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.
6 * Support for INET connection oriented protocols.
8 * Authors: See the TCP sources
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.
16 #include <linux/module.h>
17 #include <linux/jhash.h>
19 #include <net/inet_connection_sock.h>
20 #include <net/inet_hashtables.h>
21 #include <net/inet_timewait_sock.h>
23 #include <net/route.h>
24 #include <net/tcp_states.h>
28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
33 * This array holds the first and last local port number.
35 int sysctl_local_port_range[2] = { 32768, 61000 };
36 DEFINE_SEQLOCK(sysctl_port_range_lock);
38 void inet_get_local_port_range(int *low, int *high)
42 seq = read_seqbegin(&sysctl_port_range_lock);
44 *low = sysctl_local_port_range[0];
45 *high = sysctl_local_port_range[1];
46 } while (read_seqretry(&sysctl_port_range_lock, seq));
48 EXPORT_SYMBOL(inet_get_local_port_range);
50 int inet_csk_bind_conflict(const struct sock *sk,
51 const struct inet_bind_bucket *tb)
53 const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
55 struct hlist_node *node;
56 int reuse = sk->sk_reuse;
59 * Unlike other sk lookup places we do not check
60 * for sk_net here, since _all_ the socks listed
61 * in tb->owners list belong to the same net - the
62 * one this bucket belongs to.
65 sk_for_each_bound(sk2, node, &tb->owners) {
67 !inet_v6_ipv6only(sk2) &&
68 (!sk->sk_bound_dev_if ||
69 !sk2->sk_bound_dev_if ||
70 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
71 if (!reuse || !sk2->sk_reuse ||
72 sk2->sk_state == TCP_LISTEN) {
73 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
74 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
75 sk2_rcv_saddr == sk_rcv_saddr)
83 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
85 /* Obtain a reference to a local port for the given sock,
86 * if snum is zero it means select any available local port.
88 int inet_csk_get_port(struct sock *sk, unsigned short snum)
90 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
91 struct inet_bind_hashbucket *head;
92 struct hlist_node *node;
93 struct inet_bind_bucket *tb;
95 struct net *net = sock_net(sk);
99 int remaining, rover, low, high;
101 inet_get_local_port_range(&low, &high);
102 remaining = (high - low) + 1;
103 rover = net_random() % remaining + low;
106 head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)];
107 spin_lock(&head->lock);
108 inet_bind_bucket_for_each(tb, node, &head->chain)
109 if (tb->ib_net == net && tb->port == rover)
113 spin_unlock(&head->lock);
116 } while (--remaining > 0);
118 /* Exhausted local port range during search? It is not
119 * possible for us to be holding one of the bind hash
120 * locks if this test triggers, because if 'remaining'
121 * drops to zero, we broke out of the do/while loop at
122 * the top level, not from the 'break;' statement.
128 /* OK, here is the one we will use. HEAD is
129 * non-NULL and we hold it's mutex.
133 head = &hashinfo->bhash[inet_bhashfn(snum, hashinfo->bhash_size)];
134 spin_lock(&head->lock);
135 inet_bind_bucket_for_each(tb, node, &head->chain)
136 if (tb->ib_net == net && tb->port == snum)
142 if (!hlist_empty(&tb->owners)) {
143 if (tb->fastreuse > 0 &&
144 sk->sk_reuse && sk->sk_state != TCP_LISTEN) {
148 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb))
154 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
155 net, head, snum)) == NULL)
157 if (hlist_empty(&tb->owners)) {
158 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
162 } else if (tb->fastreuse &&
163 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
166 if (!inet_csk(sk)->icsk_bind_hash)
167 inet_bind_hash(sk, tb, snum);
168 BUG_TRAP(inet_csk(sk)->icsk_bind_hash == tb);
172 spin_unlock(&head->lock);
178 EXPORT_SYMBOL_GPL(inet_csk_get_port);
181 * Wait for an incoming connection, avoid race conditions. This must be called
182 * with the socket locked.
184 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
186 struct inet_connection_sock *icsk = inet_csk(sk);
191 * True wake-one mechanism for incoming connections: only
192 * one process gets woken up, not the 'whole herd'.
193 * Since we do not 'race & poll' for established sockets
194 * anymore, the common case will execute the loop only once.
196 * Subtle issue: "add_wait_queue_exclusive()" will be added
197 * after any current non-exclusive waiters, and we know that
198 * it will always _stay_ after any new non-exclusive waiters
199 * because all non-exclusive waiters are added at the
200 * beginning of the wait-queue. As such, it's ok to "drop"
201 * our exclusiveness temporarily when we get woken up without
202 * having to remove and re-insert us on the wait queue.
205 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
208 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
209 timeo = schedule_timeout(timeo);
212 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
215 if (sk->sk_state != TCP_LISTEN)
217 err = sock_intr_errno(timeo);
218 if (signal_pending(current))
224 finish_wait(sk->sk_sleep, &wait);
229 * This will accept the next outstanding connection.
231 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
233 struct inet_connection_sock *icsk = inet_csk(sk);
239 /* We need to make sure that this socket is listening,
240 * and that it has something pending.
