blob: fb09ade5897b355febf883a9ebad3f6038686501 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * Version: $Id: tcp_timer.c,v 1.88 2002/02/01 22:01:04 davem Exp $
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
16 * Linus Torvalds, <torvalds@cs.helsinki.fi>
17 * Alan Cox, <gw4pts@gw4pts.ampr.org>
18 * Matthew Dillon, <dillon@apollo.west.oic.com>
19 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
20 * Jorge Cwik, <jorge@laser.satlink.net>
21 */
22
23#include <linux/module.h>
24#include <net/tcp.h>
25
Brian Haleyab32ea52006-09-22 14:15:41 -070026int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES;
27int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES;
28int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME;
29int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES;
30int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
31int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
32int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
33int sysctl_tcp_orphan_retries __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35static void tcp_write_timer(unsigned long);
36static void tcp_delack_timer(unsigned long);
37static void tcp_keepalive_timer (unsigned long data);
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039void tcp_init_xmit_timers(struct sock *sk)
40{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -070041 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
42 &tcp_keepalive_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043}
44
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070045EXPORT_SYMBOL(tcp_init_xmit_timers);
46
Linus Torvalds1da177e2005-04-16 15:20:36 -070047static void tcp_write_err(struct sock *sk)
48{
49 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
50 sk->sk_error_report(sk);
51
52 tcp_done(sk);
53 NET_INC_STATS_BH(LINUX_MIB_TCPABORTONTIMEOUT);
54}
55
56/* Do not allow orphaned sockets to eat all our resources.
57 * This is direct violation of TCP specs, but it is required
58 * to prevent DoS attacks. It is called when a retransmission timeout
59 * or zero probe timeout occurs on orphaned socket.
60 *
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080061 * Criteria is still not confirmed experimentally and may change.
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * We kill the socket, if:
63 * 1. If number of orphaned sockets exceeds an administratively configured
64 * limit.
65 * 2. If we have strong memory pressure.
66 */
67static int tcp_out_of_resources(struct sock *sk, int do_reset)
68{
69 struct tcp_sock *tp = tcp_sk(sk);
70 int orphans = atomic_read(&tcp_orphan_count);
71
72 /* If peer does not open window for long time, or did not transmit
73 * anything for long time, penalize it. */
74 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
75 orphans <<= 1;
76
77 /* If some dubious ICMP arrived, penalize even more. */
78 if (sk->sk_err_soft)
79 orphans <<= 1;
80
81 if (orphans >= sysctl_tcp_max_orphans ||
82 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
83 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])) {
84 if (net_ratelimit())
85 printk(KERN_INFO "Out of socket memory\n");
86
87 /* Catch exceptional cases, when connection requires reset.
88 * 1. Last segment was sent recently. */
89 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
90 /* 2. Window is closed. */
91 (!tp->snd_wnd && !tp->packets_out))
92 do_reset = 1;
93 if (do_reset)
94 tcp_send_active_reset(sk, GFP_ATOMIC);
95 tcp_done(sk);
96 NET_INC_STATS_BH(LINUX_MIB_TCPABORTONMEMORY);
97 return 1;
98 }
99 return 0;
100}
101
102/* Calculate maximal number or retries on an orphaned socket. */
103static int tcp_orphan_retries(struct sock *sk, int alive)
104{
105 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
106
107 /* We know from an ICMP that something is wrong. */
108 if (sk->sk_err_soft && !alive)
109 retries = 0;
110
111 /* However, if socket sent something recently, select some safe
112 * number of retries. 8 corresponds to >100 seconds with minimal
113 * RTO of 200msec. */
114 if (retries == 0 && alive)
115 retries = 8;
116 return retries;
117}
118
119/* A write timeout has occurred. Process the after effects. */
120static int tcp_write_timeout(struct sock *sk)
121{
John Heffner5d424d52006-03-20 17:53:41 -0800122 struct inet_connection_sock *icsk = inet_csk(sk);
123 struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 int retry_until;
John Heffner5d424d52006-03-20 17:53:41 -0800125 int mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700128 if (icsk->icsk_retransmits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 dst_negative_advice(&sk->sk_dst_cache);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700130 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700132 if (icsk->icsk_retransmits >= sysctl_tcp_retries1) {
John Heffner5d424d52006-03-20 17:53:41 -0800133 /* Black hole detection */
134 if (sysctl_tcp_mtu_probing) {
135 if (!icsk->icsk_mtup.enabled) {
136 icsk->icsk_mtup.enabled = 1;
137 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
138 } else {
139 mss = min(sysctl_tcp_base_mss,
140 tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)/2);
141 mss = max(mss, 68 - tp->tcp_header_len);
142 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
143 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
144 }
145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 dst_negative_advice(&sk->sk_dst_cache);
148 }
149
150 retry_until = sysctl_tcp_retries2;
151 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700152 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
154 retry_until = tcp_orphan_retries(sk, alive);
155
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700156 if (tcp_out_of_resources(sk, alive || icsk->icsk_retransmits < retry_until))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 return 1;
158 }
159 }
160
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700161 if (icsk->icsk_retransmits >= retry_until) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 /* Has it gone just too far? */
163 tcp_write_err(sk);
164 return 1;
165 }
166 return 0;
167}
168
169static void tcp_delack_timer(unsigned long data)
170{
171 struct sock *sk = (struct sock*)data;
172 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700173 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175 bh_lock_sock(sk);
176 if (sock_owned_by_user(sk)) {
177 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700178 icsk->icsk_ack.blocked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOCKED);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700180 sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 goto out_unlock;
182 }
183
184 sk_stream_mem_reclaim(sk);
185
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700186 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 goto out;
188
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700189 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
190 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 goto out;
192 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700193 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
David S. Millerb03efcf2005-07-08 14:57:23 -0700195 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 struct sk_buff *skb;
197
David S. Millerb03efcf2005-07-08 14:57:23 -0700198 NET_INC_STATS_BH(LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
201 sk->sk_backlog_rcv(sk, skb);
202
203 tp->ucopy.memory = 0;
204 }
205
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700206 if (inet_csk_ack_scheduled(sk)) {
207 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700209 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 } else {
211 /* Delayed ACK missed: leave pingpong mode and
212 * deflate ATO.
