inet_connection_sock.c revision 5abf7f7e0f6bdbfcac737f636497d7016d9507eb
1/* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Support for INET connection oriented protocols. 7 * 8 * Authors: See the TCP sources 9 * 10 * This program is free software; you can redistribute it and/or 11 * modify it under the terms of the GNU General Public License 12 * as published by the Free Software Foundation; either version 13 * 2 of the License, or(at your option) any later version. 14 */ 15 16#include <linux/module.h> 17#include <linux/jhash.h> 18 19#include <net/inet_connection_sock.h> 20#include <net/inet_hashtables.h> 21#include <net/inet_timewait_sock.h> 22#include <net/ip.h> 23#include <net/route.h> 24#include <net/tcp_states.h> 25#include <net/xfrm.h> 26 27#ifdef INET_CSK_DEBUG 28const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n"; 29EXPORT_SYMBOL(inet_csk_timer_bug_msg); 30#endif 31 32/* 33 * This struct holds the first and last local port number. 34 */ 35struct local_ports sysctl_local_ports __read_mostly = { 36 .lock = __SEQLOCK_UNLOCKED(sysctl_local_ports.lock), 37 .range = { 32768, 61000 }, 38}; 39 40unsigned long *sysctl_local_reserved_ports; 41EXPORT_SYMBOL(sysctl_local_reserved_ports); 42 43void inet_get_local_port_range(int *low, int *high) 44{ 45 unsigned int seq; 46 47 do { 48 seq = read_seqbegin(&sysctl_local_ports.lock); 49 50 *low = sysctl_local_ports.range[0]; 51 *high = sysctl_local_ports.range[1]; 52 } while (read_seqretry(&sysctl_local_ports.lock, seq)); 53} 54EXPORT_SYMBOL(inet_get_local_port_range); 55 56int inet_csk_bind_conflict(const struct sock *sk, 57 const struct inet_bind_bucket *tb, bool relax) 58{ 59 struct sock *sk2; 60 struct hlist_node *node; 61 int reuse = sk->sk_reuse; 62 63 /* 64 * Unlike other sk lookup places we do not check 65 * for sk_net here, since _all_ the socks listed 66 * in tb->owners list belong to the same net - the 67 * one this bucket belongs to. 68 */ 69 70 sk_for_each_bound(sk2, node, &tb->owners) { 71 if (sk != sk2 && 72 !inet_v6_ipv6only(sk2) && 73 (!sk->sk_bound_dev_if || 74 !sk2->sk_bound_dev_if || 75 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) { 76 if (!reuse || !sk2->sk_reuse || 77 sk2->sk_state == TCP_LISTEN) { 78 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2); 79 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) || 80 sk2_rcv_saddr == sk_rcv_saddr(sk)) 81 break; 82 } 83 if (!relax && reuse && sk2->sk_reuse && 84 sk2->sk_state != TCP_LISTEN) { 85 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2); 86 87 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) || 88 sk2_rcv_saddr == sk_rcv_saddr(sk)) 89 break; 90 } 91 } 92 } 93 return node != NULL; 94} 95EXPORT_SYMBOL_GPL(inet_csk_bind_conflict); 96 97/* Obtain a reference to a local port for the given sock, 98 * if snum is zero it means select any available local port. 99 */ 100int inet_csk_get_port(struct sock *sk, unsigned short snum) 101{ 102 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo; 103 struct inet_bind_hashbucket *head; 104 struct hlist_node *node; 105 struct inet_bind_bucket *tb; 106 int ret, attempts = 5; 107 struct net *net = sock_net(sk); 108 int smallest_size = -1, smallest_rover; 109 110 local_bh_disable(); 111 if (!snum) { 112 int remaining, rover, low, high; 113 114again: 115 inet_get_local_port_range(&low, &high); 116 remaining = (high - low) + 1; 117 smallest_rover = rover = net_random() % remaining + low; 118 119 smallest_size = -1; 120 do { 121 if (inet_is_reserved_local_port(rover)) 122 goto next_nolock; 123 head = &hashinfo->bhash[inet_bhashfn(net, rover, 124 hashinfo->bhash_size)]; 125 spin_lock(&head->lock); 126 inet_bind_bucket_for_each(tb, node, &head->chain) 127 if (net_eq(ib_net(tb), net) && tb->port == rover) { 128 if (tb->fastreuse > 0 && 129 sk->sk_reuse && 130 sk->sk_state != TCP_LISTEN && 131 (tb->num_owners < smallest_size || smallest_size == -1)) { 132 smallest_size = tb->num_owners; 133 smallest_rover = rover; 134 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 && 135 !