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 */ 34static 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 */ 78static 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 91bool 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 105 106 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr, 107 ipv6_only_sock(sk2), match_wildcard, 108 match_wildcard); 109} 110EXPORT_SYMBOL(inet_rcv_saddr_equal); 111 112bool inet_rcv_saddr_any(const struct sock *sk) 113{ 114#if IS_ENABLED(CONFIG_IPV6) 115 if (sk->sk_family == AF_INET6) 116 return ipv6_addr_any(&sk->sk_v6_rcv_saddr); 117#endif 118 return !sk->sk_rcv_saddr; 119} 120 121void inet_get_local_port_range(struct net *net, int *low, int *high) 122{ 123 unsigned int seq; 124 125 do { 126 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock); 127 128 *low = net->ipv4.ip_local_ports.range[0]; 129 *high = net->ipv4.ip_local_ports.range[1]; 130 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq)); 131} 132EXPORT_SYMBOL(inet_get_local_port_range); 133 134static int inet_csk_bind_conflict(const struct sock *sk, 135 const struct inet_bind_bucket *tb, 136 bool relax, bool reuseport_ok) 137{ 138 struct sock *sk2; 139 bool reuse = sk->sk_reuse; 140 bool reuseport = !!sk->sk_reuseport; 141 kuid_t uid = sock_i_uid((struct sock *)sk); 142 143 /* 144 * Unlike other sk lookup places we do not check 145 * for sk_net here, since _all_ the socks listed 146 * in tb->owners list belong to the same net - the 147 * one this bucket belongs to. 148 */ 149 150 sk_for_each_bound(sk2, &tb->owners) { 151 int bound_dev_if2; 152 153 if (sk == sk2) 154 continue; 155 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if); 156 if ((!sk->sk_bound_dev_if || 157 !bound_dev_if2 || 158 sk->sk_bound_dev_if == bound_dev_if2)) { 159 if (reuse && sk2->sk_reuse && 160 sk2->sk_state != TCP_LISTEN) { 161 if ((!relax || 162 (!reuseport_ok && 163 reuseport && sk2->sk_reuseport && 164 !rcu_access_pointer(sk->sk_reuseport_cb) && 165 (sk2->sk_state == TCP_TIME_WAIT || 166 uid_eq(uid, sock_i_uid(sk2))))) && 167 inet_rcv_saddr_equal(sk, sk2, true)) 168 break; 169 } else if (!reuseport_ok || 170 !reuseport || !sk2->sk_reuseport || 171 rcu_access_pointer(sk->sk_reuseport_cb) || 172 (sk2->sk_state != TCP_TIME_WAIT && 173 !uid_eq(uid, sock_i_uid(sk2)))) { 174 if (inet_rcv_saddr_equal(sk, sk2, true)) 175 break; 176 } 177 } 178 } 179 return sk2 != NULL; 180} 181 182/* 183 * Find an open port number for the socket. Returns with the 184 * inet_bind_hashbucket lock held. 185 */ 186static struct inet_bind_hashbucket * 187inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret) 188{ 189 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; 190 int port = 0; 191 struct inet_bind_hashbucket *head; 192 struct net *net = sock_net(sk); 193 bool relax = false; 194 int i, low, high, attempt_half; 195 struct inet_bind_bucket *tb; 196 u32 remaining, offset; 197 int l3mdev; 198 199 l3mdev = inet_sk_bound_l3mdev(sk); 200ports_exhausted: 201 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0; 202other_half_scan: 203 inet_get_local_port_range(net, &low, &high); 204 high++; /* [32768, 60999] -> [32768, 61000[ */ 205 if (high - low < 4) 206 attempt_half = 0; 207 if (attempt_half) { 208 int half = low + (((high - low) >> 2) << 1); 209 210 if (attempt_half == 1) 211 high = half; 212 else 213 low = half; 214 } 215 remaining = high - low; 216 if (likely(remaining > 1)) 217 remaining &= ~1U; 218 219 offset = prandom_u32() % remaining; 220 /* __inet_hash_connect() favors ports having @low parity 221 * We do the opposite to not pollute connect() users. 