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 struct holds the first and last local port number.
35 struct local_ports sysctl_local_ports __read_mostly
= {
36 .lock
= SEQLOCK_UNLOCKED
,
37 .range
= { 32768, 61000 },
40 void inet_get_local_port_range(int *low
, int *high
)
44 seq
= read_seqbegin(&sysctl_local_ports
.lock
);
46 *low
= sysctl_local_ports
.range
[0];
47 *high
= sysctl_local_ports
.range
[1];
48 } while (read_seqretry(&sysctl_local_ports
.lock
, seq
));
50 EXPORT_SYMBOL(inet_get_local_port_range
);
52 int inet_csk_bind_conflict(const struct sock
*sk
,
53 const struct inet_bind_bucket
*tb
)
55 const __be32 sk_rcv_saddr
= inet_rcv_saddr(sk
);
57 struct hlist_node
*node
;
58 int reuse
= sk
->sk_reuse
;
61 * Unlike other sk lookup places we do not check
62 * for sk_net here, since _all_ the socks listed
63 * in tb->owners list belong to the same net - the
64 * one this bucket belongs to.
67 sk_for_each_bound(sk2
, node
, &tb
->owners
) {
69 !inet_v6_ipv6only(sk2
) &&
70 (!sk
->sk_bound_dev_if
||
71 !sk2
->sk_bound_dev_if
||
72 sk
->sk_bound_dev_if
== sk2
->sk_bound_dev_if
)) {
73 if (!reuse
|| !sk2
->sk_reuse
||
74 sk2
->sk_state
== TCP_LISTEN
) {
75 const __be32 sk2_rcv_saddr
= inet_rcv_saddr(sk2
);
76 if (!sk2_rcv_saddr
|| !sk_rcv_saddr
||
77 sk2_rcv_saddr
== sk_rcv_saddr
)
85 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict
);
87 /* Obtain a reference to a local port for the given sock,
88 * if snum is zero it means select any available local port.
90 int inet_csk_get_port(struct sock
*sk
, unsigned short snum
)
92 struct inet_hashinfo
*hashinfo
= sk
->sk_prot
->h
.hashinfo
;
93 struct inet_bind_hashbucket
*head
;
94 struct hlist_node
*node
;
95 struct inet_bind_bucket
*tb
;
96 int ret
, attempts
= 5;
97 struct net
*net
= sock_net(sk
);
98 int smallest_size
= -1, smallest_rover
;
102 int remaining
, rover
, low
, high
;
105 inet_get_local_port_range(&low
, &high
);
106 remaining
= (high
- low
) + 1;
107 smallest_rover
= rover
= net_random() % remaining
+ low
;
111 head
= &hashinfo
->bhash
[inet_bhashfn(net
, rover
,
112 hashinfo
->bhash_size
)];
113 spin_lock(&head
->lock
);
114 inet_bind_bucket_for_each(tb
, node
, &head
->chain
)
115 if (ib_net(tb
) == net
&& tb
->port
== rover
) {
116 if (tb
->fastreuse
> 0 &&
118 sk
->sk_state
!= TCP_LISTEN
&&
119 (tb
->num_owners
< smallest_size
|| smallest_size
== -1)) {
120 smallest_size
= tb
->num_owners
;
121 smallest_rover
= rover
;
122 if (atomic_read(&hashinfo
->bsockets
) > (high
- low
) + 1) {
123 spin_unlock(&head
->lock
);
124 snum
= smallest_rover
;
132 spin_unlock(&head
->lock
);
135 } while (--remaining
> 0);
137 /* Exhausted local port range during search? It is not
138 * possible for us to be holding one of the bind hash
139 * locks if this test triggers, because if 'remaining'
140 * drops to zero, we broke out of the do/while loop at
141 * the top level, not from the 'break;' statement.
144 if (remaining
<= 0) {
145 if (smallest_size
!= -1) {
146 snum
= smallest_rover
;
151 /* OK, here is the one we will use. HEAD is
152 * non-NULL and we hold it's mutex.