213 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700214 icsk->icsk_ack.pingpong = 0;
215 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 }
217 tcp_send_ack(sk);
218 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKS);
219 }
220 TCP_CHECK_TIMER(sk);
221
222out:
223 if (tcp_memory_pressure)
224 sk_stream_mem_reclaim(sk);
225out_unlock:
226 bh_unlock_sock(sk);
227 sock_put(sk);
228}
229
230static void tcp_probe_timer(struct sock *sk)
231{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300232 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 struct tcp_sock *tp = tcp_sk(sk);
234 int max_probes;
235
236 if (tp->packets_out || !sk->sk_send_head) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300237 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 return;
239 }
240
241 /* *WARNING* RFC 1122 forbids this
242 *
243 * It doesn't AFAIK, because we kill the retransmit timer -AK
244 *
245 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
246 * this behaviour in Solaris down as a bug fix. [AC]
247 *
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300248 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 * even if they advertise zero window. Hence, connection is killed only
250 * if we received no ACKs for normal connection timeout. It is not killed
251 * only because window stays zero for some time, window may be zero
252 * until armageddon and even later. We are in full accordance
253 * with RFCs, only probe timer combines both retransmission timeout
254 * and probe timeout in one bottle. --ANK
255 */
256 max_probes = sysctl_tcp_retries2;
257
258 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700259 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 max_probes = tcp_orphan_retries(sk, alive);
262
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300263 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return;
265 }
266
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300267 if (icsk->icsk_probes_out > max_probes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 tcp_write_err(sk);
269 } else {
270 /* Only send another probe if we didn't close things up. */
271 tcp_send_probe0(sk);
272 }
273}
274
275/*
276 * The TCP retransmit timer.
277 */
278
279static void tcp_retransmit_timer(struct sock *sk)
280{
281 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700282 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 if (!tp->packets_out)
285 goto out;
286
287 BUG_TRAP(!skb_queue_empty(&sk->sk_write_queue));
288
289 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
290 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
291 /* Receiver dastardly shrinks window. Our retransmits
292 * become zero probes, but we should not timeout this
293 * connection. If the socket is an orphan, time it out,
294 * we cannot allow such beasts to hang infinitely.
295 */
296#ifdef TCP_DEBUG
297 if (net_ratelimit()) {
298 struct inet_sock *inet = inet_sk(sk);
299 printk(KERN_DEBUG "TCP: Treason uncloaked! Peer %u.%u.%u.%u:%u/%u shrinks window %u:%u. Repaired.\n",
300 NIPQUAD(inet->daddr), htons(inet->dport),
301 inet->num, tp->snd_una, tp->snd_nxt);
302 }
303#endif
304 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
305 tcp_write_err(sk);
306 goto out;
307 }
308 tcp_enter_loss(sk, 0);
309 tcp_retransmit_skb(sk, skb_peek(&sk->sk_write_queue));
310 __sk_dst_reset(sk);
311 goto out_reset_timer;
312 }
313
314 if (tcp_write_timeout(sk))
315 goto out;
316
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700317 if (icsk->icsk_retransmits == 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300318 if (icsk->icsk_ca_state == TCP_CA_Disorder ||
319 icsk->icsk_ca_state == TCP_CA_Recovery) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 if (tp->rx_opt.sack_ok) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300321 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 NET_INC_STATS_BH(LINUX_MIB_TCPSACKRECOVERYFAIL);
323 else
324 NET_INC_STATS_BH(LINUX_MIB_TCPSACKFAILURES);
325 } else {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300326 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 NET_INC_STATS_BH(LINUX_MIB_TCPRENORECOVERYFAIL);
328 else
329 NET_INC_STATS_BH(LINUX_MIB_TCPRENOFAILURES);
330 }
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300331 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSFAILURES);
333 } else {
334 NET_INC_STATS_BH(LINUX_MIB_TCPTIMEOUTS);
335 }
336 }
337
338 if (tcp_use_frto(sk)) {
339 tcp_enter_frto(sk);
340 } else {
341 tcp_enter_loss(sk, 0);
342 }
343
344 if (tcp_retransmit_skb(sk, skb_peek(&sk->sk_write_queue)) > 0) {
345 /* Retransmission failed because of local congestion,
346 * do not backoff.
