blob: 1829c7fbc77e4ded47744a60e0b55c11c635e82d [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070035static void tcp_write_err(struct sock *sk)
36{
37 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
38 sk->sk_error_report(sk);
39
40 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070041 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042}
43
44/* Do not allow orphaned sockets to eat all our resources.
45 * This is direct violation of TCP specs, but it is required
46 * to prevent DoS attacks. It is called when a retransmission timeout
47 * or zero probe timeout occurs on orphaned socket.
48 *
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080049 * Criteria is still not confirmed experimentally and may change.
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * We kill the socket, if:
51 * 1. If number of orphaned sockets exceeds an administratively configured
52 * limit.
53 * 2. If we have strong memory pressure.
54 */
Yuchung Chengb2482302014-09-29 13:20:38 -070055static int tcp_out_of_resources(struct sock *sk, bool do_reset)
Linus Torvalds1da177e2005-04-16 15:20:36 -070056{
57 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerad1af0f2010-08-25 02:27:49 -070058 int shift = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090060 /* If peer does not open window for long time, or did not transmit
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 * anything for long time, penalize it. */
62 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
David S. Millerad1af0f2010-08-25 02:27:49 -070063 shift++;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
65 /* If some dubious ICMP arrived, penalize even more. */
66 if (sk->sk_err_soft)
David S. Millerad1af0f2010-08-25 02:27:49 -070067 shift++;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Arun Sharmaefcdbf22012-01-30 14:16:06 -080069 if (tcp_check_oom(sk, shift)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 /* Catch exceptional cases, when connection requires reset.
71 * 1. Last segment was sent recently. */
72 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
73 /* 2. Window is closed. */
74 (!tp->snd_wnd && !tp->packets_out))
Yuchung Chengb2482302014-09-29 13:20:38 -070075 do_reset = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 if (do_reset)
77 tcp_send_active_reset(sk, GFP_ATOMIC);
78 tcp_done(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -070079 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 return 1;
81 }
82 return 0;
83}
84
85/* Calculate maximal number or retries on an orphaned socket. */
86static int tcp_orphan_retries(struct sock *sk, int alive)
87{
88 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
89
90 /* We know from an ICMP that something is wrong. */
91 if (sk->sk_err_soft && !alive)
92 retries = 0;
93
94 /* However, if socket sent something recently, select some safe
95 * number of retries. 8 corresponds to >100 seconds with minimal
96 * RTO of 200msec. */
97 if (retries == 0 && alive)
98 retries = 8;
99 return retries;
100}
101
Eric Dumazetce55dd32007-12-21 01:50:43 -0800102static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
103{
Eric Dumazetce55dd32007-12-21 01:50:43 -0800104 /* Black hole detection */
105 if (sysctl_tcp_mtu_probing) {
106 if (!icsk->icsk_mtup.enabled) {
107 icsk->icsk_mtup.enabled = 1;
108 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
109 } else {
110 struct tcp_sock *tp = tcp_sk(sk);
David S. Miller829942c2007-12-21 04:29:16 -0800111 int mss;
112
Eric Dumazet8beb5c52007-12-21 05:58:29 -0800113 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
Eric Dumazetce55dd32007-12-21 01:50:43 -0800114 mss = min(sysctl_tcp_base_mss, mss);
115 mss = max(mss, 68 - tp->tcp_header_len);
116 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
117 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
118 }
119 }
120}
121
Damian Lukowski2f7de572009-12-07 06:06:16 +0000122/* This function calculates a "timeout" which is equivalent to the timeout of a
Daniel Mack3ad2f3f2010-02-03 08:01:28 +0800123 * TCP connection after "boundary" unsuccessful, exponentially backed-off
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700124 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
125 * syn_set flag is set.
