~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ipv4/tcp_timer.c

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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  *
  8  * Authors:     Ross Biro
  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>
 22 #include <linux/gfp.h>
 23 #include <net/tcp.h>
 24 
 25 int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES;
 26 int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES;
 27 int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME;
 28 int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES;
 29 int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
 30 int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
 31 int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
 32 int sysctl_tcp_orphan_retries __read_mostly;
 33 int sysctl_tcp_thin_linear_timeouts __read_mostly;
 34 
 35 static 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);
 41         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
 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  *
 49  * Criteria is still not confirmed experimentally and may change.
 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  */
 55 static int tcp_out_of_resources(struct sock *sk, bool do_reset)
 56 {
 57         struct tcp_sock *tp = tcp_sk(sk);
 58         int shift = 0;
 59 
 60         /* If peer does not open window for long time, or did not transmit
 61          * anything for long time, penalize it. */
 62         if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
 63                 shift++;
 64 
 65         /* If some dubious ICMP arrived, penalize even more. */
 66         if (sk->sk_err_soft)
 67                 shift++;
 68 
 69         if (tcp_check_oom(sk, shift)) {
 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 = true;
 76                 if (do_reset)
 77                         tcp_send_active_reset(sk, GFP_ATOMIC);
 78                 tcp_done(sk);
 79                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
 80                 return 1;
 81         }
 82         return 0;
 83 }
 84 
 85 /* Calculate maximal number or retries on an orphaned socket. */
 86 static 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 
102 static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
103 {
104         struct net *net = sock_net(sk);
105 
106         /* Black hole detection */
107         if (net->ipv4.sysctl_tcp_mtu_probing) {
108                 if (!icsk->icsk_mtup.enabled) {
109                         icsk->icsk_mtup.enabled = 1;
110                         tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
111                 } else {
112                         struct net *net = sock_net(sk);
113                         struct tcp_sock *tp = tcp_sk(sk);
114                         int mss;
115 
116                         mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
117                         mss = min(net->ipv4.sysctl_tcp_base_mss, mss);
118                         mss = max(mss, 68 - tp->tcp_header_len);
119                         icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
120                         tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
121                 }
122         }
123 }
124 
125 /* This function calculates a "timeout" which is equivalent to the timeout of a
126  * TCP connection after "boundary" unsuccessful, exponentially backed-off
127  * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
128  * syn_set flag is set.
129  */
130 static bool retransmits_timed_out(struct sock *sk,
131                                   unsigned int boundary,
132                                   unsigned int timeout,
133                                   bool syn_set)
134 {
135         unsigned int linear_backoff_thresh, start_ts;
136         unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
137 
138         if (!inet_csk(sk)->icsk_retransmits)
139                 return false;
140 
141         start_ts = tcp_sk(sk)->retrans_stamp;
142         if (unlikely(!start_ts))
143                 start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk));
144 
145         if (likely(timeout == 0)) {
146                 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
147 
148                 if (boundary <= linear_backoff_thresh)
149                         timeout = ((2 << boundary) - 1) * rto_base;
150                 else
151                         timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
152                                 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
153         }
154         return (tcp_time_stamp - start_ts) >= timeout;
155 }
156 
157 /* A write timeout has occurred. Process the after effects. */
158 static int tcp_write_timeout(struct sock *sk)
159 {
160         struct inet_connection_sock *icsk = inet_csk(sk);
161         struct tcp_sock *tp = tcp_sk(sk);
162         int retry_until;
163         bool do_reset, syn_set = false;
164 
165         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
