADDR_ANY && 1874 inp->inp_laddr.s_addr == laddr.s_addr && 1875 inp->inp_lport == lport) { 1876 /* 1877 * Found? 1878 */ 1879 if (cred == NULL || 1880 prison_equal_ip4(cred->cr_prison, 1881 inp->inp_cred->cr_prison)) 1882 return (inp); 1883 } 1884 } 1885 /* 1886 * Not found. 1887 */ 1888 return (NULL); 1889 } else { 1890 struct inpcbporthead *porthash; 1891 struct inpcbport *phd; 1892 struct inpcb *match = NULL; 1893 /* 1894 * Best fit PCB lookup. 1895 * 1896 * First see if this local port is in use by looking on the 1897 * port hash list. 1898 */ 1899 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, 1900 pcbinfo->ipi_porthashmask)]; 1901 CK_LIST_FOREACH(phd, porthash, phd_hash) { 1902 if (phd->phd_port == lport) 1903 break; 1904 } 1905 if (phd != NULL) { 1906 /* 1907 * Port is in use by one or more PCBs. Look for best 1908 * fit. 1909 */ 1910 CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { 1911 wildcard = 0; 1912 if (cred != NULL && 1913 !prison_equal_ip4(inp->inp_cred->cr_prison, 1914 cred->cr_prison)) 1915 continue; 1916 #ifdef INET6 1917 /* XXX inp locking */ 1918 if ((inp->inp_vflag & INP_IPV4) == 0) 1919 continue; 1920 /* 1921 * We never select the PCB that has 1922 * INP_IPV6 flag and is bound to :: if 1923 * we have another PCB which is bound 1924 * to 0.0.0.0. If a PCB has the 1925 * INP_IPV6 flag, then we set its cost 1926 * higher than IPv4 only PCBs. 1927 * 1928 * Note that the case only happens 1929 * when a socket is bound to ::, under 1930 * the condition that the use of the 1931 * mapped address is allowed. 1932 */ 1933 if ((inp->inp_vflag & INP_IPV6) != 0) 1934 wildcard += INP_LOOKUP_MAPPED_PCB_COST; 1935 #endif 1936 if (inp->inp_faddr.s_addr != INADDR_ANY) 1937 wildcard++; 1938 if (inp->inp_laddr.s_addr != INADDR_ANY) { 1939 if (laddr.s_addr == INADDR_ANY) 1940 wildcard++; 1941 else if (inp->inp_laddr.s_addr != laddr.s_addr) 1942 continue; 1943 } else { 1944 if (laddr.s_addr != INADDR_ANY) 1945 wildcard++; 1946 } 1947 if (wildcard < matchwild) { 1948 match = inp; 1949 matchwild = wildcard; 1950 if (matchwild == 0) 1951 break; 1952 } 1953 } 1954 } 1955 return (match); 1956 } 1957 } 1958 #undef INP_LOOKUP_MAPPED_PCB_COST 1959 1960 static struct inpcb * 1961 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, 1962 const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, 1963 uint16_t fport, int lookupflags) 1964 { 1965 struct inpcb *local_wild; 1966 const struct inpcblbgrouphead *hdr; 1967 struct inpcblbgroup *grp; 1968 uint32_t idx; 1969 1970 INP_HASH_LOCK_ASSERT(pcbinfo); 1971 1972 hdr = &pcbinfo->ipi_lbgrouphashbase[INP_PCBLBGROUP_PORTHASH(lport, 1973 pcbinfo->ipi_lbgrouphashmask)]; 1974 1975 /* 1976 * Order of socket selection: 1977 * 1. non-wild. 1978 * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). 1979 * 1980 * NOTE: 1981 * - Load balanced group does not contain jailed sockets 1982 * - Load balanced group does not contain IPv4 mapped INET6 wild sockets 1983 */ 1984 local_wild = NULL; 1985 CK_LIST_FOREACH(grp, hdr, il_list) { 1986 #ifdef INET6 1987 if (!(grp->il_vflag & INP_IPV4)) 1988 continue; 1989 #endif 1990 if (grp->il_lport != lport) 1991 continue; 1992 1993 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % 1994 grp->il_inpcnt; 1995 if (grp->il_laddr.s_addr == laddr->s_addr) 1996 return (grp->il_inp[idx]); 1997 if (grp->il_laddr.s_addr == INADDR_ANY && 1998 (lookupflags & INPLOOKUP_WILDCARD) != 0) 1999 local_wild = grp->il_inp[idx]; 2000 } 2001 return (local_wild); 2002 } 2003 2004 #ifdef PCBGROUP 2005 /* 2006 * Lookup PCB in hash list, using pcbgroup tables. 2007 */ 2008 static struct inpcb * 2009 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, 2010 struct in_addr faddr, u_int fport_arg, struct in_addr laddr, 2011 u_int lport_arg, int lookupflags, struct ifnet *ifp) 2012 { 2013 struct inpcbhead *head; 2014 struct inpcb *inp, *tmpinp; 2015 u_short fport = fport_arg, lport = lport_arg; 2016 bool locked; 2017 2018 /* 2019 * First look for an exact match. 2020 */ 2021 tmpinp = NULL; 2022 INP_GROUP_LOCK(pcbgroup); 2023 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2024 pcbgroup->ipg_hashmask)]; 2025 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2026 #ifdef INET6 2027 /* XXX inp locking */ 2028 if ((inp->inp_vflag & INP_IPV4) == 0) 2029 continue; 2030 #endif 2031 if (inp->inp_faddr.s_addr == faddr.s_addr && 2032 inp->inp_laddr.s_addr == laddr.s_addr && 2033 inp->inp_fport == fport && 2034 inp->inp_lport == lport) { 2035 /* 2036 * XXX We should be able to directly return 2037 * the inp here, without any checks. 2038 * Well unless both bound with SO_REUSEPORT? 2039 */ 2040 if (prison_flag(inp->inp_cred, PR_IP4)) 2041 goto found; 2042 if (tmpinp == NULL) 2043 tmpinp = inp; 2044 } 2045 } 2046 if (tmpinp != NULL) { 2047 inp = tmpinp; 2048 goto found; 2049 } 2050 2051 #ifdef RSS 2052 /* 2053 * For incoming connections, we may wish to do a wildcard 2054 * match for an RSS-local socket. 2055 */ 2056 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2057 struct inpcb *local_wild = NULL, *local_exact = NULL; 2058 #ifdef INET6 2059 struct inpcb *local_wild_mapped = NULL; 2060 #endif 2061 struct inpcb *jail_wild = NULL; 2062 struct inpcbhead *head; 2063 int injail; 2064 2065 /* 2066 * Order of socket selection - we always prefer jails. 