blob: debdd8b33e69478f7fa108c72e3787b55af6af34 [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
Andreas Petlund36e31b0a2010-02-18 02:47:01 +000025int sysctl_tcp_thin_linear_timeouts __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Linus Torvalds1da177e2005-04-16 15:20:36 -070027static void tcp_write_err(struct sock *sk)
28{
29 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
30 sk->sk_error_report(sk);
31
32 tcp_done(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -070033 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034}
35
36/* Do not allow orphaned sockets to eat all our resources.
37 * This is direct violation of TCP specs, but it is required
38 * to prevent DoS attacks. It is called when a retransmission timeout
39 * or zero probe timeout occurs on orphaned socket.
40 *
Stephen Hemmingercaa20d9a2005-11-10 17:13:47 -080041 * Criteria is still not confirmed experimentally and may change.
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 * We kill the socket, if:
43 * 1. If number of orphaned sockets exceeds an administratively configured
44 * limit.
45 * 2. If we have strong memory pressure.
46 */
Yuchung Chengb2482302014-09-29 13:20:38 -070047static int tcp_out_of_resources(struct sock *sk, bool do_reset)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048{
49 struct tcp_sock *tp = tcp_sk(sk);
David S. Millerad1af0f2010-08-25 02:27:49 -070050 int shift = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090052 /* If peer does not open window for long time, or did not transmit
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * anything for long time, penalize it. */
54 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
David S. Millerad1af0f2010-08-25 02:27:49 -070055 shift++;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
57 /* If some dubious ICMP arrived, penalize even more. */
58 if (sk->sk_err_soft)
David S. Millerad1af0f2010-08-25 02:27:49 -070059 shift++;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Arun Sharmaefcdbf22012-01-30 14:16:06 -080061 if (tcp_check_oom(sk, shift)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 /* Catch exceptional cases, when connection requires reset.
63 * 1. Last segment was sent recently. */
64 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
65 /* 2. Window is closed. */
66 (!tp->snd_wnd && !tp->packets_out))
Yuchung Chengb2482302014-09-29 13:20:38 -070067 do_reset = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 if (do_reset)
69 tcp_send_active_reset(sk, GFP_ATOMIC);
70 tcp_done(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -070071 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 return 1;
73 }
74 return 0;
75}
76
77/* Calculate maximal number or retries on an orphaned socket. */
Richard Sailer7533ce32015-10-09 02:41:37 +020078static int tcp_orphan_retries(struct sock *sk, bool alive)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
Nikolay Borisovc402d9b2016-02-03 09:46:55 +020080 int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82 /* We know from an ICMP that something is wrong. */
83 if (sk->sk_err_soft && !alive)
84 retries = 0;
85
86 /* However, if socket sent something recently, select some safe
87 * number of retries. 8 corresponds to >100 seconds with minimal
88 * RTO of 200msec. */
89 if (retries == 0 && alive)
90 retries = 8;
91 return retries;
92}
93
Eric Dumazetce55dd32007-12-21 01:50:43 -080094static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
95{
Fan Dub0f9ca52015-02-10 09:53:16 +080096 struct net *net = sock_net(sk);
97
Eric Dumazetce55dd32007-12-21 01:50:43 -080098 /* Black hole detection */
Fan Dub0f9ca52015-02-10 09:53:16 +080099 if (net->ipv4.sysctl_tcp_mtu_probing) {
Eric Dumazetce55dd32007-12-21 01:50:43 -0800100 if (!icsk->icsk_mtup.enabled) {
101 icsk->icsk_mtup.enabled = 1;
Fan Du05cbc0d2015-03-06 11:18:24 +0800102 icsk->icsk_mtup.probe_timestamp = tcp_time_stamp;
Eric Dumazetce55dd32007-12-21 01:50:43 -0800103 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
104 } else {
Fan Dub0f9ca52015-02-10 09:53:16 +0800105 struct net *net = sock_net(sk);
Eric Dumazetce55dd32007-12-21 01:50:43 -0800106 struct tcp_sock *tp = tcp_sk(sk);
David S. Miller829942c2007-12-21 04:29:16 -0800107 int mss;
108
Eric Dumazet8beb5c52007-12-21 05:58:29 -0800109 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
Fan Dub0f9ca52015-02-10 09:53:16 +0800110 mss = min(net->ipv4.sysctl_tcp_base_mss, mss);
Eric Dumazetce55dd32007-12-21 01:50:43 -0800111 mss = max(mss, 68 - tp->tcp_header_len);
112 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
113 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
114 }
115 }
116}
117
Damian Lukowski2f7de572009-12-07 06:06:16 +0000118/* This function calculates a "timeout" which is equivalent to the timeout of a
Daniel Mack3ad2f3f2010-02-03 08:01:28 +0800119 * TCP connection after "boundary" unsuccessful, exponentially backed-off
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700120 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
121 * syn_set flag is set.