166                 if (icsk->icsk_retransmits) {
167                         dst_negative_advice(sk);
168                         if (tp->syn_fastopen || tp->syn_data)
169                                 tcp_fastopen_cache_set(sk, 0, NULL, true);
170                         if (tp->syn_data)
171                                 NET_INC_STATS_BH(sock_net(sk),
172                                                  LINUX_MIB_TCPFASTOPENACTIVEFAIL);
173                 }
174                 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
175                 syn_set = true;
176         } else {
177                 if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
178                         /* Black hole detection */
179                         tcp_mtu_probing(icsk, sk);
180 
181                         dst_negative_advice(sk);
182                 }
183 
184                 retry_until = sysctl_tcp_retries2;
185                 if (sock_flag(sk, SOCK_DEAD)) {
186                         const int alive = icsk->icsk_rto < TCP_RTO_MAX;
187 
188                         retry_until = tcp_orphan_retries(sk, alive);
189                         do_reset = alive ||
190                                 !retransmits_timed_out(sk, retry_until, 0, 0);
191 
192                         if (tcp_out_of_resources(sk, do_reset))
193                                 return 1;
194                 }
195         }
196 
197         if (retransmits_timed_out(sk, retry_until,
198                                   syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
199                 /* Has it gone just too far? */
200                 tcp_write_err(sk);
201                 return 1;
202         }
203         return 0;
204 }
205 
206 void tcp_delack_timer_handler(struct sock *sk)
207 {
208         struct tcp_sock *tp = tcp_sk(sk);
209         struct inet_connection_sock *icsk = inet_csk(sk);
210 
211         sk_mem_reclaim_partial(sk);
212 
213         if (sk->sk_state == TCP_CLOSE || !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
214                 goto out;
215 
216         if (time_after(icsk->icsk_ack.timeout, jiffies)) {
217                 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
218                 goto out;
219         }
220         icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
221 
222         if (!skb_queue_empty(&tp->ucopy.prequeue)) {
223                 struct sk_buff *skb;
224 
225                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
226 
227                 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
228                         sk_backlog_rcv(sk, skb);
229 
230                 tp->ucopy.memory = 0;
231         }
232 
233         if (inet_csk_ack_scheduled(sk)) {
234                 if (!icsk->icsk_ack.pingpong) {
235                         /* Delayed ACK missed: inflate ATO. */
236                         icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
237                 } else {
238                         /* Delayed ACK missed: leave pingpong mode and
239                          * deflate ATO.
240                          */
241                         icsk->icsk_ack.pingpong = 0;
242                         icsk->icsk_ack.ato      = TCP_ATO_MIN;
243                 }
244                 tcp_send_ack(sk);
245                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
246         }
247 
248 out:
249         if (sk_under_memory_pressure(sk))
250                 sk_mem_reclaim(sk);
251 }
252 
253 static void tcp_delack_timer(unsigned long data)
254 {
255         struct sock *sk = (struct sock *)data;
256 
257         bh_lock_sock(sk);
258         if (!sock_owned_by_user(sk)) {
259                 tcp_delack_timer_handler(sk);
260         } else {
261                 inet_csk(sk)->icsk_ack.blocked = 1;
262                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
263                 /* deleguate our work to tcp_release_cb() */
264                 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
265                         sock_hold(sk);
266         }
267         bh_unlock_sock(sk);
268         sock_put(sk);
269 }
270 
271 static void tcp_probe_timer(struct sock *sk)
272 {
273         struct inet_connection_sock *icsk = inet_csk(sk);
274         struct tcp_sock *tp = tcp_sk(sk);
275         int max_probes;
276         u32 start_ts;
277 
278         if (tp->packets_out || !tcp_send_head(sk)) {
279                 icsk->icsk_probes_out = 0;
280                 return;
281         }
282 
283         /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
284          * long as the receiver continues to respond probes. We support this by
285          * default and reset icsk_probes_out with incoming ACKs. But if the
286          * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
287          * kill the socket when the retry count and the time exceeds the
288          * corresponding system limit. We also implement similar policy when
289          * we use RTO to probe window in tcp_retransmit_timer().