157 head
= &hashinfo
->bhash
[inet_bhashfn(net
, snum
,
158 hashinfo
->bhash_size
)];
159 spin_lock(&head
->lock
);
160 inet_bind_bucket_for_each(tb
, node
, &head
->chain
)
161 if (ib_net(tb
) == net
&& tb
->port
== snum
)
167 if (!hlist_empty(&tb
->owners
)) {
168 if (tb
->fastreuse
> 0 &&
169 sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
&&
170 smallest_size
== -1) {
174 if (inet_csk(sk
)->icsk_af_ops
->bind_conflict(sk
, tb
)) {
175 if (sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
&&
176 smallest_size
!= -1 && --attempts
>= 0) {
177 spin_unlock(&head
->lock
);
186 if (!tb
&& (tb
= inet_bind_bucket_create(hashinfo
->bind_bucket_cachep
,
187 net
, head
, snum
)) == NULL
)
189 if (hlist_empty(&tb
->owners
)) {
190 if (sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
)
194 } else if (tb
->fastreuse
&&
195 (!sk
->sk_reuse
|| sk
->sk_state
== TCP_LISTEN
))
198 if (!inet_csk(sk
)->icsk_bind_hash
)
199 inet_bind_hash(sk
, tb
, snum
);
200 WARN_ON(inet_csk(sk
)->icsk_bind_hash
!= tb
);
204 spin_unlock(&head
->lock
);
210 EXPORT_SYMBOL_GPL(inet_csk_get_port
);
213 * Wait for an incoming connection, avoid race conditions. This must be called
214 * with the socket locked.
216 static int inet_csk_wait_for_connect(struct sock
*sk
, long timeo
)
218 struct inet_connection_sock
*icsk
= inet_csk(sk
);
223 * True wake-one mechanism for incoming connections: only
224 * one process gets woken up, not the 'whole herd'.
225 * Since we do not 'race & poll' for established sockets
226 * anymore, the common case will execute the loop only once.
228 * Subtle issue: "add_wait_queue_exclusive()" will be added
229 * after any current non-exclusive waiters, and we know that
230 * it will always _stay_ after any new non-exclusive waiters
231 * because all non-exclusive waiters are added at the
232 * beginning of the wait-queue. As such, it's ok to "drop"
233 * our exclusiveness temporarily when we get woken up without
234 * having to remove and re-insert us on the wait queue.
237 prepare_to_wait_exclusive(sk
->sk_sleep
, &wait
,
240 if (reqsk_queue_empty(&icsk
->icsk_accept_queue
))
241 timeo
= schedule_timeout(timeo
);
244 if (!reqsk_queue_empty(&icsk
->icsk_accept_queue
))
247 if (sk
->sk_state
!= TCP_LISTEN
)
249 err
= sock_intr_errno(timeo
);
250 if (signal_pending(current
))
256 finish_wait(sk
->sk_sleep
, &wait
);
261 * This will accept the next outstanding connection.
263 struct sock
*inet_csk_accept(struct sock
*sk
, int flags
, int *err
)
265 struct inet_connection_sock
*icsk
= inet_csk(sk
);
271 /* We need to make sure that this socket is listening,
272 * and that it has something pending.
275 if (sk
->sk_state
!= TCP_LISTEN
)
278 /* Find already established connection */
279 if (reqsk_queue_empty(&icsk
->icsk_accept_queue
)) {
280 long timeo
= sock_rcvtimeo(sk
, flags
& O_NONBLOCK
);
282 /* If this is a non blocking socket don't sleep */
287 error
= inet_csk_wait_for_connect(sk
, timeo
);
292 newsk
= reqsk_queue_get_child(&icsk
->icsk_accept_queue
, sk
);
293 WARN_ON(newsk
->sk_state
== TCP_SYN_RECV
);
303 EXPORT_SYMBOL(inet_csk_accept
);
306 * Using different timers for retransmit, delayed acks and probes
307 * We may wish use just one timer maintaining a list of expire jiffies
310 void inet_csk_init_xmit_timers(struct sock
*sk
,
311 void (*retransmit_handler
)(unsigned long),
312 void (*delack_handler
)(unsigned long),
313 void (*keepalive_handler
)(unsigned long))
315 struct inet_connection_sock
*icsk
= inet_csk(sk
);
317 setup_timer(&icsk
->icsk_retransmit_timer
, retransmit_handler
,
319 setup_timer(&icsk
->icsk_delack_timer
, delack_handler
,
321 setup_timer(&sk
->sk_timer
, keepalive_handler
, (unsigned long)sk
);
322 icsk
->icsk_pending
= icsk
->icsk_ack
.pending
= 0;
325 EXPORT_SYMBOL(inet_csk_init_xmit_timers
);
327 void inet_csk_clear_xmit_timers(struct sock
*sk
)
329 struct inet_connection_sock
*icsk
= inet_csk(sk
);
331 icsk
->icsk_pending
= icsk
->icsk_ack
.pending
= icsk
->icsk_ack
.blocked
= 0;
333 sk_stop_timer(sk
, &icsk
->icsk_retransmit_timer
);
334 sk_stop_timer(sk
, &icsk
->icsk_delack_timer
);
335 sk_stop_timer(sk
, &sk
->sk_timer
);
338 EXPORT_SYMBOL(inet_csk_clear_xmit_timers
);
340 void inet_csk_delete_keepalive_timer(struct sock
*sk
)
342 sk_stop_timer(sk
, &sk
->sk_timer
);
345 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer
);
347 void inet_csk_reset_keepalive_timer(struct sock
*sk
, unsigned long len
)
349 sk_reset_timer(sk
, &sk
->sk_timer
, jiffies
+ len
);
352 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer
);
354 struct dst_entry
*inet_csk_route_req(struct sock
*sk
,
355 const struct request_sock
*req
)
358 const struct inet_request_sock
*ireq
= inet_rsk(req
);
359 struct ip_options
*opt
= inet_rsk(req
)->opt
;
360 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
362 { .daddr
= ((opt
&& opt
->srr
) ?
365 .saddr
= ireq
->loc_addr
,
366 .tos
= RT_CONN_FLAGS(sk
) } },
367 .proto
= sk
->sk_protocol
,
368 .flags
= inet_sk_flowi_flags(sk
),
370 { .sport
= inet_sk(sk
)->sport
,
371 .dport
= ireq
->rmt_port
} } };
372 struct net
*net
= sock_net(sk
);
374 security_req_classify_flow(req
, &fl
);
375 if (ip_route_output_flow(net
, &rt
, &fl
, sk
, 0))
377 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
)
384 IP_INC_STATS_BH(net
, IPSTATS_MIB_OUTNOROUTES
);
388 EXPORT_SYMBOL_GPL(inet_csk_route_req
);
390 static inline u32
inet_synq_hash(const __be32 raddr
, const __be16 rport
,
391 const u32 rnd
, const u32 synq_hsize
)
393 return jhash_2words((__force u32
)raddr
, (__force u32
)rport
, rnd
) & (synq_hsize
- 1);
396 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
397 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
399 #define AF_INET_FAMILY(fam) 1
402 struct request_sock
*inet_csk_search_req(const struct sock
*sk
,
403 struct request_sock
***prevp
,
404 const __be16 rport
, const __be32 raddr
,
407 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
408 struct listen_sock
*lopt
= icsk
->icsk_accept_queue
.listen_opt
;
409 struct request_sock
*req
, **prev
;
411 for (prev
= &lopt
->syn_table
[inet_synq_hash(raddr
, rport
, lopt
->hash_rnd
,
412 lopt
->nr_table_entries
)];
413 (req
= *prev
) != NULL
;
414 prev
= &req
->dl_next
) {
415 const struct inet_request_sock
*ireq
= inet_rsk(req
);
417 if (ireq
->rmt_port
== rport
&&
418 ireq
->rmt_addr
== raddr
&&
419 ireq
->loc_addr
== laddr
&&
420 AF_INET_FAMILY(req
->rsk_ops
->family
)) {
430 EXPORT_SYMBOL_GPL(inet_csk_search_req
);
432 void inet_csk_reqsk_queue_hash_add(struct sock
*sk
, struct request_sock
*req
,
433 unsigned long timeout
)
435 struct inet_connection_sock
*icsk
= inet_csk(sk
);
436 struct listen_sock
*lopt
= icsk
->icsk_accept_queue
.listen_opt
;
437 const u32 h
= inet_synq_hash(inet_rsk(req
)->rmt_addr
, inet_rsk(req
)->rmt_port
,
438 lopt
->hash_rnd
, lopt
->nr_table_entries
);
440 reqsk_queue_hash_req(&icsk
->icsk_accept_queue
, h
, req
, timeout
);
441 inet_csk_reqsk_queue_added(sk
, timeout
);
444 /* Only thing we need from tcp.h */
445 extern int sysctl_tcp_synack_retries
;
447 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add
);
449 /* Decide when to expire the request and when to resend SYN-ACK */
450 static inline void syn_ack_recalc(struct request_sock
*req
, const int thresh
,
451 const int max_retries
,
452 const u8 rskq_defer_accept
,
453 int *expire
, int *resend
)
455 if (!rskq_defer_accept
) {
456 *expire
= req
->retrans
>= thresh
;
460 *expire
= req
->retrans
>= thresh
&&
461 (!inet_rsk(req
)->acked
|| req
->retrans
>= max_retries
);
463 * Do not resend while waiting for data after ACK,
464 * start to resend on end of deferring period to give
465 * last chance for data or ACK to create established socket.