Damian Lukowski2f7de572009-12-07 06:06:16 +0000122 */
123static bool retransmits_timed_out(struct sock *sk,
Jerry Chudca43c72010-08-27 19:13:28 +0000124 unsigned int boundary,
David S. Miller21a180c2010-10-04 11:56:38 -0700125 unsigned int timeout,
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700126 bool syn_set)
Damian Lukowski2f7de572009-12-07 06:06:16 +0000127{
Jerry Chudca43c72010-08-27 19:13:28 +0000128 unsigned int linear_backoff_thresh, start_ts;
Damian Lukowski4d22f7d2010-09-28 13:08:32 -0700129 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
Damian Lukowski2f7de572009-12-07 06:06:16 +0000130
131 if (!inet_csk(sk)->icsk_retransmits)
132 return false;
133
Eric Dumazet7faee5c2014-09-05 15:33:33 -0700134 start_ts = tcp_sk(sk)->retrans_stamp;
135 if (unlikely(!start_ts))
136 start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk));
Damian Lukowski2f7de572009-12-07 06:06:16 +0000137
Jerry Chudca43c72010-08-27 19:13:28 +0000138 if (likely(timeout == 0)) {
David S. Miller21a180c2010-10-04 11:56:38 -0700139 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
Damian Lukowski2f7de572009-12-07 06:06:16 +0000140
Jerry Chudca43c72010-08-27 19:13:28 +0000141 if (boundary <= linear_backoff_thresh)
David S. Miller21a180c2010-10-04 11:56:38 -0700142 timeout = ((2 << boundary) - 1) * rto_base;
Jerry Chudca43c72010-08-27 19:13:28 +0000143 else
David S. Miller21a180c2010-10-04 11:56:38 -0700144 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
Jerry Chudca43c72010-08-27 19:13:28 +0000145 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
146 }
Damian Lukowski2f7de572009-12-07 06:06:16 +0000147 return (tcp_time_stamp - start_ts) >= timeout;
148}
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150/* A write timeout has occurred. Process the after effects. */
151static int tcp_write_timeout(struct sock *sk)
152{
John Heffner5d424d52006-03-20 17:53:41 -0800153 struct inet_connection_sock *icsk = inet_csk(sk);
Yuchung Chengc9686012013-10-29 10:09:05 -0700154 struct tcp_sock *tp = tcp_sk(sk);
Nikolay Borisov6fa25162016-02-03 09:46:49 +0200155 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 int retry_until;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000157 bool do_reset, syn_set = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
Yuchung Chengc9686012013-10-29 10:09:05 -0700160 if (icsk->icsk_retransmits) {
Eric Dumazetb6c67122010-04-08 23:03:29 +0000161 dst_negative_advice(sk);
Yuchung Chengc9686012013-10-29 10:09:05 -0700162 if (tp->syn_fastopen || tp->syn_data)
Daniel Lee2646c832015-04-06 14:37:27 -0700163 tcp_fastopen_cache_set(sk, 0, NULL, true, 0);
Yuchung Chengdd52bc22015-11-18 18:17:31 -0800164 if (tp->syn_data && icsk->icsk_retransmits == 1)
Eric Dumazetc10d9312016-04-29 14:16:47 -0700165 NET_INC_STATS(sock_net(sk),
166 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
Yuchung Chengc9686012013-10-29 10:09:05 -0700167 }
Nikolay Borisov6fa25162016-02-03 09:46:49 +0200168 retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
Rusty Russell3db1cd52011-12-19 13:56:45 +0000169 syn_set = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 } else {
Nikolay Borisovae5c3f42016-02-03 09:46:53 +0200171 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0, 0)) {
Yuchung Cheng0e45f4d2015-11-18 18:17:30 -0800172 /* Some middle-boxes may black-hole Fast Open _after_
173 * the handshake. Therefore we conservatively disable
174 * Fast Open on this path on recurring timeouts with
175 * few or zero bytes acked after Fast Open.
