1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Support for INET connection oriented protocols. 8 * 9 * Authors: See the TCP sources 10 */ 11 12 #include <linux/module.h> 13 #include <linux/jhash.h> 14 15 #include <net/inet_connection_sock.h> 16 #include <net/inet_hashtables.h> 17 #include <net/inet_timewait_sock.h> 18 #include <net/ip.h> 19 #include <net/route.h> 20 #include <net/tcp_states.h> 21 #include <net/xfrm.h> 22 #include <net/tcp.h> 23 #include <net/sock_reuseport.h> 24 #include <net/addrconf.h> 25 26 #if IS_ENABLED(CONFIG_IPV6) 27 /* match_sk*_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses 28 * if IPv6 only, and any IPv4 addresses 29 * if not IPv6 only 30 * match_sk*_wildcard == false: addresses must be exactly the same, i.e. 31 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY, 32 * and 0.0.0.0 equals to 0.0.0.0 only 33 */ 34 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6, 35 const struct in6_addr *sk2_rcv_saddr6, 36 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr, 37 bool sk1_ipv6only, bool sk2_ipv6only, 38 bool match_sk1_wildcard, 39 bool match_sk2_wildcard) 40 { 41 int addr_type = ipv6_addr_type(sk1_rcv_saddr6); 42 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; 43 44 /* if both are mapped, treat as IPv4 */ 45 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) { 46 if (!sk2_ipv6only) { 47 if (sk1_rcv_saddr == sk2_rcv_saddr) 48 return true; 49 return (match_sk1_wildcard && !sk1_rcv_saddr) || 50 (match_sk2_wildcard && !sk2_rcv_saddr); 51 } 52 return false; 53 } 54 55 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY) 56 return true; 57 58 if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard && 59 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) 60 return true; 61 62 if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard && 63 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED)) 64 return true; 65 66 if (sk2_rcv_saddr6 && 67 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6)) 68 return true; 69 70 return false; 71 } 72 #endif 73 74 /* match_sk*_wildcard == true: 0.0.0.0 equals to any IPv4 addresses 75 * match_sk*_wildcard == false: addresses must be exactly the same, i.e. 76 * 0.0.0.0 only equals to 0.0.0.0 77 */ 78 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr, 79 bool sk2_ipv6only, bool match_sk1_wildcard, 80 bool match_sk2_wildcard) 81 { 82 if (!sk2_ipv6only) { 83 if (sk1_rcv_saddr == sk2_rcv_saddr) 84 return true; 85 return (match_sk1_wildcard && !sk1_rcv_saddr) || 86 (match_sk2_wildcard && !sk2_rcv_saddr); 87 } 88 return false; 89 } 90 91 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2, 92 bool match_wildcard) 93 { 94 #if IS_ENABLED(CONFIG_IPV6) 95 if (sk->sk_family == AF_INET6) 96 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr, 97 inet6_rcv_saddr(sk2), 98 sk->sk_rcv_saddr, 99 sk2->sk_rcv_saddr, 100 ipv6_only_sock(sk), 101 ipv6_only_sock(sk2), 102 match_wildcard, 103 match_wildcard); 104 #endif 105 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr, 106 ipv6_only_sock(sk2), match_wildcard, 107 match_wildcard); 108 } 109 EXPORT_SYMBOL(inet_rcv_saddr_equal); 110 111 bool inet_rcv_saddr_any(const struct sock *sk) 112 { 113 #if IS_ENABLED(CONFIG_IPV6) 114 if (sk->sk_family == AF_INET6) 115 return ipv6_addr_any(&sk->sk_v6_rcv_saddr); 116 #endif 117 return !sk->sk_rcv_saddr; 118 } 119 120 void inet_get_local_port_range(struct net *net, int *low, int *high) 121 { 122 unsigned int seq; 123 124 do { 125 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock); 126 127 *low = net->ipv4.ip_local_ports.range[0]; 128 *high = net->ipv4.ip_local_ports.range[1]; 129 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq)); 130 } 131 EXPORT_SYMBOL(inet_get_local_port_range); 132 133 static int inet_csk_bind_conflict(const struct sock *sk, 134 const struct inet_bind_bucket *tb, 135 bool relax, bool reuseport_ok) 136 { 137 struct sock *sk2; 138 bool reuse = sk->sk_reuse; 139 bool reuseport = !!sk->sk_reuseport; 140 kuid_t uid = sock_i_uid((struct sock *)sk); 141 142 /* 143 * Unlike other sk lookup places we do not check 144 * for sk_net here, since _all_ the socks