blob: 63ed9d6830e7b80f77f3619168756742f6ef3bc9 [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
Eric Dumazetce55dd32007-12-21 01:50:43 -0800117static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
118{
Eric Dumazetce55dd32007-12-21 01:50:43 -0800119 /* Black hole detection */
120 if (sysctl_tcp_mtu_probing) {
121 if (!icsk->icsk_mtup.enabled) {
122 icsk->icsk_mtup.enabled = 1;
123 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
124 } else {
125 struct tcp_sock *tp = tcp_sk(sk);
David S. Miller829942c2007-12-21 04:29:16 -0800126 int mss;
127
Eric Dumazet8beb5c52007-12-21 05:58:29 -0800128 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
Eric Dumazetce55dd32007-12-21 01:50:43 -0800129 mss = min(sysctl_tcp_base_mss, mss);
130 mss = max(mss, 68 - tp->tcp_header_len);
131 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
132 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
133 }
134 }
135}
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137/* A write timeout has occurred. Process the after effects. */
138static int tcp_write_timeout(struct sock *sk)
139{
John Heffner5d424d52006-03-20 17:53:41 -0800140 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 int retry_until;
142
143 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700144 if (icsk->icsk_retransmits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 dst_negative_advice(&sk->sk_dst_cache);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700146 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 } else {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700148 if (icsk->icsk_retransmits >= sysctl_tcp_retries1) {
John Heffner5d424d52006-03-20 17:53:41 -0800149 /* Black hole detection */
Eric Dumazetce55dd32007-12-21 01:50:43 -0800150 tcp_mtu_probing(icsk, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
152 dst_negative_advice(&sk->sk_dst_cache);
153 }
154
155 retry_until = sysctl_tcp_retries2;
156 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700157 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 retry_until = tcp_orphan_retries(sk, alive);
160
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700161 if (tcp_out_of_resources(sk, alive || icsk->icsk_retransmits < retry_until))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 return 1;
163 }
164 }
165
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700166 if (icsk->icsk_retransmits >= retry_until) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 /* Has it gone just too far? */
168 tcp_write_err(sk);
169 return 1;
170 }
171 return 0;
172}
173
174static void tcp_delack_timer(unsigned long data)
175{
176 struct sock *sk = (struct sock*)data;
177 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700178 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
180 bh_lock_sock(sk);
181 if (sock_owned_by_user(sk)) {
182 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700183 icsk->icsk_ack.blocked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOCKED);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700185 sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 goto out_unlock;
187 }
188
David S. Miller9993e7d2008-01-10 21:56:38 -0800189 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700191 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 goto out;
193
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700194 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
195 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 goto out;
197 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700198 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
David S. Millerb03efcf2005-07-08 14:57:23 -0700200 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 struct sk_buff *skb;
202
David S. Millerb03efcf2005-07-08 14:57:23 -0700203 NET_INC_STATS_BH(LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
206 sk->sk_backlog_rcv(sk, skb);
207
208 tp->ucopy.memory = 0;
209 }
210
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700211 if (inet_csk_ack_scheduled(sk)) {
212 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700214 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 } else {
216 /* Delayed ACK missed: leave pingpong mode and
217 * deflate ATO.
218 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700219 icsk->icsk_ack.pingpong = 0;
220 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 }
222 tcp_send_ack(sk);
223 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKS);
224 }
225 TCP_CHECK_TIMER(sk);
226
227out:
228 if (tcp_memory_pressure)
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800229 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230out_unlock:
231 bh_unlock_sock(sk);
232 sock_put(sk);
233}
234
235static void tcp_probe_timer(struct sock *sk)
236{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300237 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 struct tcp_sock *tp = tcp_sk(sk);
239 int max_probes;
240
David S. Millerfe067e82007-03-07 12:12:44 -0800241 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300242 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 return;
244 }
245
246 /* *WARNING* RFC 1122 forbids this
247 *
248 * It doesn't AFAIK, because we kill the retransmit timer -AK
249 *
250 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
251 * this behaviour in Solaris down as a bug fix. [AC]
252 *
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300253 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 * even if they advertise zero window. Hence, connection is killed only
255 * if we received no ACKs for normal connection timeout. It is not killed
256 * only because window stays zero for some time, window may be zero
257 * until armageddon and even later. We are in full accordance
258 * with RFCs, only probe timer combines both retransmission timeout
259 * and probe timeout in one bottle. --ANK
260 */
261 max_probes = sysctl_tcp_retries2;
262
263 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700264 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 max_probes = tcp_orphan_retries(sk, alive);
267
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300268 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return;
270 }
271
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300272 if (icsk->icsk_probes_out > max_probes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 tcp_write_err(sk);
274 } else {
275 /* Only send another probe if we didn't close things up. */
276 tcp_send_probe0(sk);
277 }
278}
279
280/*
281 * The TCP retransmit timer.
