blob: e9b151b3a598e46dfd162bdc4645bdb083d48e83 [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
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090072 /* If peer does not open window for long time, or did not transmit
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * 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
Pavel Emelianove4fd5da2007-05-29 13:19:18 -070081 if (tcp_too_many_orphans(sk, orphans)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 if (net_ratelimit())
83 printk(KERN_INFO "Out of socket memory\n");
84
85 /* Catch exceptional cases, when connection requires reset.
86 * 1. Last segment was sent recently. */
87 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
88 /* 2. Window is closed. */
89 (!tp->snd_wnd && !tp->packets_out))
90 do_reset = 1;
91 if (do_reset)
92 tcp_send_active_reset(sk, GFP_ATOMIC);
93 tcp_done(sk);
94 NET_INC_STATS_BH(LINUX_MIB_TCPABORTONMEMORY);
95 return 1;
96 }
97 return 0;
98}
99
100/* Calculate maximal number or retries on an orphaned socket. */
101static int tcp_orphan_retries(struct sock *sk, int alive)
102{
103 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
104
105 /* We know from an ICMP that something is wrong. */
106 if (sk->sk_err_soft && !alive)
107 retries = 0;
108
109 /* However, if socket sent something recently, select some safe
110 * number of retries. 8 corresponds to >100 seconds with minimal
111 * RTO of 200msec. */
112 if (retries == 0 && alive)
113 retries = 8;
114 return retries;
115}
116
117/* A write timeout has occurred. Process the after effects. */
118static int tcp_write_timeout(struct sock *sk)
119{
John Heffner5d424d52006-03-20 17:53:41 -0800120 struct inet_connection_sock *icsk = inet_csk(sk);
121 struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 int retry_until;
John Heffner5d424d52006-03-20 17:53:41 -0800123 int mss;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700126 if (icsk->icsk_retransmits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 dst_negative_advice(&sk->sk_dst_cache);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700128 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700130 if (icsk->icsk_retransmits >= sysctl_tcp_retries1) {
John Heffner5d424d52006-03-20 17:53:41 -0800131 /* Black hole detection */
132 if (sysctl_tcp_mtu_probing) {
133 if (!icsk->icsk_mtup.enabled) {
134 icsk->icsk_mtup.enabled = 1;
135 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
136 } else {
137 mss = min(sysctl_tcp_base_mss,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900138 tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low)/2);
John Heffner5d424d52006-03-20 17:53:41 -0800139 mss = max(mss, 68 - tp->tcp_header_len);
140 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
141 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
142 }
143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145 dst_negative_advice(&sk->sk_dst_cache);
146 }
147
148 retry_until = sysctl_tcp_retries2;
149 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700150 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 retry_until = tcp_orphan_retries(sk, alive);
153
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700154 if (tcp_out_of_resources(sk, alive || icsk->icsk_retransmits < retry_until))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 return 1;
156 }
157 }
158
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700159 if (icsk->icsk_retransmits >= retry_until) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 /* Has it gone just too far? */
161 tcp_write_err(sk);
162 return 1;
163 }
164 return 0;
165}
166
167static void tcp_delack_timer(unsigned long data)
168{
169 struct sock *sk = (struct sock*)data;
170 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700171 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
173 bh_lock_sock(sk);
174 if (sock_owned_by_user(sk)) {
175 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700176 icsk->icsk_ack.blocked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOCKED);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700178 sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 goto out_unlock;
180 }
181
182 sk_stream_mem_reclaim(sk);
183
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700184 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 goto out;
186
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700187 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
188 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 goto out;
190 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700191 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
David S. Millerb03efcf2005-07-08 14:57:23 -0700193 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 struct sk_buff *skb;
195
David S. Millerb03efcf2005-07-08 14:57:23 -0700196 NET_INC_STATS_BH(LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
199 sk->sk_backlog_rcv(sk, skb);
200
201 tp->ucopy.memory = 0;
202 }
203
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700204 if (inet_csk_ack_scheduled(sk)) {
205 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700207 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 } else {
209 /* Delayed ACK missed: leave pingpong mode and
210 * deflate ATO.