2067 * 1. jailed, non-wild. 2068 * 2. jailed, wild. 2069 * 3. non-jailed, non-wild. 2070 * 4. non-jailed, wild. 2071 */ 2072 2073 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, 2074 lport, 0, pcbgroup->ipg_hashmask)]; 2075 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { 2076 #ifdef INET6 2077 /* XXX inp locking */ 2078 if ((inp->inp_vflag & INP_IPV4) == 0) 2079 continue; 2080 #endif 2081 if (inp->inp_faddr.s_addr != INADDR_ANY || 2082 inp->inp_lport != lport) 2083 continue; 2084 2085 injail = prison_flag(inp->inp_cred, PR_IP4); 2086 if (injail) { 2087 if (prison_check_ip4(inp->inp_cred, 2088 &laddr) != 0) 2089 continue; 2090 } else { 2091 if (local_exact != NULL) 2092 continue; 2093 } 2094 2095 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2096 if (injail) 2097 goto found; 2098 else 2099 local_exact = inp; 2100 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2101 #ifdef INET6 2102 /* XXX inp locking, NULL check */ 2103 if (inp->inp_vflag & INP_IPV6PROTO) 2104 local_wild_mapped = inp; 2105 else 2106 #endif 2107 if (injail) 2108 jail_wild = inp; 2109 else 2110 local_wild = inp; 2111 } 2112 } /* LIST_FOREACH */ 2113 2114 inp = jail_wild; 2115 if (inp == NULL) 2116 inp = local_exact; 2117 if (inp == NULL) 2118 inp = local_wild; 2119 #ifdef INET6 2120 if (inp == NULL) 2121 inp = local_wild_mapped; 2122 #endif 2123 if (inp != NULL) 2124 goto found; 2125 } 2126 #endif 2127 2128 /* 2129 * Then look for a wildcard match, if requested. 2130 */ 2131 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2132 struct inpcb *local_wild = NULL, *local_exact = NULL; 2133 #ifdef INET6 2134 struct inpcb *local_wild_mapped = NULL; 2135 #endif 2136 struct inpcb *jail_wild = NULL; 2137 struct inpcbhead *head; 2138 int injail; 2139 2140 /* 2141 * Order of socket selection - we always prefer jails. 2142 * 1. jailed, non-wild. 2143 * 2. jailed, wild. 2144 * 3. non-jailed, non-wild. 2145 * 4. non-jailed, wild. 2146 */ 2147 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 2148 0, pcbinfo->ipi_wildmask)]; 2149 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { 2150 #ifdef INET6 2151 /* XXX inp locking */ 2152 if ((inp->inp_vflag & INP_IPV4) == 0) 2153 continue; 2154 #endif 2155 if (inp->inp_faddr.s_addr != INADDR_ANY || 2156 inp->inp_lport != lport) 2157 continue; 2158 2159 injail = prison_flag(inp->inp_cred, PR_IP4); 2160 if (injail) { 2161 if (prison_check_ip4(inp->inp_cred, 2162 &laddr) != 0) 2163 continue; 2164 } else { 2165 if (local_exact != NULL) 2166 continue; 2167 } 2168 2169 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2170 if (injail) 2171 goto found; 2172 else 2173 local_exact = inp; 2174 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2175 #ifdef INET6 2176 /* XXX inp locking, NULL check */ 2177 if (inp->inp_vflag & INP_IPV6PROTO) 2178 local_wild_mapped = inp; 2179 else 2180 #endif 2181 if (injail) 2182 jail_wild = inp; 2183 else 2184 local_wild = inp; 2185 } 2186 } /* LIST_FOREACH */ 2187 inp = jail_wild; 2188 if (inp == NULL) 2189 inp = local_exact; 2190 if (inp == NULL) 2191 inp = local_wild; 2192 #ifdef INET6 2193 if (inp == NULL) 2194 inp = local_wild_mapped; 2195 #endif 2196 if (inp != NULL) 2197 goto found; 2198 } /* if (lookupflags & INPLOOKUP_WILDCARD) */ 2199 INP_GROUP_UNLOCK(pcbgroup); 2200 return (NULL); 2201 2202 found: 2203 if (lookupflags & INPLOOKUP_WLOCKPCB) 2204 locked = INP_TRY_WLOCK(inp); 2205 else if (lookupflags & INPLOOKUP_RLOCKPCB) 2206 locked = INP_TRY_RLOCK(inp); 2207 else 2208 panic("%s: locking bug", __func__); 2209 if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { 2210 if (lookupflags & INPLOOKUP_WLOCKPCB) 2211 INP_WUNLOCK(inp); 2212 else 2213 INP_RUNLOCK(inp); 2214 return (NULL); 2215 } else if (!locked) 2216 in_pcbref(inp); 2217 INP_GROUP_UNLOCK(pcbgroup); 2218 if (!locked) { 2219 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2220 INP_WLOCK(inp); 2221 if (in_pcbrele_wlocked(inp)) 2222 return (NULL); 2223 } else { 2224 INP_RLOCK(inp); 2225 if (in_pcbrele_rlocked(inp)) 2226 return (NULL); 2227 } 2228 } 2229 #ifdef INVARIANTS 2230 if (lookupflags & INPLOOKUP_WLOCKPCB) 2231 INP_WLOCK_ASSERT(inp); 2232 else 2233 INP_RLOCK_ASSERT(inp); 2234 #endif 2235 return (inp); 2236 } 2237 #endif /* PCBGROUP */ 2238 2239 /* 2240 * Lookup PCB in hash list, using pcbinfo tables. This variation assumes 2241 * that the caller has locked the hash list, and will not perform any further 2242 * locking or reference operations on either the hash list or the connection. 2243 */ 2244 static struct inpcb * 2245 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2246 u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, 2247 struct ifnet *ifp) 2248 { 2249 struct inpcbhead *head; 2250 struct inpcb *inp, *tmpinp; 2251 u_short fport = fport_arg, lport = lport_arg; 2252 2253 #ifdef INVARIANTS 2254 KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, 2255 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2256 if (!mtx_owned(&pcbinfo->ipi_hash_lock)) 2257 MPASS(in_epoch_verbose(net_epoch_preempt, 1)); 2258 #endif 2259 /* 2260 * First look for an exact match. 2261 */ 2262 tmpinp = NULL; 2263 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, 2264 pcbinfo->ipi_hashmask)]; 2265 CK_LIST_FOREACH(inp, head, inp_hash) { 2266 #ifdef INET6 2267 /* XXX inp locking */ 2268 if ((inp->inp_vflag & INP_IPV4) == 0) 2269 continue; 2270 #endif 2271 if (inp->inp_faddr.s_addr == faddr.s_addr && 2272 inp->inp_laddr.s_addr == laddr.s_addr && 2273 inp->inp_fport == fport && 2274 inp->inp_lport == lport) { 2275 /* 2276 * XXX We should be able to directly return 2277 * the inp here, without any checks. 