listed 145 * in tb->owners list belong to the same net - the 146 * one this bucket belongs to. 147 */ 148 149 sk_for_each_bound(sk2, &tb->owners) { 150 if (sk != sk2 && 151 (!sk->sk_bound_dev_if || 152 !sk2->sk_bound_dev_if || 153 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { 154 if (reuse && sk2->sk_reuse && 155 sk2->sk_state != TCP_LISTEN) { 156 if ((!relax || 157 (!reuseport_ok && 158 reuseport && sk2->sk_reuseport && 159 !rcu_access_pointer(sk->sk_reuseport_cb) && 160 (sk2->sk_state == TCP_TIME_WAIT || 161 uid_eq(uid, sock_i_uid(sk2))))) && 162 inet_rcv_saddr_equal(sk, sk2, true)) 163 break; 164 } else if (!reuseport_ok || 165 !reuseport || !sk2->sk_reuseport || 166 rcu_access_pointer(sk->sk_reuseport_cb) || 167 (sk2->sk_state != TCP_TIME_WAIT && 168 !uid_eq(uid, sock_i_uid(sk2)))) { 169 if (inet_rcv_saddr_equal(sk, sk2, true)) 170 break; 171 } 172 } 173 } 174 return sk2 != NULL; 175 } 176 177 /* 178 * Find an open port number for the socket. Returns with the 179 * inet_bind_hashbucket lock held. 180 */ 181 static struct inet_bind_hashbucket * 182 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret) 183 { 184 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; 185 int port = 0; 186 struct inet_bind_hashbucket *head; 187 struct net *net = sock_net(sk); 188 bool relax = false; 189 int i, low, high, attempt_half; 190 struct inet_bind_bucket *tb; 191 u32 remaining, offset; 192 int l3mdev; 193 194 l3mdev = inet_sk_bound_l3mdev(sk); 195 ports_exhausted: 196 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0; 197 other_half_scan: 198 inet_get_local_port_range(net, &low, &high); 199 high++; /* [32768, 60999] -> [32768, 61000[ */ 200 if (high - low < 4) 201 attempt_half = 0; 202 if (attempt_half) { 203 int half = low + (((high - low) >> 2) << 1); 204 205 if (attempt_half == 1) 206 high = half; 207 else 208 low = half; 209 } 210 remaining = high - low; 211 if (likely(remaining > 1)) 212 remaining &= ~1U; 213 214 offset = prandom_u32() % remaining; 215 /* __inet_hash_connect() favors ports having @low parity 216 * We do the opposite to not pollute connect() users. 217 */ 218 offset |= 1U; 219 220 other_parity_scan: 221 port = low + offset; 222 for (i = 0; i < remaining; i += 2, port += 2) { 223 if (unlikely(port >= high)) 224 port -= remaining; 225 if (inet_is_local_reserved_port(net, port)) 226 continue; 227 head = &hinfo->bhash[inet_bhashfn(net, port, 228 hinfo->bhash_size)]; 229 spin_lock_bh(&head->lock); 230 inet_bind_bucket_for_each(tb, &head->chain) 231 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev && 232 tb->port == port) { 233 if (!inet_csk_bind_conflict(sk, tb, relax, false)) 234 goto success; 235 goto next_port; 236 } 237 tb = NULL; 238 goto success; 239 next_port: 240 spin_unlock_bh(&head->lock); 241 cond_resched(); 242 } 243 244 offset--; 245 if (!(offset & 1)) 246 goto other_parity_scan; 247 248 if (attempt_half == 1) { 249 /* OK we now try the upper half of the range */ 250 attempt_half = 2; 251 goto other_half_scan; 252 } 253 254 if (net->ipv4.sysctl_ip_autobind_reuse && !relax) { 255 /* We still have a chance to connect to different destinations */ 256 relax = true; 257 goto ports_exhausted; 258 } 259 return NULL; 260 success: 261 *port_ret = port; 262 *tb_ret = tb; 263 return head; 264 } 265 266 static inline int sk_reuseport_match(struct inet_bind_bucket *tb, 267 struct sock *sk) 268 { 269 kuid_t uid = sock_i_uid(sk); 270 271 if (tb->fastreuseport <= 0) 272 return 0; 273 if (!sk->sk_reuseport) 274 return 0; 275 if (rcu_access_pointer(sk->sk_reuseport_cb)) 276 return 0; 277 if (!uid_eq(tb->fastuid, uid)) 278 return 0; 279 /* We only need to check the rcv_saddr if this tb was once marked 280 * without fastreuseport and then was reset, as we can only know that 281 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the 282 * owners list. 283 */ 284 if (tb->fastreuseport == FASTREUSEPORT_ANY) 285 return 1; 286 #if IS_ENABLED(CONFIG_IPV6) 287 if (tb->fast_sk_family == AF_INET6) 288 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr, 289 inet6_rcv_saddr(sk), 290 tb->fast_rcv_saddr, 291 sk->sk_rcv_saddr, 292 tb->fast_ipv6_only, 293 