290          */
291         start_ts = tcp_skb_timestamp(tcp_send_head(sk));
292         if (!start_ts)
293                 skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp);
294         else if (icsk->icsk_user_timeout &&
295                  (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout)
296                 goto abort;
297 
298         max_probes = sysctl_tcp_retries2;
299         if (sock_flag(sk, SOCK_DEAD)) {
300                 const int alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
301 
302                 max_probes = tcp_orphan_retries(sk, alive);
303                 if (!alive && icsk->icsk_backoff >= max_probes)
304                         goto abort;
305                 if (tcp_out_of_resources(sk, true))
306                         return;
307         }
308 
309         if (icsk->icsk_probes_out > max_probes) {
310 abort:          tcp_write_err(sk);
311         } else {
312                 /* Only send another probe if we didn't close things up. */
313                 tcp_send_probe0(sk);
314         }
315 }
316 
317 /*
318  *      Timer for Fast Open socket to retransmit SYNACK. Note that the
319  *      sk here is the child socket, not the parent (listener) socket.
320  */
321 static void tcp_fastopen_synack_timer(struct sock *sk)
322 {
323         struct inet_connection_sock *icsk = inet_csk(sk);
324         int max_retries = icsk->icsk_syn_retries ? :
325             sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
326         struct request_sock *req;
327 
328         req = tcp_sk(sk)->fastopen_rsk;
329         req->rsk_ops->syn_ack_timeout(sk, req);
330 
331         if (req->num_timeout >= max_retries) {
332                 tcp_write_err(sk);
333                 return;
334         }
335         /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
336          * returned from rtx_syn_ack() to make it more persistent like
337          * regular retransmit because if the child socket has been accepted
338          * it's not good to give up too easily.
339          */
340         inet_rtx_syn_ack(sk, req);
341         req->num_timeout++;
342         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
343                           TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
344 }
345 
346 /*
347  *      The TCP retransmit timer.
348  */
349 
350 void tcp_retransmit_timer(struct sock *sk)
351 {
352         struct tcp_sock *tp = tcp_sk(sk);
353         struct inet_connection_sock *icsk = inet_csk(sk);
354 
355         if (tp->fastopen_rsk) {
356                 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
357                              sk->sk_state != TCP_FIN_WAIT1);
358                 tcp_fastopen_synack_timer(sk);
359                 /* Before we receive ACK to our SYN-ACK don't retransmit
360                  * anything else (e.g., data or FIN segments).
361                  */
362                 return;
363         }
364         if (!tp->packets_out)
365                 goto out;
366 
367         WARN_ON(tcp_write_queue_empty(sk));
368 
369         tp->tlp_high_seq = 0;
370 
371         if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
372             !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
373                 /* Receiver dastardly shrinks window. Our retransmits
374                  * become zero probes, but we should not timeout this
375                  * connection. If the socket is an orphan, time it out,
376                  * we cannot allow such beasts to hang infinitely.
377                  */
378                 struct inet_sock *inet = inet_sk(sk);
379                 if (sk->sk_family == AF_INET) {
380                         net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
381                                             &inet->inet_daddr,
382                                             ntohs(inet->inet_dport),
383                                             inet->inet_num,
384                                             tp->snd_una, tp->snd_nxt);
385                 }
386 #if IS_ENABLED(CONFIG_IPV6)
387                 else if (sk->sk_family == AF_INET6) {
388                         net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
389                                             &sk->sk_v6_daddr,
390                                             ntohs(inet->inet_dport),
391                                             inet->inet_num,
392                                             tp->snd_una, tp->snd_nxt);
393                 }
394 #endif
395                 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
396                         tcp_write_err(sk);
397                         goto out;
398                 }
399                 tcp_enter_loss(sk);
400                 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
401                 __sk_dst_reset(sk);
402                 goto out_reset_timer;
403         }
404 
405         if (tcp_write_timeout(sk))
406                 goto out;
407 
408         if (icsk->icsk_retransmits == 0) {
409                 int mib_idx;
410 
411                 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
412                         if (tcp_is_sack(tp))
413                                 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
414                         else
415                                 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
416                 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
417                         mib_idx = LINUX_MIB_TCPLOSSFAILURES;
418                 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
419                            tp->sacked_out) {
420                         if (tcp_is_sack(tp))
421                                 mib_idx = LINUX_MIB_TCPSACKFAILURES;
422                         else
423                                 mib_idx = LINUX_MIB_TCPRENOFAILURES;
424                 } else {
425                         mib_idx = LINUX_MIB_TCPTIMEOUTS;
426                 }
427                 NET_INC_STATS_BH(sock_net(sk), mib_idx);
428         }
429 
430         tcp_enter_loss(sk);
431 
432         if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
433                 /* Retransmission failed because of local congestion,
434                  * do not backoff.