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) { 136 snum = smallest_rover; 137 goto tb_found; 138 } 139 } 140 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) { 141 snum = rover; 142 goto tb_found; 143 } 144 goto next; 145 } 146 break; 147 next: 148 spin_unlock(&head->lock); 149 next_nolock: 150 if (++rover > high) 151 rover = low; 152 } while (--remaining > 0); 153 154 /* Exhausted local port range during search? It is not 155 * possible for us to be holding one of the bind hash 156 * locks if this test triggers, because if 'remaining' 157 * drops to zero, we broke out of the do/while loop at 158 * the top level, not from the 'break;' statement. 159 */ 160 ret = 1; 161 if (remaining <= 0) { 162 if (smallest_size != -1) { 163 snum = smallest_rover; 164 goto have_snum; 165 } 166 goto fail; 167 } 168 /* OK, here is the one we will use. HEAD is 169 * non-NULL and we hold it's mutex. 170 */ 171 snum = rover; 172 } else { 173have_snum: 174 head = &hashinfo->bhash[inet_bhashfn(net, snum, 175 hashinfo->bhash_size)]; 176 spin_lock(&head->lock); 177 inet_bind_bucket_for_each(tb, node, &head->chain) 178 if (net_eq(ib_net(tb), net) && tb->port == snum) 179 goto tb_found; 180 } 181 tb = NULL; 182 goto tb_not_found; 183tb_found: 184 if (!hlist_empty(&tb->owners)) { 185 if (sk->sk_reuse == SK_FORCE_REUSE) 186 goto success; 187 188 if (tb->fastreuse > 0 && 189 sk->sk_reuse && sk->sk_state != TCP_LISTEN && 190 smallest_size == -1) { 191 goto success; 192 } else { 193 ret = 1; 194 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) { 195 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN && 196 smallest_size != -1 && --attempts >= 0) { 197 spin_unlock(&head->lock); 198 goto again; 199 } 200 201 goto fail_unlock; 202 } 203 } 204 } 205tb_not_found: 206 ret = 1; 207 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, 208 net, head, snum)) == NULL) 209 goto fail_unlock; 210 if (hlist_empty(&tb->owners)) { 211 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN) 212 tb->fastreuse = 1; 213 else 214 tb->fastreuse = 0; 215 } else if (tb->fastreuse && 216 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN)) 217 tb->fastreuse = 0; 218success: 219 if (!inet_csk(sk)->icsk_bind_hash) 220 inet_bind_hash(sk, tb, snum); 221 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); 222 ret = 0; 223 224fail_unlock: 225 spin_unlock(&head->lock); 226fail: 227 local_bh_enable(); 228 return ret; 229} 230EXPORT_SYMBOL_GPL(inet_csk_get_port); 231 232/* 233 * Wait for an incoming connection, avoid race conditions. This must be called 234 * with the socket locked. 235 */ 236static int inet_csk_wait_for_connect(struct sock *sk, long timeo) 237{ 238 struct inet_connection_sock *icsk = inet_csk(sk); 239 DEFINE_WAIT(wait); 240 int err; 241 242 /* 243 * True wake-one mechanism for incoming connections: only 244 * one process gets woken up, not the 'whole herd'. 245 * Since we do not 'race & poll' for established sockets 246 * anymore, the common case will execute the loop only once. 247 * 248 * Subtle issue: "add_wait_queue_exclusive()" will be added 249 * after any current non-exclusive waiters, and we know that 250 * it will always _stay_ after any new non-exclusive waiters 251 * because all non-exclusive waiters are added at the 252 * beginning of the wait-queue. As such, it's ok to "drop" 253 * our exclusiveness temporarily when we get woken up without 254 * having to remove and re-insert us on the wait queue. 