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.
6 * Implementation of the Transmission Control Protocol(TCP).
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>
21 #include <linux/module.h>
22 #include <linux/gfp.h>
25 static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
27 struct inet_connection_sock *icsk = inet_csk(sk);
28 u32 elapsed, start_ts;
31 start_ts = tcp_sk(sk)->retrans_stamp;
32 if (!icsk->icsk_user_timeout)
33 return icsk->icsk_rto;
34 elapsed = tcp_time_stamp(tcp_sk(sk)) - start_ts;
35 remaining = icsk->icsk_user_timeout - elapsed;
37 return 1; /* user timeout has passed; fire ASAP */
39 return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(remaining));
43 * tcp_write_err() - close socket and save error info
44 * @sk: The socket the error has appeared on.
46 * Returns: Nothing (void)
49 static void tcp_write_err(struct sock *sk)
51 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
52 sk->sk_error_report(sk);
54 tcp_write_queue_purge(sk);
56 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
60 * tcp_out_of_resources() - Close socket if out of resources
61 * @sk: pointer to current socket
62 * @do_reset: send a last packet with reset flag
64 * Do not allow orphaned sockets to eat all our resources.
65 * This is direct violation of TCP specs, but it is required
66 * to prevent DoS attacks. It is called when a retransmission timeout
67 * or zero probe timeout occurs on orphaned socket.
69 * Also close if our net namespace is exiting; in that case there is no
70 * hope of ever communicating again since all netns interfaces are already
71 * down (or about to be down), and we need to release our dst references,
72 * which have been moved to the netns loopback interface, so the namespace
73 * can finish exiting. This condition is only possible if we are a kernel
74 * socket, as those do not hold references to the namespace.
76 * Criteria is still not confirmed experimentally and may change.
77 * We kill the socket, if:
78 * 1. If number of orphaned sockets exceeds an administratively configured
80 * 2. If we have strong memory pressure.
81 * 3. If our net namespace is exiting.
83 static int tcp_out_of_resources(struct sock *sk, bool do_reset)
85 struct tcp_sock *tp = tcp_sk(sk);
88 /* If peer does not open window for long time, or did not transmit
89 * anything for long time, penalize it. */
90 if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
93 /* If some dubious ICMP arrived, penalize even more. */
97 if (tcp_check_oom(sk, shift)) {
98 /* Catch exceptional cases, when connection requires reset.
99 * 1. Last segment was sent recently. */
100 if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
101 /* 2. Window is closed. */
102 (!tp->snd_wnd && !tp->packets_out))
105 tcp_send_active_reset(sk, GFP_ATOMIC);
107 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
111 if (!check_net(sock_net(sk))) {
112 /* Not possible to send reset; just close */
121 * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket
122 * @sk: Pointer to the current socket.
123 * @alive: bool, socket alive state
125 static int tcp_orphan_retries(struct sock *sk, bool alive)
127 int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */
129 /* We know from an ICMP that something is wrong. */
130 if (sk->sk_err_soft && !alive)
133 /* However, if socket sent something recently, select some safe
134 * number of retries. 8 corresponds to >100 seconds with minimal
136 if (retries == 0 && alive)
141 static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
143 const struct net *net = sock_net(sk);
146 /* Black hole detection */
147 if (!net->ipv4.sysctl_tcp_mtu_probing)
150 if (!icsk->icsk_mtup.enabled) {
151 icsk->icsk_mtup.enabled = 1;
152 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
154 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
155 mss = min(net->ipv4.sysctl_tcp_base_mss, mss);
156 mss = max(mss, 68 - tcp_sk(sk)->tcp_header_len);
157 mss = max(mss, net->ipv4.sysctl_tcp_min_snd_mss);
158 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
160 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
163 static unsigned int tcp_model_timeout(struct sock *sk,
164 unsigned int boundary,
165 unsigned int rto_base)
167 unsigned int linear_backoff_thresh, timeout;
169 linear_backoff_thresh = ilog2(TCP_RTO_MAX / rto_base);
170 if (boundary <= linear_backoff_thresh)
171 timeout = ((2 << boundary) - 1) * rto_base;
173 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
174 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
175 return jiffies_to_msecs(timeout);
178 * retransmits_timed_out() - returns true if this connection has timed out
179 * @sk: The current socket
180 * @boundary: max number of retransmissions
181 * @timeout: A custom timeout value.