2278 * Well unless both bound with SO_REUSEPORT? 2279 */ 2280 if (prison_flag(inp->inp_cred, PR_IP4)) 2281 return (inp); 2282 if (tmpinp == NULL) 2283 tmpinp = inp; 2284 } 2285 } 2286 if (tmpinp != NULL) 2287 return (tmpinp); 2288 2289 /* 2290 * Then look in lb group (for wildcard match). 2291 */ 2292 if (pcbinfo->ipi_lbgrouphashbase != NULL && 2293 (lookupflags & INPLOOKUP_WILDCARD)) { 2294 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, 2295 fport, lookupflags); 2296 if (inp != NULL) { 2297 return (inp); 2298 } 2299 } 2300 2301 /* 2302 * Then look for a wildcard match, if requested. 2303 */ 2304 if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 2305 struct inpcb *local_wild = NULL, *local_exact = NULL; 2306 #ifdef INET6 2307 struct inpcb *local_wild_mapped = NULL; 2308 #endif 2309 struct inpcb *jail_wild = NULL; 2310 int injail; 2311 2312 /* 2313 * Order of socket selection - we always prefer jails. 2314 * 1. jailed, non-wild. 2315 * 2. jailed, wild. 2316 * 3. non-jailed, non-wild. 2317 * 4. non-jailed, wild. 2318 */ 2319 2320 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 2321 0, pcbinfo->ipi_hashmask)]; 2322 CK_LIST_FOREACH(inp, head, inp_hash) { 2323 #ifdef INET6 2324 /* XXX inp locking */ 2325 if ((inp->inp_vflag & INP_IPV4) == 0) 2326 continue; 2327 #endif 2328 if (inp->inp_faddr.s_addr != INADDR_ANY || 2329 inp->inp_lport != lport) 2330 continue; 2331 2332 injail = prison_flag(inp->inp_cred, PR_IP4); 2333 if (injail) { 2334 if (prison_check_ip4(inp->inp_cred, 2335 &laddr) != 0) 2336 continue; 2337 } else { 2338 if (local_exact != NULL) 2339 continue; 2340 } 2341 2342 if (inp->inp_laddr.s_addr == laddr.s_addr) { 2343 if (injail) 2344 return (inp); 2345 else 2346 local_exact = inp; 2347 } else if (inp->inp_laddr.s_addr == INADDR_ANY) { 2348 #ifdef INET6 2349 /* XXX inp locking, NULL check */ 2350 if (inp->inp_vflag & INP_IPV6PROTO) 2351 local_wild_mapped = inp; 2352 else 2353 #endif 2354 if (injail) 2355 jail_wild = inp; 2356 else 2357 local_wild = inp; 2358 } 2359 } /* LIST_FOREACH */ 2360 if (jail_wild != NULL) 2361 return (jail_wild); 2362 if (local_exact != NULL) 2363 return (local_exact); 2364 if (local_wild != NULL) 2365 return (local_wild); 2366 #ifdef INET6 2367 if (local_wild_mapped != NULL) 2368 return (local_wild_mapped); 2369 #endif 2370 } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ 2371 2372 return (NULL); 2373 } 2374 2375 /* 2376 * Lookup PCB in hash list, using pcbinfo tables. This variation locks the 2377 * hash list lock, and will return the inpcb locked (i.e., requires 2378 * INPLOOKUP_LOCKPCB). 2379 */ 2380 static struct inpcb * 2381 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2382 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2383 struct ifnet *ifp) 2384 { 2385 struct inpcb *inp; 2386 2387 INP_HASH_RLOCK(pcbinfo); 2388 inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, 2389 (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); 2390 if (inp != NULL) { 2391 if (lookupflags & INPLOOKUP_WLOCKPCB) { 2392 INP_WLOCK(inp); 2393 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2394 INP_WUNLOCK(inp); 2395 inp = NULL; 2396 } 2397 } else if (lookupflags & INPLOOKUP_RLOCKPCB) { 2398 INP_RLOCK(inp); 2399 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 2400 INP_RUNLOCK(inp); 2401 inp = NULL; 2402 } 2403 } else 2404 panic("%s: locking bug", __func__); 2405 #ifdef INVARIANTS 2406 if (inp != NULL) { 2407 if (lookupflags & INPLOOKUP_WLOCKPCB) 2408 INP_WLOCK_ASSERT(inp); 2409 else 2410 INP_RLOCK_ASSERT(inp); 2411 } 2412 #endif 2413 } 2414 INP_HASH_RUNLOCK(pcbinfo); 2415 return (inp); 2416 } 2417 2418 /* 2419 * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf 2420 * from which a pre-calculated hash value may be extracted. 2421 * 2422 * Possibly more of this logic should be in in_pcbgroup.c. 2423 */ 2424 struct inpcb * 2425 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, 2426 struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) 2427 { 2428 #if defined(PCBGROUP) && !defined(RSS) 2429 struct inpcbgroup *pcbgroup; 2430 #endif 2431 2432 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2433 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2434 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2435 ("%s: LOCKPCB not set", __func__)); 2436 2437 /* 2438 * When not using RSS, use connection groups in preference to the 2439 * reservation table when looking up 4-tuples. When using RSS, just 2440 * use the reservation table, due to the cost of the Toeplitz hash 2441 * in software. 2442 * 2443 * XXXRW: This policy belongs in the pcbgroup code, as in principle 2444 * we could be doing RSS with a non-Toeplitz hash that is affordable 2445 * in software. 