Damian Lukowski2f7de572009-12-07 06:06:16 +0000126 */
127static bool retransmits_timed_out(struct sock *sk,
Jerry Chudca43c72010-08-27 19:13:28 +0000128 unsigned int boundary,
David S. Miller21a180c2010-10-04 11:56:38 -0700129 unsigned int timeout,
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700130 bool syn_set)
Damian Lukowski2f7de572009-12-07 06:06:16 +0000131{
Jerry Chudca43c72010-08-27 19:13:28 +0000132 unsigned int linear_backoff_thresh, start_ts;
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700133 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
Damian Lukowski2f7de572009-12-07 06:06:16 +0000134
135 if (!inet_csk(sk)->icsk_retransmits)
136 return false;
137
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700138 start_ts = tcp_sk(sk)->retrans_stamp;
139 if (unlikely(!start_ts))
140 start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk));
Damian Lukowski2f7de572009-12-07 06:06:16 +0000141
Jerry Chudca43c72010-08-27 19:13:28 +0000142 if (likely(timeout == 0)) {
David S. Miller21a180c2010-10-04 11:56:38 -0700143 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
Damian Lukowski2f7de572009-12-07 06:06:16 +0000144
Jerry Chudca43c72010-08-27 19:13:28 +0000145 if (boundary <= linear_backoff_thresh)
David S. Miller21a180c2010-10-04 11:56:38 -0700146 timeout = ((2 << boundary) - 1) * rto_base;
Jerry Chudca43c72010-08-27 19:13:28 +0000147 else
David S. Miller21a180c2010-10-04 11:56:38 -0700148 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
Jerry Chudca43c72010-08-27 19:13:28 +0000149 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
150 }
Damian Lukowski2f7de572009-12-07 06:06:16 +0000151 return (tcp_time_stamp - start_ts) >= timeout;
152}
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154/* A write timeout has occurred. Process the after effects. */
155static int tcp_write_timeout(struct sock *sk)
156{
John Heffner5d424d52006-03-20 17:53:41 -0800157 struct inet_connection_sock *icsk = inet_csk(sk);
Yuchung Chengc9686012013-10-29 10:09:05 -0700158 struct tcp_sock *tp = tcp_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 int retry_until;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000160 bool do_reset, syn_set = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Yuchung Chengc9686012013-10-29 10:09:05 -0700163 if (icsk->icsk_retransmits) {
Eric Dumazetb6c67122010-04-08 23:03:29 +0000164 dst_negative_advice(sk);
Yuchung Chengc9686012013-10-29 10:09:05 -0700165 if (tp->syn_fastopen || tp->syn_data)
166 tcp_fastopen_cache_set(sk, 0, NULL, true);
Yuchung Chengf19c29e2014-03-03 12:31:36 -0800167 if (tp->syn_data)
168 NET_INC_STATS_BH(sock_net(sk),
169 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
Yuchung Chengc9686012013-10-29 10:09:05 -0700170 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700171 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000172 syn_set = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 } else {
David S. Miller21a180c2010-10-04 11:56:38 -0700174 if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
John Heffner5d424d52006-03-20 17:53:41 -0800175 /* Black hole detection */
Eric Dumazetce55dd32007-12-21 01:50:43 -0800176 tcp_mtu_probing(icsk, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Eric Dumazetb6c67122010-04-08 23:03:29 +0000178 dst_negative_advice(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 }
180
181 retry_until = sysctl_tcp_retries2;
182 if (sock_flag(sk, SOCK_DEAD)) {
Eric Dumazetfcdd1cf2014-09-22 13:19:44 -0700183 const int alive = icsk->icsk_rto < TCP_RTO_MAX;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 retry_until = tcp_orphan_retries(sk, alive);
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000186 do_reset = alive ||
David S. Miller21a180c2010-10-04 11:56:38 -0700187 !retransmits_timed_out(sk, retry_until, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000189 if (tcp_out_of_resources(sk, do_reset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 return 1;
191 }
192 }
193
Jerry Chudca43c72010-08-27 19:13:28 +0000194 if (retransmits_timed_out(sk, retry_until,
David S. Miller21a180c2010-10-04 11:56:38 -0700195 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 /* Has it gone just too far? */
197 tcp_write_err(sk);
198 return 1;
199 }
200 return 0;
201}
202
Eric Dumazet6f458df2012-07-20 05:45:50 +0000203void tcp_delack_timer_handler(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700206 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
David S. Miller9993e7d2008-01-10 21:56:38 -0800208 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700210 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 goto out;
212
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700213 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
214 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 goto out;
216 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700217 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
David S. Millerb03efcf2005-07-08 14:57:23 -0700219 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 struct sk_buff *skb;
221
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700222 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
224 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -0700225 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
227 tp->ucopy.memory = 0;
228 }
229
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700230 if (inet_csk_ack_scheduled(sk)) {
231 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700233 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 } else {
235 /* Delayed ACK missed: leave pingpong mode and
236 * deflate ATO.