347 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700348 if (!icsk->icsk_retransmits)
349 icsk->icsk_retransmits = 1;
350 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700351 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
352 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 goto out;
354 }
355
356 /* Increase the timeout each time we retransmit. Note that
357 * we do not increase the rtt estimate. rto is initialized
358 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
359 * that doubling rto each time is the least we can get away with.
360 * In KA9Q, Karn uses this for the first few times, and then
361 * goes to quadratic. netBSD doubles, but only goes up to *64,
362 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
363 * defined in the protocol as the maximum possible RTT. I guess
364 * we'll have to use something other than TCP to talk to the
365 * University of Mars.
366 *
367 * PAWS allows us longer timeouts and large windows, so once
368 * implemented ftp to mars will work nicely. We will have to fix
369 * the 120 second clamps though!
370 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700371 icsk->icsk_backoff++;
372 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
374out_reset_timer:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700375 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700376 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700377 if (icsk->icsk_retransmits > sysctl_tcp_retries1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 __sk_dst_reset(sk);
379
380out:;
381}
382
383static void tcp_write_timer(unsigned long data)
384{
385 struct sock *sk = (struct sock*)data;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700386 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 int event;
388
389 bh_lock_sock(sk);
390 if (sock_owned_by_user(sk)) {
391 /* Try again later */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700392 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 goto out_unlock;
394 }
395
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700396 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 goto out;
398
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700399 if (time_after(icsk->icsk_timeout, jiffies)) {
400 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 goto out;
402 }
403
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700404 event = icsk->icsk_pending;
405 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 switch (event) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700408 case ICSK_TIME_RETRANS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 tcp_retransmit_timer(sk);
410 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700411 case ICSK_TIME_PROBE0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 tcp_probe_timer(sk);
413 break;
414 }
415 TCP_CHECK_TIMER(sk);
416
417out:
418 sk_stream_mem_reclaim(sk);
419out_unlock:
420 bh_unlock_sock(sk);
421 sock_put(sk);
422}
423
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700424/*
425 * Timer for listening sockets
426 */
427
428static void tcp_synack_timer(struct sock *sk)
429{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700430 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
431 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434void tcp_set_keepalive(struct sock *sk, int val)
435{
436 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
437 return;
438
439 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700440 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700442 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
445
446static void tcp_keepalive_timer (unsigned long data)
447{
448 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300449 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 struct tcp_sock *tp = tcp_sk(sk);
451 __u32 elapsed;
452
453 /* Only process if socket is not in use. */
454 bh_lock_sock(sk);
455 if (sock_owned_by_user(sk)) {
456 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700457 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 goto out;
459 }
460
461 if (sk->sk_state == TCP_LISTEN) {
462 tcp_synack_timer(sk);
463 goto out;
464 }
465
466 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
467 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700468 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 if (tmo > 0) {
471 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
472 goto out;
473 }
474 }
475 tcp_send_active_reset(sk, GFP_ATOMIC);
476 goto death;
477 }
478
479 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
480 goto out;
481
482 elapsed = keepalive_time_when(tp);
483
484 /* It is alive without keepalive 8) */
485 if (tp->packets_out || sk->sk_send_head)
486 goto resched;
487
488 elapsed = tcp_time_stamp - tp->rcv_tstamp;
489
490 if (elapsed >= keepalive_time_when(tp)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300491 if ((!tp->keepalive_probes && icsk->icsk_probes_out >= sysctl_tcp_keepalive_probes) ||
492 (tp->keepalive_probes && icsk->icsk_probes_out >= tp->keepalive_probes)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 tcp_send_active_reset(sk, GFP_ATOMIC);
494 tcp_write_err(sk);
495 goto out;
496 }
497 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300498 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 elapsed = keepalive_intvl_when(tp);
500 } else {
501 /* If keepalive was lost due to local congestion,
502 * try harder.
503 */
504 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
505 }
506 } else {
507 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
508 elapsed = keepalive_time_when(tp) - elapsed;
509 }
510
511 TCP_CHECK_TIMER(sk);
512 sk_stream_mem_reclaim(sk);
513
514resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700515 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 goto out;
517
518death:
519 tcp_done(sk);
520
521out:
522 bh_unlock_sock(sk);
523 sock_put(sk);
524}