ipv6_only_sock(sk), true, false); 294 #endif 295 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr, 296 ipv6_only_sock(sk), true, false); 297 } 298 299 /* Obtain a reference to a local port for the given sock, 300 * if snum is zero it means select any available local port. 301 * We try to allocate an odd port (and leave even ports for connect()) 302 */ 303 int inet_csk_get_port(struct sock *sk, unsigned short snum) 304 { 305 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; 306 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; 307 int ret = 1, port = snum; 308 struct inet_bind_hashbucket *head; 309 struct net *net = sock_net(sk); 310 struct inet_bind_bucket *tb = NULL; 311 kuid_t uid = sock_i_uid(sk); 312 int l3mdev; 313 314 l3mdev = inet_sk_bound_l3mdev(sk); 315 316 if (!port) { 317 head = inet_csk_find_open_port(sk, &tb, &port); 318 if (!head) 319 return ret; 320 if (!tb) 321 goto tb_not_found; 322 goto success; 323 } 324 head = &hinfo->bhash[inet_bhashfn(net, port, 325 hinfo->bhash_size)]; 326 spin_lock_bh(&head->lock); 327 inet_bind_bucket_for_each(tb, &head->chain) 328 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev && 329 tb->port == port) 330 goto tb_found; 331 tb_not_found: 332 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, 333 net, head, port, l3mdev); 334 if (!tb) 335 goto fail_unlock; 336 tb_found: 337 if (!hlist_empty(&tb->owners)) { 338 if (sk->sk_reuse == SK_FORCE_REUSE) 339 goto success; 340 341 if ((tb->fastreuse > 0 && reuse) || 342 sk_reuseport_match(tb, sk)) 343 goto success; 344 if (inet_csk_bind_conflict(sk, tb, true, true)) 345 goto fail_unlock; 346 } 347 success: 348 if (hlist_empty(&tb->owners)) { 349 tb->fastreuse = reuse; 350 if (sk->sk_reuseport) { 351 tb->fastreuseport = FASTREUSEPORT_ANY; 352 tb->fastuid = uid; 353 tb->fast_rcv_saddr = sk->sk_rcv_saddr; 354 tb->fast_ipv6_only = ipv6_only_sock(sk); 355 tb->fast_sk_family = sk->sk_family; 356 #if IS_ENABLED(CONFIG_IPV6) 357 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 358 #endif 359 } else { 360 tb->fastreuseport = 0; 361 } 362 } else { 363 if (!reuse) 364 tb->fastreuse = 0; 365 if (sk->sk_reuseport) { 366 /* We didn't match or we don't have fastreuseport set on 367 * the tb, but we have sk_reuseport set on this socket 368 * and we know that there are no bind conflicts with 369 * this socket in this tb, so reset our tb's reuseport 370 * settings so that any subsequent sockets that match 371 * our current socket will be put on the fast path. 372 * 373 * If we reset we need to set FASTREUSEPORT_STRICT so we 374 * do extra checking for all subsequent sk_reuseport 375 * socks. 376 */ 377 if (!sk_reuseport_match(tb, sk)) { 378 tb->fastreuseport = FASTREUSEPORT_STRICT; 379 tb->fastuid = uid; 380 tb->fast_rcv_saddr = sk->sk_rcv_saddr; 381 tb->fast_ipv6_only = ipv6_only_sock(sk); 382 tb->fast_sk_family = sk->sk_family; 383 #if IS_ENABLED(CONFIG_IPV6) 384 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 385 #endif 386 } 387 } else { 388 tb->fastreuseport = 0; 389 } 390 } 391 if (!inet_csk(sk)->icsk_bind_hash) 392 inet_bind_hash(sk, tb, port); 393 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); 394 ret = 0; 395 396 fail_unlock: 397 spin_unlock_bh(&head->lock); 398 return ret; 399 } 400 EXPORT_SYMBOL_GPL(inet_csk_get_port); 401 402 /* 403 * Wait for an incoming connection, avoid race conditions. This must be called 404 * with the socket locked. 405 */ 406 static int inet_csk_wait_for_connect(struct sock *sk, long timeo) 407 { 408 struct inet_connection_sock *icsk = inet_csk(sk); 409 DEFINE_WAIT(wait); 410 int err; 411 412 /* 413 * True wake-one mechanism for incoming connections: only 414 * one process gets woken up, not the 'whole herd'. 415 * Since we do not 'race & poll' for established sockets 416 * anymore, the common case will execute the loop only once. 417 * 418 * Subtle issue: "add_wait_queue_exclusive()" will be added 419 * after any current non-exclusive waiters, and we know that 420 * it will always _stay_ after any new non-exclusive waiters 421 * because all non-exclusive waiters are added at the 422 * beginning of the wait-queue. As such, it's ok to "drop" 423 * our exclusiveness temporarily when we get woken up without 424 * having to remove and re-insert us on the wait queue. 