ipv4: ERROR: code indent should use tabs where possible
[deliverable/linux.git] / net / ipv4 / tcp_timer.c
CommitLineData
1da177e4
LT
1/*
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 *
02c30a84 8 * Authors: Ross Biro
1da177e4
LT
9 * 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>
5a0e3ad6 22#include <linux/gfp.h>
1da177e4
LT
23#include <net/tcp.h>
24
ab32ea5d
BH
25int 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;
36e31b0a 33int sysctl_tcp_thin_linear_timeouts __read_mostly;
1da177e4 34
1da177e4
LT
35static 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);
de0744af 41 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
1da177e4
LT
42}
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 *
caa20d9a 49 * Criteria is still not confirmed experimentally and may change.
1da177e4
LT
50 * 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 */
55static int tcp_out_of_resources(struct sock *sk, int do_reset)
56{
57 struct tcp_sock *tp = tcp_sk(sk);
ad1af0fe 58 int shift = 0;
1da177e4 59
e905a9ed 60 /* If peer does not open window for long time, or did not transmit
1da177e4
LT
61 * anything for long time, penalize it. */
62 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
ad1af0fe 63 shift++;
1da177e4
LT
64
65 /* If some dubious ICMP arrived, penalize even more. */
66 if (sk->sk_err_soft)
ad1af0fe 67 shift++;
1da177e4 68
efcdbf24 69 if (tcp_check_oom(sk, shift)) {
1da177e4
LT
70 /* 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))
75 do_reset = 1;
76 if (do_reset)
77 tcp_send_active_reset(sk, GFP_ATOMIC);
78 tcp_done(sk);
de0744af 79 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
1da177e4
LT
80 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
ce55dd36
ED
102static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
103{
ce55dd36
ED
104 /* 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);
829942c1
DM
111 int mss;
112
8beb5c5f 113 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
ce55dd36
ED
114 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
2f7de571 122/* This function calculates a "timeout" which is equivalent to the timeout of a
3ad2f3fb 123 * TCP connection after "boundary" unsuccessful, exponentially backed-off
4d22f7d3
DL
124 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
125 * syn_set flag is set.
2f7de571
DL
126 */
127static bool retransmits_timed_out(struct sock *sk,
dca43c75 128 unsigned int boundary,
21a180cd 129 unsigned int timeout,
4d22f7d3 130 bool syn_set)
2f7de571 131{
dca43c75 132 unsigned int linear_backoff_thresh, start_ts;
4d22f7d3 133 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
2f7de571
DL
134
135 if (!inet_csk(sk)->icsk_retransmits)
136 return false;
137
138 if (unlikely(!tcp_sk(sk)->retrans_stamp))
139 start_ts = TCP_SKB_CB(tcp_write_queue_head(sk))->when;
140 else
141 start_ts = tcp_sk(sk)->retrans_stamp;
142
dca43c75 143 if (likely(timeout == 0)) {
21a180cd 144 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
2f7de571 145
dca43c75 146 if (boundary <= linear_backoff_thresh)
21a180cd 147 timeout = ((2 << boundary) - 1) * rto_base;
dca43c75 148 else
21a180cd 149 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
dca43c75
JC
150 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
151 }
2f7de571
DL
152 return (tcp_time_stamp - start_ts) >= timeout;
153}
154
1da177e4
LT
155/* A write timeout has occurred. Process the after effects. */
156static int tcp_write_timeout(struct sock *sk)
157{
5d424d5a 158 struct inet_connection_sock *icsk = inet_csk(sk);
c968601d 159 struct tcp_sock *tp = tcp_sk(sk);
1da177e4 160 int retry_until;
3db1cd5c 161 bool do_reset, syn_set = false;
1da177e4
LT
162
163 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
c968601d 164 if (icsk->icsk_retransmits) {
b6c6712a 165 dst_negative_advice(sk);
c968601d
YC
166 if (tp->syn_fastopen || tp->syn_data)
167 tcp_fastopen_cache_set(sk, 0, NULL, true);
168 }
463c84b9 169 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
3db1cd5c 170 syn_set = true;
1da177e4 171 } else {
21a180cd 172 if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
5d424d5a 173 /* Black hole detection */
ce55dd36 174 tcp_mtu_probing(icsk, sk);
1da177e4 175
b6c6712a 176 dst_negative_advice(sk);
1da177e4
LT
177 }
178
179 retry_until = sysctl_tcp_retries2;
180 if (sock_flag(sk, SOCK_DEAD)) {
463c84b9 181 const int alive = (icsk->icsk_rto < TCP_RTO_MAX);