435                  */
436                 if (!icsk->icsk_retransmits)
437                         icsk->icsk_retransmits = 1;
438                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
439                                           min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
440                                           TCP_RTO_MAX);
441                 goto out;
442         }
443 
444         /* Increase the timeout each time we retransmit.  Note that
445          * we do not increase the rtt estimate.  rto is initialized
446          * from rtt, but increases here.  Jacobson (SIGCOMM 88) suggests
447          * that doubling rto each time is the least we can get away with.
448          * In KA9Q, Karn uses this for the first few times, and then
449          * goes to quadratic.  netBSD doubles, but only goes up to *64,
450          * and clamps at 1 to 64 sec afterwards.  Note that 120 sec is
451          * defined in the protocol as the maximum possible RTT.  I guess
452          * we'll have to use something other than TCP to talk to the
453          * University of Mars.
454          *
455          * PAWS allows us longer timeouts and large windows, so once
456          * implemented ftp to mars will work nicely. We will have to fix
457          * the 120 second clamps though!
458          */
459         icsk->icsk_backoff++;
460         icsk->icsk_retransmits++;
461 
462 out_reset_timer:
463         /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
464          * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
465          * might be increased if the stream oscillates between thin and thick,
466          * thus the old value might already be too high compared to the value
467          * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
468          * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
469          * exponential backoff behaviour to avoid continue hammering
470          * linear-timeout retransmissions into a black hole
471          */
472         if (sk->sk_state == TCP_ESTABLISHED &&
473             (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
474             tcp_stream_is_thin(tp) &&
475             icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
476                 icsk->icsk_backoff = 0;
477                 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
478         } else {
479                 /* Use normal (exponential) backoff */
480                 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
481         }
482         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
483         if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
484                 __sk_dst_reset(sk);
485 
486 out:;
487 }
488 
489 void tcp_write_timer_handler(struct sock *sk)
490 {
491         struct inet_connection_sock *icsk = inet_csk(sk);
492         int event;
493 
494         if (sk->sk_state == TCP_CLOSE || !icsk->icsk_pending)
495                 goto out;
496 
497         if (time_after(icsk->icsk_timeout, jiffies)) {
498                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
499                 goto out;
500         }
501 
502         event = icsk->icsk_pending;
503 
504         switch (event) {
505         case ICSK_TIME_EARLY_RETRANS:
506                 tcp_resume_early_retransmit(sk);
507                 break;
508         case ICSK_TIME_LOSS_PROBE:
509                 tcp_send_loss_probe(sk);
510                 break;
511         case ICSK_TIME_RETRANS:
512                 icsk->icsk_pending = 0;
513                 tcp_retransmit_timer(sk);
514                 break;
515         case ICSK_TIME_PROBE0:
516                 icsk->icsk_pending = 0;
517                 tcp_probe_timer(sk);
518                 break;
519         }
520 
521 out:
522         sk_mem_reclaim(sk);
523 }
524 
525 static void tcp_write_timer(unsigned long data)
526 {
527         struct sock *sk = (struct sock *)data;
528 
529         bh_lock_sock(sk);
530         if (!sock_owned_by_user(sk)) {
531                 tcp_write_timer_handler(sk);
532         } else {
533                 /* deleguate our work to tcp_release_cb() */
534                 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
535                         sock_hold(sk);
536         }