425 */ 426 for (;;) { 427 prepare_to_wait_exclusive(sk_sleep(sk), &wait, 428 TASK_INTERRUPTIBLE); 429 release_sock(sk); 430 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) 431 timeo = schedule_timeout(timeo); 432 sched_annotate_sleep(); 433 lock_sock(sk); 434 err = 0; 435 if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) 436 break; 437 err = -EINVAL; 438 if (sk->sk_state != TCP_LISTEN) 439 break; 440 err = sock_intr_errno(timeo); 441 if (signal_pending(current)) 442 break; 443 err = -EAGAIN; 444 if (!timeo) 445 break; 446 } 447 finish_wait(sk_sleep(sk), &wait); 448 return err; 449 } 450 451 /* 452 * This will accept the next outstanding connection. 453 */ 454 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern) 455 { 456 struct inet_connection_sock *icsk = inet_csk(sk); 457 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 458 struct request_sock *req; 459 struct sock *newsk; 460 int error; 461 462 lock_sock(sk); 463 464 /* We need to make sure that this socket is listening, 465 * and that it has something pending. 466 */ 467 error = -EINVAL; 468 if (sk->sk_state != TCP_LISTEN) 469 goto out_err; 470 471 /* Find already established connection */ 472 if (reqsk_queue_empty(queue)) { 473 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 474 475 /* If this is a non blocking socket don't sleep */ 476 error = -EAGAIN; 477 if (!timeo) 478 goto out_err; 479 480 error = inet_csk_wait_for_connect(sk, timeo); 481 if (error) 482 goto out_err; 483 } 484 req = reqsk_queue_remove(queue, sk); 485 newsk = req->sk; 486 487 if (sk->sk_protocol == IPPROTO_TCP && 488 tcp_rsk(req)->tfo_listener) { 489 spin_lock_bh(&queue->fastopenq.lock); 490 if (tcp_rsk(req)->tfo_listener) { 491 /* We are still waiting for the final ACK from 3WHS 492 * so can't free req now. Instead, we set req->sk to 493 * NULL to signify that the child socket is taken 494 * so reqsk_fastopen_remove() will free the req 495 * when 3WHS finishes (or is aborted). 496 */ 497 req->sk = NULL; 498 req = NULL; 499 } 500 spin_unlock_bh(&queue->fastopenq.lock); 501 } 502 503 out: 504 release_sock(sk); 505 if (newsk && mem_cgroup_sockets_enabled) { 506 int amt; 507 508 /* atomically get the memory usage, set and charge the 509 * newsk->sk_memcg. 510 */ 511 lock_sock(newsk); 512 513 /* The socket has not been accepted yet, no need to look at 514 * newsk->sk_wmem_queued. 515 */ 516 amt = sk_mem_pages(newsk->sk_forward_alloc + 517 atomic_read(&newsk->sk_rmem_alloc)); 518 mem_cgroup_sk_alloc(newsk); 519 if (newsk->sk_memcg && amt) 520 mem_cgroup_charge_skmem(newsk->sk_memcg, amt); 521 522 release_sock(newsk); 523 } 524 if (req) 525 reqsk_put(req); 526 return newsk; 527 out_err: 528 newsk = NULL; 529 req = NULL; 530 *err = error; 531 goto out; 532 } 533 EXPORT_SYMBOL(inet_csk_accept); 534 535 /* 536 * Using different timers for retransmit, delayed acks and probes 537 * We may wish use just one timer maintaining a list of expire jiffies 538 * to optimize. 539 */ 540 void inet_csk_init_xmit_timers(struct sock *sk, 541 void (*retransmit_handler)(struct timer_list *t), 542 void (*delack_handler)(struct timer_list *t), 543 void (*keepalive_handler)(struct timer_list *t)) 544 { 545 struct inet_connection_sock *icsk = inet_csk(sk); 546 547 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0); 548 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0); 549 timer_setup(&sk->sk_timer, keepalive_handler, 0); 550 icsk->icsk_pending = icsk->icsk_ack.pending = 0; 551 } 552 EXPORT_SYMBOL(inet_csk_init_xmit_timers); 553 554 void inet_csk_clear_xmit_timers(struct sock *sk) 555 { 556 struct inet_connection_sock *icsk = inet_csk(sk); 557 558 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; 559 560 sk_stop_timer(sk, &icsk->icsk_retransmit_timer); 561 sk_stop_timer(sk, &icsk->icsk_delack_timer); 562 sk_stop_timer(sk, &sk->sk_timer); 563 } 564 EXPORT_SYMBOL(inet_csk_clear_xmit_timers); 565 566 void inet_csk_delete_keepalive_timer(struct sock *sk) 567 { 568 sk_stop_timer(sk, &sk->sk_timer); 