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.
34 * For high-usage systems, use sysctl to change this to
37 int sysctl_local_port_range
[2] = { 1024, 4999 };
39 int inet_csk_bind_conflict(const struct sock
*sk
,
40 const struct inet_bind_bucket
*tb
)
42 const __be32 sk_rcv_saddr
= inet_rcv_saddr(sk
);
44 struct hlist_node
*node
;
45 int reuse
= sk
->sk_reuse
;
47 sk_for_each_bound(sk2
, node
, &tb
->owners
) {
49 !inet_v6_ipv6only(sk2
) &&
50 (!sk
->sk_bound_dev_if
||
51 !sk2
->sk_bound_dev_if
||
52 sk
->sk_bound_dev_if
== sk2
->sk_bound_dev_if
)) {
53 if (!reuse
|| !sk2
->sk_reuse
||
54 sk2
->sk_state
== TCP_LISTEN
) {
55 const __be32 sk2_rcv_saddr
= inet_rcv_saddr(sk2
);
56 if (!sk2_rcv_saddr
|| !sk_rcv_saddr
||
57 sk2_rcv_saddr
== sk_rcv_saddr
)
65 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict
);
67 /* Obtain a reference to a local port for the given sock,
68 * if snum is zero it means select any available local port.
70 int inet_csk_get_port(struct inet_hashinfo
*hashinfo
,
71 struct sock
*sk
, unsigned short snum
,
72 int (*bind_conflict
)(const struct sock
*sk
,
73 const struct inet_bind_bucket
*tb
))
75 struct inet_bind_hashbucket
*head
;
76 struct hlist_node
*node
;
77 struct inet_bind_bucket
*tb
;
82 int low
= sysctl_local_port_range
[0];
83 int high
= sysctl_local_port_range
[1];
84 int remaining
= (high
- low
) + 1;
85 int rover
= net_random() % (high
- low
) + low
;
88 head
= &hashinfo
->bhash
[inet_bhashfn(rover
, hashinfo
->bhash_size
)];
89 spin_lock(&head
->lock
);
90 inet_bind_bucket_for_each(tb
, node
, &head
->chain
)
91 if (tb
->port
== rover
)
95 spin_unlock(&head
->lock
);
98 } while (--remaining
> 0);
100 /* Exhausted local port range during search? It is not
101 * possible for us to be holding one of the bind hash
102 * locks if this test triggers, because if 'remaining'
103 * drops to zero, we broke out of the do/while loop at
104 * the top level, not from the 'break;' statement.
110 /* OK, here is the one we will use. HEAD is
111 * non-NULL and we hold it's mutex.
115 head
= &hashinfo
->bhash
[inet_bhashfn(snum
, hashinfo
->bhash_size
)];
116 spin_lock(&head
->lock
);
117 inet_bind_bucket_for_each(tb
, node
, &head
->chain
)
118 if (tb
->port
== snum
)
124 if (!hlist_empty(&tb
->owners
)) {
125 if (sk
->sk_reuse
> 1)
127 if (tb
->fastreuse
> 0 &&
128 sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
) {
132 if (bind_conflict(sk
, tb
))
138 if (!tb
&& (tb
= inet_bind_bucket_create(hashinfo
->bind_bucket_cachep
, head
, snum
)) == NULL
)
140 if (hlist_empty(&tb
->owners
)) {
141 if (sk
->sk_reuse
&& sk
->sk_state
!= TCP_LISTEN
)
145 } else if (tb
->fastreuse
&&
146 (!sk
->sk_reuse
|| sk
->sk_state
== TCP_LISTEN
))
149 if (!inet_csk(sk
)->icsk_bind_hash
)
150 inet_bind_hash(sk
, tb
, snum
);
151 BUG_TRAP(inet_csk(sk
)->icsk_bind_hash
== tb
);
155 spin_unlock(&head
->lock
);
161 EXPORT_SYMBOL_GPL(inet_csk_get_port
);
164 * Wait for an incoming connection, avoid race conditions. This must be called
165 * with the socket locked.
167 static int inet_csk_wait_for_connect(struct sock
*sk
, long timeo
)
169 struct inet_connection_sock
*icsk
= inet_csk(sk
);
174 * True wake-one mechanism for incoming connections: only
175 * one process gets woken up, not the 'whole herd'.
176 * Since we do not 'race & poll' for established sockets
177 * anymore, the common case will execute the loop only once.
179 * Subtle issue: "add_wait_queue_exclusive()" will be added
180 * after any current non-exclusive waiters, and we know that
181 * it will always _stay_ after any new non-exclusive waiters
182 * because all non-exclusive waiters are added at the
183 * beginning of the wait-queue. As such, it's ok to "drop"
184 * our exclusiveness temporarily when we get woken up without
185 * having to remove and re-insert us on the wait queue.
188 prepare_to_wait_exclusive(sk
->sk_sleep
, &wait
,
191 if (reqsk_queue_empty(&icsk
->icsk_accept_queue
))
192 timeo
= schedule_timeout(timeo
);
195 if (!reqsk_queue_empty(&icsk
->icsk_accept_queue
))
198 if (sk
->sk_state
!= TCP_LISTEN
)
200 err
= sock_intr_errno(timeo
);
201 if (signal_pending(current
))
207 finish_wait(sk
->sk_sleep
, &wait
);
212 * This will accept the next outstanding connection.