176 */
177 if (tp->syn_data_acked &&
178 tp->bytes_acked <= tp->rx_opt.mss_clamp) {
179 tcp_fastopen_cache_set(sk, 0, NULL, true, 0);
Nikolay Borisovae5c3f42016-02-03 09:46:53 +0200180 if (icsk->icsk_retransmits == net->ipv4.sysctl_tcp_retries1)
Eric Dumazetc10d9312016-04-29 14:16:47 -0700181 NET_INC_STATS(sock_net(sk),
182 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
Yuchung Cheng0e45f4d2015-11-18 18:17:30 -0800183 }
John Heffner5d424d52006-03-20 17:53:41 -0800184 /* Black hole detection */
Eric Dumazetce55dd32007-12-21 01:50:43 -0800185 tcp_mtu_probing(icsk, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Eric Dumazetb6c67122010-04-08 23:03:29 +0000187 dst_negative_advice(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 }
189
Nikolay Borisovc6214a92016-02-03 09:46:54 +0200190 retry_until = net->ipv4.sysctl_tcp_retries2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 if (sock_flag(sk, SOCK_DEAD)) {
Richard Sailer7533ce32015-10-09 02:41:37 +0200192 const bool alive = icsk->icsk_rto < TCP_RTO_MAX;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 retry_until = tcp_orphan_retries(sk, alive);
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000195 do_reset = alive ||
David S. Miller21a180c2010-10-04 11:56:38 -0700196 !retransmits_timed_out(sk, retry_until, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Damian Lukowski6fa12c82009-08-26 00:16:34 +0000198 if (tcp_out_of_resources(sk, do_reset))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 return 1;
200 }
201 }
202
Jerry Chudca43c72010-08-27 19:13:28 +0000203 if (retransmits_timed_out(sk, retry_until,
David S. Miller21a180c2010-10-04 11:56:38 -0700204 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 /* Has it gone just too far? */
206 tcp_write_err(sk);
207 return 1;
208 }
209 return 0;
210}
211
Eric Dumazetc10d9312016-04-29 14:16:47 -0700212/* Called with BH disabled */
Eric Dumazet6f458df2012-07-20 05:45:50 +0000213void tcp_delack_timer_handler(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700216 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
David S. Miller9993e7d2008-01-10 21:56:38 -0800218 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700220 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 goto out;
222
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700223 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
224 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 goto out;
226 }
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700227 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
David S. Millerb03efcf2005-07-08 14:57:23 -0700229 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 struct sk_buff *skb;
231
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700232 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
Peter Zijlstrac57943a2008-10-07 14:18:42 -0700235 sk_backlog_rcv(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237 tp->ucopy.memory = 0;
238 }
239
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700240 if (inet_csk_ack_scheduled(sk)) {
241 if (!icsk->icsk_ack.pingpong) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 /* Delayed ACK missed: inflate ATO. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700243 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 } else {
245 /* Delayed ACK missed: leave pingpong mode and
246 * deflate ATO.
247 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700248 icsk->icsk_ack.pingpong = 0;
249 icsk->icsk_ack.ato = TCP_ATO_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
251 tcp_send_ack(sk);
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700252 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255out:
Eric Dumazetb8da51e2015-05-15 12:39:27 -0700256 if (tcp_under_memory_pressure(sk))
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800257 sk_mem_reclaim(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000258}
259
260static void tcp_delack_timer(unsigned long data)
261{
262 struct sock *sk = (struct sock *)data;
263
264 bh_lock_sock(sk);
265 if (!sock_owned_by_user(sk)) {
266 tcp_delack_timer_handler(sk);
267 } else {
268 inet_csk(sk)->icsk_ack.blocked = 1;
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700269 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000270 /* deleguate our work to tcp_release_cb() */
Eric Dumazet144d56e2012-08-20 00:22:46 +0000271 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
272 sock_hold(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 bh_unlock_sock(sk);
275 sock_put(sk);
276}
277
278static void tcp_probe_timer(struct sock *sk)
279{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300280 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 struct tcp_sock *tp = tcp_sk(sk);
282 int max_probes;
Yuchung Chengb2482302014-09-29 13:20:38 -0700283 u32 start_ts;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