569 } 570 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); 571 572 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) 573 { 574 sk_reset_timer(sk, &sk->sk_timer, jiffies + len); 575 } 576 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); 577 578 struct dst_entry *inet_csk_route_req(const struct sock *sk, 579 struct flowi4 *fl4, 580 const struct request_sock *req) 581 { 582 const struct inet_request_sock *ireq = inet_rsk(req); 583 struct net *net = read_pnet(&ireq->ireq_net); 584 struct ip_options_rcu *opt; 585 struct rtable *rt; 586 587 rcu_read_lock(); 588 opt = rcu_dereference(ireq->ireq_opt); 589 590 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, 591 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 592 sk->sk_protocol, inet_sk_flowi_flags(sk), 593 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, 594 ireq->ir_loc_addr, ireq->ir_rmt_port, 595 htons(ireq->ir_num), sk->sk_uid); 596 security_req_classify_flow(req, flowi4_to_flowi(fl4)); 597 rt = ip_route_output_flow(net, fl4, sk); 598 if (IS_ERR(rt)) 599 goto no_route; 600 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) 601 goto route_err; 602 rcu_read_unlock(); 603 return &rt->dst; 604 605 route_err: 606 ip_rt_put(rt); 607 no_route: 608 rcu_read_unlock(); 609 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 610 return NULL; 611 } 612 EXPORT_SYMBOL_GPL(inet_csk_route_req); 613 614 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk, 615 struct sock *newsk, 616 const struct request_sock *req) 617 { 618 const struct inet_request_sock *ireq = inet_rsk(req); 619 struct net *net = read_pnet(&ireq->ireq_net); 620 struct inet_sock *newinet = inet_sk(newsk); 621 struct ip_options_rcu *opt; 622 struct flowi4 *fl4; 623 struct rtable *rt; 624 625 opt = rcu_dereference(ireq->ireq_opt); 626 fl4 = &newinet->cork.fl.u.ip4; 627 628 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, 629 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 630 sk->sk_protocol, inet_sk_flowi_flags(sk), 631 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, 632 ireq->ir_loc_addr, ireq->ir_rmt_port, 633 htons(ireq->ir_num), sk->sk_uid); 634 security_req_classify_flow(req, flowi4_to_flowi(fl4)); 635 rt = ip_route_output_flow(net, fl4, sk); 636 if (IS_ERR(rt)) 637 goto no_route; 638 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) 639 goto route_err; 640 return &rt->dst; 641 642 route_err: 643 ip_rt_put(rt); 644 no_route: 645 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 646 return NULL; 647 } 648 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); 649 650 /* Decide when to expire the request and when to resend SYN-ACK */ 651 static void syn_ack_recalc(struct request_sock *req, 652 const int max_syn_ack_retries, 653 const u8 rskq_defer_accept, 654 int *expire, int *resend) 655 { 656 if (!rskq_defer_accept) { 657 *expire = req->num_timeout >= max_syn_ack_retries; 658 *resend = 1; 659 return; 660 } 661 *expire = req->num_timeout >= max_syn_ack_retries && 662 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept); 663 /* Do not resend while waiting for data after ACK, 664 * start to resend on end of deferring period to give 665 * last chance for data or ACK to create established socket. 666 */ 667 *resend = !inet_rsk(req)->acked || 668 req->num_timeout >= rskq_defer_accept - 1; 669 } 670 671 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req) 672 { 673 int err = req->rsk_ops->rtx_syn_ack(parent, req); 674 675 if (!err) 676 req->num_retrans++; 677 return err; 678 } 679 EXPORT_SYMBOL(inet_rtx_syn_ack); 680 681 /* return true if req was found in the ehash table */ 682 static bool reqsk_queue_unlink(struct request_sock *req) 683 { 684 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo; 685 bool found = false; 686 687 if (sk_hashed(req_to_sk(req))) { 688 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash); 689 690 spin_lock(lock); 691 found = __sk_nulls_del_node_init_rcu(req_to_sk(req)); 692 spin_unlock(lock); 693 } 694 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer)) 695 reqsk_put(req); 696 return found; 697 } 698 699 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) 