214 struct sock
*inet_csk_accept(struct sock
*sk
, int flags
, int *err
)
216 struct inet_connection_sock
*icsk
= inet_csk(sk
);
222 /* We need to make sure that this socket is listening,
223 * and that it has something pending.
226 if (sk
->sk_state
!= TCP_LISTEN
)
229 /* Find already established connection */
230 if (reqsk_queue_empty(&icsk
->icsk_accept_queue
)) {
231 long timeo
= sock_rcvtimeo(sk
, flags
& O_NONBLOCK
);
233 /* If this is a non blocking socket don't sleep */
238 error
= inet_csk_wait_for_connect(sk
, timeo
);
243 newsk
= reqsk_queue_get_child(&icsk
->icsk_accept_queue
, sk
);
244 BUG_TRAP(newsk
->sk_state
!= TCP_SYN_RECV
);
254 EXPORT_SYMBOL(inet_csk_accept
);
257 * Using different timers for retransmit, delayed acks and probes
258 * We may wish use just one timer maintaining a list of expire jiffies
261 void inet_csk_init_xmit_timers(struct sock
*sk
,
262 void (*retransmit_handler
)(unsigned long),
263 void (*delack_handler
)(unsigned long),
264 void (*keepalive_handler
)(unsigned long))
266 struct inet_connection_sock
*icsk
= inet_csk(sk
);
268 init_timer(&icsk
->icsk_retransmit_timer
);
269 init_timer(&icsk
->icsk_delack_timer
);
270 init_timer(&sk
->sk_timer
);
272 icsk
->icsk_retransmit_timer
.function
= retransmit_handler
;
273 icsk
->icsk_delack_timer
.function
= delack_handler
;
274 sk
->sk_timer
.function
= keepalive_handler
;
276 icsk
->icsk_retransmit_timer
.data
=
277 icsk
->icsk_delack_timer
.data
=
278 sk
->sk_timer
.data
= (unsigned long)sk
;
280 icsk
->icsk_pending
= icsk
->icsk_ack
.pending
= 0;
283 EXPORT_SYMBOL(inet_csk_init_xmit_timers
);
285 void inet_csk_clear_xmit_timers(struct sock
*sk
)
287 struct inet_connection_sock
*icsk
= inet_csk(sk
);
289 icsk
->icsk_pending
= icsk
->icsk_ack
.pending
= icsk
->icsk_ack
.blocked
= 0;
291 sk_stop_timer(sk
, &icsk
->icsk_retransmit_timer
);
292 sk_stop_timer(sk
, &icsk
->icsk_delack_timer
);
293 sk_stop_timer(sk
, &sk
->sk_timer
);
296 EXPORT_SYMBOL(inet_csk_clear_xmit_timers
);
298 void inet_csk_delete_keepalive_timer(struct sock
*sk
)
300 sk_stop_timer(sk
, &sk
->sk_timer
);
303 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer
);
305 void inet_csk_reset_keepalive_timer(struct sock
*sk
, unsigned long len
)
307 sk_reset_timer(sk
, &sk
->sk_timer
, jiffies
+ len
);
310 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer
);
312 struct dst_entry
* inet_csk_route_req(struct sock
*sk
,
313 const struct request_sock
*req
)
316 const struct inet_request_sock
*ireq
= inet_rsk(req
);
317 struct ip_options
*opt
= inet_rsk(req
)->opt
;
318 struct flowi fl
= { .oif
= sk
->sk_bound_dev_if
,
320 { .daddr
= ((opt
&& opt
->srr
) ?
323 .saddr
= ireq
->loc_addr
,
324 .tos
= RT_CONN_FLAGS(sk
) } },
325 .proto
= sk
->sk_protocol
,
327 { .sport
= inet_sk(sk
)->sport
,
328 .dport
= ireq
->rmt_port
} } };
330 security_req_classify_flow(req
, &fl
);
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 __be32 raddr
, const __be16 rport
,
346 const u32 rnd
, const u16 synq_hsize
)
348 return jhash_2words((__force u32
)raddr
, (__force 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 __be16 rport
, const __be32 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
));
514 security_inet_csk_clone(newsk
, req
);
519 EXPORT_SYMBOL_GPL(inet_csk_clone
);
522 * At this point, there should be no process reference to this
523 * socket, and thus no user references at all. Therefore we
524 * can assume the socket waitqueue is inactive and nobody will
525 * try to jump onto it.