222 */ 223 offset |= 1U; 224 225other_parity_scan: 226 port = low + offset; 227 for (i = 0; i < remaining; i += 2, port += 2) { 228 if (unlikely(port >= high)) 229 port -= remaining; 230 if (inet_is_local_reserved_port(net, port)) 231 continue; 232 head = &hinfo->bhash[inet_bhashfn(net, port, 233 hinfo->bhash_size)]; 234 spin_lock_bh(&head->lock); 235 inet_bind_bucket_for_each(tb, &head->chain) 236 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev && 237 tb->port == port) { 238 if (!inet_csk_bind_conflict(sk, tb, relax, false)) 239 goto success; 240 goto next_port; 241 } 242 tb = NULL; 243 goto success; 244next_port: 245 spin_unlock_bh(&head->lock); 246 cond_resched(); 247 } 248 249 offset--; 250 if (!(offset & 1)) 251 goto other_parity_scan; 252 253 if (attempt_half == 1) { 254 /* OK we now try the upper half of the range */ 255 attempt_half = 2; 256 goto other_half_scan; 257 } 258 259 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) { 260 /* We still have a chance to connect to different destinations */ 261 relax = true; 262 goto ports_exhausted; 263 } 264 return NULL; 265success: 266 *port_ret = port; 267 *tb_ret = tb; 268 return head; 269} 270 271static inline int sk_reuseport_match(struct inet_bind_bucket *tb, 272 struct sock *sk) 273{ 274 kuid_t uid = sock_i_uid(sk); 275 276 if (tb->fastreuseport <= 0) 277 return 0; 278 if (!sk->sk_reuseport) 279 return 0; 280 if (rcu_access_pointer(sk->sk_reuseport_cb)) 281 return 0; 282 if (!uid_eq(tb->fastuid, uid)) 283 return 0; 284 /* We only need to check the rcv_saddr if this tb was once marked 285 * without fastreuseport and then was reset, as we can only know that 286 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the 287 * owners list. 288 */ 289 if (tb->fastreuseport == FASTREUSEPORT_ANY) 290 return 1; 291#if IS_ENABLED(CONFIG_IPV6) 292 if (tb->fast_sk_family == AF_INET6) 293 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr, 294 inet6_rcv_saddr(sk), 295 tb->fast_rcv_saddr, 296 sk->sk_rcv_saddr, 297 tb->fast_ipv6_only, 298 ipv6_only_sock(sk), true, false); 299#endif 300 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr, 301 ipv6_only_sock(sk), true, false); 302} 303 304void inet_csk_update_fastreuse(struct inet_bind_bucket *tb, 305 struct sock *sk) 306{ 307 kuid_t uid = sock_i_uid(sk); 308 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; 309 310 if (hlist_empty(&tb->owners)) { 311 tb->fastreuse = reuse; 312 if (sk->sk_reuseport) { 313 tb->fastreuseport = FASTREUSEPORT_ANY; 314 tb->fastuid = uid; 315 tb->fast_rcv_saddr = sk->sk_rcv_saddr; 316 tb->fast_ipv6_only = ipv6_only_sock(sk); 317 tb->fast_sk_family = sk->sk_family; 318#if IS_ENABLED(CONFIG_IPV6) 319 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 320#endif 321 } else { 322 tb->fastreuseport = 0; 323 } 324 } else { 325 if (!reuse) 326 tb->fastreuse = 0; 327 if (sk->sk_reuseport) { 328 /* We didn't match or we don't have fastreuseport set on 329 * the tb, but we have sk_reuseport set on this socket 330 * and we know that there are no bind conflicts with 331 * this socket in this tb, so reset our tb's reuseport 332 * settings so that any subsequent sockets that match 333 * our current socket will be put on the fast path. 334 * 335 * If we reset we need to set FASTREUSEPORT_STRICT so we 336 * do extra checking for all subsequent sk_reuseport 337 * socks. 338 */ 339 if (!sk_reuseport_match(tb, sk)) { 340 tb->fastreuseport = FASTREUSEPORT_STRICT; 341 tb->fastuid = uid; 342 tb->fast_rcv_saddr = sk->sk_rcv_saddr; 343 tb->fast_ipv6_only = ipv6_only_sock(sk); 344 tb->fast_sk_family = sk->sk_family; 345#if IS_ENABLED(CONFIG_IPV6) 346 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; 347#endif 348 } 349 } else { 350 tb->fastreuseport = 0; 351 } 352 } 353} 354 355/* Obtain a reference to a local port for the given sock, 356 * if snum is zero it means select any available local port. 357 * We try to allocate an odd port (and leave even ports for connect()) 358 */ 359int inet_csk_get_port(struct sock *sk, unsigned short snum) 360{ 361 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; 362 struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; 363 int ret = 1, port = snum; 364 struct inet_bind_hashbucket *head; 365 struct net *net = sock_net(sk); 366 struct inet_bind_bucket *tb = NULL; 