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/config.h>
17 #include <linux/module.h>
18 #include <linux/jhash.h>
20 #include <net/inet_connection_sock.h>
21 #include <net/inet_hashtables.h>
22 #include <net/inet_timewait_sock.h>
24 #include <net/route.h>
25 #include <net/tcp_states.h>
29 const char inet_csk_timer_bug_msg
[] = "inet_csk BUG: unknown timer value\n";
30 EXPORT_SYMBOL(inet_csk_timer_bug_msg
);
34 * This array holds the first and last local port number.
35 * For high-usage systems, use sysctl to change this to
38 int sysctl_local_port_range
[2] = { 1024, 4999 };
40 int inet_csk_bind_conflict(const struct sock
*sk
,
41 const struct inet_bind_bucket
*tb
)
43 const u32 sk_rcv_saddr
= inet_rcv_saddr(sk
);
45 struct hlist_node
*node
;
46 int reuse
= sk
->sk_reuse
;
48 sk_for_each_bound(sk2
, node
, &tb
->owners
) {
50 !inet_v6_ipv6only(sk2
) &&
51 (!sk
->sk_bound_dev_if
||
52 !sk2
->sk_bound_dev_if
||
53 sk
->sk_bound_dev_if
== sk2
->sk_bound_dev_if
)) {
54 if (!reuse
|| !sk2
->sk_reuse
||
55 sk2
->sk_state
== TCP_LISTEN
) {
56 const u32 sk2_rcv_saddr
= inet_rcv_saddr(sk2
);
57 if (!sk2_rcv_saddr
|| !sk_rcv_saddr
||
58 sk2_rcv_saddr
== sk_rcv_saddr
)
66 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict
);
68 /* Obtain a reference to a local port for the given sock,
69 * if snum is zero it means select any available local port.
71 int inet_csk_get_port(struct inet_hashinfo
*hashinfo
,
72 struct sock
*sk
, unsigned short snum
,
73 int (*bind_conflict
)(const struct sock
*sk
,
74 const struct inet_bind_bucket
*tb
))
76 struct inet_bind_hashbucket
*head
;
77 struct hlist_node
*node
;
78 struct inet_bind_bucket
*tb
;
83 int low
= sysctl_local_port_range
[0];
84 int high
= sysctl_local_port_range
[1];
85 int remaining
= (high
- low
) + 1;
86 int rover
= net_random() % (high
- low
) + low
;
89 head
= &hashinfo
->bhash
[inet_bhashfn(rover
, hashinfo
->bhash_size
)];
90 spin_lock(&head
->lock
);
91 inet_bind_bucket_for_each(tb
, node
, &head
->chain
)
92 if (tb
->port
== rover
)
96 spin_unlock(&head
->lock
);
99 } while (--remaining
> 0);
101 /* Exhausted local port range during search? It is not
102 * possible for us to be holding one of the bind hash
103 * locks if this test triggers, because if 'remaining'
104 * drops to zero, we broke out of the do/while loop at
105 * the top level, not from the 'break;' statement.
111 /* OK, here is the one we will use. HEAD is
112 * non-NULL and we hold it's mutex.
116 head
= &hashinfo
->bhash
[inet_bhashfn(snum
, hashinfo
->bhash_size
)];
117 spin_lock(&head
->lock
);
118 inet_bind_bucket_for_each(tb
, node
, &head
->chain
)
119 if (tb
->port
== snum
)
125 if (!hlist_empty(&tb
->owners
)) {
126 if (sk
->sk_reuse
> 1)
128 if (tb
->fastreuse
> 0 &&
129 sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
) {
133 if (bind_conflict(sk
, tb
))
139 if (!tb
&& (tb
= inet_bind_bucket_create(hashinfo
->bind_bucket_cachep
, head
, snum
)) == NULL
)
141 if (hlist_empty(&tb
->owners
)) {
142 if (sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
)
146 } else if (tb
->fastreuse
&&
147 (!sk
->sk_reuse
|| sk
->sk_state
== TCP_LISTEN
))
150 if (!inet_csk(sk
)->icsk_bind_hash
)
151 inet_bind_hash(sk
, tb
, snum
);
152 BUG_TRAP(inet_csk(sk
)->icsk_bind_hash
== tb
);
156 spin_unlock(&head
->lock
);
162 EXPORT_SYMBOL_GPL(inet_csk_get_port
);
165 * Wait for an incoming connection, avoid race conditions. This must be called
166 * with the socket locked.
168 static int inet_csk_wait_for_connect(struct sock
*sk
, long timeo
)
170 struct inet_connection_sock
*icsk
= inet_csk(sk
);
175 * True wake-one mechanism for incoming connections: only
176 * one process gets woken up, not the 'whole herd'.
177 * Since we do not 'race & poll' for established sockets
178 * anymore, the common case will execute the loop only once.
180 * Subtle issue: "add_wait_queue_exclusive()" will be added
181 * after any current non-exclusive waiters, and we know that
182 * it will always _stay_ after any new non-exclusive waiters
183 * because all non-exclusive waiters are added at the
184 * beginning of the wait-queue. As such, it's ok to "drop"
185 * our exclusiveness temporarily when we get woken up without
186 * having to remove and re-insert us on the wait queue.
189 prepare_to_wait_exclusive(sk
->sk_sleep
, &wait
,
192 if (reqsk_queue_empty(&icsk
->icsk_accept_queue
))
193 timeo
= schedule_timeout(timeo
);
196 if (!reqsk_queue_empty(&icsk
->icsk_accept_queue
))
199 if (sk
->sk_state
!= TCP_LISTEN
)
201 err
= sock_intr_errno(timeo
);
202 if (signal_pending(current
))
208 finish_wait(sk
->sk_sleep
, &wait
);
213 * This will accept the next outstanding connection.
215 struct sock
*inet_csk_accept(struct sock
*sk
, int flags
, int *err
)
217 struct inet_connection_sock
*icsk
= inet_csk(sk
);
223 /* We need to make sure that this socket is listening,
224 * and that it has something pending.
227 if (sk
->sk_state
!= TCP_LISTEN
)
230 /* Find already established connection */
231 if (reqsk_queue_empty(&icsk
->icsk_accept_queue
)) {
232 long timeo
= sock_rcvtimeo(sk
, flags
& O_NONBLOCK
);
234 /* If this is a non blocking socket don't sleep */
239 error
= inet_csk_wait_for_connect(sk
, timeo
);
244 newsk
= reqsk_queue_get_child(&icsk
->icsk_accept_queue
, sk
);
245 BUG_TRAP(newsk
->sk_state
!= TCP_SYN_RECV
);
255 EXPORT_SYMBOL(inet_csk_accept
);
258 * Using different timers for retransmit, delayed acks and probes
259 * We may wish use just one timer maintaining a list of expire jiffies
262 void inet_csk_init_xmit_timers(struct sock
*sk
,
263 void (*retransmit_handler
)(unsigned long),
264 void (*delack_handler
)(unsigned long),
265 void (*keepalive_handler
)(unsigned long))
267 struct inet_connection_sock
*icsk
= inet_csk(sk
);
269 init_timer(&icsk
->icsk_retransmit_timer
);
270 init_timer(&icsk
->icsk_delack_timer
);
271 init_timer(&sk
->sk_timer
);
273 icsk
->icsk_retransmit_timer
.function
= retransmit_handler
;
274 icsk
->icsk_delack_timer
.function
= delack_handler
;
275 sk
->sk_timer
.function
= keepalive_handler
;
277 icsk
->icsk_retransmit_timer
.data
=
278 icsk
->icsk_delack_timer
.data
=
279 sk
->sk_timer
.data
= (unsigned long)sk
;
281 icsk
->icsk_pending
= icsk
->icsk_ack
.pending
= 0;
284 EXPORT_SYMBOL(inet_csk_init_xmit_timers
);
286 void inet_csk_clear_xmit_timers(struct sock
*sk
)
288 struct inet_connection_sock
*icsk
= inet_csk(sk
);
290 icsk
->icsk_pending
= icsk
->icsk_ack
.pending
= icsk
->icsk_ack
.blocked
= 0;
292 sk_stop_timer(sk
, &icsk
->icsk_retransmit_timer
);
293 sk_stop_timer(sk
, &icsk
->icsk_delack_timer
);
294 sk_stop_timer(sk
, &sk
->sk_timer
);
297 EXPORT_SYMBOL(inet_csk_clear_xmit_timers
);
299 void inet_csk_delete_keepalive_timer(struct sock
*sk
)
301 sk_stop_timer(sk
, &sk
->sk_timer
);
304 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer
);
306 void inet_csk_reset_keepalive_timer(struct sock
*sk
, unsigned long len
)
308 sk_reset_timer(sk
, &sk
->sk_timer
, jiffies
+ len
);
311 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer
);
313 struct dst_entry
* inet_csk_route_req(struct sock
*sk
,
314 const struct request_sock
*req
)
317 const struct inet_request_sock
*ireq
= inet_rsk(req
);
318 struct ip_options
*opt
= inet_rsk(req
)->opt
;
319 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
321 { .daddr
= ((opt
&& opt
->srr
) ?
324 .saddr
= ireq
->loc_addr
,
325 .tos
= RT_CONN_FLAGS(sk
) } },
326 .proto
= sk
->sk_protocol
,
328 { .sport
= inet_sk(sk
)->sport
,
329 .dport
= ireq
->rmt_port
} } };
331 if (ip_route_output_flow(&rt
, &fl
, sk
, 0)) {
332 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES
);
335 if (opt
&& opt
->is_strictroute
&& rt
->rt_dst
!= rt
->rt_gateway
) {
337 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES
);
343 EXPORT_SYMBOL_GPL(inet_csk_route_req
);
345 static inline u32
inet_synq_hash(const u32 raddr
, const u16 rport
,
346 const u32 rnd
, const u16 synq_hsize
)
348 return jhash_2words(raddr
, (u32
)rport
, rnd
) & (synq_hsize
- 1);
351 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
352 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
354 #define AF_INET_FAMILY(fam) 1
357 struct request_sock
*inet_csk_search_req(const struct sock
*sk
,
358 struct request_sock
***prevp
,
359 const __u16 rport
, const __u32 raddr
,
362 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
363 struct listen_sock
*lopt
= icsk
->icsk_accept_queue
.listen_opt
;
364 struct request_sock
*req
, **prev
;
366 for (prev
= &lopt
->syn_table
[inet_synq_hash(raddr
, rport
, lopt
->hash_rnd
,
367 lopt
->nr_table_entries
)];
368 (req
= *prev
) != NULL
;
369 prev
= &req
->dl_next
) {
370 const struct inet_request_sock
*ireq
= inet_rsk(req
);
372 if (ireq
->rmt_port
== rport
&&
373 ireq
->rmt_addr
== raddr
&&
374 ireq
->loc_addr
== laddr
&&
375 AF_INET_FAMILY(req
->rsk_ops
->family
)) {
385 EXPORT_SYMBOL_GPL(inet_csk_search_req
);
387 void inet_csk_reqsk_queue_hash_add(struct sock
*sk
, struct request_sock
*req
,
388 unsigned long timeout
)
390 struct inet_connection_sock
*icsk
= inet_csk(sk
);
391 struct listen_sock
*lopt
= icsk
->icsk_accept_queue
.listen_opt
;
392 const u32 h
= inet_synq_hash(inet_rsk(req
)->rmt_addr
, inet_rsk(req
)->rmt_port
,
393 lopt
->hash_rnd
, lopt
->nr_table_entries
);
395 reqsk_queue_hash_req(&icsk
->icsk_accept_queue
, h
, req
, timeout
);
396 inet_csk_reqsk_queue_added(sk
, timeout
);
399 /* Only thing we need from tcp.h */
400 extern int sysctl_tcp_synack_retries
;
402 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add
);
404 void inet_csk_reqsk_queue_prune(struct sock
*parent
,
405 const unsigned long interval
,
406 const unsigned long timeout
,
407 const unsigned long max_rto
)
409 struct inet_connection_sock
*icsk
= inet_csk(parent
);
410 struct request_sock_queue
*queue
= &icsk
->icsk_accept_queue
;
411 struct listen_sock
*lopt
= queue
->listen_opt
;
412 int max_retries
= icsk
->icsk_syn_retries
? : sysctl_tcp_synack_retries
;
413 int thresh
= max_retries
;
414 unsigned long now
= jiffies
;
415 struct request_sock
**reqp
, *req
;
418 if (lopt
== NULL
|| lopt
->qlen
== 0)
421 /* Normally all the openreqs are young and become mature
422 * (i.e. converted to established socket) for first timeout.
423 * If synack was not acknowledged for 3 seconds, it means
424 * one of the following things: synack was lost, ack was lost,
425 * rtt is high or nobody planned to ack (i.e. synflood).
426 * When server is a bit loaded, queue is populated with old
427 * open requests, reducing effective size of queue.
428 * When server is well loaded, queue size reduces to zero
429 * after several minutes of work. It is not synflood,
430 * it is normal operation. The solution is pruning
431 * too old entries overriding normal timeout, when
432 * situation becomes dangerous.
434 * Essentially, we reserve half of room for young
435 * embrions; and abort old ones without pity, if old
436 * ones are about to clog our table.
438 if (lopt
->qlen
>>(lopt
->max_qlen_log
-1)) {
439 int young
= (lopt
->qlen_young
<<1);
442 if (lopt
->qlen
< young
)
449 if (queue
->rskq_defer_accept
)
450 max_retries
= queue
->rskq_defer_accept
;
452 budget
= 2 * (lopt
->nr_table_entries
/ (timeout
/ interval
));
453 i
= lopt
->clock_hand
;
456 reqp
=&lopt
->syn_table
[i
];
457 while ((req
= *reqp
) != NULL
) {
458 if (time_after_eq(now
, req
->expires
)) {
459 if ((req
->retrans
< thresh
||
460 (inet_rsk(req
)->acked
&& req
->retrans
< max_retries
))
461 && !req
->rsk_ops
->rtx_syn_ack(parent
, req
, NULL
)) {
464 if (req
->retrans
++ == 0)
466 timeo
= min((timeout
<< req
->retrans
), max_rto
);
467 req
->expires
= now
+ timeo
;
468 reqp
= &req
->dl_next
;
472 /* Drop this request */
473 inet_csk_reqsk_queue_unlink(parent
, req
, reqp
);
474 reqsk_queue_removed(queue
, req
);
478 reqp
= &req
->dl_next
;
481 i
= (i
+ 1) & (lopt
->nr_table_entries
- 1);
483 } while (--budget
> 0);
485 lopt
->clock_hand
= i
;
488 inet_csk_reset_keepalive_timer(parent
, interval
);
491 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune
);
493 struct sock
*inet_csk_clone(struct sock
*sk
, const struct request_sock
*req
,
494 const gfp_t priority
)
496 struct sock
*newsk
= sk_clone(sk
, priority
);
499 struct inet_connection_sock
*newicsk
= inet_csk(newsk
);
501 newsk
->sk_state
= TCP_SYN_RECV
;
502 newicsk
->icsk_bind_hash
= NULL
;
504 inet_sk(newsk
)->dport
= inet_rsk(req
)->rmt_port
;
505 newsk
->sk_write_space
= sk_stream_write_space
;
507 newicsk
->icsk_retransmits
= 0;
508 newicsk
->icsk_backoff
= 0;
509 newicsk
->icsk_probes_out
= 0;
511 /* Deinitialize accept_queue to trap illegal accesses. */
512 memset(&newicsk
->icsk_accept_queue
, 0, sizeof(newicsk
->icsk_accept_queue
));
517 EXPORT_SYMBOL_GPL(inet_csk_clone
);
520 * At this point, there should be no process reference to this
521 * socket, and thus no user references at all. Therefore we
522 * can assume the socket waitqueue is inactive and nobody will
523 * try to jump onto it.
525 void inet_csk_destroy_sock(struct sock
*sk
)
527 BUG_TRAP(sk
->sk_state
== TCP_CLOSE
);
528 BUG_TRAP(sock_flag(sk
, SOCK_DEAD
));
530 /* It cannot be in hash table! */
531 BUG_TRAP(sk_unhashed(sk
));
533 /* If it has not 0 inet_sk(sk)->num, it must be bound */
534 BUG_TRAP(!inet_sk(sk
)->num
|| inet_csk(sk
)->icsk_bind_hash
);
536 sk
->sk_prot
->destroy(sk
);
538 sk_stream_kill_queues(sk
);
540 xfrm_sk_free_policy(sk
);
542 sk_refcnt_debug_release(sk
);
544 atomic_dec(sk
->sk_prot
->orphan_count
);
548 EXPORT_SYMBOL(inet_csk_destroy_sock
);
550 int inet_csk_listen_start(struct sock
*sk
, const int nr_table_entries
)
552 struct inet_sock
*inet
= inet_sk(sk
);
553 struct inet_connection_sock
*icsk
= inet_csk(sk
);
554 int rc
= reqsk_queue_alloc(&icsk
->icsk_accept_queue
, nr_table_entries
);
559 sk
->sk_max_ack_backlog
= 0;
560 sk
->sk_ack_backlog
= 0;
561 inet_csk_delack_init(sk
);
563 /* There is race window here: we announce ourselves listening,
564 * but this transition is still not validated by get_port().
565 * It is OK, because this socket enters to hash table only
566 * after validation is complete.
568 sk
->sk_state
= TCP_LISTEN
;
569 if (!sk
->sk_prot
->get_port(sk
, inet
->num
)) {
570 inet
->sport
= htons(inet
->num
);
573 sk
->sk_prot
->hash(sk
);
578 sk
->sk_state
= TCP_CLOSE
;
579 __reqsk_queue_destroy(&icsk
->icsk_accept_queue
);
583 EXPORT_SYMBOL_GPL(inet_csk_listen_start
);
586 * This routine closes sockets which have been at least partially
587 * opened, but not yet accepted.
589 void inet_csk_listen_stop(struct sock
*sk
)
591 struct inet_connection_sock
*icsk
= inet_csk(sk
);
592 struct request_sock
*acc_req
;
593 struct request_sock
*req
;
595 inet_csk_delete_keepalive_timer(sk
);
597 /* make all the listen_opt local to us */
598 acc_req
= reqsk_queue_yank_acceptq(&icsk
->icsk_accept_queue
);
600 /* Following specs, it would be better either to send FIN
601 * (and enter FIN-WAIT-1, it is normal close)
602 * or to send active reset (abort).
603 * Certainly, it is pretty dangerous while synflood, but it is
604 * bad justification for our negligence 8)
605 * To be honest, we are not able to make either
606 * of the variants now. --ANK
608 reqsk_queue_destroy(&icsk
->icsk_accept_queue
);
610 while ((req
= acc_req
) != NULL
) {
611 struct sock
*child
= req
->sk
;
613 acc_req
= req
->dl_next
;
617 BUG_TRAP(!sock_owned_by_user(child
));
620 sk
->sk_prot
->disconnect(child
, O_NONBLOCK
);
624 atomic_inc(sk
->sk_prot
->orphan_count
);
626 inet_csk_destroy_sock(child
);
628 bh_unlock_sock(child
);
632 sk_acceptq_removed(sk
);
635 BUG_TRAP(!sk
->sk_ack_backlog
);
638 EXPORT_SYMBOL_GPL(inet_csk_listen_stop
);
640 void inet_csk_addr2sockaddr(struct sock
*sk
, struct sockaddr
*uaddr
)
642 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
643 const struct inet_sock
*inet
= inet_sk(sk
);
645 sin
->sin_family
= AF_INET
;
646 sin
->sin_addr
.s_addr
= inet
->daddr
;
647 sin
->sin_port
= inet
->dport
;
650 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr
);