2446 */ 2447 #if defined(PCBGROUP) && !defined(RSS) 2448 if (in_pcbgroup_enabled(pcbinfo)) { 2449 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2450 fport); 2451 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2452 laddr, lport, lookupflags, ifp)); 2453 } 2454 #endif 2455 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2456 lookupflags, ifp)); 2457 } 2458 2459 struct inpcb * 2460 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, 2461 u_int fport, struct in_addr laddr, u_int lport, int lookupflags, 2462 struct ifnet *ifp, struct mbuf *m) 2463 { 2464 #ifdef PCBGROUP 2465 struct inpcbgroup *pcbgroup; 2466 #endif 2467 2468 KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, 2469 ("%s: invalid lookup flags %d", __func__, lookupflags)); 2470 KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, 2471 ("%s: LOCKPCB not set", __func__)); 2472 2473 #ifdef PCBGROUP 2474 /* 2475 * If we can use a hardware-generated hash to look up the connection 2476 * group, use that connection group to find the inpcb. Otherwise 2477 * fall back on a software hash -- or the reservation table if we're 2478 * using RSS. 2479 * 2480 * XXXRW: As above, that policy belongs in the pcbgroup code. 2481 */ 2482 if (in_pcbgroup_enabled(pcbinfo) && 2483 !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { 2484 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), 2485 m->m_pkthdr.flowid); 2486 if (pcbgroup != NULL) 2487 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, 2488 fport, laddr, lport, lookupflags, ifp)); 2489 #ifndef RSS 2490 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, 2491 fport); 2492 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, 2493 laddr, lport, lookupflags, ifp)); 2494 #endif 2495 } 2496 #endif 2497 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, 2498 lookupflags, ifp)); 2499 } 2500 #endif /* INET */ 2501 2502 /* 2503 * Insert PCB onto various hash lists. 2504 */ 2505 static int 2506 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) 2507 { 2508 struct inpcbhead *pcbhash; 2509 struct inpcbporthead *pcbporthash; 2510 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2511 struct inpcbport *phd; 2512 u_int32_t hashkey_faddr; 2513 int so_options; 2514 2515 INP_WLOCK_ASSERT(inp); 2516 INP_HASH_WLOCK_ASSERT(pcbinfo); 2517 2518 KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, 2519 ("in_pcbinshash: INP_INHASHLIST")); 2520 2521 #ifdef INET6 2522 if (inp->inp_vflag & INP_IPV6) 2523 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2524 else 2525 #endif 2526 hashkey_faddr = inp->inp_faddr.s_addr; 2527 2528 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2529 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2530 2531 pcbporthash = &pcbinfo->ipi_porthashbase[ 2532 INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; 2533 2534 /* 2535 * Add entry to load balance group. 2536 * Only do this if SO_REUSEPORT_LB is set. 2537 */ 2538 so_options = inp_so_options(inp); 2539 if (so_options & SO_REUSEPORT_LB) { 2540 int ret = in_pcbinslbgrouphash(inp); 2541 if (ret) { 2542 /* pcb lb group malloc fail (ret=ENOBUFS). */ 2543 return (ret); 2544 } 2545 } 2546 2547 /* 2548 * Go through port list and look for a head for this lport. 2549 */ 2550 CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { 2551 if (phd->phd_port == inp->inp_lport) 2552 break; 2553 } 2554 /* 2555 * If none exists, malloc one and tack it on. 2556 */ 2557 if (phd == NULL) { 2558 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); 2559 if (phd == NULL) { 2560 return (ENOBUFS); /* XXX */ 2561 } 2562 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); 2563 phd->phd_port = inp->inp_lport; 2564 CK_LIST_INIT(&phd->phd_pcblist); 2565 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); 2566 } 2567 inp->inp_phd = phd; 2568 CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); 2569 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); 2570 inp->inp_flags |= INP_INHASHLIST; 2571 #ifdef PCBGROUP 2572 if (do_pcbgroup_update) 2573 in_pcbgroup_update(inp); 2574 #endif 2575 return (0); 2576 } 2577 2578 /* 2579 * For now, there are two public interfaces to insert an inpcb into the hash 2580 * lists -- one that does update pcbgroups, and one that doesn't. The latter 2581 * is used only in the TCP syncache, where in_pcbinshash is called before the 2582 * full 4-tuple is set for the inpcb, and we don't want to install in the 2583 * pcbgroup until later. 2584 * 2585 * XXXRW: This seems like a misfeature. in_pcbinshash should always update 2586 * connection groups, and partially initialised inpcbs should not be exposed 2587 * to either reservation hash tables or pcbgroups. 2588 */ 2589 int 2590 in_pcbinshash(struct inpcb *inp) 2591 { 2592 2593 return (in_pcbinshash_internal(inp, 1)); 2594 } 2595 2596 int 2597 in_pcbinshash_nopcbgroup(struct inpcb *inp) 2598 { 2599 2600 return (in_pcbinshash_internal(inp, 0)); 2601 } 2602 2603 /* 2604 * Move PCB to the proper hash bucket when { faddr, fport } have been 2605 * changed. NOTE: This does not handle the case of the lport changing (the 2606 * hashed port list would have to be updated as well), so the lport must 2607 * not change after in_pcbinshash() has been called. 2608 */ 2609 void 2610 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) 2611 { 2612 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2613 struct inpcbhead *head; 2614 u_int32_t hashkey_faddr; 2615 2616 INP_WLOCK_ASSERT(inp); 2617 INP_HASH_WLOCK_ASSERT(pcbinfo); 2618 2619 KASSERT(inp->inp_flags & INP_INHASHLIST, 2620 ("in_pcbrehash: !INP_INHASHLIST")); 2621 2622 #ifdef INET6 2623 if (inp->inp_vflag & INP_IPV6) 2624 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); 2625 else 2626 #endif 2627 hashkey_faddr = inp->inp_faddr.s_addr; 2628 2629 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, 2630 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; 2631 2632 CK_LIST_REMOVE(inp, inp_hash); 2633 CK_LIST_INSERT_HEAD(head, inp, inp_hash); 2634 2635 #ifdef PCBGROUP 2636 if (m != NULL) 2637 in_pcbgroup_update_mbuf(inp, m); 2638 else 2639 in_pcbgroup_update(inp); 2640 #endif 2641 } 2642 2643 void 2644 in_pcbrehash(struct inpcb *inp) 2645 { 2646 2647 in_pcbrehash_mbuf(inp, NULL); 2648 } 2649 2650 /* 2651 * Remove PCB from various lists. 