367 int l3mdev; 368 369 l3mdev = inet_sk_bound_l3mdev(sk); 370 371 if (!port) { 372 head = inet_csk_find_open_port(sk, &tb, &port); 373 if (!head) 374 return ret; 375 if (!tb) 376 goto tb_not_found; 377 goto success; 378 } 379 head = &hinfo->bhash[inet_bhashfn(net, port, 380 hinfo->bhash_size)]; 381 spin_lock_bh(&head->lock); 382 inet_bind_bucket_for_each(tb, &head->chain) 383 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev && 384 tb->port == port) 385 goto tb_found; 386tb_not_found: 387 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, 388 net, head, port, l3mdev); 389 if (!tb) 390 goto fail_unlock; 391tb_found: 392 if (!hlist_empty(&tb->owners)) { 393 if (sk->sk_reuse == SK_FORCE_REUSE) 394 goto success; 395 396 if ((tb->fastreuse > 0 && reuse) || 397 sk_reuseport_match(tb, sk)) 398 goto success; 399 if (inet_csk_bind_conflict(sk, tb, true, true)) 400 goto fail_unlock; 401 } 402success: 403 inet_csk_update_fastreuse(tb, sk); 404 405 if (!inet_csk(sk)->icsk_bind_hash) 406 inet_bind_hash(sk, tb, port); 407 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); 408 ret = 0; 409 410fail_unlock: 411 spin_unlock_bh(&head->lock); 412 return ret; 413} 414EXPORT_SYMBOL_GPL(inet_csk_get_port); 415 416/* 417 * Wait for an incoming connection, avoid race conditions. This must be called 418 * with the socket locked. 419 */ 420static int inet_csk_wait_for_connect(struct sock *sk, long timeo) 421{ 422 struct inet_connection_sock *icsk = inet_csk(sk); 423 DEFINE_WAIT(wait); 424 int err; 425 426 /* 427 * True wake-one mechanism for incoming connections: only 428 * one process gets woken up, not the 'whole herd'. 429 * Since we do not 'race & poll' for established sockets 430 * anymore, the common case will execute the loop only once. 431 * 432 * Subtle issue: "add_wait_queue_exclusive()" will be added 433 * after any current non-exclusive waiters, and we know that 434 * it will always _stay_ after any new non-exclusive waiters 435 * because all non-exclusive waiters are added at the 436 * beginning of the wait-queue. As such, it's ok to "drop" 437 * our exclusiveness temporarily when we get woken up without 438 * having to remove and re-insert us on the wait queue. 439 */ 440 for (;;) { 441 prepare_to_wait_exclusive(sk_sleep(sk), &wait, 442 TASK_INTERRUPTIBLE); 443 release_sock(sk); 444 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) 445 timeo = schedule_timeout(timeo); 446 sched_annotate_sleep(); 447 lock_sock(sk); 448 err = 0; 449 if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) 450 break; 451 err = -EINVAL; 452 if (sk->sk_state != TCP_LISTEN) 453 break; 454 err = sock_intr_errno(timeo); 455 if (signal_pending(current)) 456 break; 457 err = -EAGAIN; 458 if (!timeo) 459 break; 460 } 461 finish_wait(sk_sleep(sk), &wait); 462 return err; 463} 464 465/* 466 * This will accept the next outstanding connection. 467 */ 468struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern) 469{ 470 struct inet_connection_sock *icsk = inet_csk(sk); 471 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 472 struct request_sock *req; 473 struct sock *newsk; 474 int error; 475 476 lock_sock(sk); 477 478 /* We need to make sure that this socket is listening, 479 * and that it has something pending. 480 */ 481 error = -EINVAL; 482 if (sk->sk_state != TCP_LISTEN) 483 goto out_err; 484 485 /* Find already established connection */ 486 if (reqsk_queue_empty(queue)) { 487 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 488 489 /* If this is a non blocking socket don't sleep */ 490 error = -EAGAIN; 491 if (!timeo) 492 goto out_err; 493 494 error = inet_csk_wait_for_connect(sk, timeo); 495 if (error) 496 goto out_err; 497 } 498 req = reqsk_queue_remove(queue, sk); 499 newsk = req->sk; 500 501 if (sk->sk_protocol == IPPROTO_TCP && 502 tcp_rsk(req)->tfo_listener) { 503 spin_lock_bh(&queue->fastopenq.lock); 504 if (tcp_rsk(req)->tfo_listener) { 505 /* We are still waiting for the final ACK from 3WHS 506 * so can't free req now. Instead, we set req->sk to 507 * NULL to signify that the child socket is taken 508 * so reqsk_fastopen_remove() will free the req 509 * when 3WHS finishes (or is aborted). 