David S. Millerfe067e82007-03-07 12:12:44 -0800285 if (tp->packets_out || !tcp_send_head(sk)) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300286 icsk->icsk_probes_out = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return;
288 }
289
Yuchung Chengb2482302014-09-29 13:20:38 -0700290 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
291 * long as the receiver continues to respond probes. We support this by
292 * default and reset icsk_probes_out with incoming ACKs. But if the
293 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
294 * kill the socket when the retry count and the time exceeds the
295 * corresponding system limit. We also implement similar policy when
296 * we use RTO to probe window in tcp_retransmit_timer().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
Yuchung Chengb2482302014-09-29 13:20:38 -0700298 start_ts = tcp_skb_timestamp(tcp_send_head(sk));
299 if (!start_ts)
300 skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp);
301 else if (icsk->icsk_user_timeout &&
302 (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout)
303 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
Nikolay Borisovc6214a92016-02-03 09:46:54 +0200305 max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 if (sock_flag(sk, SOCK_DEAD)) {
Richard Sailer7533ce32015-10-09 02:41:37 +0200307 const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900308
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 max_probes = tcp_orphan_retries(sk, alive);
Yuchung Chengb2482302014-09-29 13:20:38 -0700310 if (!alive && icsk->icsk_backoff >= max_probes)
311 goto abort;
312 if (tcp_out_of_resources(sk, true))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 return;
314 }
315
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300316 if (icsk->icsk_probes_out > max_probes) {
Yuchung Chengb2482302014-09-29 13:20:38 -0700317abort: tcp_write_err(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 } else {
319 /* Only send another probe if we didn't close things up. */
320 tcp_send_probe0(sk);
321 }
322}
323
324/*
Jerry Chu83368862012-08-31 12:29:12 +0000325 * Timer for Fast Open socket to retransmit SYNACK. Note that the
326 * sk here is the child socket, not the parent (listener) socket.
327 */
328static void tcp_fastopen_synack_timer(struct sock *sk)
329{
330 struct inet_connection_sock *icsk = inet_csk(sk);
331 int max_retries = icsk->icsk_syn_retries ? :
Nikolay Borisov7c083ec2016-02-03 09:46:50 +0200332 sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
Jerry Chu83368862012-08-31 12:29:12 +0000333 struct request_sock *req;
334
335 req = tcp_sk(sk)->fastopen_rsk;
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700336 req->rsk_ops->syn_ack_timeout(req);
Jerry Chu83368862012-08-31 12:29:12 +0000337
Eric Dumazete6c022a2012-10-27 23:16:46 +0000338 if (req->num_timeout >= max_retries) {
Jerry Chu83368862012-08-31 12:29:12 +0000339 tcp_write_err(sk);
340 return;
341 }
342 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
343 * returned from rtx_syn_ack() to make it more persistent like
344 * regular retransmit because if the child socket has been accepted
345 * it's not good to give up too easily.
346 */
Eric Dumazete6c022a2012-10-27 23:16:46 +0000347 inet_rtx_syn_ack(sk, req);
348 req->num_timeout++;
Jerry Chu83368862012-08-31 12:29:12 +0000349 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Eric Dumazete6c022a2012-10-27 23:16:46 +0000350 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
Jerry Chu83368862012-08-31 12:29:12 +0000351}
352
353/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * The TCP retransmit timer.
355 */
356
Damian Lukowskif1ecd5d2009-08-26 00:16:31 +0000357void tcp_retransmit_timer(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
359 struct tcp_sock *tp = tcp_sk(sk);
Nikolay Borisovae5c3f42016-02-03 09:46:53 +0200360 struct net *net = sock_net(sk);
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700361 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Jerry Chu83368862012-08-31 12:29:12 +0000363 if (tp->fastopen_rsk) {
Jerry Chu37561f62012-10-22 11:26:36 +0000364 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
365 sk->sk_state != TCP_FIN_WAIT1);
Jerry Chu83368862012-08-31 12:29:12 +0000366 tcp_fastopen_synack_timer(sk);
367 /* Before we receive ACK to our SYN-ACK don't retransmit
368 * anything else (e.g., data or FIN segments).
369 */
370 return;
371 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 if (!tp->packets_out)
373 goto out;
374
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700375 WARN_ON(tcp_write_queue_empty(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Nandita Dukkipati9b717a82013-03-11 10:00:44 +0000377 tp->tlp_high_seq = 0;
378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
380 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
381 /* Receiver dastardly shrinks window. Our retransmits
382 * become zero probes, but we should not timeout this
383 * connection. If the socket is an orphan, time it out,
384 * we cannot allow such beasts to hang infinitely.