e905a9ed 182
1da177e4 183 retry_until = tcp_orphan_retries(sk, alive);
6fa12c85 184 do_reset = alive ||
21a180cd 185 !retransmits_timed_out(sk, retry_until, 0, 0);
1da177e4 186
6fa12c85 187 if (tcp_out_of_resources(sk, do_reset))
1da177e4
LT
188 return 1;
189 }
190 }
191
dca43c75 192 if (retransmits_timed_out(sk, retry_until,
21a180cd 193 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
1da177e4
LT
194 /* Has it gone just too far? */
195 tcp_write_err(sk);
196 return 1;
197 }
198 return 0;
199}
200
6f458dfb 201void tcp_delack_timer_handler(struct sock *sk)
1da177e4 202{
1da177e4 203 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 204 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 205
9993e7d3 206 sk_mem_reclaim_partial(sk);
1da177e4 207
463c84b9 208 if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
1da177e4
LT
209 goto out;
210
463c84b9
ACM
211 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
212 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
1da177e4
LT
213 goto out;
214 }
463c84b9 215 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
1da177e4 216
b03efcfb 217 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
1da177e4
LT
218 struct sk_buff *skb;
219
de0744af 220 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
1da177e4
LT
221
222 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
c57943a1 223 sk_backlog_rcv(sk, skb);
1da177e4
LT
224
225 tp->ucopy.memory = 0;
226 }
227
463c84b9
ACM
228 if (inet_csk_ack_scheduled(sk)) {
229 if (!icsk->icsk_ack.pingpong) {
1da177e4 230 /* Delayed ACK missed: inflate ATO. */
463c84b9 231 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
1da177e4
LT
232 } else {
233 /* Delayed ACK missed: leave pingpong mode and
234 * deflate ATO.
235 */
463c84b9
ACM
236 icsk->icsk_ack.pingpong = 0;
237 icsk->icsk_ack.ato = TCP_ATO_MIN;
1da177e4
LT
238 }
239 tcp_send_ack(sk);
de0744af 240 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
1da177e4 241 }
1da177e4
LT
242
243out:
180d8cd9 244 if (sk_under_memory_pressure(sk))
3ab224be 245 sk_mem_reclaim(sk);
6f458dfb
ED
246}
247
248static void tcp_delack_timer(unsigned long data)
249{
250 struct sock *sk = (struct sock *)data;
251
252 bh_lock_sock(sk);
253 if (!sock_owned_by_user(sk)) {
254 tcp_delack_timer_handler(sk);
255 } else {
256 inet_csk(sk)->icsk_ack.blocked = 1;
257 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
258 /* deleguate our work to tcp_release_cb() */
144d56e9
ED
259 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
260 sock_hold(sk);
6f458dfb 261 }
1da177e4
LT
262 bh_unlock_sock(sk);
263 sock_put(sk);
264}
265
266static void tcp_probe_timer(struct sock *sk)
267{
6687e988 268 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT
269 struct tcp_sock *tp = tcp_sk(sk);
270 int max_probes;
271
fe067e8a 272 if (tp->packets_out || !tcp_send_head(sk)) {
6687e988 273 icsk->icsk_probes_out = 0;
1da177e4
LT
274 return;
275 }
276
277 /* *WARNING* RFC 1122 forbids this
278 *
279 * It doesn't AFAIK, because we kill the retransmit timer -AK
280 *
281 * FIXME: We ought not to do it, Solaris 2.5 actually has fixing
282 * this behaviour in Solaris down as a bug fix. [AC]
283 *
6687e988 284 * Let me to explain. icsk_probes_out is zeroed by incoming ACKs
1da177e4
LT
285 * even if they advertise zero window. Hence, connection is killed only
286 * if we received no ACKs for normal connection timeout. It is not killed
287 * only because window stays zero for some time, window may be zero
288 * until armageddon and even later. We are in full accordance
289 * with RFCs, only probe timer combines both retransmission timeout
290 * and probe timeout in one bottle. --ANK
291 */
292 max_probes = sysctl_tcp_retries2;
293
294 if (sock_flag(sk, SOCK_DEAD)) {
463c84b9 295 const int alive = ((icsk->icsk_rto << icsk->icsk_backoff) < TCP_RTO_MAX);
e905a9ed 296
1da177e4
LT
297 max_probes = tcp_orphan_retries(sk, alive);
298
6687e988 299 if (tcp_out_of_resources(sk, alive || icsk->icsk_probes_out <= max_probes))
1da177e4
LT
300 return;
301 }
302
6687e988 303 if (icsk->icsk_probes_out > max_probes) {
1da177e4
LT
304 tcp_write_err(sk);
305 } else {
306 /* Only send another probe if we didn't close things up. */
307 tcp_send_probe0(sk);
308 }
309}
310
8336886f
JC
311/*
312 * Timer for Fast Open socket to retransmit SYNACK. Note that the
313 * sk here is the child socket, not the parent (listener) socket.