255 */ 256 for (;;) { 257 prepare_to_wait_exclusive(sk_sleep(sk), &wait, 258 TASK_INTERRUPTIBLE); 259 release_sock(sk); 260 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) 261 timeo = schedule_timeout(timeo); 262 lock_sock(sk); 263 err = 0; 264 if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) 265 break; 266 err = -EINVAL; 267 if (sk->sk_state != TCP_LISTEN) 268 break; 269 err = sock_intr_errno(timeo); 270 if (signal_pending(current)) 271 break; 272 err = -EAGAIN; 273 if (!timeo) 274 break; 275 } 276 finish_wait(sk_sleep(sk), &wait); 277 return err; 278} 279 280/* 281 * This will accept the next outstanding connection. 282 */ 283struct sock *inet_csk_accept(struct sock *sk, int flags, int *err) 284{ 285 struct inet_connection_sock *icsk = inet_csk(sk); 286 struct sock *newsk; 287 int error; 288 289 lock_sock(sk); 290 291 /* We need to make sure that this socket is listening, 292 * and that it has something pending. 293 */ 294 error = -EINVAL; 295 if (sk->sk_state != TCP_LISTEN) 296 goto out_err; 297 298 /* Find already established connection */ 299 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) { 300 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 301 302 /* If this is a non blocking socket don't sleep */ 303 error = -EAGAIN; 304 if (!timeo) 305 goto out_err; 306 307 error = inet_csk_wait_for_connect(sk, timeo); 308 if (error) 309 goto out_err; 310 } 311 312 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk); 313 WARN_ON(newsk->sk_state == TCP_SYN_RECV); 314out: 315 release_sock(sk); 316 return newsk; 317out_err: 318 newsk = NULL; 319 *err = error; 320 goto out; 321} 322EXPORT_SYMBOL(inet_csk_accept); 323 324/* 325 * Using different timers for retransmit, delayed acks and probes 326 * We may wish use just one timer maintaining a list of expire jiffies 327 * to optimize. 328 */ 329void inet_csk_init_xmit_timers(struct sock *sk, 330 void (*retransmit_handler)(unsigned long), 331 void (*delack_handler)(unsigned long), 332 void (*keepalive_handler)(unsigned long)) 333{ 334 struct inet_connection_sock *icsk = inet_csk(sk); 335 336 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler, 337 (unsigned long)sk); 338 setup_timer(&icsk->icsk_delack_timer, delack_handler, 339 (unsigned long)sk); 340 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk); 341 icsk->icsk_pending = icsk->icsk_ack.pending = 0; 342} 343EXPORT_SYMBOL(inet_csk_init_xmit_timers); 344 345void inet_csk_clear_xmit_timers(struct sock *sk) 346{ 347 struct inet_connection_sock *icsk = inet_csk(sk); 348 349 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0; 350 351 sk_stop_timer(sk, &icsk->icsk_retransmit_timer); 352 sk_stop_timer(sk, &icsk->icsk_delack_timer); 353 sk_stop_timer(sk, &sk->sk_timer); 354} 355EXPORT_SYMBOL(inet_csk_clear_xmit_timers); 356 357void inet_csk_delete_keepalive_timer(struct sock *sk) 358{ 359 sk_stop_timer(sk, &sk->sk_timer); 360} 361EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); 362 363void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) 364{ 365 sk_reset_timer(sk, &sk->sk_timer, jiffies + len); 366} 367EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); 368 369struct dst_entry *inet_csk_route_req(struct sock *sk, 370 struct flowi4 *fl4, 371 const struct request_sock *req, 372 bool nocache) 373{ 374 struct rtable *rt; 375 const struct inet_request_sock *ireq = inet_rsk(req); 376 struct ip_options_rcu *opt = inet_rsk(req)->opt; 377 struct net *net = sock_net(sk); 378 int flags = inet_sk_flowi_flags(sk); 379 380 if (nocache) 381 flags |= FLOWI_FLAG_RT_NOCACHE; 382 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 383 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 384 sk->sk_protocol, 385 flags, 386 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr, 387 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport); 388 security_req_classify_flow(req, flowi4_to_flowi(fl4)); 389 rt = ip_route_output_flow(net, fl4, sk); 390 if (IS_ERR(rt)) 391 goto no_route; 392 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) 393 goto route_err; 394 return &rt->dst; 395 396route_err: 397 ip_rt_put(rt); 398no_route: 399 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 400 return NULL; 401} 402EXPORT_SYMBOL_GPL(inet_csk_route_req); 403 404struct dst_entry *inet_csk_route_child_sock(struct sock *sk, 405 struct sock *newsk, 406 const struct request_sock *req) 407{ 408 const struct inet_request_sock *ireq = inet_rsk(req); 409 struct inet_sock *newinet = inet_sk(newsk); 410 struct ip_options_rcu *opt = ireq->opt; 411 struct net *net = sock_net(sk); 412 struct flowi4 *fl4; 413 struct rtable *rt; 414 415 fl4 = &newinet->cork.fl.u.ip4; 416 flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark, 417 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, 418 sk->sk_protocol, inet_sk_flowi_flags(sk), 419 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr, 420 ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport); 421 security_req_classify_flow(req, flowi4_to_flowi(fl4)); 422 rt = ip_route_output_flow(net, fl4, sk); 423 if (IS_ERR(rt)) 424 goto no_route; 425 if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway) 426 goto route_err; 427 return &rt->dst; 428 429route_err: 430 ip_rt_put(rt); 431no_route: 432 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 433 return NULL; 434} 435EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); 436 437static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport, 438 const u32 rnd, const u32 synq_hsize) 439{ 440 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1); 441} 442 443#if IS_ENABLED(CONFIG_IPV6) 444#define AF_INET_FAMILY(fam) ((fam) == AF_INET) 445#else 446#define AF_INET_FAMILY(fam) 1 447#endif 448 449struct request_sock *inet_csk_search_req(const struct sock *sk, 450 struct request_sock ***prevp, 451 const __be16 rport, const __be32 raddr, 452 const __be32 laddr) 453{ 454 const struct inet_connection_sock *icsk = inet_csk(sk); 455 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; 456 struct request_sock *req, **prev; 457 458 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd, 459 lopt->nr_table_entries)]; 460 (req = *prev) != NULL; 461 prev = &req->dl_next) { 462 const struct inet_request_sock *ireq = inet_rsk(req); 463 464 if (ireq->rmt_port == rport && 465 ireq->rmt_addr == raddr && 466 ireq->loc_addr == laddr && 467 AF_INET_FAMILY(req->rsk_ops->family)) { 468 WARN_ON(req->sk); 469 *prevp = prev; 470 break; 471 } 472 } 473 474 return req; 475} 476EXPORT_SYMBOL_GPL(inet_csk_search_req); 477 478void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, 479 unsigned long timeout) 480{ 481 struct inet_connection_sock *icsk = inet_csk(sk); 482 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt; 483 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port, 484 lopt->hash_rnd, lopt->nr_table_entries); 485 486 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout); 487 inet_csk_reqsk_queue_added(sk, timeout); 488} 489EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); 490 491/* Only thing we need from tcp.h */ 492extern int sysctl_tcp_synack_retries; 493 494 495/* Decide when to expire the request and when to resend SYN-ACK */ 496static inline void syn_ack_recalc(struct request_sock *req, const int thresh, 497 const int max_retries, 498 const u8 rskq_defer_accept, 499 int *expire, int *resend) 500{ 501 if (!rskq_defer_accept) { 502 *expire = req->retrans >= thresh; 503 *resend = 1; 504 return; 505 } 506 *expire = req->retrans >= thresh && 507 (!inet_rsk(req)->acked || req->retrans >= max_retries); 508 /* 509 * Do not resend while waiting for data after ACK, 510 * start to resend on end of deferring period to give 511 * last chance for data or ACK to create established socket. 512 */ 513 *resend = !inet_rsk(req)->acked || 514 req->retrans >= rskq_defer_accept - 1; 515} 516 517void inet_csk_reqsk_queue_prune(struct sock *parent, 518 const unsigned long interval, 519 const unsigned long timeout, 520 const unsigned long max_rto) 521{ 522 struct inet_connection_sock *icsk = inet_csk(parent); 523 struct request_sock_queue *queue = &icsk->icsk_accept_queue; 524 struct listen_sock *lopt = queue->listen_opt; 525 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries; 526 int thresh = max_retries; 527 unsigned long now = jiffies; 528 struct request_sock **reqp, *req; 529 int i, budget; 530 531 if (lopt == NULL || lopt->qlen == 0) 532 return; 533 534 /* Normally all the openreqs are young and become mature 535 * (i.e. converted to established socket) for first timeout. 536 * If synack was not acknowledged for 1 second, it means 537 * one of the following things: synack was lost, ack was lost, 538 * rtt is high or nobody planned to ack (i.e. synflood). 539 * When server is a bit loaded, queue is populated with old 540 * open requests, reducing effective size of queue. 541 * When server is well loaded, queue size reduces to zero 542 * after several minutes of work. It is not synflood, 543 * it is normal operation. The solution is pruning 544 * too old entries overriding normal timeout, when 545 * situation becomes dangerous. 546 * 547 * Essentially, we reserve half of room for young 548 * embrions; and abort old ones without pity, if old 549 * ones are about to clog our table. 550 */ 551 if (lopt->qlen>>(lopt->max_qlen_log-1)) { 552 int young = (lopt->qlen_young<<1); 553 554 while (thresh > 2) { 555 if (lopt->qlen < young) 556 break; 557 thresh--; 558 young <<= 1; 559 } 560 } 561 562 if (queue->rskq_defer_accept) 563 max_retries = queue->rskq_defer_accept; 564 565 budget = 2 * (lopt->nr_table_entries / (timeout / interval)); 566 i = lopt->clock_hand; 567 568 do { 569 reqp=&lopt->syn_table[i]; 570 while ((req = *reqp) != NULL) { 571 if (time_after_eq(now, req->expires)) { 572 int expire = 0, resend = 0; 573 574 syn_ack_recalc(req, thresh, max_retries, 575 queue->rskq_defer_accept, 576 &expire, &resend); 577 req->rsk_ops->syn_ack_timeout(parent, req); 578 if (!expire && 579 (!resend || 580 !req->rsk_ops->rtx_syn_ack(parent, req, NULL) || 581 inet_rsk(req)->acked)) { 582 unsigned long timeo; 583 584 if (req->retrans++ == 0) 585 lopt->qlen_young--; 586 timeo = min((timeout << req->retrans), max_rto); 587 req->expires = now + timeo; 588 reqp = &req->dl_next; 589 continue; 590 } 591 592 /* Drop this request */ 593 inet_csk_reqsk_queue_unlink(parent, req, reqp); 594 reqsk_queue_removed(queue, req); 595 reqsk_free(req); 596 continue; 597 } 598 reqp = &req->dl_next; 599 } 600 601 i = (i + 1) & (lopt->nr_table_entries - 1); 602 603 } while (--budget > 0); 604 605 lopt->clock_hand = i; 606 607 if (lopt->qlen) 608 inet_csk_reset_keepalive_timer(parent, interval); 609} 610EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune); 611 612/** 613 * inet_csk_clone_lock - clone an inet socket, and lock its clone 614 * @sk: the socket to clone 615 * @req: request_sock 616 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 617 * 618 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 619 */ 620struct sock *inet_csk_clone_lock(const struct sock *sk, 621 const struct request_sock *req, 622 const gfp_t priority) 623{ 624 struct sock *newsk = sk_clone_lock(sk, priority); 625 626 if (newsk != NULL) { 627 struct inet_connection_sock *newicsk = inet_csk(newsk); 628 629 newsk->sk_state = TCP_SYN_RECV; 630 newicsk->icsk_bind_hash = NULL; 631 632 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port; 633 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port); 634 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port; 635 newsk->sk_write_space = sk_stream_write_space; 636 637 newicsk->icsk_retransmits = 0; 638 newicsk->icsk_backoff = 0; 639 newicsk->icsk_probes_out = 0; 640 641 /* Deinitialize accept_queue to trap illegal accesses. */ 642 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); 643 644 security_inet_csk_clone(newsk, req); 645 } 646 return newsk; 647} 648EXPORT_SYMBOL_GPL(inet_csk_clone_lock); 649 650/* 651 * At this point, there should be no process reference to this 652 * socket, and thus no user references at all. Therefore we 653 * can assume the socket waitqueue is inactive and nobody will 654 * try to jump onto it. 655 */ 656void inet_csk_destroy_sock(struct sock *sk) 657{ 658 WARN_ON(sk->sk_state != TCP_CLOSE); 659 WARN_ON(!sock_flag(sk, SOCK_DEAD)); 660 661 /* It cannot be in hash table! */ 662 WARN_ON(!sk_unhashed(sk)); 663 664 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ 665 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); 666 667 sk->sk_prot->destroy(sk); 668 669 sk_stream_kill_queues(sk); 670 671 xfrm_sk_free_policy(sk); 672 673 sk_refcnt_debug_release(sk); 674 675 percpu_counter_dec(sk->sk_prot->orphan_count); 676 sock_put(sk); 677} 678EXPORT_SYMBOL(inet_csk_destroy_sock); 679 680int inet_csk_listen_start(struct sock *sk, const int nr_table_entries) 681{ 682 struct inet_sock *inet = inet_sk(sk); 683 struct inet_connection_sock *icsk = inet_csk(sk); 684 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries); 685 686 if (rc != 0) 687 return rc; 688 689 sk->sk_max_ack_backlog = 0; 690 sk->sk_ack_backlog = 0; 691 inet_csk_delack_init(sk); 692 693 /* There is race window here: we announce ourselves listening, 694 * but this transition is still not validated by get_port(). 695 * It is OK, because this socket enters to hash table only 696 * after validation is complete. 697 */ 698 sk->sk_state = TCP_LISTEN; 699 if (!sk->sk_prot->get_port(sk, inet->inet_num)) { 700 inet->inet_sport = htons(inet->inet_num); 701 702 sk_dst_reset(sk); 703 sk->sk_prot->hash(sk); 704 705 return 0; 706 } 707 708 sk->sk_state = TCP_CLOSE; 709 __reqsk_queue_destroy(&icsk->icsk_accept_queue); 710 return -EADDRINUSE; 711} 712EXPORT_SYMBOL_GPL(inet_csk_listen_start); 713 714/* 715 * This routine closes sockets which have been at least partially 716 * opened, but not yet accepted. 717 */ 718void inet_csk_listen_stop(struct sock *sk) 719{ 720 struct inet_connection_sock *icsk = inet_csk(sk); 721 struct request_sock *acc_req; 722 struct request_sock *req; 723 724 inet_csk_delete_keepalive_timer(sk); 725 726 /* make all the listen_opt local to us */ 727 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue); 728 729 /* Following specs, it would be better either to send FIN 730 * (and enter FIN-WAIT-1, it is normal close) 731 * or to send active reset (abort). 