blob: a516d1e399dfcb1a22ca39e452f3dbe725455ade [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
David S. Millerad1af0f2010-08-25 02:27:49 -070080 if (tcp_too_many_orphans(sk, shift)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 if (net_ratelimit())
82 printk(KERN_INFO "Out of socket memory\n");
83
84 /* Catch exceptional cases, when connection requires reset.
85 * 1. Last segment was sent recently. */
86 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
87 /* 2. Window is closed. */
88 (!tp->snd_wnd && !tp->packets_out))
89 do_reset = 1;
90 if (do_reset)
91 tcp_send_active_reset(sk, GFP_ATOMIC);
92 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070093 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 return 1;
95 }
96 return 0;
97}
98
99/* Calculate maximal number or retries on an orphaned socket. */
100static int tcp_orphan_retries(struct sock *sk, int alive)
101{
102 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
103
104 /* We know from an ICMP that something is wrong. */
105 if (sk->sk_err_soft && !alive)
106 retries = 0;
107
108 /* However, if socket sent something recently, select some safe
109 * number of retries. 8 corresponds to >100 seconds with minimal
110 * RTO of 200msec. */
111 if (retries == 0 && alive)
112 retries = 8;
113 return retries;
114}
115
Eric Dumazetce55dd32007-12-21 01:50:43 -0800116static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
117{
Eric Dumazetce55dd32007-12-21 01:50:43 -0800118 /* Black hole detection */
119 if (sysctl_tcp_mtu_probing) {
120 if (!icsk->icsk_mtup.enabled) {
121 icsk->icsk_mtup.enabled = 1;
122 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
123 } else {
124 struct tcp_sock *tp = tcp_sk(sk);
David S. Miller829942c2007-12-21 04:29:16 -0800125 int mss;
126
Eric Dumazet8beb5c52007-12-21 05:58:29 -0800127 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
Eric Dumazetce55dd32007-12-21 01:50:43 -0800128 mss = min(sysctl_tcp_base_mss, mss);
129 mss = max(mss, 68 - tp->tcp_header_len);
130 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
131 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
132 }
133 }
134}
135
Damian Lukowski2f7de572009-12-07 06:06:16 +0000136/* This function calculates a "timeout" which is equivalent to the timeout of a
Daniel Mack3ad2f3f2010-02-03 08:01:28 +0800137 * TCP connection after "boundary" unsuccessful, exponentially backed-off
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700138 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
139 * syn_set flag is set.
Damian Lukowski2f7de572009-12-07 06:06:16 +0000140 */
141static bool retransmits_timed_out(struct sock *sk,
Jerry Chudca43c72010-08-27 19:13:28 +0000142 unsigned int boundary,
David S. Miller21a180c2010-10-04 11:56:38 -0700143 unsigned int timeout,
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700144 bool syn_set)
Damian Lukowski2f7de572009-12-07 06:06:16 +0000145{
Jerry Chudca43c72010-08-27 19:13:28 +0000146 unsigned int linear_backoff_thresh, start_ts;
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700147 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
Damian Lukowski2f7de572009-12-07 06:06:16 +0000148
149 if (!inet_csk(sk)->icsk_retransmits)
150 return false;
151
152 if (unlikely(!tcp_sk(sk)->retrans_stamp))
153 start_ts = TCP_SKB_CB(tcp_write_queue_head(sk))->when;
154 else
155 start_ts = tcp_sk(sk)->retrans_stamp;
156
Jerry Chudca43c72010-08-27 19:13:28 +0000157 if (likely(timeout == 0)) {
David S. Miller21a180c2010-10-04 11:56:38 -0700158 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
Damian Lukowski2f7de572009-12-07 06:06:16 +0000159
Jerry Chudca43c72010-08-27 19:13:28 +0000160 if (boundary <= linear_backoff_thresh)
David S. Miller21a180c2010-10-04 11:56:38 -0700161 timeout = ((2 << boundary) - 1) * rto_base;
Jerry Chudca43c72010-08-27 19:13:28 +0000162 else
David S. Miller21a180c2010-10-04 11:56:38 -0700163 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
Jerry Chudca43c72010-08-27 19:13:28 +0000164 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
165 }
Damian Lukowski2f7de572009-12-07 06:06:16 +0000166 return (tcp_time_stamp - start_ts) >= timeout;
167}
168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169/* A write timeout has occurred. Process the after effects. */
170static int tcp_write_timeout(struct sock *sk)
171{
John Heffner5d424d52006-03-20 17:53:41 -0800172 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 int retry_until;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000174 bool do_reset, syn_set = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700177 if (icsk->icsk_retransmits)
Eric Dumazetb6c67122010-04-08 23:03:29 +0000178 dst_negative_advice(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700179 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000180 syn_set = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 } else {
David S. Miller21a180c2010-10-04 11:56:38 -0700182 if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
John Heffner5d424d52006-03-20 17:53:41 -0800183 /* Black hole detection */