314 */
315static void tcp_fastopen_synack_timer(struct sock *sk)
316{
317 struct inet_connection_sock *icsk = inet_csk(sk);
318 int max_retries = icsk->icsk_syn_retries ? :
319 sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
320 struct request_sock *req;
321
322 req = tcp_sk(sk)->fastopen_rsk;
323 req->rsk_ops->syn_ack_timeout(sk, req);
324
e6c022a4 325 if (req->num_timeout >= max_retries) {
8336886f
JC
326 tcp_write_err(sk);
327 return;
328 }
329 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
330 * returned from rtx_syn_ack() to make it more persistent like
331 * regular retransmit because if the child socket has been accepted
332 * it's not good to give up too easily.
333 */
e6c022a4
ED
334 inet_rtx_syn_ack(sk, req);
335 req->num_timeout++;
8336886f 336 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
e6c022a4 337 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
8336886f
JC
338}
339
1da177e4
LT
340/*
341 * The TCP retransmit timer.
342 */
343
f1ecd5d9 344void tcp_retransmit_timer(struct sock *sk)
1da177e4
LT
345{
346 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 347 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 348
8336886f 349 if (tp->fastopen_rsk) {
37561f68
JC
350 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
351 sk->sk_state != TCP_FIN_WAIT1);
8336886f
JC
352 tcp_fastopen_synack_timer(sk);
353 /* Before we receive ACK to our SYN-ACK don't retransmit
354 * anything else (e.g., data or FIN segments).
355 */
356 return;
357 }
1da177e4
LT
358 if (!tp->packets_out)
359 goto out;
360
547b792c 361 WARN_ON(tcp_write_queue_empty(sk));
1da177e4 362
9b717a8d
ND
363 tp->tlp_high_seq = 0;
364
1da177e4
LT
365 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
366 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
367 /* Receiver dastardly shrinks window. Our retransmits
368 * become zero probes, but we should not timeout this
369 * connection. If the socket is an orphan, time it out,
370 * we cannot allow such beasts to hang infinitely.
371 */
569508c9
YH
372 struct inet_sock *inet = inet_sk(sk);
373 if (sk->sk_family == AF_INET) {
afd46503
JP
374 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
375 &inet->inet_daddr,
376 ntohs(inet->inet_dport), inet->inet_num,
377 tp->snd_una, tp->snd_nxt);
1da177e4 378 }
dfd56b8b 379#if IS_ENABLED(CONFIG_IPV6)
569508c9 380 else if (sk->sk_family == AF_INET6) {
afd46503 381 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
efe4208f 382 &sk->sk_v6_daddr,
afd46503
JP
383 ntohs(inet->inet_dport), inet->inet_num,
384 tp->snd_una, tp->snd_nxt);
569508c9 385 }
1da177e4
LT
386#endif
387 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
388 tcp_write_err(sk);
389 goto out;
390 }
391 tcp_enter_loss(sk, 0);
fe067e8a 392 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
1da177e4
LT
393 __sk_dst_reset(sk);
394 goto out_reset_timer;
395 }
396
397 if (tcp_write_timeout(sk))
398 goto out;
399
463c84b9 400 if (icsk->icsk_retransmits == 0) {
40b215e5
PE
401 int mib_idx;
402
c60ce4e2 403 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
bc079e9e
IJ
404 if (tcp_is_sack(tp))
405 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
406 else
407 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
6687e988 408 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
40b215e5 409 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
c60ce4e2
IJ
410 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
411 tp->sacked_out) {
412 if (tcp_is_sack(tp))
413 mib_idx = LINUX_MIB_TCPSACKFAILURES;
414 else
415 mib_idx = LINUX_MIB_TCPRENOFAILURES;
1da177e4 416 } else {
40b215e5 417 mib_idx = LINUX_MIB_TCPTIMEOUTS;
1da177e4 418 }
de0744af 419 NET_INC_STATS_BH(sock_net(sk), mib_idx);
1da177e4
LT
420 }
421
9b44190d 422 tcp_enter_loss(sk, 0);
1da177e4 423
fe067e8a 424 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
1da177e4
LT
425 /* Retransmission failed because of local congestion,
426 * do not backoff.