2652 */ 2653 static void 2654 in_pcbremlists(struct inpcb *inp) 2655 { 2656 struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; 2657 2658 #ifdef INVARIANTS 2659 if (pcbinfo == &V_tcbinfo) { 2660 INP_INFO_RLOCK_ASSERT(pcbinfo); 2661 } else { 2662 INP_INFO_WLOCK_ASSERT(pcbinfo); 2663 } 2664 #endif 2665 2666 INP_WLOCK_ASSERT(inp); 2667 INP_LIST_WLOCK_ASSERT(pcbinfo); 2668 2669 inp->inp_gencnt = ++pcbinfo->ipi_gencnt; 2670 if (inp->inp_flags & INP_INHASHLIST) { 2671 struct inpcbport *phd = inp->inp_phd; 2672 2673 INP_HASH_WLOCK(pcbinfo); 2674 2675 /* XXX: Only do if SO_REUSEPORT_LB set? */ 2676 in_pcbremlbgrouphash(inp); 2677 2678 CK_LIST_REMOVE(inp, inp_hash); 2679 CK_LIST_REMOVE(inp, inp_portlist); 2680 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { 2681 CK_LIST_REMOVE(phd, phd_hash); 2682 epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); 2683 } 2684 INP_HASH_WUNLOCK(pcbinfo); 2685 inp->inp_flags &= ~INP_INHASHLIST; 2686 } 2687 CK_LIST_REMOVE(inp, inp_list); 2688 pcbinfo->ipi_count--; 2689 #ifdef PCBGROUP 2690 in_pcbgroup_remove(inp); 2691 #endif 2692 } 2693 2694 /* 2695 * Check for alternatives when higher level complains 2696 * about service problems. For now, invalidate cached 2697 * routing information. If the route was created dynamically 2698 * (by a redirect), time to try a default gateway again. 2699 */ 2700 void 2701 in_losing(struct inpcb *inp) 2702 { 2703 2704 RO_INVALIDATE_CACHE(&inp->inp_route); 2705 return; 2706 } 2707 2708 /* 2709 * A set label operation has occurred at the socket layer, propagate the 2710 * label change into the in_pcb for the socket. 2711 */ 2712 void 2713 in_pcbsosetlabel(struct socket *so) 2714 { 2715 #ifdef MAC 2716 struct inpcb *inp; 2717 2718 inp = sotoinpcb(so); 2719 KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); 2720 2721 INP_WLOCK(inp); 2722 SOCK_LOCK(so); 2723 mac_inpcb_sosetlabel(so, inp); 2724 SOCK_UNLOCK(so); 2725 INP_WUNLOCK(inp); 2726 #endif 2727 } 2728 2729 /* 2730 * ipport_tick runs once per second, determining if random port allocation 2731 * should be continued. If more than ipport_randomcps ports have been 2732 * allocated in the last second, then we return to sequential port 2733 * allocation. We return to random allocation only once we drop below 2734 * ipport_randomcps for at least ipport_randomtime seconds. 2735 */ 2736 static void 2737 ipport_tick(void *xtp) 2738 { 2739 VNET_ITERATOR_DECL(vnet_iter); 2740 2741 VNET_LIST_RLOCK_NOSLEEP(); 2742 VNET_FOREACH(vnet_iter) { 2743 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ 2744 if (V_ipport_tcpallocs <= 2745 V_ipport_tcplastcount + V_ipport_randomcps) { 2746 if (V_ipport_stoprandom > 0) 2747 V_ipport_stoprandom--; 2748 } else 2749 V_ipport_stoprandom = V_ipport_randomtime; 2750 V_ipport_tcplastcount = V_ipport_tcpallocs; 2751 CURVNET_RESTORE(); 2752 } 2753 VNET_LIST_RUNLOCK_NOSLEEP(); 2754 callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); 2755 } 2756 2757 static void 2758 ip_fini(void *xtp) 2759 { 2760 2761 callout_stop(&ipport_tick_callout); 2762 } 2763 2764 /* 2765 * The ipport_callout should start running at about the time we attach the 2766 * inet or inet6 domains. 2767 */ 2768 static void 2769 ipport_tick_init(const void *unused __unused) 2770 { 2771 2772 /* Start ipport_tick. */ 2773 callout_init(&ipport_tick_callout, 1); 2774 callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); 2775 EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, 2776 SHUTDOWN_PRI_DEFAULT); 2777 } 2778 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, 2779 ipport_tick_init, NULL); 2780 2781 void 2782 inp_wlock(struct inpcb *inp) 2783 { 2784 2785 INP_WLOCK(inp); 2786 } 2787 2788 void 2789 inp_wunlock(struct inpcb *inp) 2790 { 2791 2792 INP_WUNLOCK(inp); 2793 } 2794 2795 void 2796 inp_rlock(struct inpcb *inp) 2797 { 2798 2799 INP_RLOCK(inp); 2800 } 2801 2802 void 2803 inp_runlock(struct inpcb *inp) 2804 { 2805 2806 INP_RUNLOCK(inp); 2807 } 2808 2809 #ifdef INVARIANT_SUPPORT 2810 void 2811 inp_lock_assert(struct inpcb *inp) 2812 { 2813 2814 INP_WLOCK_ASSERT(inp); 2815 } 2816 2817 void 2818 inp_unlock_assert(struct inpcb *inp) 2819 { 2820 2821 INP_UNLOCK_ASSERT(inp); 2822 } 2823 #endif 2824 2825 void 2826 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) 2827 { 2828 struct inpcb *inp; 2829 2830 INP_INFO_WLOCK(&V_tcbinfo); 2831 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 2832 INP_WLOCK(inp); 2833 func(inp, arg); 2834 INP_WUNLOCK(inp); 2835 } 2836 INP_INFO_WUNLOCK(&V_tcbinfo); 2837 } 2838 2839 struct socket * 2840 inp_inpcbtosocket(struct inpcb *inp) 2841 { 2842 2843 INP_WLOCK_ASSERT(inp); 2844 return (inp->inp_socket); 2845 } 2846 2847 struct tcpcb * 2848 inp_inpcbtotcpcb(struct inpcb *inp) 2849 { 2850 2851 INP_WLOCK_ASSERT(inp); 2852 return ((struct tcpcb *)inp->inp_ppcb); 2853 } 2854 2855 int 2856 inp_ip_tos_get(const struct