385 */
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700386 struct inet_sock *inet = inet_sk(sk);
387 if (sk->sk_family == AF_INET) {
Joe Perchesba7a46f2014-11-11 10:59:17 -0800388 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
389 &inet->inet_daddr,
390 ntohs(inet->inet_dport),
391 inet->inet_num,
392 tp->snd_una, tp->snd_nxt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
Eric Dumazetdfd56b82011-12-10 09:48:31 +0000394#if IS_ENABLED(CONFIG_IPV6)
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700395 else if (sk->sk_family == AF_INET6) {
Joe Perchesba7a46f2014-11-11 10:59:17 -0800396 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
397 &sk->sk_v6_daddr,
398 ntohs(inet->inet_dport),
399 inet->inet_num,
400 tp->snd_una, tp->snd_nxt);
YOSHIFUJI Hideaki569508c2008-04-14 04:09:36 -0700401 }
402#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
404 tcp_write_err(sk);
405 goto out;
406 }
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400407 tcp_enter_loss(sk);
Eric Dumazet10d3be52016-04-21 10:55:23 -0700408 tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 __sk_dst_reset(sk);
410 goto out_reset_timer;
411 }
412
413 if (tcp_write_timeout(sk))
414 goto out;
415
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700416 if (icsk->icsk_retransmits == 0) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700417 int mib_idx;
418
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000419 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
Ilpo Järvinenbc079e92009-02-28 04:44:34 +0000420 if (tcp_is_sack(tp))
421 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
422 else
423 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300424 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700425 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
Ilpo Järvinenc60ce4e2010-10-14 01:52:09 +0000426 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
427 tp->sacked_out) {
428 if (tcp_is_sack(tp))
429 mib_idx = LINUX_MIB_TCPSACKFAILURES;
430 else
431 mib_idx = LINUX_MIB_TCPRENOFAILURES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 } else {
Pavel Emelyanov40b215e2008-07-03 01:05:41 -0700433 mib_idx = LINUX_MIB_TCPTIMEOUTS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 }
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700435 __NET_INC_STATS(sock_net(sk), mib_idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 }
437
Neal Cardwell5ae344c2014-08-04 19:12:29 -0400438 tcp_enter_loss(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Eric Dumazet10d3be52016-04-21 10:55:23 -0700440 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1) > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 /* Retransmission failed because of local congestion,
442 * do not backoff.
443 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700444 if (!icsk->icsk_retransmits)
445 icsk->icsk_retransmits = 1;
446 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700447 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
448 TCP_RTO_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 goto out;
450 }
451
452 /* Increase the timeout each time we retransmit. Note that
453 * we do not increase the rtt estimate. rto is initialized
454 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
455 * that doubling rto each time is the least we can get away with.
456 * In KA9Q, Karn uses this for the first few times, and then
457 * goes to quadratic. netBSD doubles, but only goes up to *64,
458 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
459 * defined in the protocol as the maximum possible RTT. I guess
460 * we'll have to use something other than TCP to talk to the
461 * University of Mars.
462 *
463 * PAWS allows us longer timeouts and large windows, so once
464 * implemented ftp to mars will work nicely. We will have to fix
465 * the 120 second clamps though!
466 */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700467 icsk->icsk_backoff++;
468 icsk->icsk_retransmits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470out_reset_timer:
Andreas Petlund36e31b0a2010-02-18 02:47:01 +0000471 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
472 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
473 * might be increased if the stream oscillates between thin and thick,
474 * thus the old value might already be too high compared to the value
475 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
476 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
477 * exponential backoff behaviour to avoid continue hammering
478 * linear-timeout retransmissions into a black hole
479 */
480 if (sk->sk_state == TCP_ESTABLISHED &&
481 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
482 tcp_stream_is_thin(tp) &&
483 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
484 icsk->icsk_backoff = 0;
485 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
486 } else {
487 /* Use normal (exponential) backoff */
488 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
489 }
Arnaldo Carvalho de Melo3f421ba2005-08-09 20:11:08 -0700490 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
Nikolay Borisovae5c3f42016-02-03 09:46:53 +0200491 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 __sk_dst_reset(sk);
493
494out:;
495}
496
Eric Dumazetc10d9312016-04-29 14:16:47 -0700497/* Called with BH disabled */
Eric Dumazet6f458df2012-07-20 05:45:50 +0000498void tcp_write_timer_handler(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700500 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 int event;
502
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700503 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 goto out;
505
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700506 if (time_after(icsk->icsk_timeout, jiffies)) {
507 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 goto out;
509 }
510
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700511 event = icsk->icsk_pending;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