467 *resend
= !inet_rsk(req
)->acked
||
468 req
->retrans
>= rskq_defer_accept
- 1;
471 void inet_csk_reqsk_queue_prune(struct sock
*parent
,
472 const unsigned long interval
,
473 const unsigned long timeout
,
474 const unsigned long max_rto
)
476 struct inet_connection_sock
*icsk
= inet_csk(parent
);
477 struct request_sock_queue
*queue
= &icsk
->icsk_accept_queue
;
478 struct listen_sock
*lopt
= queue
->listen_opt
;
479 int max_retries
= icsk
->icsk_syn_retries
? : sysctl_tcp_synack_retries
;
480 int thresh
= max_retries
;
481 unsigned long now
= jiffies
;
482 struct request_sock
**reqp
, *req
;
485 if (lopt
== NULL
|| lopt
->qlen
== 0)
488 /* Normally all the openreqs are young and become mature
489 * (i.e. converted to established socket) for first timeout.
490 * If synack was not acknowledged for 3 seconds, it means
491 * one of the following things: synack was lost, ack was lost,
492 * rtt is high or nobody planned to ack (i.e. synflood).
493 * When server is a bit loaded, queue is populated with old
494 * open requests, reducing effective size of queue.
495 * When server is well loaded, queue size reduces to zero
496 * after several minutes of work. It is not synflood,
497 * it is normal operation. The solution is pruning
498 * too old entries overriding normal timeout, when
499 * situation becomes dangerous.
501 * Essentially, we reserve half of room for young
502 * embrions; and abort old ones without pity, if old
503 * ones are about to clog our table.
505 if (lopt
->qlen
>>(lopt
->max_qlen_log
-1)) {
506 int young
= (lopt
->qlen_young
<<1);
509 if (lopt
->qlen
< young
)
516 if (queue
->rskq_defer_accept
)
517 max_retries
= queue
->rskq_defer_accept
;
519 budget
= 2 * (lopt
->nr_table_entries
/ (timeout
/ interval
));
520 i
= lopt
->clock_hand
;
523 reqp
=&lopt
->syn_table
[i
];
524 while ((req
= *reqp
) != NULL
) {
525 if (time_after_eq(now
, req
->expires
)) {
526 int expire
= 0, resend
= 0;
528 syn_ack_recalc(req
, thresh
, max_retries
,
529 queue
->rskq_defer_accept
,
533 !req
->rsk_ops
->rtx_syn_ack(parent
, req
) ||
534 inet_rsk(req
)->acked
)) {
537 if (req
->retrans
++ == 0)
539 timeo
= min((timeout
<< req
->retrans
), max_rto
);
540 req
->expires
= now
+ timeo
;
541 reqp
= &req
->dl_next
;
545 /* Drop this request */
546 inet_csk_reqsk_queue_unlink(parent
, req
, reqp
);
547 reqsk_queue_removed(queue
, req
);
551 reqp
= &req
->dl_next
;
554 i
= (i
+ 1) & (lopt
->nr_table_entries
- 1);
556 } while (--budget
> 0);
558 lopt
->clock_hand
= i
;
561 inet_csk_reset_keepalive_timer(parent
, interval
);
564 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune
);
566 struct sock
*inet_csk_clone(struct sock
*sk
, const struct request_sock
*req
,
567 const gfp_t priority
)
569 struct sock
*newsk
= sk_clone(sk
, priority
);
572 struct inet_connection_sock
*newicsk
= inet_csk(newsk
);
574 newsk
->sk_state
= TCP_SYN_RECV
;
575 newicsk
->icsk_bind_hash
= NULL
;
577 inet_sk(newsk
)->dport
= inet_rsk(req
)->rmt_port
;
578 inet_sk(newsk
)->num
= ntohs(inet_rsk(req
)->loc_port
);
579 inet_sk(newsk
)->sport
= inet_rsk(req
)->loc_port
;
580 newsk
->sk_write_space
= sk_stream_write_space
;
582 newicsk
->icsk_retransmits
= 0;
583 newicsk
->icsk_backoff
= 0;
584 newicsk
->icsk_probes_out
= 0;
586 /* Deinitialize accept_queue to trap illegal accesses. */
587 memset(&newicsk
->icsk_accept_queue
, 0, sizeof(newicsk
->icsk_accept_queue
));
589 security_inet_csk_clone(newsk
, req
);
594 EXPORT_SYMBOL_GPL(inet_csk_clone
);
597 * At this point, there should be no process reference to this
598 * socket, and thus no user references at all. Therefore we
599 * can assume the socket waitqueue is inactive and nobody will
600 * try to jump onto it.