Eric Dumazetce55dd32007-12-21 01:50:43 -0800184 tcp_mtu_probing(icsk, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Eric Dumazetb6c67122010-04-08 23:03:29 +0000186 dst_negative_advice(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 }
188
189 retry_until = sysctl_tcp_retries2;
190 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700191 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 retry_until = tcp_orphan_retries(sk, alive);
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000194 do_reset = alive ||
David S. Miller21a180c2010-10-04 11:56:38 -0700195 !retransmits_timed_out(sk, retry_until, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000197 if (tcp_out_of_resources(sk, do_reset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 return 1;
199 }
200 }
201
Jerry Chudca43c72010-08-27 19:13:28 +0000202 if (retransmits_timed_out(sk, retry_until,
David S. Miller21a180c2010-10-04 11:56:38 -0700203 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 /* Has it gone just too far? */
205 tcp_write_err(sk);
206 return 1;
207 }
208 return 0;
209}
210
211static void tcp_delack_timer(unsigned long data)
212{
Jianjun Kongfd3f8c42008-11-03 02:47:38 -0800213 struct sock *sk = (struct sock *)data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700215 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217 bh_lock_sock(sk);
218 if (sock_owned_by_user(sk)) {
219 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700220 icsk->icsk_ack.blocked = 1;
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700221 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700222 sk_reset_timer(sk, &icsk->icsk_delack_timer, jiffies + TCP_DELACK_MIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 goto out_unlock;
224 }
225
David S. Miller9993e7d2008-01-10 21:56:38 -0800226 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700228 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 goto out;
230
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700231 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
232 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 goto out;
234 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700235 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
David S. Millerb03efcf2005-07-08 14:57:23 -0700237 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 struct sk_buff *skb;
239
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700240 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -0700243 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 tp->ucopy.memory = 0;
246 }
247
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700248 if (inet_csk_ack_scheduled(sk)) {
249 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700251 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 } else {
253 /* Delayed ACK missed: leave pingpong mode and
254 * deflate ATO.
255 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700256 icsk->icsk_ack.pingpong = 0;
257 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 }
259 tcp_send_ack(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700260 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263out:
Glauber Costa180d8cd2011-12-11 21:47:02 +0000264 if (sk_under_memory_pressure(sk))
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800265 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266out_unlock:
267 bh_unlock_sock(sk);
268 sock_put(sk);
269}
270
271static void tcp_probe_timer(struct sock *sk)
272{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300273 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 struct tcp_sock *tp = tcp_sk(sk);
275 int max_probes;
276
David S. Millerfe067e82007-03-07 12:12:44 -0800277 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300278 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return;
280 }
281
282 /* *WARNING* RFC 1122 forbids this
283 *
284 * It doesn't AFAIK, because we kill the retransmit timer -AK
285 *
286 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
287 * this behaviour in Solaris down as a bug fix. [AC]
288 *
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300289 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * even if they advertise zero window. Hence, connection is killed only
291 * if we received no ACKs for normal connection timeout. It is not killed
292 * only because window stays zero for some time, window may be zero
293 * until armageddon and even later. We are in full accordance
294 * with RFCs, only probe timer combines both retransmission timeout
295 * and probe timeout in one bottle. --ANK
296 */
297 max_probes = sysctl_tcp_retries2;
298
299 if (sock_flag(sk, SOCK_DEAD)) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700300 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 max_probes = tcp_orphan_retries(sk, alive);
303
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300304 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 return;
306 }
307
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300308 if (icsk->icsk_probes_out > max_probes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 tcp_write_err(sk);
310 } else {
311 /* Only send another probe if we didn't close things up. */
312 tcp_send_probe0(sk);
313 }
314}
315
316/*
317 * The TCP retransmit timer.
