blob: 34d4a02c2f16d867cb95827ee5ebc2256d0745b3 [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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Corey Minyard <wf-rch!minyard@relay.EU.net>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14 * Linus Torvalds, <torvalds@cs.helsinki.fi>
15 * Alan Cox, <gw4pts@gw4pts.ampr.org>
16 * Matthew Dillon, <dillon@apollo.west.oic.com>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 * Jorge Cwik, <jorge@laser.satlink.net>
19 */
20
21#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090022#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <net/tcp.h>
24
Brian Haleyab32ea52006-09-22 14:15:41 -070025int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES;
26int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES;
27int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME;
28int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES;
29int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
30int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
31int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
32int sysctl_tcp_orphan_retries __read_mostly;
Andreas Petlund36e31b0a2010-02-18 02:47:01 +000033int sysctl_tcp_thin_linear_timeouts __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}
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -070044EXPORT_SYMBOL(tcp_init_xmit_timers);
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046static void tcp_write_err(struct sock *sk)
47{
48 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
49 sk->sk_error_report(sk);
50
51 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070052 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053}
54
55/* Do not allow orphaned sockets to eat all our resources.
56 * This is direct violation of TCP specs, but it is required
57 * to prevent DoS attacks. It is called when a retransmission timeout
58 * or zero probe timeout occurs on orphaned socket.
59 *
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080060 * Criteria is still not confirmed experimentally and may change.
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * We kill the socket, if:
62 * 1. If number of orphaned sockets exceeds an administratively configured
63 * limit.
64 * 2. If we have strong memory pressure.
65 */
66static int tcp_out_of_resources(struct sock *sk, int do_reset)
67{
68 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerad1af0f2010-08-25 02:27:49 -070069 int shift = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090071 /* If peer does not open window for long time, or did not transmit
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * anything for long time, penalize it. */
73 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
David S. Millerad1af0f2010-08-25 02:27:49 -070074 shift++;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 /* If some dubious ICMP arrived, penalize even more. */
77 if (sk->sk_err_soft)
David S. Millerad1af0f2010-08-25 02:27:49 -070078 shift++;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Arun Sharmaefcdbf22012-01-30 14:16:06 -080080 if (tcp_check_oom(sk, shift)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 /* Catch exceptional cases, when connection requires reset.
82 * 1. Last segment was sent recently. */
83 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
84 /* 2. Window is closed. */
85 (!tp->snd_wnd && !tp->packets_out))
86 do_reset = 1;
87 if (do_reset)
88 tcp_send_active_reset(sk, GFP_ATOMIC);
89 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070090 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 return 1;
92 }
93 return 0;
94}
95
96/* Calculate maximal number or retries on an orphaned socket. */
97static int tcp_orphan_retries(struct sock *sk, int alive)
98{
99 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
100
101 /* We know from an ICMP that something is wrong. */
102 if (sk->sk_err_soft && !alive)
103 retries = 0;
104
105 /* However, if socket sent something recently, select some safe
106 * number of retries. 8 corresponds to >100 seconds with minimal
107 * RTO of 200msec. */
108 if (retries == 0 && alive)
109 retries = 8;
110 return retries;
111}
112
Eric Dumazetce55dd32007-12-21 01:50:43 -0800113static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
114{
Eric Dumazetce55dd32007-12-21 01:50:43 -0800115 /* Black hole detection */
116 if (sysctl_tcp_mtu_probing) {
117 if (!icsk->icsk_mtup.enabled) {
118 icsk->icsk_mtup.enabled = 1;
119 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
120 } else {
121 struct tcp_sock *tp = tcp_sk(sk);
David S. Miller829942c2007-12-21 04:29:16 -0800122 int mss;
123
Eric Dumazet8beb5c52007-12-21 05:58:29 -0800124 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
Eric Dumazetce55dd32007-12-21 01:50:43 -0800125 mss = min(sysctl_tcp_base_mss, mss);
126 mss = max(mss, 68 - tp->tcp_header_len);
127 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
128 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
129 }
130 }
131}
132
Damian Lukowski2f7de572009-12-07 06:06:16 +0000133/* This function calculates a "timeout" which is equivalent to the timeout of a
Daniel Mack3ad2f3f2010-02-03 08:01:28 +0800134 * TCP connection after "boundary" unsuccessful, exponentially backed-off
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700135 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
136 * syn_set flag is set.