732 * Certainly, it is pretty dangerous while synflood, but it is 733 * bad justification for our negligence 8) 734 * To be honest, we are not able to make either 735 * of the variants now. --ANK 736 */ 737 reqsk_queue_destroy(&icsk->icsk_accept_queue); 738 739 while ((req = acc_req) != NULL) { 740 struct sock *child = req->sk; 741 742 acc_req = req->dl_next; 743 744 local_bh_disable(); 745 bh_lock_sock(child); 746 WARN_ON(sock_owned_by_user(child)); 747 sock_hold(child); 748 749 sk->sk_prot->disconnect(child, O_NONBLOCK); 750 751 sock_orphan(child); 752 753 percpu_counter_inc(sk->sk_prot->orphan_count); 754 755 inet_csk_destroy_sock(child); 756 757 bh_unlock_sock(child); 758 local_bh_enable(); 759 sock_put(child); 760 761 sk_acceptq_removed(sk); 762 __reqsk_free(req); 763 } 764 WARN_ON(sk->sk_ack_backlog); 765} 766EXPORT_SYMBOL_GPL(inet_csk_listen_stop); 767 768void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) 769{ 770 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; 771 const struct inet_sock *inet = inet_sk(sk); 772 773 sin->sin_family = AF_INET; 774 sin->sin_addr.s_addr = inet->inet_daddr; 775 sin->sin_port = inet->inet_dport; 776} 777EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); 778 779#ifdef CONFIG_COMPAT 780int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname, 781 char __user *optval, int __user *optlen) 782{ 783 const struct inet_connection_sock *icsk = inet_csk(sk); 784 785 if (icsk->icsk_af_ops->compat_getsockopt != NULL) 786 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname, 787 optval, optlen); 788 return icsk->icsk_af_ops->getsockopt(sk, level, optname, 789 optval, optlen); 790} 791EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt); 792 793int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname, 794 char __user *optval, unsigned int optlen) 795{ 796 const struct inet_connection_sock *icsk = inet_csk(sk); 797 798 if (icsk->icsk_af_ops->compat_setsockopt != NULL) 799 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname, 800 optval, optlen); 801 return icsk->icsk_af_ops->setsockopt(sk, level, optname, 802 optval, optlen); 803} 804EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt); 805#endif 806 807static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) 808{ 809 const struct inet_sock *inet = inet_sk(sk); 810 const struct ip_options_rcu *inet_opt; 811 __be32 daddr = inet->inet_daddr; 812 struct flowi4 *fl4; 813 struct rtable *rt; 814 815 rcu_read_lock(); 816 inet_opt = rcu_dereference(inet->inet_opt); 817 if (inet_opt && inet_opt->opt.srr) 818 daddr = inet_opt->opt.faddr; 819 fl4 = &fl->u.ip4; 820 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, 821 inet->inet_saddr, inet->inet_dport, 822 inet->inet_sport, sk->sk_protocol, 823 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); 824 if (IS_ERR(rt)) 825 rt = NULL; 826 if (rt) 827 sk_setup_caps(sk, &rt->dst); 828 rcu_read_unlock(); 829 830 return &rt->dst; 831} 832 833struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) 834{ 835 struct dst_entry *dst = __sk_dst_check(sk, 0); 836 struct inet_sock *inet = inet_sk(sk); 837 838 if (!dst) { 839 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 840 if (!dst) 841 goto out; 842 } 843 dst->ops->update_pmtu(dst, sk, NULL, mtu); 844 845 dst = __sk_dst_check(sk, 0); 846 if (!dst) 847 dst = inet_csk_rebuild_route(sk, &inet->cork.fl); 848out: 849 return dst; 850} 851EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); 852