237 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700238 icsk->icsk_ack.pingpong = 0;
239 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 }
241 tcp_send_ack(sk);
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700242 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245out:
Glauber Costa180d8cd2011-12-11 21:47:02 +0000246 if (sk_under_memory_pressure(sk))
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800247 sk_mem_reclaim(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000248}
249
250static void tcp_delack_timer(unsigned long data)
251{
252 struct sock *sk = (struct sock *)data;
253
254 bh_lock_sock(sk);
255 if (!sock_owned_by_user(sk)) {
256 tcp_delack_timer_handler(sk);
257 } else {
258 inet_csk(sk)->icsk_ack.blocked = 1;
259 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
260 /* deleguate our work to tcp_release_cb() */
Eric Dumazet144d56e2012-08-20 00:22:46 +0000261 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
262 sock_hold(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 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;
Yuchung Chengb2482302014-09-29 13:20:38 -0700273 u32 start_ts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
David S. Millerfe067e82007-03-07 12:12:44 -0800275 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300276 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 return;
278 }
279
Yuchung Chengb2482302014-09-29 13:20:38 -0700280 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
281 * long as the receiver continues to respond probes. We support this by
282 * default and reset icsk_probes_out with incoming ACKs. But if the
283 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
284 * kill the socket when the retry count and the time exceeds the
285 * corresponding system limit. We also implement similar policy when
286 * we use RTO to probe window in tcp_retransmit_timer().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 */
Yuchung Chengb2482302014-09-29 13:20:38 -0700288 start_ts = tcp_skb_timestamp(tcp_send_head(sk));
289 if (!start_ts)
290 skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp);
291 else if (icsk->icsk_user_timeout &&
292 (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout)
293 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Yuchung Chengb2482302014-09-29 13:20:38 -0700295 max_probes = sysctl_tcp_retries2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 if (sock_flag(sk, SOCK_DEAD)) {
Eric Dumazetfcdd1cf2014-09-22 13:19:44 -0700297 const int alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < 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);
Yuchung Chengb2482302014-09-29 13:20:38 -0700300 if (!alive && icsk->icsk_backoff >= max_probes)
301 goto abort;
302 if (tcp_out_of_resources(sk, true))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 return;
304 }
305
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300306 if (icsk->icsk_probes_out > max_probes) {
Yuchung Chengb2482302014-09-29 13:20:38 -0700307abort: tcp_write_err(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 } else {
309 /* Only send another probe if we didn't close things up. */
310 tcp_send_probe0(sk);
311 }
312}
313
314/*
Jerry Chu83368862012-08-31 12:29:12 +0000315 * Timer for Fast Open socket to retransmit SYNACK. Note that the
316 * sk here is the child socket, not the parent (listener) socket.
317 */
318static void tcp_fastopen_synack_timer(struct sock *sk)
319{
320 struct inet_connection_sock *icsk = inet_csk(sk);
321 int max_retries = icsk->icsk_syn_retries ? :
322 sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
323 struct request_sock *req;
324
325 req = tcp_sk(sk)->fastopen_rsk;
326 req->rsk_ops->syn_ack_timeout(sk, req);
327
Eric Dumazete6c022a2012-10-27 23:16:46 +0000328 if (req->num_timeout >= max_retries) {
Jerry Chu83368862012-08-31 12:29:12 +0000329 tcp_write_err(sk);
330 return;
331 }
332 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
333 * returned from rtx_syn_ack() to make it more persistent like
334 * regular retransmit because if the child socket has been accepted
335 * it's not good to give up too easily.
336 */
Eric Dumazete6c022a2012-10-27 23:16:46 +0000337 inet_rtx_syn_ack(sk, req);
338 req->num_timeout++;
Jerry Chu83368862012-08-31 12:29:12 +0000339 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Eric Dumazete6c022a2012-10-27 23:16:46 +0000340 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
Jerry Chu83368862012-08-31 12:29:12 +0000341}
342
343/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 * The TCP retransmit timer.