Damian Lukowski2f7de572009-12-07 06:06:16 +0000137 */
138static bool retransmits_timed_out(struct sock *sk,
Jerry Chudca43c72010-08-27 19:13:28 +0000139 unsigned int boundary,
David S. Miller21a180c2010-10-04 11:56:38 -0700140 unsigned int timeout,
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700141 bool syn_set)
Damian Lukowski2f7de572009-12-07 06:06:16 +0000142{
Jerry Chudca43c72010-08-27 19:13:28 +0000143 unsigned int linear_backoff_thresh, start_ts;
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700144 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
Damian Lukowski2f7de572009-12-07 06:06:16 +0000145
146 if (!inet_csk(sk)->icsk_retransmits)
147 return false;
148
149 if (unlikely(!tcp_sk(sk)->retrans_stamp))
150 start_ts = TCP_SKB_CB(tcp_write_queue_head(sk))->when;
151 else
152 start_ts = tcp_sk(sk)->retrans_stamp;
153
Jerry Chudca43c72010-08-27 19:13:28 +0000154 if (likely(timeout == 0)) {
David S. Miller21a180c2010-10-04 11:56:38 -0700155 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
Damian Lukowski2f7de572009-12-07 06:06:16 +0000156
Jerry Chudca43c72010-08-27 19:13:28 +0000157 if (boundary <= linear_backoff_thresh)
David S. Miller21a180c2010-10-04 11:56:38 -0700158 timeout = ((2 << boundary) - 1) * rto_base;
Jerry Chudca43c72010-08-27 19:13:28 +0000159 else
David S. Miller21a180c2010-10-04 11:56:38 -0700160 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
Jerry Chudca43c72010-08-27 19:13:28 +0000161 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
162 }
Damian Lukowski2f7de572009-12-07 06:06:16 +0000163 return (tcp_time_stamp - start_ts) >= timeout;
164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166/* A write timeout has occurred. Process the after effects. */
167static int tcp_write_timeout(struct sock *sk)
168{
John Heffner5d424d52006-03-20 17:53:41 -0800169 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 int retry_until;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000171 bool do_reset, syn_set = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
173 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700174 if (icsk->icsk_retransmits)
Eric Dumazetb6c67122010-04-08 23:03:29 +0000175 dst_negative_advice(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700176 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000177 syn_set = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 } else {
David S. Miller21a180c2010-10-04 11:56:38 -0700179 if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
John Heffner5d424d52006-03-20 17:53:41 -0800180 /* Black hole detection */
Eric Dumazetce55dd32007-12-21 01:50:43 -0800181 tcp_mtu_probing(icsk, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Eric Dumazetb6c67122010-04-08 23:03:29 +0000183 dst_negative_advice(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 }
185
186 retry_until = sysctl_tcp_retries2;
187 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700188 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 retry_until = tcp_orphan_retries(sk, alive);
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000191 do_reset = alive ||
David S. Miller21a180c2010-10-04 11:56:38 -0700192 !retransmits_timed_out(sk, retry_until, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000194 if (tcp_out_of_resources(sk, do_reset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 return 1;
196 }
197 }
198
Jerry Chudca43c72010-08-27 19:13:28 +0000199 if (retransmits_timed_out(sk, retry_until,
David S. Miller21a180c2010-10-04 11:56:38 -0700200 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 /* Has it gone just too far? */
202 tcp_write_err(sk);
203 return 1;
204 }
205 return 0;
206}
207
208static void tcp_delack_timer(unsigned long data)
209{
Jianjun Kongfd3f8c42008-11-03 02:47:38 -0800210 struct sock *sk = (struct sock *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700212 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214 bh_lock_sock(sk);
215 if (sock_owned_by_user(sk)) {
216 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700217 icsk->icsk_ack.blocked = 1;
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700218 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700219 sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 goto out_unlock;
221 }
222
David S. Miller9993e7d2008-01-10 21:56:38 -0800223 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700225 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 goto out;
227
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700228 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
229 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 goto out;
231 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700232 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
David S. Millerb03efcf2005-07-08 14:57:23 -0700234 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 struct sk_buff *skb;
236
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700237 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -0700240 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242 tp->ucopy.memory = 0;
243 }
244
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700245 if (inet_csk_ack_scheduled(sk)) {
246 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700248 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 } else {
250 /* Delayed ACK missed: leave pingpong mode and
251 * deflate ATO.