318 */
319
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000320void tcp_retransmit_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
322 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700323 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325 if (!tp->packets_out)
326 goto out;
327
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700328 WARN_ON(tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
330 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
331 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
332 /* Receiver dastardly shrinks window. Our retransmits
333 * become zero probes, but we should not timeout this
334 * connection. If the socket is an orphan, time it out,
335 * we cannot allow such beasts to hang infinitely.
336 */
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700337 struct inet_sock *inet = inet_sk(sk);
338 if (sk->sk_family == AF_INET) {
Matt Mackall6086ebc2008-12-18 19:54:22 -0800339 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000340 &inet->inet_daddr, ntohs(inet->inet_dport),
341 inet->inet_num, tp->snd_una, tp->snd_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 }
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000343#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700344 else if (sk->sk_family == AF_INET6) {
345 struct ipv6_pinfo *np = inet6_sk(sk);
Matt Mackall6086ebc2008-12-18 19:54:22 -0800346 LIMIT_NETDEBUG(KERN_DEBUG "TCP: Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000347 &np->daddr, ntohs(inet->inet_dport),
348 inet->inet_num, tp->snd_una, tp->snd_nxt);
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700349 }
350#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
352 tcp_write_err(sk);
353 goto out;
354 }
355 tcp_enter_loss(sk, 0);
David S. Millerfe067e82007-03-07 12:12:44 -0800356 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 __sk_dst_reset(sk);
358 goto out_reset_timer;
359 }
360
361 if (tcp_write_timeout(sk))
362 goto out;
363
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700364 if (icsk->icsk_retransmits == 0) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700365 int mib_idx;
366
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000367 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
Ilpo Järvinenbc079e92009-02-28 04:44:34 +0000368 if (tcp_is_sack(tp))
369 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
370 else
371 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300372 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700373 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000374 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
375 tp->sacked_out) {
376 if (tcp_is_sack(tp))
377 mib_idx = LINUX_MIB_TCPSACKFAILURES;
378 else
379 mib_idx = LINUX_MIB_TCPRENOFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 } else {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700381 mib_idx = LINUX_MIB_TCPTIMEOUTS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700383 NET_INC_STATS_BH(sock_net(sk), mib_idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
385
386 if (tcp_use_frto(sk)) {
387 tcp_enter_frto(sk);
388 } else {
389 tcp_enter_loss(sk, 0);
390 }
391
David S. Millerfe067e82007-03-07 12:12:44 -0800392 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 /* Retransmission failed because of local congestion,
394 * do not backoff.
395 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700396 if (!icsk->icsk_retransmits)
397 icsk->icsk_retransmits = 1;
398 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700399 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
400 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 goto out;
402 }
403
404 /* Increase the timeout each time we retransmit. Note that
405 * we do not increase the rtt estimate. rto is initialized
406 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
407 * that doubling rto each time is the least we can get away with.
408 * In KA9Q, Karn uses this for the first few times, and then
409 * goes to quadratic. netBSD doubles, but only goes up to *64,
410 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
411 * defined in the protocol as the maximum possible RTT. I guess
412 * we'll have to use something other than TCP to talk to the
413 * University of Mars.
414 *
415 * PAWS allows us longer timeouts and large windows, so once
416 * implemented ftp to mars will work nicely. We will have to fix
417 * the 120 second clamps though!