243 if (sk->sk_state != TCP_LISTEN)
246 /* Find already established connection */
247 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
248 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
250 /* If this is a non blocking socket don't sleep */
255 error = inet_csk_wait_for_connect(sk, timeo);
260 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
261 BUG_TRAP(newsk->sk_state != TCP_SYN_RECV);
271 EXPORT_SYMBOL(inet_csk_accept);
274 * Using different timers for retransmit, delayed acks and probes
275 * We may wish use just one timer maintaining a list of expire jiffies
278 void inet_csk_init_xmit_timers(struct sock *sk,
279 void (*retransmit_handler)(unsigned long),
280 void (*delack_handler)(unsigned long),
281 void (*keepalive_handler)(unsigned long))
283 struct inet_connection_sock *icsk = inet_csk(sk);
285 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
287 setup_timer(&icsk->icsk_delack_timer, delack_handler,
289 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
290 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
293 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
295 void inet_csk_clear_xmit_timers(struct sock *sk)
297 struct inet_connection_sock *icsk = inet_csk(sk);
299 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
301 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
302 sk_stop_timer(sk, &icsk->icsk_delack_timer);
303 sk_stop_timer(sk, &sk->sk_timer);
306 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
308 void inet_csk_delete_keepalive_timer(struct sock *sk)
310 sk_stop_timer(sk, &sk->sk_timer);
313 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
315 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
317 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
320 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
322 struct dst_entry* inet_csk_route_req(struct sock *sk,
323 const struct request_sock *req)
326 const struct inet_request_sock *ireq = inet_rsk(req);
327 struct ip_options *opt = inet_rsk(req)->opt;
328 struct flowi fl = { .oif = sk->sk_bound_dev_if,
330 { .daddr = ((opt && opt->srr) ?
333 .saddr = ireq->loc_addr,
334 .tos = RT_CONN_FLAGS(sk) } },
335 .proto = sk->sk_protocol,
337 { .sport = inet_sk(sk)->sport,
338 .dport = ireq->rmt_port } } };
340 security_req_classify_flow(req, &fl);
341 if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0)) {
342 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
345 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) {
347 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
353 EXPORT_SYMBOL_GPL(inet_csk_route_req);
355 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
356 const u32 rnd, const u32 synq_hsize)
358 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
361 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
362 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
364 #define AF_INET_FAMILY(fam) 1
367 struct request_sock *inet_csk_search_req(const struct sock *sk,
368 struct request_sock ***prevp,
369 const __be16 rport, const __be32 raddr,
372 const struct inet_connection_sock *icsk = inet_csk(sk);
373 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
374 struct request_sock *req, **prev;
376 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
377 lopt->nr_table_entries)];
378 (req = *prev) != NULL;
379 prev = &req->dl_next) {
380 const struct inet_request_sock *ireq = inet_rsk(req);
382 if (ireq->rmt_port == rport &&
383 ireq->rmt_addr == raddr &&
384 ireq->loc_addr == laddr &&
385 AF_INET_FAMILY(req->rsk_ops->family)) {
395 EXPORT_SYMBOL_GPL(inet_csk_search_req);
397 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
398 unsigned long timeout)
400 struct inet_connection_sock *icsk = inet_csk(sk);
401 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
402 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
403 lopt->hash_rnd, lopt->nr_table_entries);
405 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
406 inet_csk_reqsk_queue_added(sk, timeout);
409 /* Only thing we need from tcp.h */
410 extern int sysctl_tcp_synack_retries;
412 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
414 void inet_csk_reqsk_queue_prune(struct sock *parent,
415 const unsigned long interval,
416 const unsigned long timeout,
417 const unsigned long max_rto)
419 struct inet_connection_sock *icsk = inet_csk(parent);
420 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
421 struct listen_sock *lopt = queue->listen_opt;
422 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
423 int thresh = max_retries;
424 unsigned long now = jiffies;
425 struct request_sock **reqp, *req;
428 if (lopt == NULL || lopt->qlen == 0)
431 /* Normally all the openreqs are young and become mature
432 * (i.e. converted to established socket) for first timeout.
433 * If synack was not acknowledged for 3 seconds, it means
434 * one of the following things: synack was lost, ack was lost,
435 * rtt is high or nobody planned to ack (i.e. synflood).
436 * When server is a bit loaded, queue is populated with old
437 * open requests, reducing effective size of queue.
438 * When server is well loaded, queue size reduces to zero
439 * after several minutes of work. It is not synflood,
440 * it is normal operation. The solution is pruning
441 * too old entries overriding normal timeout, when
442 * situation becomes dangerous.
444 * Essentially, we reserve half of room for young
445 * embrions; and abort old ones without pity, if old
446 * ones are about to clog our table.