252 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700253 icsk->icsk_ack.pingpong = 0;
254 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 }
256 tcp_send_ack(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700257 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260out:
Glauber Costa180d8cd2011-12-11 21:47:02 +0000261 if (sk_under_memory_pressure(sk))
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800262 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263out_unlock:
264 bh_unlock_sock(sk);
265 sock_put(sk);
266}
267
268static void tcp_probe_timer(struct sock *sk)
269{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300270 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 struct tcp_sock *tp = tcp_sk(sk);
272 int max_probes;
273
David S. Millerfe067e82007-03-07 12:12:44 -0800274 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300275 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 return;
277 }
278
279 /* *WARNING* RFC 1122 forbids this
280 *
281 * It doesn't AFAIK, because we kill the retransmit timer -AK
282 *
283 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
284 * this behaviour in Solaris down as a bug fix. [AC]
285 *
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300286 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 * even if they advertise zero window. Hence, connection is killed only
288 * if we received no ACKs for normal connection timeout. It is not killed
289 * only because window stays zero for some time, window may be zero
290 * until armageddon and even later. We are in full accordance
291 * with RFCs, only probe timer combines both retransmission timeout
292 * and probe timeout in one bottle. --ANK
293 */
294 max_probes = sysctl_tcp_retries2;
295
296 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700297 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 max_probes = tcp_orphan_retries(sk, alive);
300
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300301 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 return;
303 }
304
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300305 if (icsk->icsk_probes_out > max_probes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 tcp_write_err(sk);
307 } else {
308 /* Only send another probe if we didn't close things up. */
309 tcp_send_probe0(sk);
310 }
311}
312
313/*
314 * The TCP retransmit timer.
315 */
316
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000317void tcp_retransmit_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700320 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 if (!tp->packets_out)
323 goto out;
324
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700325 WARN_ON(tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
328 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
329 /* Receiver dastardly shrinks window. Our retransmits
330 * become zero probes, but we should not timeout this
331 * connection. If the socket is an orphan, time it out,
332 * we cannot allow such beasts to hang infinitely.
333 */
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700334 struct inet_sock *inet = inet_sk(sk);
335 if (sk->sk_family == AF_INET) {
Joe Perchesafd465032012-03-12 07:03:32 +0000336 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
337 &inet->inet_daddr,
338 ntohs(inet->inet_dport), inet->inet_num,
339 tp->snd_una, tp->snd_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 }
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000341#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700342 else if (sk->sk_family == AF_INET6) {
343 struct ipv6_pinfo *np = inet6_sk(sk);
Joe Perchesafd465032012-03-12 07:03:32 +0000344 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
345 &np->daddr,
346 ntohs(inet->inet_dport), inet->inet_num,
347 tp->snd_una, tp->snd_nxt);
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700348 }
349#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
351 tcp_write_err(sk);
352 goto out;
353 }
354 tcp_enter_loss(sk, 0);
David S. Millerfe067e82007-03-07 12:12:44 -0800355 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 __sk_dst_reset(sk);
357 goto out_reset_timer;
358 }
359
360 if (tcp_write_timeout(sk))
361 goto out;
362
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700363 if (icsk->icsk_retransmits == 0) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700364 int mib_idx;
365
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000366 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
Ilpo Järvinenbc079e92009-02-28 04:44:34 +0000367 if (tcp_is_sack(tp))
368 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
369 else
370 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300371 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700372 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000373 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
374 tp->sacked_out) {
375 if (tcp_is_sack(tp))
376 mib_idx = LINUX_MIB_TCPSACKFAILURES;
377 else
378 mib_idx = LINUX_MIB_TCPRENOFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 } else {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700380 mib_idx = LINUX_MIB_TCPTIMEOUTS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700382 NET_INC_STATS_BH(sock_net(sk), mib_idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
384
385 if (tcp_use_frto(sk)) {
386 tcp_enter_frto(sk);
387 } else {
388 tcp_enter_loss(sk, 0);
389 }
390
David S. Millerfe067e82007-03-07 12:12:44 -0800391 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 /* Retransmission failed because of local congestion,
393 * do not backoff.
394 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700395 if (!icsk->icsk_retransmits)
396 icsk->icsk_retransmits = 1;
397 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700398 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
399 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 goto out;
401 }
402
403 /* Increase the timeout each time we retransmit. Note that
404 * we do not increase the rtt estimate. rto is initialized
405 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
406 * that doubling rto each time is the least we can get away with.