700 { 701 if (reqsk_queue_unlink(req)) { 702 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); 703 reqsk_put(req); 704 } 705 } 706 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); 707 708 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req) 709 { 710 inet_csk_reqsk_queue_drop(sk, req); 711 reqsk_put(req); 712 } 713 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put); 714 715 static void reqsk_timer_handler(struct timer_list *t) 716 { 717 struct request_sock *req = from_timer(req, t, rsk_timer); 718 struct sock *sk_listener = req->rsk_listener; 719 struct net *net = sock_net(sk_listener); 720 struct inet_connection_sock *icsk = inet_csk(sk_listener); 721 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 722 int max_syn_ack_retries, qlen, expire = 0, resend = 0; 723 724 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) 725 goto drop; 726 727 max_syn_ack_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries; 728 /* Normally all the openreqs are young and become mature 729 * (i.e. converted to established socket) for first timeout. 730 * If synack was not acknowledged for 1 second, it means 731 * one of the following things: synack was lost, ack was lost, 732 * rtt is high or nobody planned to ack (i.e. synflood). 733 * When server is a bit loaded, queue is populated with old 734 * open requests, reducing effective size of queue. 735 * When server is well loaded, queue size reduces to zero 736 * after several minutes of work. It is not synflood, 737 * it is normal operation. The solution is pruning 738 * too old entries overriding normal timeout, when 739 * situation becomes dangerous. 740 * 741 * Essentially, we reserve half of room for young 742 * embrions; and abort old ones without pity, if old 743 * ones are about to clog our table. 744 */ 745 qlen = reqsk_queue_len(queue); 746 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) { 747 int young = reqsk_queue_len_young(queue) << 1; 748 749 while (max_syn_ack_retries > 2) { 750 if (qlen < young) 751 break; 752 max_syn_ack_retries--; 753 young <<= 1; 754 } 755 } 756 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept), 757 &expire, &resend); 758 req->rsk_ops->syn_ack_timeout(req); 759 if (!expire && 760 (!resend || 761 !inet_rtx_syn_ack(sk_listener, req) || 762 inet_rsk(req)->acked)) { 763 unsigned long timeo; 764 765 if (req->num_timeout++ == 0) 766 atomic_dec(&queue->young); 767 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); 768 mod_timer(&req->rsk_timer, jiffies + timeo); 769 return; 770 } 771 drop: 772 inet_csk_reqsk_queue_drop_and_put(sk_listener, req); 773 } 774 775 static void reqsk_queue_hash_req(struct request_sock *req, 776 unsigned long timeout) 777 { 778 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED); 779 mod_timer(&req->rsk_timer, jiffies + timeout); 780 781 inet_ehash_insert(req_to_sk(req), NULL); 782 /* before letting lookups find us, make sure all req fields 783 * are committed to memory and refcnt initialized. 784 */ 785 smp_wmb(); 786 refcount_set(&req->rsk_refcnt, 2 + 1); 787 } 788 789 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, 790 unsigned long timeout) 791 { 792 reqsk_queue_hash_req(req, timeout); 793 inet_csk_reqsk_queue_added(sk); 794 } 795 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); 796 797 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk, 798 const gfp_t priority) 799 { 800 struct inet_connection_sock *icsk = inet_csk(newsk); 801 802 if (!icsk->icsk_ulp_ops) 803 return; 804 805 if (icsk->icsk_ulp_ops->clone) 806 icsk->icsk_ulp_ops->clone(req, newsk, priority); 807 } 808 809 /** 810 * inet_csk_clone_lock - clone an inet socket, and lock its clone 811 * @sk: the socket to clone 812 * @req: request_sock 813 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 814 * 815 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 816 */ 817 struct sock *inet_csk_clone_lock(const struct sock *sk, 818 const struct request_sock *req, 819 const gfp_t priority) 820 { 821 struct sock *newsk = sk_clone_lock(sk, priority); 822 823 if (newsk) { 824 struct inet_connection_sock *newicsk = inet_csk(newsk); 825 826 