513 switch (event) {
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000514 case ICSK_TIME_EARLY_RETRANS:
515 tcp_resume_early_retransmit(sk);
516 break;
517 case ICSK_TIME_LOSS_PROBE:
518 tcp_send_loss_probe(sk);
519 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700520 case ICSK_TIME_RETRANS:
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000521 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 tcp_retransmit_timer(sk);
523 break;
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700524 case ICSK_TIME_PROBE0:
Nandita Dukkipati6ba8a3b2013-03-11 10:00:43 +0000525 icsk->icsk_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 tcp_probe_timer(sk);
527 break;
528 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
530out:
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800531 sk_mem_reclaim(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000532}
533
534static void tcp_write_timer(unsigned long data)
535{
536 struct sock *sk = (struct sock *)data;
537
538 bh_lock_sock(sk);
539 if (!sock_owned_by_user(sk)) {
540 tcp_write_timer_handler(sk);
541 } else {
542 /* deleguate our work to tcp_release_cb() */
Eric Dumazet144d56e2012-08-20 00:22:46 +0000543 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
544 sock_hold(sk);
Eric Dumazet6f458df2012-07-20 05:45:50 +0000545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 bh_unlock_sock(sk);
547 sock_put(sk);
548}
549
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700550void tcp_syn_ack_timeout(const struct request_sock *req)
Octavian Purdila72659ec2010-01-17 19:09:39 -0800551{
Eric Dumazet42cb80a2015-03-22 10:22:19 -0700552 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
553
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700554 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
Octavian Purdila72659ec2010-01-17 19:09:39 -0800555}
556EXPORT_SYMBOL(tcp_syn_ack_timeout);
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558void tcp_set_keepalive(struct sock *sk, int val)
559{
560 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
561 return;
562
563 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700564 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 else if (!val)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700566 inet_csk_delete_keepalive_timer(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
569
570static void tcp_keepalive_timer (unsigned long data)
571{
572 struct sock *sk = (struct sock *) data;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300573 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 struct tcp_sock *tp = tcp_sk(sk);
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000575 u32 elapsed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 /* Only process if socket is not in use. */
578 bh_lock_sock(sk);
579 if (sock_owned_by_user(sk)) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900580 /* Try again later. */
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700581 inet_csk_reset_keepalive_timer (sk, HZ/20);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 goto out;
583 }
584
585 if (sk->sk_state == TCP_LISTEN) {
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700586 pr_err("Hmm... keepalive on a LISTEN ???\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 goto out;
588 }
589
590 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
591 if (tp->linger2 >= 0) {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700592 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 if (tmo > 0) {
595 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
596 goto out;
597 }
598 }
599 tcp_send_active_reset(sk, GFP_ATOMIC);
600 goto death;
601 }
602
603 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
604 goto out;
605
606 elapsed = keepalive_time_when(tp);
607
608 /* It is alive without keepalive 8) */
David S. Millerfe067e82007-03-07 12:12:44 -0800609 if (tp->packets_out || tcp_send_head(sk))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 goto resched;
611
Flavio Leitner6c37e5d2010-04-26 18:33:27 +0000612 elapsed = keepalive_time_elapsed(tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614 if (elapsed >= keepalive_time_when(tp)) {
Jerry Chudca43c72010-08-27 19:13:28 +0000615 /* If the TCP_USER_TIMEOUT option is enabled, use that
616 * to determine when to timeout instead.
617 */
618 if ((icsk->icsk_user_timeout != 0 &&
619 elapsed >= icsk->icsk_user_timeout &&
620 icsk->icsk_probes_out > 0) ||
621 (icsk->icsk_user_timeout == 0 &&
622 icsk->icsk_probes_out >= keepalive_probes(tp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 tcp_send_active_reset(sk, GFP_ATOMIC);
624 tcp_write_err(sk);
625 goto out;
626 }
Eric Dumazete520af42015-05-06 14:26:25 -0700627 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300628 icsk->icsk_probes_out++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 elapsed = keepalive_intvl_when(tp);
630 } else {
631 /* If keepalive was lost due to local congestion,
632 * try harder.
633 */
634 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
635 }
636 } else {
637 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
638 elapsed = keepalive_time_when(tp) - elapsed;
639 }
640
Hideo Aoki3ab224b2007-12-31 00:11:19 -0800641 sk_mem_reclaim(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643resched:
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700644 inet_csk_reset_keepalive_timer (sk, elapsed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 goto out;
646
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900647death:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 tcp_done(sk);
649
650out:
651 bh_unlock_sock(sk);
652 sock_put(sk);
653}
Eric Dumazet6f458df2012-07-20 05:45:50 +0000654
655void tcp_init_xmit_timers(struct sock *sk)
656{
657 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
658 &tcp_keepalive_timer);
659}