407 * In KA9Q, Karn uses this for the first few times, and then
408 * goes to quadratic. netBSD doubles, but only goes up to *64,
409 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
410 * defined in the protocol as the maximum possible RTT. I guess
411 * we'll have to use something other than TCP to talk to the
412 * University of Mars.
413 *
414 * PAWS allows us longer timeouts and large windows, so once
415 * implemented ftp to mars will work nicely. We will have to fix
416 * the 120 second clamps though!
417 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700418 icsk->icsk_backoff++;
419 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421out_reset_timer:
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000422 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
423 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
424 * might be increased if the stream oscillates between thin and thick,
425 * thus the old value might already be too high compared to the value
426 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
427 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
428 * exponential backoff behaviour to avoid continue hammering
429 * linear-timeout retransmissions into a black hole
430 */
431 if (sk->sk_state == TCP_ESTABLISHED &&
432 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
433 tcp_stream_is_thin(tp) &&
434 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
435 icsk->icsk_backoff = 0;
436 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
437 } else {
438 /* Use normal (exponential) backoff */
439 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
440 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700441 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
David S. Miller21a180c2010-10-04 11:56:38 -0700442 if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 __sk_dst_reset(sk);
444
445out:;
446}
447
448static void tcp_write_timer(unsigned long data)
449{
Jianjun Kongfd3f8c42008-11-03 02:47:38 -0800450 struct sock *sk = (struct sock *)data;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700451 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 int event;
453
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 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 goto out_unlock;
459 }
460
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700461 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 goto out;
463
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700464 if (time_after(icsk->icsk_timeout, jiffies)) {
465 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 goto out;
467 }
468
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700469 event = icsk->icsk_pending;
470 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 switch (event) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700473 case ICSK_TIME_RETRANS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 tcp_retransmit_timer(sk);
475 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700476 case ICSK_TIME_PROBE0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 tcp_probe_timer(sk);
478 break;
479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481out:
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800482 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483out_unlock:
484 bh_unlock_sock(sk);
485 sock_put(sk);
486}
487
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700488/*
489 * Timer for listening sockets
490 */
491
492static void tcp_synack_timer(struct sock *sk)
493{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700494 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
495 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496}
497
Octavian Purdila72659ec2010-01-17 19:09:39 -0800498void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
499{
500 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
501}
502EXPORT_SYMBOL(tcp_syn_ack_timeout);
503
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504void tcp_set_keepalive(struct sock *sk, int val)
505{
506 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
507 return;
508
509 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700510 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700512 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513}
514
515
516static void tcp_keepalive_timer (unsigned long data)
517{
518 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300519 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 struct tcp_sock *tp = tcp_sk(sk);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000521 u32 elapsed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
523 /* Only process if socket is not in use. */
524 bh_lock_sock(sk);
525 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900526 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700527 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 goto out;
529 }
530
531 if (sk->sk_state == TCP_LISTEN) {
532 tcp_synack_timer(sk);
533 goto out;
534 }
535
536 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
537 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700538 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540 if (tmo > 0) {
541 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
542 goto out;
543 }
544 }
545 tcp_send_active_reset(sk, GFP_ATOMIC);
546 goto death;
547 }
548
549 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
550 goto out;
551
552 elapsed = keepalive_time_when(tp);
553
554 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800555 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 goto resched;
557
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000558 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 if (elapsed >= keepalive_time_when(tp)) {
Jerry Chudca43c72010-08-27 19:13:28 +0000561 /* If the TCP_USER_TIMEOUT option is enabled, use that
562 * to determine when to timeout instead.
563 */
564 if ((icsk->icsk_user_timeout != 0 &&
565 elapsed >= icsk->icsk_user_timeout &&
566 icsk->icsk_probes_out > 0) ||
567 (icsk->icsk_user_timeout == 0 &&
568 icsk->icsk_probes_out >= keepalive_probes(tp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 tcp_send_active_reset(sk, GFP_ATOMIC);
570 tcp_write_err(sk);
571 goto out;
572 }
573 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300574 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 elapsed = keepalive_intvl_when(tp);
576 } else {
577 /* If keepalive was lost due to local congestion,
578 * try harder.
579 */
580 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
581 }
582 } else {
583 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
584 elapsed = keepalive_time_when(tp) - elapsed;
585 }
586
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800587 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700590 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 goto out;
592
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900593death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 tcp_done(sk);
595
596out:
597 bh_unlock_sock(sk);
598 sock_put(sk);
599}