inet_sk_set_state(newsk, TCP_SYN_RECV); 827 newicsk->icsk_bind_hash = NULL; 828 829 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port; 830 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num; 831 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); 832 833 /* listeners have SOCK_RCU_FREE, not the children */ 834 sock_reset_flag(newsk, SOCK_RCU_FREE); 835 836 inet_sk(newsk)->mc_list = NULL; 837 838 newsk->sk_mark = inet_rsk(req)->ir_mark; 839 atomic64_set(&newsk->sk_cookie, 840 atomic64_read(&inet_rsk(req)->ir_cookie)); 841 842 newicsk->icsk_retransmits = 0; 843 newicsk->icsk_backoff = 0; 844 newicsk->icsk_probes_out = 0; 845 846 /* Deinitialize accept_queue to trap illegal accesses. */ 847 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); 848 849 inet_clone_ulp(req, newsk, priority); 850 851 security_inet_csk_clone(newsk, req); 852 } 853 return newsk; 854 } 855 EXPORT_SYMBOL_GPL(inet_csk_clone_lock); 856 857 /* 858 * At this point, there should be no process reference to this 859 * socket, and thus no user references at all. Therefore we 860 * can assume the socket waitqueue is inactive and nobody will 861 * try to jump onto it. 862 */ 863 void inet_csk_destroy_sock(struct sock *sk) 864 { 865 WARN_ON(sk->sk_state != TCP_CLOSE); 866 WARN_ON(!sock_flag(sk, SOCK_DEAD)); 867 868 /* It cannot be in hash table! */ 869 WARN_ON(!sk_unhashed(sk)); 870 871 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ 872 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); 873 874 sk->sk_prot->destroy(sk); 875 876 sk_stream_kill_queues(sk); 877 878 xfrm_sk_free_policy(sk); 879 880 sk_refcnt_debug_release(sk); 881 882 percpu_counter_dec(sk->sk_prot->orphan_count); 883 884 sock_put(sk); 885 } 886 EXPORT_SYMBOL(inet_csk_destroy_sock); 887 888 /* This function allows to force a closure of a socket after the call to 889 * tcp/dccp_create_openreq_child(). 890 */ 891 void inet_csk_prepare_forced_close(struct sock *sk) 892 __releases(&sk->sk_lock.slock) 893 { 894 /* sk_clone_lock locked the socket and set refcnt to 2 */ 895 bh_unlock_sock(sk); 896 sock_put(sk); 897 inet_csk_prepare_for_destroy_sock(sk); 898 inet_sk(sk)->inet_num = 0; 899 } 900 EXPORT_SYMBOL(inet_csk_prepare_forced_close); 901 902 int inet_csk_listen_start(struct sock *sk, int backlog) 903 { 904 struct inet_connection_sock *icsk = inet_csk(sk); 905 struct inet_sock *inet = inet_sk(sk); 906 int err = -EADDRINUSE; 907 908 reqsk_queue_alloc(&icsk->icsk_accept_queue); 909 910 sk->sk_ack_backlog = 0; 911 inet_csk_delack_init(sk); 912 913 /* There is race window here: we announce ourselves listening, 914 * but this transition is still not validated by get_port(). 915 * It is OK, because this socket enters to hash table only 916 * after validation is complete. 917 */ 918 inet_sk_state_store(sk, TCP_LISTEN); 919 if (!sk->sk_prot->get_port(sk, inet->inet_num)) { 920 inet->inet_sport = htons(inet->inet_num); 921 922 sk_dst_reset(sk); 923 err = sk->sk_prot->hash(sk); 924 925 if (likely(!err)) 926 return 0; 927 } 928 929 inet_sk_set_state(sk, TCP_CLOSE); 930 return err; 931 } 932 EXPORT_SYMBOL_GPL(inet_csk_listen_start); 933 934 static void inet_child_forget(struct sock *sk, struct request_sock *req, 935 struct sock *child) 936 { 937 sk->sk_prot->disconnect(child, O_NONBLOCK); 938 939 sock_orphan(child); 940 941 percpu_counter_inc(sk->sk_prot->orphan_count); 942 943 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { 944 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req); 945 BUG_ON(sk != req->rsk_listener); 946 947 /* Paranoid, to prevent race condition if 948 * an inbound pkt destined for child is 949 * blocked by sock lock in tcp_v4_rcv(). 950 * Also to satisfy an assertion in 951 * tcp_v4_destroy_sock(). 