510 */ 511 req->sk = NULL; 512 req = NULL; 513 } 514 spin_unlock_bh(&queue->fastopenq.lock); 515 } 516 517out: 518 release_sock(sk); 519 if (newsk && mem_cgroup_sockets_enabled) { 520 int amt; 521 522 /* atomically get the memory usage, set and charge the 523 * newsk->sk_memcg. 524 */ 525 lock_sock(newsk); 526 527 /* The socket has not been accepted yet, no need to look at 528 * newsk->sk_wmem_queued. 529 */ 530 amt = sk_mem_pages(newsk->sk_forward_alloc + 531 atomic_read(&newsk->sk_rmem_alloc)); 532 mem_cgroup_sk_alloc(newsk); 533 if (newsk->sk_memcg && amt) 534 mem_cgroup_charge_skmem(newsk->sk_memcg, amt); 535 536 release_sock(newsk); 537 } 538 if (req) 539 reqsk_put(req); 540 541 if (newsk) 542 inet_init_csk_locks(newsk); 543 544 return newsk; 545out_err: 546 newsk = NULL; 547 req = NULL; 548 *err = error; 549 goto out; 550} 551EXPORT_SYMBOL(inet_csk_accept); 552 553/* 554 * Using different timers for retransmit, delayed acks and probes 555 * We may wish use just one timer maintaining a list of expire jiffies 556 * to optimize. 557 */ 558void inet_csk_init_xmit_timers(struct sock *sk, 559 void (*retransmit_handler)(struct timer_list *t), 560 void (*delack_handler)(struct timer_list *t), 561 void (*keepalive_handler)(struct timer_list *t)) 562{ 563 struct inet_connection_sock *icsk = inet_csk(sk); 564 565 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0); 566 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0); 567 timer_setup(&sk->sk_timer, keepalive_handler, 0); 568 icsk->icsk_pending = icsk->icsk_ack.pending = 0; 569} 570EXPORT_SYMBOL(inet_csk_init_xmit_timers); 571 572void inet_csk_clear_xmit_timers(struct sock *sk) 573{ 574 struct inet_connection_sock *icsk = inet_csk(sk); 575 576 icsk->icsk_pending = icsk->icsk_ack.pending = 0; 577 578 sk_stop_timer(sk, &icsk->icsk_retransmit_timer); 579 sk_stop_timer(sk, &icsk->icsk_delack_timer); 580 sk_stop_timer(sk, &sk->sk_timer); 581} 582EXPORT_SYMBOL(inet_csk_clear_xmit_timers); 583 584void inet_csk_clear_xmit_timers_sync(struct sock *sk) 585{ 586 struct inet_connection_sock *icsk = inet_csk(sk); 587 588 /* ongoing timer handlers need to acquire socket lock. */ 589 sock_not_owned_by_me(sk); 590 591 icsk->icsk_pending = icsk->icsk_ack.pending = 0; 592 593 sk_stop_timer_sync(sk, &icsk->icsk_retransmit_timer); 594 sk_stop_timer_sync(sk, &icsk->icsk_delack_timer); 595 sk_stop_timer_sync(sk, &sk->sk_timer); 596} 597 598void inet_csk_delete_keepalive_timer(struct sock *sk) 599{ 600 sk_stop_timer(sk, &sk->sk_timer); 601} 602EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); 603 604void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) 605{ 606 sk_reset_timer(sk, &sk->sk_timer, jiffies + len); 607} 608EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); 609 610struct dst_entry *inet_csk_route_req(const struct sock *sk, 611 struct flowi4 *fl4, 612 const struct request_sock *req) 613{ 614 const struct inet_request_sock *ireq = inet_rsk(req); 615 struct net *net = read_pnet(&ireq->ireq_net); 616 struct ip_options_rcu *opt; 617 struct rtable *rt; 618 619 rcu_read_lock(); 620 opt = rcu_dereference(ireq->ireq_opt); 621 622 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, 623 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 624 sk->sk_protocol, inet_sk_flowi_flags(sk), 625 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, 626 ireq->ir_loc_addr, ireq->ir_rmt_port, 627 htons(ireq->ir_num), sk->sk_uid); 628 security_req_classify_flow(req, flowi4_to_flowi_common(fl4)); 629 rt = ip_route_output_flow(net, fl4, sk); 630 if (IS_ERR(rt)) 631 goto no_route; 632 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) 633 goto route_err; 634 rcu_read_unlock(); 635 return &rt->dst; 636 637route_err: 638 ip_rt_put(rt); 639no_route: 640 rcu_read_unlock(); 641 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 642 return NULL; 643} 644EXPORT_SYMBOL_GPL(inet_csk_route_req); 645 646struct dst_entry *inet_csk_route_child_sock(const struct sock *sk, 647 struct sock *newsk, 648 const struct request_sock *req) 649{ 650 const struct inet_request_sock *ireq = inet_rsk(req); 651 struct net *net = read_pnet(&ireq->ireq_net); 652 struct inet_sock *newinet = inet_sk(newsk); 653 struct ip_options_rcu *opt; 654 struct flowi4 *fl4; 655 struct rtable *rt; 656 657 opt = rcu_dereference(ireq->ireq_opt); 658 fl4 = &newinet->cork.fl.u.ip4; 659 660 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, 661 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 662 sk->sk_protocol, inet_sk_flowi_flags(sk), 663 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, 664 ireq->ir_loc_addr, ireq->ir_rmt_port, 665 htons(ireq->ir_num), sk->sk_uid); 666 security_req_classify_flow(req, flowi4_to_flowi_common(fl4)); 667 rt = ip_route_output_flow(net, fl4, sk); 668 if (IS_ERR(rt)) 669 goto no_route; 670 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) 671 goto route_err; 672 return &rt->dst; 673 674route_err: 675 ip_rt_put(rt); 676no_route: 677 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); 678 return NULL; 679} 680EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); 681 682/* Decide when to expire the request and when to resend SYN-ACK */ 683static void syn_ack_recalc(struct request_sock *req, 684 const int max_syn_ack_retries, 685 const u8 rskq_defer_accept, 686 int *expire, int *resend) 687{ 688 if (!rskq_defer_accept) { 689 *expire = req->num_timeout >= max_syn_ack_retries; 690 *resend = 1; 691 return; 692 } 693 *expire = req->num_timeout >= max_syn_ack_retries && 694 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept); 695 /* Do not resend while waiting for data after ACK, 696 * start to resend on end of deferring period to give 697 * last chance for data or ACK to create established socket. 698 */ 699 *resend = !inet_rsk(req)->acked || 700 req->num_timeout >= rskq_defer_accept - 1; 701} 702 703int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req) 704{ 705 int err = req->rsk_ops->rtx_syn_ack(parent, req); 706 707 if (!err) 708 req->num_retrans++; 709 return err; 710} 711EXPORT_SYMBOL(inet_rtx_syn_ack); 712 713/* return true if req was found in the ehash table */ 714static bool reqsk_queue_unlink(struct request_sock *req) 715{ 716 struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo; 717 bool found = false; 718 719 if (sk_hashed(req_to_sk(req))) { 720 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash); 721 722 spin_lock(lock); 723 found = __sk_nulls_del_node_init_rcu(req_to_sk(req)); 724 spin_unlock(lock); 725 } 726 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer)) 727 reqsk_put(req); 728 return found; 729} 730 731bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) 732{ 733 bool unlinked = reqsk_queue_unlink(req); 734 735 if (unlinked) { 736 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); 737 reqsk_put(req); 738 } 739 return unlinked; 740} 741EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); 742 743void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req) 744{ 745 inet_csk_reqsk_queue_drop(sk, req); 746 reqsk_put(req); 747} 748EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put); 749 750static void reqsk_timer_handler(struct timer_list *t) 751{ 752 struct request_sock *req = from_timer(req, t, rsk_timer); 753 struct sock *sk_listener = req->rsk_listener; 754 struct net *net = sock_net(sk_listener); 755 struct inet_connection_sock *icsk = inet_csk(sk_listener); 756 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 757 int max_syn_ack_retries, qlen, expire = 0, resend = 0; 758 759 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) 760 goto drop; 761 762 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? : 763 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries); 764 /* Normally all the openreqs are young and become mature 765 * (i.e. converted to established socket) for first timeout. 