inpcb *inp) 2857 { 2858 2859 return (inp->inp_ip_tos); 2860 } 2861 2862 void 2863 inp_ip_tos_set(struct inpcb *inp, int val) 2864 { 2865 2866 inp->inp_ip_tos = val; 2867 } 2868 2869 void 2870 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, 2871 uint32_t *faddr, uint16_t *fp) 2872 { 2873 2874 INP_LOCK_ASSERT(inp); 2875 *laddr = inp->inp_laddr.s_addr; 2876 *faddr = inp->inp_faddr.s_addr; 2877 *lp = inp->inp_lport; 2878 *fp = inp->inp_fport; 2879 } 2880 2881 struct inpcb * 2882 so_sotoinpcb(struct socket *so) 2883 { 2884 2885 return (sotoinpcb(so)); 2886 } 2887 2888 struct tcpcb * 2889 so_sototcpcb(struct socket *so) 2890 { 2891 2892 return (sototcpcb(so)); 2893 } 2894 2895 /* 2896 * Create an external-format (``xinpcb'') structure using the information in 2897 * the kernel-format in_pcb structure pointed to by inp. This is done to 2898 * reduce the spew of irrelevant information over this interface, to isolate 2899 * user code from changes in the kernel structure, and potentially to provide 2900 * information-hiding if we decide that some of this information should be 2901 * hidden from users. 2902 */ 2903 void 2904 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) 2905 { 2906 2907 xi->xi_len = sizeof(struct xinpcb); 2908 if (inp->inp_socket) 2909 sotoxsocket(inp->inp_socket, &xi->xi_socket); 2910 else 2911 bzero(&xi->xi_socket, sizeof(struct xsocket)); 2912 bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); 2913 xi->inp_gencnt = inp->inp_gencnt; 2914 xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; 2915 xi->inp_flow = inp->inp_flow; 2916 xi->inp_flowid = inp->inp_flowid; 2917 xi->inp_flowtype = inp->inp_flowtype; 2918 xi->inp_flags = inp->inp_flags; 2919 xi->inp_flags2 = inp->inp_flags2; 2920 xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; 2921 xi->in6p_cksum = inp->in6p_cksum; 2922 xi->in6p_hops = inp->in6p_hops; 2923 xi->inp_ip_tos = inp->inp_ip_tos; 2924 xi->inp_vflag = inp->inp_vflag; 2925 xi->inp_ip_ttl = inp->inp_ip_ttl; 2926 xi->inp_ip_p = inp->inp_ip_p; 2927 xi->inp_ip_minttl = inp->inp_ip_minttl; 2928 } 2929 2930 #ifdef DDB 2931 static void 2932 db_print_indent(int indent) 2933 { 2934 int i; 2935 2936 for (i = 0; i < indent; i++) 2937 db_printf(" "); 2938 } 2939 2940 static void 2941 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) 2942 { 2943 char faddr_str[48], laddr_str[48]; 2944 2945 db_print_indent(indent); 2946 db_printf("%s at %p\n", name, inc); 2947 2948 indent += 2; 2949 2950 #ifdef INET6 2951 if (inc->inc_flags & INC_ISIPV6) { 2952 /* IPv6. */ 2953 ip6_sprintf(laddr_str, &inc->inc6_laddr); 2954 ip6_sprintf(faddr_str, &inc->inc6_faddr); 2955 } else 2956 #endif 2957 { 2958 /* IPv4. */ 2959 inet_ntoa_r(inc->inc_laddr, laddr_str); 2960 inet_ntoa_r(inc->inc_faddr, faddr_str); 2961 } 2962 db_print_indent(indent); 2963 db_printf("inc_laddr %s inc_lport %u\n", laddr_str, 2964 ntohs(inc->inc_lport)); 2965 db_print_indent(indent); 2966 db_printf("inc_faddr %s inc_fport %u\n", faddr_str, 2967 ntohs(inc->inc_fport)); 2968 } 2969 2970 static void 2971 db_print_inpflags(int inp_flags) 2972 { 2973 int comma; 2974 2975 comma = 0; 2976 if (inp_flags & INP_RECVOPTS) { 2977 db_printf("%sINP_RECVOPTS", comma ? ", " : ""); 2978 comma = 1; 2979 } 2980 if (inp_flags & INP_RECVRETOPTS) { 2981 db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); 2982 comma = 1; 2983 } 2984 if (inp_flags & INP_RECVDSTADDR) { 2985 db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); 2986 comma = 1; 2987 } 2988 if (inp_flags & INP_ORIGDSTADDR) { 2989 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); 2990 comma = 1; 2991 } 2992 if (inp_flags & INP_HDRINCL) { 2993 db_printf("%sINP_HDRINCL", comma ? ", " : ""); 2994 comma = 1; 2995 } 2996 if (inp_flags & INP_HIGHPORT) { 2997 db_printf("%sINP_HIGHPORT", comma ? ", " : ""); 2998 comma = 1; 2999 } 3000 if (inp_flags & INP_LOWPORT) { 3001 db_printf("%sINP_LOWPORT", comma ? ", " : ""); 3002 comma = 1; 3003 } 3004 if (inp_flags & INP_ANONPORT) { 3005 db_printf("%sINP_ANONPORT", comma ? ", " : ""); 3006 comma = 1; 3007 } 3008 if (inp_flags & INP_RECVIF) { 3009 db_printf("%sINP_RECVIF", comma ? ", " : ""); 3010 comma = 1; 3011 } 3012 if (inp_flags & INP_MTUDISC) { 3013 db_printf("%sINP_MTUDISC", comma ? ", " : ""); 3014 comma = 1; 3015 } 3016 if (inp_flags & INP_RECVTTL) { 3017 db_printf("%sINP_RECVTTL", comma ? ", " : ""); 3018 comma = 1; 3019 } 3020 if (inp_flags & INP_DONTFRAG) { 3021 db_printf("%sINP_DONTFRAG", comma ? ", " : ""); 3022 comma = 1; 3023 } 3024 if (inp_flags & INP_RECVTOS) { 3025 db_printf("%sINP_RECVTOS", comma ? ", " : ""); 3026 comma = 1; 3027 } 3028 if (inp_flags & IN6P_IPV6_V6ONLY) { 3029 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); 3030 comma = 1; 3031 } 3032 if (inp_flags & IN6P_PKTINFO) { 3033 db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); 3034 comma = 1; 3035 } 3036 if (inp_flags & IN6P_HOPLIMIT) { 3037 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); 3038 comma = 1; 3039 } 3040 if (inp_flags & IN6P_HOPOPTS) { 3041 db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); 3042 comma = 1; 3043 } 3044 if (inp_flags & IN6P_DSTOPTS) { 3045 db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); 3046 comma = 1; 3047 } 3048 if (inp_flags & IN6P_RTHDR) { 3049 db_printf("%sIN6P_RTHDR", comma ? ", " : ""); 3050 comma = 1; 3051 } 3052 if (inp_flags & IN6P_RTHDRDSTOPTS) { 3053 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); 3054 comma = 1; 3055 } 3056 if (inp_flags & IN6P_TCLASS) { 3057 db_printf("%sIN6P_TCLASS", comma ? ", " : ""); 3058 comma = 1; 3059 } 3060 if (inp_flags & IN6P_AUTOFLOWLABEL) { 3061 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); 3062 comma = 1; 3063 } 3064 if (inp_flags & INP_TIMEWAIT) { 3065 db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); 3066 comma = 1; 3067 } 3068 if (inp_flags & INP_ONESBCAST) { 3069 db_printf("%sINP_ONESBCAST", comma ? ", " : ""); 3070 comma = 1; 3071 } 3072 if (inp_flags & INP_DROPPED) { 3073 db_printf("%sINP_DROPPED", comma ? ", " : ""); 3074 comma = 1; 3075 } 3076 if (inp_flags & INP_SOCKREF) { 3077 db_printf("%sINP_SOCKREF", comma ? ", " : ""); 3078 comma = 1; 3079 } 3080 if (inp_flags & IN6P_RFC2292) { 3081 db_printf("%sIN6P_RFC2292", comma ? ", " : ""); 3082 comma = 1; 3083 } 3084 if (inp_flags & IN6P_MTU) { 3085 db_printf("IN6P_MTU%s", comma ? ", " : ""); 3086 comma = 1; 3087 } 3088 } 3089 3090 static void 3091 db_print_inpvflag(u_char inp_vflag) 3092 { 3093 int comma; 3094 3095 comma = 0; 3096 if (inp_vflag & INP_IPV4) { 3097 db_printf("%sINP_IPV4", comma ? ", " : ""); 3098 comma = 1; 3099 } 3100 if (inp_vflag & INP_IPV6) { 3101 db_printf("%sINP_IPV6", comma ? ", " : ""); 3102 comma = 1; 3103 } 3104 if (inp_vflag & INP_IPV6PROTO) { 3105 db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); 3106 comma = 1; 3107 } 3108 } 3109 3110 static void 3111 db_print_inpcb(struct inpcb *inp, const char *name, int indent) 3112 { 3113 3114 db_print_indent(indent); 3115 db_printf("%s at %p\n", name, inp); 3116 3117 indent += 2; 3118 3119 db_print_indent(indent); 3120 db_printf("inp_flow: 0x%x\n", inp->inp_flow); 3121 3122 db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); 3123 3124 db_print_indent(indent); 3125 db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", 3126 inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); 3127 3128 db_print_indent(indent); 3129 db_printf("inp_label: %p inp_flags: 0x%x (", 3130 inp->inp_label, inp->inp_flags); 3131 db_print_inpflags(inp->inp_flags); 3132 db_printf(")\n"); 3133 3134 db_print_indent(indent); 3135 db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, 3136 inp->inp_vflag); 3137 db_print_inpvflag(inp->inp_vflag); 3138 db_printf(")\n"); 3139 3140 db_print_indent(indent); 3141 db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", 3142 inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); 3143 3144 db_print_indent(indent); 3145 #ifdef INET6 3146 if (inp->inp_vflag & INP_IPV6) { 3147 db_printf("in6p_options: %p in6p_outputopts: %p " 3148 "in6p_moptions: %p\n", inp->in6p_options, 3149 inp->in6p_outputopts, inp->in6p_moptions); 3150 db_printf("in6p_icmp6filt: %p in6p_cksum %d " 3151 "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, 3152 inp->in6p_hops); 3153 } else 3154 #endif 3155 { 3156 db_printf("inp_ip_tos: %d inp_ip_options: %p " 3157 "inp_ip_moptions: %p\n", inp->inp_ip_tos, 3158 inp->inp_options, inp->inp_moptions); 3159 } 3160 3161 db_print_indent(indent); 3162 db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, 3163 (uintmax_t)inp->inp_gencnt); 3164 } 3165 3166 DB_SHOW_COMMAND(inpcb, db_show_inpcb) 3167 { 3168 struct inpcb *inp; 3169 3170 if (!have_addr) { 3171 db_printf("usage: show inpcb <addr>\n"); 3172 return; 3173 } 3174 inp = (struct inpcb *)addr; 3175 3176 db_print_inpcb(inp, "inpcb", 0); 3177 } 3178 #endif /* DDB */ 3179 3180 #ifdef RATELIMIT 3181 /* 3182 * Modify TX rate limit based on the existing "inp->inp_snd_tag", 3183 * if any. 3184 */ 3185 int 3186 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) 3187 { 3188 union if_snd_tag_modify_params params = { 3189 .rate_limit.max_rate = max_pacing_rate, 3190 }; 3191 struct m_snd_tag *mst; 3192 struct ifnet *ifp; 3193 int error; 3194 3195 mst = inp->inp_snd_tag; 3196 if (mst == NULL) 3197 return (EINVAL); 3198 3199 ifp = mst->ifp; 3200 if (ifp == NULL) 3201 return (EINVAL); 3202 3203 if (ifp->if_snd_tag_modify == NULL) { 3204 error = EOPNOTSUPP; 3205 } else { 3206 error = ifp->if_snd_tag_modify(mst, ¶ms); 3207 } 3208 return (error); 3209 } 3210 3211 /* 3212 * Query existing TX rate limit based on the existing 3213 * "inp->inp_snd_tag", if any. 3214 */ 3215 int 3216 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) 3217 { 3218 union if_snd_tag_query_params params = { }; 3219 struct m_snd_tag *mst; 3220 struct ifnet *ifp; 3221 int error; 3222 3223 mst = inp->inp_snd_tag; 3224 if (mst == NULL) 3225 return (EINVAL); 3226 3227 ifp = mst->ifp; 3228 if (ifp == NULL) 3229 return (EINVAL); 3230 3231 if (ifp->if_snd_tag_query == NULL) { 3232 error = EOPNOTSUPP; 3233 } else { 3234 error = ifp->if_snd_tag_query(mst, ¶ms); 3235 if (error == 0 && p_max_pacing_rate != NULL) 3236 *p_max_pacing_rate = params.rate_limit.max_rate; 3237 } 3238 return (error); 3239 } 3240 3241 /* 3242 * Query existing TX queue level based on the existing 3243 * "inp->inp_snd_tag", if any. 3244 */ 3245 int 3246 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) 3247 { 3248 union if_snd_tag_query_params params = { }; 3249 struct m_snd_tag *mst; 3250 struct ifnet *ifp; 3251 int error; 3252 3253 mst = inp->inp_snd_tag; 3254 if (mst == NULL) 3255 return (EINVAL); 3256 3257 ifp = mst->ifp; 3258 if (ifp == NULL) 3259 return (EINVAL); 3260 3261 if (ifp->if_snd_tag_query == NULL) 3262 return (EOPNOTSUPP); 3263 3264 error = ifp->if_snd_tag_query(mst, ¶ms); 3265 if (error == 0 && p_txqueue_level != NULL) 3266 *p_txqueue_level = params.rate_limit.queue_level; 3267 return (error); 3268 } 3269 3270 /* 3271 * Allocate a new TX rate limit send tag from the network interface 3272 * given by the "ifp" argument and save it in "inp->inp_snd_tag": 3273 */ 3274 int 3275 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, 3276 uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate) 3277 { 3278 union if_snd_tag_alloc_params params = { 3279 .rate_limit.hdr.type = (max_pacing_rate == -1U) ? 