527 void inet_csk_destroy_sock(struct sock
*sk
)
529 BUG_TRAP(sk
->sk_state
== TCP_CLOSE
);
530 BUG_TRAP(sock_flag(sk
, SOCK_DEAD
));
532 /* It cannot be in hash table! */
533 BUG_TRAP(sk_unhashed(sk
));
535 /* If it has not 0 inet_sk(sk)->num, it must be bound */
536 BUG_TRAP(!inet_sk(sk
)->num
|| inet_csk(sk
)->icsk_bind_hash
);
538 sk
->sk_prot
->destroy(sk
);
540 sk_stream_kill_queues(sk
);
542 xfrm_sk_free_policy(sk
);
544 sk_refcnt_debug_release(sk
);
546 atomic_dec(sk
->sk_prot
->orphan_count
);
550 EXPORT_SYMBOL(inet_csk_destroy_sock
);
552 int inet_csk_listen_start(struct sock
*sk
, const int nr_table_entries
)
554 struct inet_sock
*inet
= inet_sk(sk
);
555 struct inet_connection_sock
*icsk
= inet_csk(sk
);
556 int rc
= reqsk_queue_alloc(&icsk
->icsk_accept_queue
, nr_table_entries
);
561 sk
->sk_max_ack_backlog
= 0;
562 sk
->sk_ack_backlog
= 0;
563 inet_csk_delack_init(sk
);
565 /* There is race window here: we announce ourselves listening,
566 * but this transition is still not validated by get_port().
567 * It is OK, because this socket enters to hash table only
568 * after validation is complete.
570 sk
->sk_state
= TCP_LISTEN
;
571 if (!sk
->sk_prot
->get_port(sk
, inet
->num
)) {
572 inet
->sport
= htons(inet
->num
);
575 sk
->sk_prot
->hash(sk
);
580 sk
->sk_state
= TCP_CLOSE
;
581 __reqsk_queue_destroy(&icsk
->icsk_accept_queue
);
585 EXPORT_SYMBOL_GPL(inet_csk_listen_start
);
588 * This routine closes sockets which have been at least partially
589 * opened, but not yet accepted.
591 void inet_csk_listen_stop(struct sock
*sk
)
593 struct inet_connection_sock
*icsk
= inet_csk(sk
);
594 struct request_sock
*acc_req
;
595 struct request_sock
*req
;
597 inet_csk_delete_keepalive_timer(sk
);
599 /* make all the listen_opt local to us */
600 acc_req
= reqsk_queue_yank_acceptq(&icsk
->icsk_accept_queue
);
602 /* Following specs, it would be better either to send FIN
603 * (and enter FIN-WAIT-1, it is normal close)
604 * or to send active reset (abort).
605 * Certainly, it is pretty dangerous while synflood, but it is
606 * bad justification for our negligence 8)
607 * To be honest, we are not able to make either
608 * of the variants now. --ANK
610 reqsk_queue_destroy(&icsk
->icsk_accept_queue
);
612 while ((req
= acc_req
) != NULL
) {
613 struct sock
*child
= req
->sk
;
615 acc_req
= req
->dl_next
;
619 BUG_TRAP(!sock_owned_by_user(child
));
622 sk
->sk_prot
->disconnect(child
, O_NONBLOCK
);
626 atomic_inc(sk
->sk_prot
->orphan_count
);
628 inet_csk_destroy_sock(child
);
630 bh_unlock_sock(child
);
634 sk_acceptq_removed(sk
);
637 BUG_TRAP(!sk
->sk_ack_backlog
);
640 EXPORT_SYMBOL_GPL(inet_csk_listen_stop
);
642 void inet_csk_addr2sockaddr(struct sock
*sk
, struct sockaddr
*uaddr
)
644 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
645 const struct inet_sock
*inet
= inet_sk(sk
);
647 sin
->sin_family
= AF_INET
;
648 sin
->sin_addr
.s_addr
= inet
->daddr
;
649 sin
->sin_port
= inet
->dport
;
652 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr
);
654 int inet_csk_ctl_sock_create(struct socket
**sock
, unsigned short family
,
655 unsigned short type
, unsigned char protocol
)
657 int rc
= sock_create_kern(family
, type
, protocol
, sock
);
660 (*sock
)->sk
->sk_allocation
= GFP_ATOMIC
;
661 inet_sk((*sock
)->sk
)->uc_ttl
= -1;
663 * Unhash it so that IP input processing does not even see it,
664 * we do not wish this socket to see incoming packets.
666 (*sock
)->sk
->sk_prot
->unhash((*sock
)->sk
);
671 EXPORT_SYMBOL_GPL(inet_csk_ctl_sock_create
);
674 int inet_csk_compat_getsockopt(struct sock
*sk
, int level
, int optname
,
675 char __user
*optval
, int __user
*optlen
)
677 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
679 if (icsk
->icsk_af_ops
->compat_getsockopt
!= NULL
)
680 return icsk
->icsk_af_ops
->compat_getsockopt(sk
, level
, optname
,
682 return icsk
->icsk_af_ops
->getsockopt(sk
, level
, optname
,
686 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt
);
688 int inet_csk_compat_setsockopt(struct sock
*sk
, int level
, int optname
,
689 char __user
*optval
, int optlen
)
691 const struct inet_connection_sock
*icsk
= inet_csk(sk
);
693 if (icsk
->icsk_af_ops
->compat_setsockopt
!= NULL
)
694 return icsk
->icsk_af_ops
->compat_setsockopt(sk
, level
, optname
,
696 return icsk
->icsk_af_ops
->setsockopt(sk
, level
, optname
,
700 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt
);