766 * If synack was not acknowledged for 1 second, it means 767 * one of the following things: synack was lost, ack was lost, 768 * rtt is high or nobody planned to ack (i.e. synflood). 769 * When server is a bit loaded, queue is populated with old 770 * open requests, reducing effective size of queue. 771 * When server is well loaded, queue size reduces to zero 772 * after several minutes of work. It is not synflood, 773 * it is normal operation. The solution is pruning 774 * too old entries overriding normal timeout, when 775 * situation becomes dangerous. 776 * 777 * Essentially, we reserve half of room for young 778 * embrions; and abort old ones without pity, if old 779 * ones are about to clog our table. 780 */ 781 qlen = reqsk_queue_len(queue); 782 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) { 783 int young = reqsk_queue_len_young(queue) << 1; 784 785 while (max_syn_ack_retries > 2) { 786 if (qlen < young) 787 break; 788 max_syn_ack_retries--; 789 young <<= 1; 790 } 791 } 792 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept), 793 &expire, &resend); 794 req->rsk_ops->syn_ack_timeout(req); 795 if (!expire && 796 (!resend || 797 !inet_rtx_syn_ack(sk_listener, req) || 798 inet_rsk(req)->acked)) { 799 unsigned long timeo; 800 801 if (req->num_timeout++ == 0) 802 atomic_dec(&queue->young); 803 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX); 804 mod_timer(&req->rsk_timer, jiffies + timeo); 805 return; 806 } 807drop: 808 inet_csk_reqsk_queue_drop_and_put(sk_listener, req); 809} 810 811static void reqsk_queue_hash_req(struct request_sock *req, 812 unsigned long timeout) 813{ 814 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED); 815 mod_timer(&req->rsk_timer, jiffies + timeout); 816 817 inet_ehash_insert(req_to_sk(req), NULL, NULL); 818 /* before letting lookups find us, make sure all req fields 819 * are committed to memory and refcnt initialized. 820 */ 821 smp_wmb(); 822 refcount_set(&req->rsk_refcnt, 2 + 1); 823} 824 825void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, 826 unsigned long timeout) 827{ 828 reqsk_queue_hash_req(req, timeout); 829 inet_csk_reqsk_queue_added(sk); 830} 831EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); 832 833static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk, 834 const gfp_t priority) 835{ 836 struct inet_connection_sock *icsk = inet_csk(newsk); 837 838 if (!icsk->icsk_ulp_ops) 839 return; 840 841 icsk->icsk_ulp_ops->clone(req, newsk, priority); 842} 843 844/** 845 * inet_csk_clone_lock - clone an inet socket, and lock its clone 846 * @sk: the socket to clone 847 * @req: request_sock 848 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 849 * 850 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 851 */ 852struct sock *inet_csk_clone_lock(const struct sock *sk, 853 const struct request_sock *req, 854 const gfp_t priority) 855{ 856 struct sock *newsk = sk_clone_lock(sk, priority); 857 858 if (newsk) { 859 struct inet_connection_sock *newicsk = inet_csk(newsk); 860 861 newsk->sk_wait_pending = 0; 862 inet_sk_set_state(newsk, TCP_SYN_RECV); 863 newicsk->icsk_bind_hash = NULL; 864 865 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port; 866 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num; 867 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); 868 869 /* listeners have SOCK_RCU_FREE, not the children */ 870 sock_reset_flag(newsk, SOCK_RCU_FREE); 871 872 inet_sk(newsk)->mc_list = NULL; 873 874 newsk->sk_mark = inet_rsk(req)->ir_mark; 875 atomic64_set(&newsk->sk_cookie, 876 atomic64_read(&inet_rsk(req)->ir_cookie)); 877 878 newicsk->icsk_retransmits = 0; 879 newicsk->icsk_backoff = 0; 880 newicsk->icsk_probes_out = 0; 881 newicsk->icsk_probes_tstamp = 0; 882 883 /* Deinitialize accept_queue to trap illegal accesses. */ 884 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); 885 886 inet_clone_ulp(req, newsk, priority); 887 888 security_inet_csk_clone(newsk, req); 889 } 890 return newsk; 891} 892EXPORT_SYMBOL_GPL(inet_csk_clone_lock); 893 894/* 895 * At this point, there should be no process reference to this 896 * socket, and thus no user references at all. Therefore we 897 * can assume the socket waitqueue is inactive and nobody will 898 * try to jump onto it. 899 */ 900void inet_csk_destroy_sock(struct sock *sk) 901{ 902 WARN_ON(sk->sk_state != TCP_CLOSE); 903 WARN_ON(!sock_flag(sk, SOCK_DEAD)); 904 905 /* It cannot be in hash table! */ 906 WARN_ON(!sk_unhashed(sk)); 907 908 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ 909 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); 910 911 sk->sk_prot->destroy(sk); 912 913 sk_stream_kill_queues(sk); 914 915 xfrm_sk_free_policy(sk); 916 917 sk_refcnt_debug_release(sk); 918 919 this_cpu_dec(*sk->sk_prot->orphan_count); 920 921 sock_put(sk); 922} 923EXPORT_SYMBOL(inet_csk_destroy_sock); 924 925/* This function allows to force a closure of a socket after the call to 926 * tcp/dccp_create_openreq_child(). 927 */ 928void inet_csk_prepare_forced_close(struct sock *sk) 929 __releases(&sk->sk_lock.slock) 930{ 931 /* sk_clone_lock locked the socket and set refcnt to 2 */ 932 bh_unlock_sock(sk); 933 sock_put(sk); 934 inet_csk_prepare_for_destroy_sock(sk); 935 inet_sk(sk)->inet_num = 0; 936} 937EXPORT_SYMBOL(inet_csk_prepare_forced_close); 938 939static int inet_ulp_can_listen(const struct sock *sk) 940{ 941 const struct inet_connection_sock *icsk = inet_csk(sk); 942 943 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone) 944 return -EINVAL; 945 946 return 0; 947} 948 949int inet_csk_listen_start(struct sock *sk, int backlog) 950{ 951 struct inet_connection_sock *icsk = inet_csk(sk); 952 struct inet_sock *inet = inet_sk(sk); 953 int err; 954 955 err = inet_ulp_can_listen(sk); 956 if (unlikely(err)) 957 return err; 958 959 reqsk_queue_alloc(&icsk->icsk_accept_queue); 960 961 sk->sk_ack_backlog = 0; 962 inet_csk_delack_init(sk); 963 964 /* There is race window here: we announce ourselves listening, 965 * but this transition is still not validated by get_port(). 966 * It is OK, because this socket enters to hash table only 967 * after validation is complete. 968 */ 969 err = -EADDRINUSE; 970 inet_sk_state_store(sk, TCP_LISTEN); 971 if (!sk->sk_prot->get_port(sk, inet->inet_num)) { 972 inet->inet_sport = htons(inet->inet_num); 973 974 sk_dst_reset(sk); 975 err = sk->sk_prot->hash(sk); 976 977 if (likely(!err)) 978 return 0; 979 } 980 981 inet_sk_set_state(sk, TCP_CLOSE); 982 return err; 983} 984EXPORT_SYMBOL_GPL(inet_csk_listen_start); 985 986static void inet_child_forget(struct sock *sk, struct request_sock *req, 987 struct sock *child) 988{ 989 sk->sk_prot->disconnect(child, O_NONBLOCK); 990 991 sock_orphan(child); 992 993 this_cpu_inc(*sk->sk_prot->orphan_count); 994 995 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { 996 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req); 997 BUG_ON(sk != req->rsk_listener); 998 999 /* Paranoid, to prevent race condition if 1000 * an inbound pkt destined for child is 1001 * blocked by sock lock in tcp_v4_rcv(). 1002 * Also to satisfy an assertion in 1003 * tcp_v4_destroy_sock(). 1004 */ 1005 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL); 1006 } 1007 inet_csk_destroy_sock(child); 1008} 1009 1010struct sock *inet_csk_reqsk_queue_add(struct sock *sk, 1011 struct request_sock *req, 1012 struct sock *child) 1013{ 1014 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; 1015 1016 spin_lock(&queue->rskq_lock); 1017 if (unlikely(sk->sk_state != TCP_LISTEN)) { 1018 inet_child_forget(sk, req, child); 1019 child = NULL; 1020 } else { 1021 req->sk = child; 1022 req->dl_next = NULL; 1023 if (queue->rskq_accept_head == NULL) 1024 WRITE_ONCE(queue->rskq_accept_head, req); 1025 else 1026 queue->rskq_accept_tail->dl_next = req; 1027 queue->rskq_accept_tail = req; 1028 sk_acceptq_added(sk); 1029 } 1030 spin_unlock(&queue->rskq_lock); 1031 return child; 1032} 1033EXPORT_SYMBOL(inet_csk_reqsk_queue_add); 1034 1035struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child, 1036 struct request_sock *req, bool own_req) 1037{ 1038 if (own_req) { 1039 inet_csk_reqsk_queue_drop(sk, req); 1040 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); 1041 if (inet_csk_reqsk_queue_add(sk, req, child)) 1042 return child; 1043 } 1044 /* Too bad, another child took ownership of the request, undo. */ 1045 bh_unlock_sock(child); 1046 sock_put(child); 1047 return NULL; 1048} 1049EXPORT_SYMBOL(inet_csk_complete_hashdance); 1050 1051/* 1052 * This routine closes sockets which have been at least partially 1053 * opened, but not yet accepted. 1054 */ 1055void inet_csk_listen_stop(struct sock *sk) 1056{ 1057 struct inet_connection_sock *icsk = inet_csk(sk); 1058 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 1059 struct request_sock *next, *req; 1060 1061 /* Following specs, it would be better either to send FIN 1062 * (and enter FIN-WAIT-1, it is normal close) 1063 * or to send active reset (abort). 1064 * Certainly, it is pretty dangerous while synflood, but it is 1065 * bad justification for our negligence 8) 1066 * To be honest, we are not able to make either 1067 * of the variants now. --ANK 1068 */ 1069 while ((req = reqsk_queue_remove(queue, sk)) != NULL) { 1070 struct sock *child = req->sk; 1071 1072 local_bh_disable(); 1073 bh_lock_sock(child); 1074 WARN_ON(sock_owned_by_user(child)); 1075 sock_hold(child); 1076 1077 inet_child_forget(sk, req, child); 1078 reqsk_put(req); 1079 bh_unlock_sock(child); 1080 local_bh_enable(); 1081 sock_put(child); 1082 1083 cond_resched(); 1084 } 1085 if (queue->fastopenq.rskq_rst_head) { 1086 /* Free all the reqs queued in rskq_rst_head. */ 1087 spin_lock_bh(&queue->fastopenq.lock); 1088 req = queue->fastopenq.rskq_rst_head; 1089 queue->fastopenq.rskq_rst_head = NULL; 1090 spin_unlock_bh(&queue->fastopenq.lock); 1091 while (req != NULL) { 1092 next = req->dl_next; 1093 reqsk_put(req); 1094 req = next; 1095 } 1096 } 1097 WARN_ON_ONCE(sk->sk_ack_backlog); 1098} 1099EXPORT_SYMBOL_GPL(inet_csk_listen_stop); 1100 1101void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) 1102{ 1103 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 1104 const struct inet_sock *inet = inet_sk(sk); 1105 1106 sin->sin_family = AF_INET; 1107 sin->sin_addr.s_addr = inet->inet_daddr; 1108 sin->sin_port = inet->inet_dport; 1109} 1110EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); 1111 1112static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) 1113{ 1114 const struct inet_sock *inet = inet_sk(sk); 1115 const struct ip_options_rcu *inet_opt; 1116 __be32 daddr = inet->inet_daddr; 1117 struct flowi4 *fl4; 1118 struct rtable *rt; 1119 1120 rcu_read_lock(); 1121 inet_opt = rcu_dereference(inet->inet_opt); 1122 if (inet_opt && inet_opt->opt.srr) 1123 daddr = inet_opt->opt.faddr; 1124 fl4 = &fl->u.ip4; 1125 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, 1126 inet->inet_saddr, inet->inet_dport, 1127 inet->inet_sport, sk->sk_protocol, 1128 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); 1129 if (IS_ERR(rt)) 1130 rt = NULL; 1131 if (rt) 1132 sk_setup_caps(sk, &rt->dst); 1133 rcu_read_unlock(); 1134 1135 return &rt->dst; 1136} 1137 1138struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) 1139{ 1140 struct dst_entry *dst = __sk_dst_check(sk, 0); 1141 struct inet_sock *inet = inet_sk(sk); 1142 1143 if (!dst) { 1144 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 1145 if (!dst) 1146 goto out; 1147 } 1148 dst->ops->update_pmtu(dst, sk, NULL, mtu, true); 1149 1150 dst = __sk_dst_check(sk, 0); 1151 if (!dst) 1152 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 1153out: 1154 return dst; 1155} 1156EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); 1157