3280 IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, 3281 .rate_limit.hdr.flowid = flowid, 3282 .rate_limit.hdr.flowtype = flowtype, 3283 .rate_limit.max_rate = max_pacing_rate, 3284 }; 3285 int error; 3286 3287 INP_WLOCK_ASSERT(inp); 3288 3289 if (inp->inp_snd_tag != NULL) 3290 return (EINVAL); 3291 3292 if (ifp->if_snd_tag_alloc == NULL) { 3293 error = EOPNOTSUPP; 3294 } else { 3295 error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); 3296 3297 /* 3298 * At success increment the refcount on 3299 * the send tag's network interface: 3300 */ 3301 if (error == 0) 3302 if_ref(inp->inp_snd_tag->ifp); 3303 } 3304 return (error); 3305 } 3306 3307 /* 3308 * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", 3309 * if any: 3310 */ 3311 void 3312 in_pcbdetach_txrtlmt(struct inpcb *inp) 3313 { 3314 struct m_snd_tag *mst; 3315 struct ifnet *ifp; 3316 3317 INP_WLOCK_ASSERT(inp); 3318 3319 mst = inp->inp_snd_tag; 3320 inp->inp_snd_tag = NULL; 3321 3322 if (mst == NULL) 3323 return; 3324 3325 ifp = mst->ifp; 3326 if (ifp == NULL) 3327 return; 3328 3329 /* 3330 * If the device was detached while we still had reference(s) 3331 * on the ifp, we assume if_snd_tag_free() was replaced with 3332 * stubs. 3333 */ 3334 ifp->if_snd_tag_free(mst); 3335 3336 /* release reference count on network interface */ 3337 if_rele(ifp); 3338 } 3339 3340 /* 3341 * This function should be called when the INP_RATE_LIMIT_CHANGED flag 3342 * is set in the fast path and will attach/detach/modify the TX rate 3343 * limit send tag based on the socket's so_max_pacing_rate value. 3344 */ 3345 void 3346 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) 3347 { 3348 struct socket *socket; 3349 uint32_t max_pacing_rate; 3350 bool did_upgrade; 3351 int error; 3352 3353 if (inp == NULL) 3354 return; 3355 3356 socket = inp->inp_socket; 3357 if (socket == NULL) 3358 return; 3359 3360 if (!INP_WLOCKED(inp)) { 3361 /* 3362 * NOTE: If the write locking fails, we need to bail 3363 * out and use the non-ratelimited ring for the 3364 * transmit until there is a new chance to get the 3365 * write lock. 3366 */ 3367 if (!INP_TRY_UPGRADE(inp)) 3368 return; 3369 did_upgrade = 1; 3370 } else { 3371 did_upgrade = 0; 3372 } 3373 3374 /* 3375 * NOTE: The so_max_pacing_rate value is read unlocked, 3376 * because atomic updates are not required since the variable 3377 * is checked at every mbuf we send. It is assumed that the 3378 * variable read itself will be atomic. 3379 */ 3380 max_pacing_rate = socket->so_max_pacing_rate; 3381 3382 /* 3383 * NOTE: When attaching to a network interface a reference is 3384 * made to ensure the network interface doesn't go away until 3385 * all ratelimit connections are gone. The network interface 3386 * pointers compared below represent valid network interfaces, 3387 * except when comparing towards NULL. 3388 */ 3389 if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { 3390 error = 0; 3391 } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { 3392 if (inp->inp_snd_tag != NULL) 3393 in_pcbdetach_txrtlmt(inp); 3394 error = 0; 3395 } else if (inp->inp_snd_tag == NULL) { 3396 /* 3397 * In order to utilize packet pacing with RSS, we need 3398 * to wait until there is a valid RSS hash before we 3399 * can proceed: 3400 */ 3401 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { 3402 error = EAGAIN; 3403 } else { 3404 error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), 3405 mb->m_pkthdr.flowid, max_pacing_rate); 3406 } 3407 } else { 3408 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); 3409 } 3410 if (error == 0 || error == EOPNOTSUPP) 3411 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; 3412 if (did_upgrade) 3413 INP_DOWNGRADE(inp); 3414 } 3415 3416 /* 3417 * Track route changes for TX rate limiting. 3418 */ 3419 void 3420 in_pcboutput_eagain(struct inpcb *inp) 3421 { 3422 struct socket *socket; 3423 bool did_upgrade; 3424 3425 if (inp == NULL) 3426 return; 3427 3428 socket = inp->inp_socket; 3429 if (socket == NULL) 3430 return; 3431 3432 if (inp->inp_snd_tag == NULL) 3433 return; 3434 3435 if (!INP_WLOCKED(inp)) { 3436 /* 3437 * NOTE: If the write locking fails, we need to bail 3438 * out and use the non-ratelimited ring for the 3439 * transmit until there is a new chance to get the 3440 * write lock. 3441 */ 3442 if (!INP_TRY_UPGRADE(inp)) 3443 return; 3444 did_upgrade = 1; 3445 } else { 3446 did_upgrade = 0; 3447 } 3448 3449 /* detach rate limiting */ 3450 in_pcbdetach_txrtlmt(inp); 3451 3452 /* make sure new mbuf send tag allocation is made */ 3453 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; 3454 3455 if (did_upgrade) 3456 INP_DOWNGRADE(inp); 3457 } 3458 #endif /* RATELIMIT */ 3459