427 */
463c84b9
ACM
428 if (!icsk->icsk_retransmits)
429 icsk->icsk_retransmits = 1;
430 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
3f421baa
ACM
431 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
432 TCP_RTO_MAX);
1da177e4
LT
433 goto out;
434 }
435
436 /* Increase the timeout each time we retransmit. Note that
437 * we do not increase the rtt estimate. rto is initialized
438 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
439 * that doubling rto each time is the least we can get away with.
440 * In KA9Q, Karn uses this for the first few times, and then
441 * goes to quadratic. netBSD doubles, but only goes up to *64,
442 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
443 * defined in the protocol as the maximum possible RTT. I guess
444 * we'll have to use something other than TCP to talk to the
445 * University of Mars.
446 *
447 * PAWS allows us longer timeouts and large windows, so once
448 * implemented ftp to mars will work nicely. We will have to fix
449 * the 120 second clamps though!
450 */
463c84b9
ACM
451 icsk->icsk_backoff++;
452 icsk->icsk_retransmits++;
1da177e4
LT
453
454out_reset_timer:
36e31b0a
AP
455 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
456 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
457 * might be increased if the stream oscillates between thin and thick,
458 * thus the old value might already be too high compared to the value
459 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
460 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
461 * exponential backoff behaviour to avoid continue hammering
462 * linear-timeout retransmissions into a black hole
463 */
464 if (sk->sk_state == TCP_ESTABLISHED &&
465 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
466 tcp_stream_is_thin(tp) &&
467 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
468 icsk->icsk_backoff = 0;
469 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
470 } else {
471 /* Use normal (exponential) backoff */
472 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
473 }
3f421baa 474 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
21a180cd 475 if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
1da177e4
LT
476 __sk_dst_reset(sk);
477
478out:;
479}
480
6f458dfb 481void tcp_write_timer_handler(struct sock *sk)
1da177e4 482{
463c84b9 483 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4
LT
484 int event;
485
463c84b9 486 if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
1da177e4
LT
487 goto out;
488
463c84b9
ACM
489 if (time_after(icsk->icsk_timeout, jiffies)) {
490 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
1da177e4
LT
491 goto out;
492 }
493
463c84b9 494 event = icsk->icsk_pending;
1da177e4
LT
495
496 switch (event) {
6ba8a3b1
ND
497 case ICSK_TIME_EARLY_RETRANS:
498 tcp_resume_early_retransmit(sk);
499 break;
500 case ICSK_TIME_LOSS_PROBE:
501 tcp_send_loss_probe(sk);
502 break;
463c84b9 503 case ICSK_TIME_RETRANS:
6ba8a3b1 504 icsk->icsk_pending = 0;
1da177e4
LT
505 tcp_retransmit_timer(sk);
506 break;
463c84b9 507 case ICSK_TIME_PROBE0:
6ba8a3b1 508 icsk->icsk_pending = 0;
1da177e4
LT
509 tcp_probe_timer(sk);
510 break;
511 }
1da177e4
LT
512
513out:
3ab224be 514 sk_mem_reclaim(sk);
6f458dfb
ED
515}
516
517static void tcp_write_timer(unsigned long data)
518{
519 struct sock *sk = (struct sock *)data;
520
521 bh_lock_sock(sk);
522 if (!sock_owned_by_user(sk)) {
523 tcp_write_timer_handler(sk);
524 } else {
525 /* deleguate our work to tcp_release_cb() */
144d56e9
ED
526 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