602 void inet_csk_destroy_sock(struct sock
*sk
)
604 WARN_ON(sk
->sk_state
!= TCP_CLOSE
);
605 WARN_ON(!sock_flag(sk
, SOCK_DEAD
));
607 /* It cannot be in hash table! */
608 WARN_ON(!sk_unhashed(sk
));
610 /* If it has not 0 inet_sk(sk)->num, it must be bound */
611 WARN_ON(inet_sk(sk
)->num
&& !inet_csk(sk
)->icsk_bind_hash
);
613 sk
->sk_prot
->destroy(sk
);
615 sk_stream_kill_queues(sk
);
617 xfrm_sk_free_policy(sk
);
619 sk_refcnt_debug_release(sk
);
621 percpu_counter_dec(sk
->sk_prot
->orphan_count
);
625 EXPORT_SYMBOL(inet_csk_destroy_sock
);
627 int inet_csk_listen_start(struct sock
*sk
, const int nr_table_entries
)
629 struct inet_sock
*inet
= inet_sk(sk
);
630 struct inet_connection_sock
*icsk
= inet_csk(sk
);
631 int rc
= reqsk_queue_alloc(&icsk
->icsk_accept_queue
, nr_table_entries
);
636 sk
->sk_max_ack_backlog
= 0;
637 sk
->sk_ack_backlog
= 0;
638 inet_csk_delack_init(sk
);
640 /* There is race window here: we announce ourselves listening,
641 * but this transition is still not validated by get_port().
642 * It is OK, because this socket enters to hash table only
643 * after validation is complete.
645 sk
->sk_state
= TCP_LISTEN
;
646 if (!sk
->sk_prot
->get_port(sk
, inet
->num
)) {
647 inet
->sport
= htons(inet
->num
);
650 sk
->sk_prot
->hash(sk
);
655 sk
->sk_state
= TCP_CLOSE
;
656 __reqsk_queue_destroy(&icsk
->icsk_accept_queue
);
660 EXPORT_SYMBOL_GPL(inet_csk_listen_start
);
663 * This routine closes sockets which have been at least partially
664 * opened, but not yet accepted.
666 void inet_csk_listen_stop(struct sock
*sk
)
668 struct inet_connection_sock
*icsk
= inet_csk(sk
);
669 struct request_sock
*acc_req
;
670 struct request_sock
*req
;
672 inet_csk_delete_keepalive_timer(sk
);
674 /* make all the listen_opt local to us */
675 acc_req
= reqsk_queue_yank_acceptq(&icsk
->icsk_accept_queue
);
677 /* Following specs, it would be better either to send FIN
678 * (and enter FIN-WAIT-1, it is normal close)
679 * or to send active reset (abort).
680 * Certainly, it is pretty dangerous while synflood, but it is
681 * bad justification for our negligence 8)
682 * To be honest, we are not able to make either
683 * of the variants now. --ANK
685 reqsk_queue_destroy(&icsk
->icsk_accept_queue
);
687 while ((req
= acc_req
) != NULL
) {
688 struct sock
*child
= req
->sk
;
690 acc_req
= req
->dl_next
;
694 WARN_ON(sock_owned_by_user(child
));
697 sk
->sk_prot
->disconnect(child
, O_NONBLOCK
);
701 percpu_counter_inc(sk
->sk_prot
->orphan_count
);
703 inet_csk_destroy_sock(child
);
705 bh_unlock_sock(child
);
709 sk_acceptq_removed(sk
);
712 WARN_ON(sk
->sk_ack_backlog
);
715 EXPORT_SYMBOL_GPL(inet_csk_listen_stop
);
717 void inet_csk_addr2sockaddr(struct sock
*sk
, struct sockaddr
*uaddr
)
719 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
720 const struct inet_sock
*inet
= inet_sk(sk
);
722 sin
->sin_family
= AF_INET
;
723 sin
->sin_addr
.s_addr
= inet
->daddr
;
724 sin
->sin_port
= inet
->dport
;
727 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr
);
730 int inet_csk_compat_getsockopt(struct sock
*sk
, int level
, int optname
,
731 char __user
*optval
, int __user
*optlen
)
733 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
735 if (icsk
->icsk_af_ops
->compat_getsockopt
!= NULL
)
736 return icsk
->icsk_af_ops
->compat_getsockopt(sk
, level
, optname
,
738 return icsk
->icsk_af_ops
->getsockopt(sk
, level
, optname
,
742 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt
);
744 int inet_csk_compat_setsockopt(struct sock
*sk
, int level
, int optname
,
745 char __user
*optval
, unsigned int optlen
)
747 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
749 if (icsk
->icsk_af_ops
->compat_setsockopt
!= NULL
)
750 return icsk
->icsk_af_ops
->compat_setsockopt(sk
, level
, optname
,
752 return icsk
->icsk_af_ops
->setsockopt(sk
, level
, optname
,
756 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt
);