282 */
283
284static void tcp_retransmit_timer(struct sock *sk)
285{
286 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700287 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
289 if (!tp->packets_out)
290 goto out;
291
David S. Millerfe067e82007-03-07 12:12:44 -0800292 BUG_TRAP(!tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
294 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
295 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
296 /* Receiver dastardly shrinks window. Our retransmits
297 * become zero probes, but we should not timeout this
298 * connection. If the socket is an orphan, time it out,
299 * we cannot allow such beasts to hang infinitely.
300 */
301#ifdef TCP_DEBUG
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700302 struct inet_sock *inet = inet_sk(sk);
303 if (sk->sk_family == AF_INET) {
YOSHIFUJI Hideakia7d632b2008-04-14 04:09:00 -0700304 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Treason uncloaked! Peer " NIPQUAD_FMT ":%u/%u shrinks window %u:%u. Repaired.\n",
Al Viro714e85b2006-11-14 20:51:49 -0800305 NIPQUAD(inet->daddr), ntohs(inet->dport),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 inet->num, tp->snd_una, tp->snd_nxt);
307 }
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700308#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
309 else if (sk->sk_family == AF_INET6) {
310 struct ipv6_pinfo *np = inet6_sk(sk);
311 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Treason uncloaked! Peer " NIP6_FMT ":%u/%u shrinks window %u:%u. Repaired.\n",
312 NIP6(np->daddr), ntohs(inet->dport),
313 inet->num, tp->snd_una, tp->snd_nxt);
314 }
315#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316#endif
317 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
318 tcp_write_err(sk);
319 goto out;
320 }
321 tcp_enter_loss(sk, 0);
David S. Millerfe067e82007-03-07 12:12:44 -0800322 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 __sk_dst_reset(sk);
324 goto out_reset_timer;
325 }
326
327 if (tcp_write_timeout(sk))
328 goto out;
329
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700330 if (icsk->icsk_retransmits == 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300331 if (icsk->icsk_ca_state == TCP_CA_Disorder ||
332 icsk->icsk_ca_state == TCP_CA_Recovery) {
Ilpo Järvinene60402d2007-08-09 15:14:46 +0300333 if (tcp_is_sack(tp)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300334 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 NET_INC_STATS_BH(LINUX_MIB_TCPSACKRECOVERYFAIL);
336 else
337 NET_INC_STATS_BH(LINUX_MIB_TCPSACKFAILURES);
338 } else {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300339 if (icsk->icsk_ca_state == TCP_CA_Recovery)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 NET_INC_STATS_BH(LINUX_MIB_TCPRENORECOVERYFAIL);
341 else
342 NET_INC_STATS_BH(LINUX_MIB_TCPRENOFAILURES);
343 }
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300344 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 NET_INC_STATS_BH(LINUX_MIB_TCPLOSSFAILURES);
346 } else {
347 NET_INC_STATS_BH(LINUX_MIB_TCPTIMEOUTS);
348 }
349 }
350
351 if (tcp_use_frto(sk)) {
352 tcp_enter_frto(sk);
353 } else {
354 tcp_enter_loss(sk, 0);
355 }
356
David S. Millerfe067e82007-03-07 12:12:44 -0800357 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 /* Retransmission failed because of local congestion,
359 * do not backoff.