345 */
346
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000347void tcp_retransmit_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
349 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700350 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Jerry Chu83368862012-08-31 12:29:12 +0000352 if (tp->fastopen_rsk) {
Jerry Chu37561f62012-10-22 11:26:36 +0000353 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
354 sk->sk_state != TCP_FIN_WAIT1);
Jerry Chu83368862012-08-31 12:29:12 +0000355 tcp_fastopen_synack_timer(sk);
356 /* Before we receive ACK to our SYN-ACK don't retransmit
357 * anything else (e.g., data or FIN segments).
358 */
359 return;
360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 if (!tp->packets_out)
362 goto out;
363
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700364 WARN_ON(tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Nandita Dukkipati9b717a82013-03-11 10:00:44 +0000366 tp->tlp_high_seq = 0;
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
369 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
370 /* Receiver dastardly shrinks window. Our retransmits
371 * become zero probes, but we should not timeout this
372 * connection. If the socket is an orphan, time it out,
373 * we cannot allow such beasts to hang infinitely.
374 */
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700375 struct inet_sock *inet = inet_sk(sk);
376 if (sk->sk_family == AF_INET) {
Joe Perchesba7a46f2014-11-11 10:59:17 -0800377 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
378 &inet->inet_daddr,
379 ntohs(inet->inet_dport),
380 inet->inet_num,
381 tp->snd_una, tp->snd_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000383#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700384 else if (sk->sk_family == AF_INET6) {
Joe Perchesba7a46f2014-11-11 10:59:17 -0800385 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
386 &sk->sk_v6_daddr,
387 ntohs(inet->inet_dport),
388 inet->inet_num,
389 tp->snd_una, tp->snd_nxt);
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700390 }
391#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
393 tcp_write_err(sk);
394 goto out;
395 }
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400396 tcp_enter_loss(sk);
David S. Millerfe067e82007-03-07 12:12:44 -0800397 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 __sk_dst_reset(sk);
399 goto out_reset_timer;
400 }
401
402 if (tcp_write_timeout(sk))
403 goto out;
404
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700405 if (icsk->icsk_retransmits == 0) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700406 int mib_idx;
407
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000408 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
Ilpo Järvinenbc079e92009-02-28 04:44:34 +0000409 if (tcp_is_sack(tp))
410 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
411 else
412 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300413 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700414 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000415 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
416 tp->sacked_out) {
417 if (tcp_is_sack(tp))
418 mib_idx = LINUX_MIB_TCPSACKFAILURES;
419 else
420 mib_idx = LINUX_MIB_TCPRENOFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 } else {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700422 mib_idx = LINUX_MIB_TCPTIMEOUTS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700424 NET_INC_STATS_BH(sock_net(sk), mib_idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 }
426
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400427 tcp_enter_loss(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
David S. Millerfe067e82007-03-07 12:12:44 -0800429 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 /* Retransmission failed because of local congestion,
431 * do not backoff.
432 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700433 if (!icsk->icsk_retransmits)
434 icsk->icsk_retransmits = 1;
435 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700436 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
437 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 goto out;
439 }
440
441 /* Increase the timeout each time we retransmit. Note that
442 * we do not increase the rtt estimate. rto is initialized
443 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
444 * that doubling rto each time is the least we can get away with.
445 * In KA9Q, Karn uses this for the first few times, and then
446 * goes to quadratic. netBSD doubles, but only goes up to *64,
447 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
448 * defined in the protocol as the maximum possible RTT. I guess
449 * we'll have to use something other than TCP to talk to the
450 * University of Mars.
451 *
452 * PAWS allows us longer timeouts and large windows, so once
453 * implemented ftp to mars will work nicely. We will have to fix
454 * the 120 second clamps though!