211 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700212 icsk->icsk_ack.pingpong = 0;
213 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215 tcp_send_ack(sk);
216 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKS);
217 }
218 TCP_CHECK_TIMER(sk);
219
220out:
221 if (tcp_memory_pressure)
222 sk_stream_mem_reclaim(sk);
223out_unlock:
224 bh_unlock_sock(sk);
225 sock_put(sk);
226}
227
228static void tcp_probe_timer(struct sock *sk)
229{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300230 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 struct tcp_sock *tp = tcp_sk(sk);
232 int max_probes;
233
David S. Millerfe067e82007-03-07 12:12:44 -0800234 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300235 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 return;
237 }
238
239 /* *WARNING* RFC 1122 forbids this
240 *
241 * It doesn't AFAIK, because we kill the retransmit timer -AK
242 *
243 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
244 * this behaviour in Solaris down as a bug fix. [AC]
245 *
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300246 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 * even if they advertise zero window. Hence, connection is killed only
248 * if we received no ACKs for normal connection timeout. It is not killed
249 * only because window stays zero for some time, window may be zero
250 * until armageddon and even later. We are in full accordance
251 * with RFCs, only probe timer combines both retransmission timeout
252 * and probe timeout in one bottle. --ANK
253 */
254 max_probes = sysctl_tcp_retries2;
255
256 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700257 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 max_probes = tcp_orphan_retries(sk, alive);
260
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300261 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 return;
263 }
264
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300265 if (icsk->icsk_probes_out > max_probes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 tcp_write_err(sk);
267 } else {
268 /* Only send another probe if we didn't close things up. */
269 tcp_send_probe0(sk);
270 }
271}
272
273/*
274 * The TCP retransmit timer.
275 */
276
277static void tcp_retransmit_timer(struct sock *sk)
278{
279 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700280 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282 if (!tp->packets_out)
283 goto out;
284
David S. Millerfe067e82007-03-07 12:12:44 -0800285 BUG_TRAP(!tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
288 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
289 /* Receiver dastardly shrinks window. Our retransmits
290 * become zero probes, but we should not timeout this
291 * connection. If the socket is an orphan, time it out,
292 * we cannot allow such beasts to hang infinitely.
293 */
294#ifdef TCP_DEBUG
Eric Dumazet274707c2007-06-05 00:16:12 -0700295 if (1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 struct inet_sock *inet = inet_sk(sk);
Eric Dumazet274707c2007-06-05 00:16:12 -0700297 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Treason uncloaked! Peer %u.%u.%u.%u:%u/%u shrinks window %u:%u. Repaired.\n",
Al Viro714e85b2006-11-14 20:51:49 -0800298 NIPQUAD(inet->daddr), ntohs(inet->dport),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 inet->num, tp->snd_una, tp->snd_nxt);
300 }
301#endif
302 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
303 tcp_write_err(sk);
304 goto out;
305 }
306 tcp_enter_loss(sk, 0);
David S. Millerfe067e82007-03-07 12:12:44 -0800307 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 __sk_dst_reset(sk);
309 goto out_reset_timer;
310 }
311
312 if (tcp_write_timeout(sk))
313 goto out;
314
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700315 if (icsk->icsk_retransmits == 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300316 if (icsk->icsk_ca_state == TCP_CA_Disorder ||
317 icsk->icsk_ca_state == TCP_CA_Recovery) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 if (tp->rx_opt.sack_ok) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300319 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 NET_INC_STATS_BH(LINUX_MIB_TCPSACKRECOVERYFAIL);
321 else
322 NET_INC_STATS_BH(LINUX_MIB_TCPSACKFAILURES);
323 } else {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300324 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 NET_INC_STATS_BH(LINUX_MIB_TCPRENORECOVERYFAIL);
326 else
327 NET_INC_STATS_BH(LINUX_MIB_TCPRENOFAILURES);
328 }
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300329 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSFAILURES);
331 } else {
332 NET_INC_STATS_BH(LINUX_MIB_TCPTIMEOUTS);
333 }
334 }
335
336 if (tcp_use_frto(sk)) {
337 tcp_enter_frto(sk);
338 } else {
339 tcp_enter_loss(sk, 0);
340 }
341
David S. Millerfe067e82007-03-07 12:12:44 -0800342 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 /* Retransmission failed because of local congestion,
344 * do not backoff.