360 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700361 if (!icsk->icsk_retransmits)
362 icsk->icsk_retransmits = 1;
363 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700364 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
365 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 goto out;
367 }
368
369 /* Increase the timeout each time we retransmit. Note that
370 * we do not increase the rtt estimate. rto is initialized
371 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
372 * that doubling rto each time is the least we can get away with.
373 * In KA9Q, Karn uses this for the first few times, and then
374 * goes to quadratic. netBSD doubles, but only goes up to *64,
375 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
376 * defined in the protocol as the maximum possible RTT. I guess
377 * we'll have to use something other than TCP to talk to the
378 * University of Mars.
379 *
380 * PAWS allows us longer timeouts and large windows, so once
381 * implemented ftp to mars will work nicely. We will have to fix
382 * the 120 second clamps though!
383 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700384 icsk->icsk_backoff++;
385 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387out_reset_timer:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700388 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700389 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700390 if (icsk->icsk_retransmits > sysctl_tcp_retries1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 __sk_dst_reset(sk);
392
393out:;
394}
395
396static void tcp_write_timer(unsigned long data)
397{
398 struct sock *sk = (struct sock*)data;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700399 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 int event;
401
402 bh_lock_sock(sk);
403 if (sock_owned_by_user(sk)) {
404 /* Try again later */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700405 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 goto out_unlock;
407 }
408
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700409 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 goto out;
411
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700412 if (time_after(icsk->icsk_timeout, jiffies)) {
413 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 goto out;
415 }
416
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700417 event = icsk->icsk_pending;
418 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 switch (event) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700421 case ICSK_TIME_RETRANS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 tcp_retransmit_timer(sk);
423 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700424 case ICSK_TIME_PROBE0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 tcp_probe_timer(sk);
426 break;
427 }
428 TCP_CHECK_TIMER(sk);
429
430out:
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800431 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432out_unlock:
433 bh_unlock_sock(sk);
434 sock_put(sk);
435}
436
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700437/*
438 * Timer for listening sockets
439 */
440
441static void tcp_synack_timer(struct sock *sk)
442{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700443 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
444 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447void tcp_set_keepalive(struct sock *sk, int val)
448{
449 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
450 return;
451
452 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700453 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700455 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
458
459static void tcp_keepalive_timer (unsigned long data)
460{
461 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300462 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 struct tcp_sock *tp = tcp_sk(sk);
464 __u32 elapsed;
465
466 /* Only process if socket is not in use. */
467 bh_lock_sock(sk);
468 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900469 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700470 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 goto out;
472 }
473
474 if (sk->sk_state == TCP_LISTEN) {
475 tcp_synack_timer(sk);
476 goto out;
477 }
478
479 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
480 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700481 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 if (tmo > 0) {
484 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
485 goto out;
486 }
487 }
488 tcp_send_active_reset(sk, GFP_ATOMIC);
489 goto death;
490 }
491
492 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
493 goto out;
494
495 elapsed = keepalive_time_when(tp);
496
497 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800498 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 goto resched;
500
501 elapsed = tcp_time_stamp - tp->rcv_tstamp;
502
503 if (elapsed >= keepalive_time_when(tp)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300504 if ((!tp->keepalive_probes && icsk->icsk_probes_out >= sysctl_tcp_keepalive_probes) ||
505 (tp->keepalive_probes && icsk->icsk_probes_out >= tp->keepalive_probes)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 tcp_send_active_reset(sk, GFP_ATOMIC);
507 tcp_write_err(sk);
508 goto out;
509 }
510 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300511 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 elapsed = keepalive_intvl_when(tp);
513 } else {
514 /* If keepalive was lost due to local congestion,
515 * try harder.
516 */
517 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
518 }
519 } else {
520 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
521 elapsed = keepalive_time_when(tp) - elapsed;
522 }
523
524 TCP_CHECK_TIMER(sk);
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800525 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
527resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700528 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 goto out;
530
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900531death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 tcp_done(sk);
533
534out:
535 bh_unlock_sock(sk);
536 sock_put(sk);
537}