527 sock_hold(sk);
6f458dfb 528 }
1da177e4
LT
529 bh_unlock_sock(sk);
530 sock_put(sk);
531}
532
295f7324
ACM
533/*
534 * Timer for listening sockets
535 */
536
537static void tcp_synack_timer(struct sock *sk)
538{
a019d6fe
ACM
539 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
540 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
1da177e4
LT
541}
542
72659ecc
OP
543void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
544{
545 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
546}
547EXPORT_SYMBOL(tcp_syn_ack_timeout);
548
1da177e4
LT
549void tcp_set_keepalive(struct sock *sk, int val)
550{
551 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
552 return;
553
554 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
463c84b9 555 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
1da177e4 556 else if (!val)
463c84b9 557 inet_csk_delete_keepalive_timer(sk);
1da177e4
LT
558}
559
560
561static void tcp_keepalive_timer (unsigned long data)
562{
563 struct sock *sk = (struct sock *) data;
6687e988 564 struct inet_connection_sock *icsk = inet_csk(sk);
1da177e4 565 struct tcp_sock *tp = tcp_sk(sk);
6c37e5de 566 u32 elapsed;
1da177e4
LT
567
568 /* Only process if socket is not in use. */
569 bh_lock_sock(sk);
570 if (sock_owned_by_user(sk)) {
e905a9ed 571 /* Try again later. */
463c84b9 572 inet_csk_reset_keepalive_timer (sk, HZ/20);
1da177e4
LT
573 goto out;
574 }
575
576 if (sk->sk_state == TCP_LISTEN) {
577 tcp_synack_timer(sk);
578 goto out;
579 }
580
581 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
582 if (tp->linger2 >= 0) {
463c84b9 583 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
1da177e4
LT
584
585 if (tmo > 0) {
586 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
587 goto out;
588 }
589 }
590 tcp_send_active_reset(sk, GFP_ATOMIC);
591 goto death;
592 }
593
594 if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
595 goto out;
596
597 elapsed = keepalive_time_when(tp);
598
599 /* It is alive without keepalive 8) */
fe067e8a 600 if (tp->packets_out || tcp_send_head(sk))
1da177e4
LT
601 goto resched;
602
6c37e5de 603 elapsed = keepalive_time_elapsed(tp);
1da177e4
LT
604
605 if (elapsed >= keepalive_time_when(tp)) {
dca43c75
JC
606 /* If the TCP_USER_TIMEOUT option is enabled, use that
607 * to determine when to timeout instead.
608 */
609 if ((icsk->icsk_user_timeout != 0 &&
610 elapsed >= icsk->icsk_user_timeout &&
611 icsk->icsk_probes_out > 0) ||
612 (icsk->icsk_user_timeout == 0 &&
613 icsk->icsk_probes_out >= keepalive_probes(tp))) {
1da177e4
LT
614 tcp_send_active_reset(sk, GFP_ATOMIC);
615 tcp_write_err(sk);
616 goto out;
617 }
618 if (tcp_write_wakeup(sk) <= 0) {
6687e988 619 icsk->icsk_probes_out++;
1da177e4
LT
620 elapsed = keepalive_intvl_when(tp);
621 } else {
622 /* If keepalive was lost due to local congestion,
623 * try harder.
624 */
625 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
626 }
627 } else {
628 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
629 elapsed = keepalive_time_when(tp) - elapsed;
630 }
631
3ab224be 632 sk_mem_reclaim(sk);
1da177e4
LT
633
634resched:
463c84b9 635 inet_csk_reset_keepalive_timer (sk, elapsed);
1da177e4
LT
636 goto out;
637
e905a9ed 638death:
1da177e4
LT
639 tcp_done(sk);
640
641out:
642 bh_unlock_sock(sk);
643 sock_put(sk);
644}
6f458dfb
ED
645
646void tcp_init_xmit_timers(struct sock *sk)
647{
648 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
649 &tcp_keepalive_timer);
650}
651EXPORT_SYMBOL(tcp_init_xmit_timers);
This page took 1.048263 seconds and 5 git commands to generate.