418 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700419 icsk->icsk_backoff++;
420 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
422out_reset_timer:
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000423 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
424 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
425 * might be increased if the stream oscillates between thin and thick,
426 * thus the old value might already be too high compared to the value
427 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
428 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
429 * exponential backoff behaviour to avoid continue hammering
430 * linear-timeout retransmissions into a black hole
431 */
432 if (sk->sk_state == TCP_ESTABLISHED &&
433 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
434 tcp_stream_is_thin(tp) &&
435 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
436 icsk->icsk_backoff = 0;
437 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
438 } else {
439 /* Use normal (exponential) backoff */
440 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
441 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700442 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
David S. Miller21a180c2010-10-04 11:56:38 -0700443 if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 __sk_dst_reset(sk);
445
446out:;
447}
448
449static void tcp_write_timer(unsigned long data)
450{
Jianjun Kongfd3f8c42008-11-03 02:47:38 -0800451 struct sock *sk = (struct sock *)data;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700452 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 int event;
454
455 bh_lock_sock(sk);
456 if (sock_owned_by_user(sk)) {
457 /* Try again later */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700458 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, jiffies + (HZ / 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 goto out_unlock;
460 }
461
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700462 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 goto out;
464
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700465 if (time_after(icsk->icsk_timeout, jiffies)) {
466 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 goto out;
468 }
469
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700470 event = icsk->icsk_pending;
471 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 switch (event) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700474 case ICSK_TIME_RETRANS:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 tcp_retransmit_timer(sk);
476 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700477 case ICSK_TIME_PROBE0:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 tcp_probe_timer(sk);
479 break;
480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482out:
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800483 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484out_unlock:
485 bh_unlock_sock(sk);
486 sock_put(sk);
487}
488
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700489/*
490 * Timer for listening sockets
491 */
492
493static void tcp_synack_timer(struct sock *sk)
494{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700495 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
496 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
Octavian Purdila72659ec2010-01-17 19:09:39 -0800499void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
500{
501 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
502}
503EXPORT_SYMBOL(tcp_syn_ack_timeout);
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505void tcp_set_keepalive(struct sock *sk, int val)
506{
507 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
508 return;
509
510 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700511 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700513 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
516
517static void tcp_keepalive_timer (unsigned long data)
518{
519 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300520 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 struct tcp_sock *tp = tcp_sk(sk);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000522 u32 elapsed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
524 /* Only process if socket is not in use. */
525 bh_lock_sock(sk);
526 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900527 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700528 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 goto out;
530 }
531
532 if (sk->sk_state == TCP_LISTEN) {
533 tcp_synack_timer(sk);
534 goto out;
535 }
536
537 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
538 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700539 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
541 if (tmo > 0) {
542 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
543 goto out;
544 }
545 }
546 tcp_send_active_reset(sk, GFP_ATOMIC);
547 goto death;
548 }
549
550 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
551 goto out;
552
553 elapsed = keepalive_time_when(tp);
554
555 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800556 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 goto resched;
558
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000559 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 if (elapsed >= keepalive_time_when(tp)) {
Jerry Chudca43c72010-08-27 19:13:28 +0000562 /* If the TCP_USER_TIMEOUT option is enabled, use that
563 * to determine when to timeout instead.
564 */
565 if ((icsk->icsk_user_timeout != 0 &&
566 elapsed >= icsk->icsk_user_timeout &&
567 icsk->icsk_probes_out > 0) ||
568 (icsk->icsk_user_timeout == 0 &&
569 icsk->icsk_probes_out >= keepalive_probes(tp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 tcp_send_active_reset(sk, GFP_ATOMIC);
571 tcp_write_err(sk);
572 goto out;
573 }
574 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300575 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 elapsed = keepalive_intvl_when(tp);
577 } else {
578 /* If keepalive was lost due to local congestion,
579 * try harder.
580 */
581 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
582 }
583 } else {
584 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
585 elapsed = keepalive_time_when(tp) - elapsed;
586 }
587
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800588 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700591 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 goto out;
593
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900594death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 tcp_done(sk);
596
597out:
598 bh_unlock_sock(sk);
599 sock_put(sk);
600}