448 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
449 int young = (lopt->qlen_young<<1);
452 if (lopt->qlen < young)
459 if (queue->rskq_defer_accept)
460 max_retries = queue->rskq_defer_accept;
462 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
463 i = lopt->clock_hand;
466 reqp=&lopt->syn_table[i];
467 while ((req = *reqp) != NULL) {
468 if (time_after_eq(now, req->expires)) {
469 if ((req->retrans < thresh ||
470 (inet_rsk(req)->acked && req->retrans < max_retries))
471 && !req->rsk_ops->rtx_syn_ack(parent, req)) {
474 if (req->retrans++ == 0)
476 timeo = min((timeout << req->retrans), max_rto);
477 req->expires = now + timeo;
478 reqp = &req->dl_next;
482 /* Drop this request */
483 inet_csk_reqsk_queue_unlink(parent, req, reqp);
484 reqsk_queue_removed(queue, req);
488 reqp = &req->dl_next;
491 i = (i + 1) & (lopt->nr_table_entries - 1);
493 } while (--budget > 0);
495 lopt->clock_hand = i;
498 inet_csk_reset_keepalive_timer(parent, interval);
501 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
503 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
504 const gfp_t priority)
506 struct sock *newsk = sk_clone(sk, priority);
509 struct inet_connection_sock *newicsk = inet_csk(newsk);
511 newsk->sk_state = TCP_SYN_RECV;
512 newicsk->icsk_bind_hash = NULL;
514 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
515 newsk->sk_write_space = sk_stream_write_space;
517 newicsk->icsk_retransmits = 0;
518 newicsk->icsk_backoff = 0;
519 newicsk->icsk_probes_out = 0;
521 /* Deinitialize accept_queue to trap illegal accesses. */
522 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
524 security_inet_csk_clone(newsk, req);
529 EXPORT_SYMBOL_GPL(inet_csk_clone);
532 * At this point, there should be no process reference to this
533 * socket, and thus no user references at all. Therefore we
534 * can assume the socket waitqueue is inactive and nobody will
535 * try to jump onto it.
537 void inet_csk_destroy_sock(struct sock *sk)
539 BUG_TRAP(sk->sk_state == TCP_CLOSE);
540 BUG_TRAP(sock_flag(sk, SOCK_DEAD));
542 /* It cannot be in hash table! */
543 BUG_TRAP(sk_unhashed(sk));
545 /* If it has not 0 inet_sk(sk)->num, it must be bound */
546 BUG_TRAP(!inet_sk(sk)->num || inet_csk(sk)->icsk_bind_hash);
548 sk->sk_prot->destroy(sk);
550 sk_stream_kill_queues(sk);
552 xfrm_sk_free_policy(sk);
554 sk_refcnt_debug_release(sk);
556 atomic_dec(sk->sk_prot->orphan_count);
560 EXPORT_SYMBOL(inet_csk_destroy_sock);
562 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
564 struct inet_sock *inet = inet_sk(sk);
565 struct inet_connection_sock *icsk = inet_csk(sk);
566 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
571 sk->sk_max_ack_backlog = 0;
572 sk->sk_ack_backlog = 0;
573 inet_csk_delack_init(sk);
575 /* There is race window here: we announce ourselves listening,
576 * but this transition is still not validated by get_port().
577 * It is OK, because this socket enters to hash table only
578 * after validation is complete.
580 sk->sk_state = TCP_LISTEN;
581 if (!sk->sk_prot->get_port(sk, inet->num)) {
582 inet->sport = htons(inet->num);
585 sk->sk_prot->hash(sk);
590 sk->sk_state = TCP_CLOSE;
591 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
595 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
598 * This routine closes sockets which have been at least partially
599 * opened, but not yet accepted.
601 void inet_csk_listen_stop(struct sock *sk)
603 struct inet_connection_sock *icsk = inet_csk(sk);
604 struct request_sock *acc_req;
605 struct request_sock *req;
607 inet_csk_delete_keepalive_timer(sk);
609 /* make all the listen_opt local to us */
610 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
612 /* Following specs, it would be better either to send FIN
613 * (and enter FIN-WAIT-1, it is normal close)
614 * or to send active reset (abort).
615 * Certainly, it is pretty dangerous while synflood, but it is
616 * bad justification for our negligence 8)
617 * To be honest, we are not able to make either
618 * of the variants now. --ANK
620 reqsk_queue_destroy(&icsk->icsk_accept_queue);
622 while ((req = acc_req) != NULL) {
623 struct sock *child = req->sk;
625 acc_req = req->dl_next;
629 BUG_TRAP(!sock_owned_by_user(child));
632 sk->sk_prot->disconnect(child, O_NONBLOCK);
636 atomic_inc(sk->sk_prot->orphan_count);
638 inet_csk_destroy_sock(child);
640 bh_unlock_sock(child);
644 sk_acceptq_removed(sk);
647 BUG_TRAP(!sk->sk_ack_backlog);
650 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
652 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
654 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
655 const struct inet_sock *inet = inet_sk(sk);
657 sin->sin_family = AF_INET;
658 sin->sin_addr.s_addr = inet->daddr;
659 sin->sin_port = inet->dport;
662 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
665 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
666 char __user *optval, int __user *optlen)
668 const struct inet_connection_sock *icsk = inet_csk(sk);
670 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
671 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
673 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
677 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
679 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
680 char __user *optval, int optlen)
682 const struct inet_connection_sock *icsk = inet_csk(sk);
684 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
685 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
687 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
691 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);