455 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700456 icsk->icsk_backoff++;
457 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459out_reset_timer:
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000460 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
461 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
462 * might be increased if the stream oscillates between thin and thick,
463 * thus the old value might already be too high compared to the value
464 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
465 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
466 * exponential backoff behaviour to avoid continue hammering
467 * linear-timeout retransmissions into a black hole
468 */
469 if (sk->sk_state == TCP_ESTABLISHED &&
470 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
471 tcp_stream_is_thin(tp) &&
472 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
473 icsk->icsk_backoff = 0;
474 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
475 } else {
476 /* Use normal (exponential) backoff */
477 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
478 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700479 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
David S. Miller21a180c2010-10-04 11:56:38 -0700480 if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 __sk_dst_reset(sk);
482
483out:;
484}
485
Eric Dumazet6f458df2012-07-20 05:45:50 +0000486void tcp_write_timer_handler(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700488 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 int event;
490
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700491 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 goto out;
493
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700494 if (time_after(icsk->icsk_timeout, jiffies)) {
495 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 goto out;
497 }
498
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700499 event = icsk->icsk_pending;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
501 switch (event) {
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000502 case ICSK_TIME_EARLY_RETRANS:
503 tcp_resume_early_retransmit(sk);
504 break;
505 case ICSK_TIME_LOSS_PROBE:
506 tcp_send_loss_probe(sk);
507 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700508 case ICSK_TIME_RETRANS:
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000509 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 tcp_retransmit_timer(sk);
511 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700512 case ICSK_TIME_PROBE0:
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000513 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 tcp_probe_timer(sk);
515 break;
516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
518out:
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800519 sk_mem_reclaim(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000520}
521
522static void tcp_write_timer(unsigned long data)
523{
524 struct sock *sk = (struct sock *)data;
525
526 bh_lock_sock(sk);
527 if (!sock_owned_by_user(sk)) {
528 tcp_write_timer_handler(sk);
529 } else {
530 /* deleguate our work to tcp_release_cb() */
Eric Dumazet144d56e2012-08-20 00:22:46 +0000531 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
532 sock_hold(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 bh_unlock_sock(sk);
535 sock_put(sk);
536}
537
Arnaldo Carvalho de Melo295f7322005-08-09 20:11:56 -0700538/*
539 * Timer for listening sockets
540 */
541
542static void tcp_synack_timer(struct sock *sk)
543{
Arnaldo Carvalho de Meloa019d6f2005-08-09 20:15:09 -0700544 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
545 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
547
Octavian Purdila72659ec2010-01-17 19:09:39 -0800548void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
549{
550 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
551}
552EXPORT_SYMBOL(tcp_syn_ack_timeout);
553
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554void tcp_set_keepalive(struct sock *sk, int val)
555{
556 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
557 return;
558
559 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700560 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700562 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563}
564
565
566static void tcp_keepalive_timer (unsigned long data)
567{
568 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300569 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 struct tcp_sock *tp = tcp_sk(sk);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000571 u32 elapsed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 /* Only process if socket is not in use. */
574 bh_lock_sock(sk);
575 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900576 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700577 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 goto out;
579 }
580
581 if (sk->sk_state == TCP_LISTEN) {
582 tcp_synack_timer(sk);
583 goto out;
584 }
585
586 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
587 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700588 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 if (tmo > 0) {
591 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
592 goto out;
593 }
594 }
595 tcp_send_active_reset(sk, GFP_ATOMIC);
596 goto death;
597 }
598
599 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
600 goto out;
601
602 elapsed = keepalive_time_when(tp);
603
604 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800605 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 goto resched;
607
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000608 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610 if (elapsed >= keepalive_time_when(tp)) {
Jerry Chudca43c72010-08-27 19:13:28 +0000611 /* If the TCP_USER_TIMEOUT option is enabled, use that
612 * to determine when to timeout instead.
613 */
614 if ((icsk->icsk_user_timeout != 0 &&
615 elapsed >= icsk->icsk_user_timeout &&
616 icsk->icsk_probes_out > 0) ||
617 (icsk->icsk_user_timeout == 0 &&
618 icsk->icsk_probes_out >= keepalive_probes(tp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 tcp_send_active_reset(sk, GFP_ATOMIC);
620 tcp_write_err(sk);
621 goto out;
622 }
623 if (tcp_write_wakeup(sk) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300624 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 elapsed = keepalive_intvl_when(tp);
626 } else {
627 /* If keepalive was lost due to local congestion,
628 * try harder.
629 */
630 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
631 }
632 } else {
633 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
634 elapsed = keepalive_time_when(tp) - elapsed;
635 }
636
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800637 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700640 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 goto out;
642
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900643death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 tcp_done(sk);
645
646out:
647 bh_unlock_sock(sk);
648 sock_put(sk);
649}
Eric Dumazet6f458df2012-07-20 05:45:50 +0000650
651void tcp_init_xmit_timers(struct sock *sk)
652{
653 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
654 &tcp_keepalive_timer);
655}
656EXPORT_SYMBOL(tcp_init_xmit_timers);