952 */ 953 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL); 954 } 955 inet_csk_destroy_sock(child); 956 } 957 958 struct sock *inet_csk_reqsk_queue_add(struct sock *sk, 959 struct request_sock *req, 960 struct sock *child) 961 { 962 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; 963 964 spin_lock(&queue->rskq_lock); 965 if (unlikely(sk->sk_state != TCP_LISTEN)) { 966 inet_child_forget(sk, req, child); 967 child = NULL; 968 } else { 969 req->sk = child; 970 req->dl_next = NULL; 971 if (queue->rskq_accept_head == NULL) 972 WRITE_ONCE(queue->rskq_accept_head, req); 973 else 974 queue->rskq_accept_tail->dl_next = req; 975 queue->rskq_accept_tail = req; 976 sk_acceptq_added(sk); 977 } 978 spin_unlock(&queue->rskq_lock); 979 return child; 980 } 981 EXPORT_SYMBOL(inet_csk_reqsk_queue_add); 982 983 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child, 984 struct request_sock *req, bool own_req) 985 { 986 if (own_req) { 987 inet_csk_reqsk_queue_drop(sk, req); 988 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); 989 if (inet_csk_reqsk_queue_add(sk, req, child)) 990 return child; 991 } 992 /* Too bad, another child took ownership of the request, undo. */ 993 bh_unlock_sock(child); 994 sock_put(child); 995 return NULL; 996 } 997 EXPORT_SYMBOL(inet_csk_complete_hashdance); 998 999 /* 1000 * This routine closes sockets which have been at least partially 1001 * opened, but not yet accepted. 1002 */ 1003 void inet_csk_listen_stop(struct sock *sk) 1004 { 1005 struct inet_connection_sock *icsk = inet_csk(sk); 1006 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 1007 struct request_sock *next, *req; 1008 1009 /* Following specs, it would be better either to send FIN 1010 * (and enter FIN-WAIT-1, it is normal close) 1011 * or to send active reset (abort). 1012 * Certainly, it is pretty dangerous while synflood, but it is 1013 * bad justification for our negligence 8) 1014 * To be honest, we are not able to make either 1015 * of the variants now. --ANK 1016 */ 1017 while ((req = reqsk_queue_remove(queue, sk)) != NULL) { 1018 struct sock *child = req->sk; 1019 1020 local_bh_disable(); 1021 bh_lock_sock(child); 1022 WARN_ON(sock_owned_by_user(child)); 1023 sock_hold(child); 1024 1025 inet_child_forget(sk, req, child); 1026 reqsk_put(req); 1027 bh_unlock_sock(child); 1028 local_bh_enable(); 1029 sock_put(child); 1030 1031 cond_resched(); 1032 } 1033 if (queue->fastopenq.rskq_rst_head) { 1034 /* Free all the reqs queued in rskq_rst_head. */ 1035 spin_lock_bh(&queue->fastopenq.lock); 1036 req = queue->fastopenq.rskq_rst_head; 1037 queue->fastopenq.rskq_rst_head = NULL; 1038 spin_unlock_bh(&queue->fastopenq.lock); 1039 while (req != NULL) { 1040 next = req->dl_next; 1041 reqsk_put(req); 1042 req = next; 1043 } 1044 } 1045 WARN_ON_ONCE(sk->sk_ack_backlog); 1046 } 1047 EXPORT_SYMBOL_GPL(inet_csk_listen_stop); 1048 1049 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) 1050 { 1051 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 1052 const struct inet_sock *inet = inet_sk(sk); 1053 1054 sin->sin_family = AF_INET; 1055 sin->sin_addr.s_addr = inet->inet_daddr; 1056 sin->sin_port = inet->inet_dport; 1057 } 1058 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); 1059 1060 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) 1061 { 1062 const struct inet_sock *inet = inet_sk(sk); 1063 const struct ip_options_rcu *inet_opt; 1064 __be32 daddr = inet->inet_daddr; 1065 struct flowi4 *fl4; 1066 struct rtable *rt; 1067 1068 rcu_read_lock(); 1069 inet_opt = rcu_dereference(inet->inet_opt); 1070 if (inet_opt && inet_opt->opt.srr) 1071 daddr = inet_opt->opt.faddr; 1072 fl4 = &fl->u.ip4; 1073 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, 1074 inet->inet_saddr, inet->inet_dport, 1075 inet->inet_sport, sk->sk_protocol, 1076 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); 1077 if (IS_ERR(rt)) 1078 rt = NULL; 1079 if (rt) 1080 sk_setup_caps(sk, &rt->dst); 1081 rcu_read_unlock(); 1082 1083 return &rt->dst; 1084 } 1085 1086 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) 1087 { 1088 struct dst_entry *dst = __sk_dst_check(sk, 0); 1089 struct inet_sock *inet = inet_sk(sk); 1090 1091 if (!dst) { 1092 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 1093 if (!dst) 1094 goto out; 1095 } 1096 dst->ops->update_pmtu(dst, sk, NULL, mtu, true); 1097 1098 dst = __sk_dst_check(sk, 0); 1099 if (!dst) 1100 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 1101 out: 1102 return dst; 1103 } 1104 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); 1105