345 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700346 if (!icsk->icsk_retransmits)
347 icsk->icsk_retransmits = 1;
348 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700349 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
350 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 goto out;
352 }
353
354 /* Increase the timeout each time we retransmit. Note that
355 * we do not increase the rtt estimate. rto is initialized
356 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
357 * that doubling rto each time is the least we can get away with.
358 * In KA9Q, Karn uses this for the first few times, and then
359 * goes to quadratic. netBSD doubles, but only goes up to *64,
360 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
361 * defined in the protocol as the maximum possible RTT. I guess
362 * we'll have to use something other than TCP to talk to the
363 * University of Mars.
364 *
365 * PAWS allows us longer timeouts and large windows, so once
366 * implemented ftp to mars will work nicely. We will have to fix
367 * the 120 second clamps though!
368 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700369 icsk->icsk_backoff++;
370 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372out_reset_timer:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700373 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700374 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700375 if (icsk->icsk_retransmits > sysctl_tcp_retries1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 __sk_dst_reset(sk);
377
378out:;
379}
380
381static void tcp_write_timer(unsigned long data)
382{
383 struct sock *sk = (struct sock*)data;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700384 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 int event;
386
387 bh_lock_sock(sk);
388 if (sock_owned_by_user(sk)) {
389 /* Try again later */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700390 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 goto out_unlock;
392 }
393
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700394 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 goto out;
396
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700397 if (time_after(icsk->icsk_timeout, jiffies)) {
398 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 goto out;
400 }
401
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700402 event = icsk->icsk_pending;
403 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405 switch (event) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700406 case ICSK_TIME_RETRANS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 tcp_retransmit_timer(sk);
408 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700409 case ICSK_TIME_PROBE0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 tcp_probe_timer(sk);
411 break;
412 }
413 TCP_CHECK_TIMER(sk);
414
415out:
416 sk_stream_mem_reclaim(sk);
417out_unlock:
418 bh_unlock_sock(sk);
419 sock_put(sk);
420}
421
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700422/*
423 * Timer for listening sockets
424 */
425
426static void tcp_synack_timer(struct sock *sk)
427{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700428 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
429 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432void tcp_set_keepalive(struct sock *sk, int val)
433{
434 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
435 return;
436
437 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700438 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700440 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441}
442
443
444static void tcp_keepalive_timer (unsigned long data)
445{
446 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300447 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 struct tcp_sock *tp = tcp_sk(sk);
449 __u32 elapsed;
450
451 /* Only process if socket is not in use. */
452 bh_lock_sock(sk);
453 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900454 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700455 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 goto out;
457 }
458
459 if (sk->sk_state == TCP_LISTEN) {
460 tcp_synack_timer(sk);
461 goto out;
462 }
463
464 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
465 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700466 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468 if (tmo > 0) {
469 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
470 goto out;
471 }
472 }
473 tcp_send_active_reset(sk, GFP_ATOMIC);
474 goto death;
475 }
476
477 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
478 goto out;
479
480 elapsed = keepalive_time_when(tp);
481
482 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800483 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 goto resched;
485
486 elapsed = tcp_time_stamp - tp->rcv_tstamp;
487
488 if (elapsed >= keepalive_time_when(tp)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300489 if ((!tp->keepalive_probes && icsk->icsk_probes_out >= sysctl_tcp_keepalive_probes) ||
490 (tp->keepalive_probes && icsk->icsk_probes_out >= tp->keepalive_probes)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 tcp_send_active_reset(sk, GFP_ATOMIC);
492 tcp_write_err(sk);
493 goto out;
494 }
495 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300496 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 elapsed = keepalive_intvl_when(tp);
498 } else {
499 /* If keepalive was lost due to local congestion,
500 * try harder.
501 */
502 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
503 }
504 } else {
505 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
506 elapsed = keepalive_time_when(tp) - elapsed;
507 }
508
509 TCP_CHECK_TIMER(sk);
510 sk_stream_mem_reclaim(sk);
511
512resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700513 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 goto out;
515
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900516death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 tcp_done(sk);
518
519out:
520 bh_unlock_sock(sk);
521 sock_put(sk);
522}