537         bh_unlock_sock(sk);
538         sock_put(sk);
539 }
540 
541 /*
542  *      Timer for listening sockets
543  */
544 
545 static void tcp_synack_timer(struct sock *sk)
546 {
547         inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
548                                    TCP_TIMEOUT_INIT, TCP_RTO_MAX);
549 }
550 
551 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
552 {
553         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
554 }
555 EXPORT_SYMBOL(tcp_syn_ack_timeout);
556 
557 void tcp_set_keepalive(struct sock *sk, int val)
558 {
559         if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
560                 return;
561 
562         if (val && !sock_flag(sk, SOCK_KEEPOPEN))
563                 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
564         else if (!val)
565                 inet_csk_delete_keepalive_timer(sk);
566 }
567 
568 
569 static void tcp_keepalive_timer (unsigned long data)
570 {
571         struct sock *sk = (struct sock *) data;
572         struct inet_connection_sock *icsk = inet_csk(sk);
573         struct tcp_sock *tp = tcp_sk(sk);
574         u32 elapsed;
575 
576         /* Only process if socket is not in use. */
577         bh_lock_sock(sk);
578         if (sock_owned_by_user(sk)) {
579                 /* Try again later. */
580                 inet_csk_reset_keepalive_timer (sk, HZ/20);
581                 goto out;
582         }
583 
584         if (sk->sk_state == TCP_LISTEN) {
585                 tcp_synack_timer(sk);
586                 goto out;
587         }
588 
589         if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
590                 if (tp->linger2 >= 0) {
591                         const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
592 
593                         if (tmo > 0) {
594                                 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
595                                 goto out;
596                         }
597                 }
598                 tcp_send_active_reset(sk, GFP_ATOMIC);
599                 goto death;
600         }
601 
602         if (!sock_flag(sk, SOCK_KEEPOPEN) || sk->sk_state == TCP_CLOSE)
603                 goto out;
604 
605         elapsed = keepalive_time_when(tp);
606 
607         /* It is alive without keepalive 8) */
608         if (tp->packets_out || tcp_send_head(sk))
609                 goto resched;
610 
611         elapsed = keepalive_time_elapsed(tp);
612 
613         if (elapsed >= keepalive_time_when(tp)) {
614                 /* If the TCP_USER_TIMEOUT option is enabled, use that
615                  * to determine when to timeout instead.
616                  */
617                 if ((icsk->icsk_user_timeout != 0 &&
618                     elapsed >= icsk->icsk_user_timeout &&
619                     icsk->icsk_probes_out > 0) ||
620                     (icsk->icsk_user_timeout == 0 &&
621                     icsk->icsk_probes_out >= keepalive_probes(tp))) {
622                         tcp_send_active_reset(sk, GFP_ATOMIC);
623                         tcp_write_err(sk);
624                         goto out;
625                 }
626                 if (tcp_write_wakeup(sk) <= 0) {
627                         icsk->icsk_probes_out++;
628                         elapsed = keepalive_intvl_when(tp);
629                 } else {
630                         /* If keepalive was lost due to local congestion,
631                          * try harder.
632                          */
633                         elapsed = TCP_RESOURCE_PROBE_INTERVAL;
634                 }
635         } else {
636                 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
637                 elapsed = keepalive_time_when(tp) - elapsed;
638         }
639 
640         sk_mem_reclaim(sk);
641 
642 resched:
643         inet_csk_reset_keepalive_timer (sk, elapsed);
644         goto out;
645 
646 death:
647         tcp_done(sk);
648 
649 out:
650         bh_unlock_sock(sk);
651         sock_put(sk);
652 }
653 
654 void tcp_init_xmit_timers(struct sock *sk)
655 {
656         inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
657                                   &tcp_keepalive_timer);
658 }
659 EXPORT_SYMBOL(tcp_init_xmit_timers);
660 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp