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(net, rover,
107 hashinfo->bhash_size)];
108 spin_lock(&head->lock);
109 inet_bind_bucket_for_each(tb, node, &head->chain)
110 if (tb->ib_net == net && tb->port == rover)
114 spin_unlock(&head->lock);
117 } while (--remaining > 0);
119 /* Exhausted local port range during search? It is not
120 * possible for us to be holding one of the bind hash
121 * locks if this test triggers, because if 'remaining'
122 * drops to zero, we broke out of the do/while loop at
123 * the top level, not from the 'break;' statement.
129 /* OK, here is the one we will use. HEAD is
130 * non-NULL and we hold it's mutex.
134 head = &hashinfo->bhash[inet_bhashfn(net, snum,
135 hashinfo->bhash_size)];
136 spin_lock(&head->lock);
137 inet_bind_bucket_for_each(tb, node, &head->chain)
138 if (tb->ib_net == net && tb->port == snum)
144 if (!hlist_empty(&tb->owners)) {
145 if (tb->fastreuse > 0 &&
146 sk->sk_reuse && sk->sk_state != TCP_LISTEN) {
150 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb))
156 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
157 net, head, snum)) == NULL)
159 if (hlist_empty(&tb->owners)) {
160 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
164 } else if (tb->fastreuse &&
165 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
168 if (!inet_csk(sk)->icsk_bind_hash)
169 inet_bind_hash(sk, tb, snum);
170 BUG_TRAP(inet_csk(sk)->icsk_bind_hash == tb);
174 spin_unlock(&head->lock);
180 EXPORT_SYMBOL_GPL(inet_csk_get_port);
183 * Wait for an incoming connection, avoid race conditions. This must be called
184 * with the socket locked.
186 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
188 struct inet_connection_sock *icsk = inet_csk(sk);
193 * True wake-one mechanism for incoming connections: only
194 * one process gets woken up, not the 'whole herd'.
195 * Since we do not 'race & poll' for established sockets
196 * anymore, the common case will execute the loop only once.
198 * Subtle issue: "add_wait_queue_exclusive()" will be added
199 * after any current non-exclusive waiters, and we know that
200 * it will always _stay_ after any new non-exclusive waiters
201 * because all non-exclusive waiters are added at the
202 * beginning of the wait-queue. As such, it's ok to "drop"
203 * our exclusiveness temporarily when we get woken up without
204 * having to remove and re-insert us on the wait queue.
207 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
210 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
211 timeo = schedule_timeout(timeo);
214 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
217 if (sk->sk_state != TCP_LISTEN)
219 err = sock_intr_errno(timeo);
220 if (signal_pending(current))
226 finish_wait(sk->sk_sleep, &wait);
231 * This will accept the next outstanding connection.
233 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
235 struct inet_connection_sock *icsk = inet_csk(sk);
241 /* We need to make sure that this socket is listening,
242 * and that it has something pending.
245 if (sk->sk_state != TCP_LISTEN)
248 /* Find already established connection */
249 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
250 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
252 /* If this is a non blocking socket don't sleep */
257 error = inet_csk_wait_for_connect(sk, timeo);
262 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
263 BUG_TRAP(newsk->sk_state != TCP_SYN_RECV);
273 EXPORT_SYMBOL(inet_csk_accept);
276 * Using different timers for retransmit, delayed acks and probes
277 * We may wish use just one timer maintaining a list of expire jiffies
280 void inet_csk_init_xmit_timers(struct sock *sk,
281 void (*retransmit_handler)(unsigned long),
282 void (*delack_handler)(unsigned long),
283 void (*keepalive_handler)(unsigned long))
285 struct inet_connection_sock *icsk = inet_csk(sk);
287 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
289 setup_timer(&icsk->icsk_delack_timer, delack_handler,
291 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
292 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
295 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
297 void inet_csk_clear_xmit_timers(struct sock *sk)
299 struct inet_connection_sock *icsk = inet_csk(sk);
301 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
303 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
304 sk_stop_timer(sk, &icsk->icsk_delack_timer);
305 sk_stop_timer(sk, &sk->sk_timer);
308 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
310 void inet_csk_delete_keepalive_timer(struct sock *sk)
312 sk_stop_timer(sk, &sk->sk_timer);
315 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
317 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
319 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
322 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
324 struct dst_entry* inet_csk_route_req(struct sock *sk,
325 const struct request_sock *req)
328 const struct inet_request_sock *ireq = inet_rsk(req);
329 struct ip_options *opt = inet_rsk(req)->opt;
330 struct flowi fl = { .oif = sk->sk_bound_dev_if,
332 { .daddr = ((opt && opt->srr) ?
335 .saddr = ireq->loc_addr,
336 .tos = RT_CONN_FLAGS(sk) } },
337 .proto = sk->sk_protocol,
339 { .sport = inet_sk(sk)->sport,
340 .dport = ireq->rmt_port } } };
341 struct net *net = sock_net(sk);
343 security_req_classify_flow(req, &fl);
344 if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
345 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
348 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) {
350 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
356 EXPORT_SYMBOL_GPL(inet_csk_route_req);
358 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
359 const u32 rnd, const u32 synq_hsize)
361 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
364 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
365 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
367 #define AF_INET_FAMILY(fam) 1
370 struct request_sock *inet_csk_search_req(const struct sock *sk,
371 struct request_sock ***prevp,
372 const __be16 rport, const __be32 raddr,
375 const struct inet_connection_sock *icsk = inet_csk(sk);
376 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
377 struct request_sock *req, **prev;
379 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
380 lopt->nr_table_entries)];
381 (req = *prev) != NULL;
382 prev = &req->dl_next) {
383 const struct inet_request_sock *ireq = inet_rsk(req);
385 if (ireq->rmt_port == rport &&
386 ireq->rmt_addr == raddr &&
387 ireq->loc_addr == laddr &&
388 AF_INET_FAMILY(req->rsk_ops->family)) {
398 EXPORT_SYMBOL_GPL(inet_csk_search_req);
400 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
401 unsigned long timeout)
403 struct inet_connection_sock *icsk = inet_csk(sk);
404 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
405 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
406 lopt->hash_rnd, lopt->nr_table_entries);
408 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
409 inet_csk_reqsk_queue_added(sk, timeout);
412 /* Only thing we need from tcp.h */
413 extern int sysctl_tcp_synack_retries;
415 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
417 void inet_csk_reqsk_queue_prune(struct sock *parent,
418 const unsigned long interval,
419 const unsigned long timeout,
420 const unsigned long max_rto)
422 struct inet_connection_sock *icsk = inet_csk(parent);
423 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
424 struct listen_sock *lopt = queue->listen_opt;
425 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
426 int thresh = max_retries;
427 unsigned long now = jiffies;
428 struct request_sock **reqp, *req;
431 if (lopt == NULL || lopt->qlen == 0)
434 /* Normally all the openreqs are young and become mature
435 * (i.e. converted to established socket) for first timeout.
436 * If synack was not acknowledged for 3 seconds, it means
437 * one of the following things: synack was lost, ack was lost,
438 * rtt is high or nobody planned to ack (i.e. synflood).
439 * When server is a bit loaded, queue is populated with old
440 * open requests, reducing effective size of queue.
441 * When server is well loaded, queue size reduces to zero
442 * after several minutes of work. It is not synflood,
443 * it is normal operation. The solution is pruning
444 * too old entries overriding normal timeout, when
445 * situation becomes dangerous.
447 * Essentially, we reserve half of room for young
448 * embrions; and abort old ones without pity, if old
449 * ones are about to clog our table.
451 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
452 int young = (lopt->qlen_young<<1);
455 if (lopt->qlen < young)
462 if (queue->rskq_defer_accept)
463 max_retries = queue->rskq_defer_accept;
465 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
466 i = lopt->clock_hand;
469 reqp=&lopt->syn_table[i];
470 while ((req = *reqp) != NULL) {
471 if (time_after_eq(now, req->expires)) {
472 if ((req->retrans < thresh ||
473 (inet_rsk(req)->acked && req->retrans < max_retries))
474 && !req->rsk_ops->rtx_syn_ack(parent, req)) {
477 if (req->retrans++ == 0)
479 timeo = min((timeout << req->retrans), max_rto);
480 req->expires = now + timeo;
481 reqp = &req->dl_next;
485 /* Drop this request */
486 inet_csk_reqsk_queue_unlink(parent, req, reqp);
487 reqsk_queue_removed(queue, req);
491 reqp = &req->dl_next;
494 i = (i + 1) & (lopt->nr_table_entries - 1);
496 } while (--budget > 0);
498 lopt->clock_hand = i;
501 inet_csk_reset_keepalive_timer(parent, interval);
504 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
506 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
507 const gfp_t priority)
509 struct sock *newsk = sk_clone(sk, priority);
512 struct inet_connection_sock *newicsk = inet_csk(newsk);
514 newsk->sk_state = TCP_SYN_RECV;
515 newicsk->icsk_bind_hash = NULL;
517 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
518 newsk->sk_write_space = sk_stream_write_space;
520 newicsk->icsk_retransmits = 0;
521 newicsk->icsk_backoff = 0;
522 newicsk->icsk_probes_out = 0;
524 /* Deinitialize accept_queue to trap illegal accesses. */
525 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
527 security_inet_csk_clone(newsk, req);
532 EXPORT_SYMBOL_GPL(inet_csk_clone);
535 * At this point, there should be no process reference to this
536 * socket, and thus no user references at all. Therefore we
537 * can assume the socket waitqueue is inactive and nobody will
538 * try to jump onto it.
540 void inet_csk_destroy_sock(struct sock *sk)
542 BUG_TRAP(sk->sk_state == TCP_CLOSE);
543 BUG_TRAP(sock_flag(sk, SOCK_DEAD));
545 /* It cannot be in hash table! */
546 BUG_TRAP(sk_unhashed(sk));
548 /* If it has not 0 inet_sk(sk)->num, it must be bound */
549 BUG_TRAP(!inet_sk(sk)->num || inet_csk(sk)->icsk_bind_hash);
551 sk->sk_prot->destroy(sk);
553 sk_stream_kill_queues(sk);
555 xfrm_sk_free_policy(sk);
557 sk_refcnt_debug_release(sk);
559 atomic_dec(sk->sk_prot->orphan_count);
563 EXPORT_SYMBOL(inet_csk_destroy_sock);
565 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
567 struct inet_sock *inet = inet_sk(sk);
568 struct inet_connection_sock *icsk = inet_csk(sk);
569 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
574 sk->sk_max_ack_backlog = 0;
575 sk->sk_ack_backlog = 0;
576 inet_csk_delack_init(sk);
578 /* There is race window here: we announce ourselves listening,
579 * but this transition is still not validated by get_port().
580 * It is OK, because this socket enters to hash table only
581 * after validation is complete.
583 sk->sk_state = TCP_LISTEN;
584 if (!sk->sk_prot->get_port(sk, inet->num)) {
585 inet->sport = htons(inet->num);
588 sk->sk_prot->hash(sk);
593 sk->sk_state = TCP_CLOSE;
594 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
598 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
601 * This routine closes sockets which have been at least partially
602 * opened, but not yet accepted.
604 void inet_csk_listen_stop(struct sock *sk)
606 struct inet_connection_sock *icsk = inet_csk(sk);
607 struct request_sock *acc_req;
608 struct request_sock *req;
610 inet_csk_delete_keepalive_timer(sk);
612 /* make all the listen_opt local to us */
613 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
615 /* Following specs, it would be better either to send FIN
616 * (and enter FIN-WAIT-1, it is normal close)
617 * or to send active reset (abort).
618 * Certainly, it is pretty dangerous while synflood, but it is
619 * bad justification for our negligence 8)
620 * To be honest, we are not able to make either
621 * of the variants now. --ANK
623 reqsk_queue_destroy(&icsk->icsk_accept_queue);
625 while ((req = acc_req) != NULL) {
626 struct sock *child = req->sk;
628 acc_req = req->dl_next;
632 BUG_TRAP(!sock_owned_by_user(child));
635 sk->sk_prot->disconnect(child, O_NONBLOCK);
639 atomic_inc(sk->sk_prot->orphan_count);
641 inet_csk_destroy_sock(child);
643 bh_unlock_sock(child);
647 sk_acceptq_removed(sk);
650 BUG_TRAP(!sk->sk_ack_backlog);
653 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
655 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
657 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
658 const struct inet_sock *inet = inet_sk(sk);
660 sin->sin_family = AF_INET;
661 sin->sin_addr.s_addr = inet->daddr;
662 sin->sin_port = inet->dport;
665 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
668 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
669 char __user *optval, int __user *optlen)
671 const struct inet_connection_sock *icsk = inet_csk(sk);
673 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
674 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
676 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
680 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
682 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
683 char __user *optval, int optlen)
685 const struct inet_connection_sock *icsk = inet_csk(sk);
687 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
688 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
690 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
694 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);