182 * If set to 0 the default timeout is calculated and used.
183 * Using TCP_RTO_MIN and the number of unsuccessful retransmits.
185 * The default "timeout" value this function can calculate and use
186 * is equivalent to the timeout of a TCP Connection
187 * after "boundary" unsuccessful, exponentially backed-off
188 * retransmissions with an initial RTO of TCP_RTO_MIN.
190 static bool retransmits_timed_out(struct sock *sk,
191 unsigned int boundary,
192 unsigned int timeout)
194 unsigned int start_ts;
196 if (!inet_csk(sk)->icsk_retransmits)
199 start_ts = tcp_sk(sk)->retrans_stamp;
200 if (likely(timeout == 0)) {
201 unsigned int rto_base = TCP_RTO_MIN;
203 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
204 rto_base = tcp_timeout_init(sk);
205 timeout = tcp_model_timeout(sk, boundary, rto_base);
208 return (s32)(tcp_time_stamp(tcp_sk(sk)) - start_ts - timeout) >= 0;
211 /* A write timeout has occurred. Process the after effects. */
212 static int tcp_write_timeout(struct sock *sk)
214 struct inet_connection_sock *icsk = inet_csk(sk);
215 struct tcp_sock *tp = tcp_sk(sk);
216 struct net *net = sock_net(sk);
217 bool expired = false, do_reset;
220 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
221 if (icsk->icsk_retransmits) {
222 dst_negative_advice(sk);
224 sk_rethink_txhash(sk);
226 retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
227 expired = icsk->icsk_retransmits >= retry_until;
229 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0)) {
230 /* Black hole detection */
231 tcp_mtu_probing(icsk, sk);
233 dst_negative_advice(sk);
235 sk_rethink_txhash(sk);
238 retry_until = net->ipv4.sysctl_tcp_retries2;
239 if (sock_flag(sk, SOCK_DEAD)) {
240 const bool alive = icsk->icsk_rto < TCP_RTO_MAX;
242 retry_until = tcp_orphan_retries(sk, alive);
244 !retransmits_timed_out(sk, retry_until, 0);
246 if (tcp_out_of_resources(sk, do_reset))
251 expired = retransmits_timed_out(sk, retry_until,
252 icsk->icsk_user_timeout);
253 tcp_fastopen_active_detect_blackhole(sk, expired);
255 if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RTO_CB_FLAG))
256 tcp_call_bpf_3arg(sk, BPF_SOCK_OPS_RTO_CB,
257 icsk->icsk_retransmits,
258 icsk->icsk_rto, (int)expired);
261 /* Has it gone just too far? */
269 /* Called with BH disabled */
270 void tcp_delack_timer_handler(struct sock *sk)
272 struct inet_connection_sock *icsk = inet_csk(sk);
274 sk_mem_reclaim_partial(sk);
276 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
277 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
280 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
281 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
284 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
286 if (inet_csk_ack_scheduled(sk)) {
287 if (!icsk->icsk_ack.pingpong) {
288 /* Delayed ACK missed: inflate ATO. */
289 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
291 /* Delayed ACK missed: leave pingpong mode and
294 icsk->icsk_ack.pingpong = 0;
295 icsk->icsk_ack.ato = TCP_ATO_MIN;
297 tcp_mstamp_refresh(tcp_sk(sk));
299 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
303 if (tcp_under_memory_pressure(sk))
309 * tcp_delack_timer() - The TCP delayed ACK timeout handler
310 * @data: Pointer to the current socket. (gets casted to struct sock *)
312 * This function gets (indirectly) called when the kernel timer for a TCP packet
313 * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work.
315 * Returns: Nothing (void)
317 static void tcp_delack_timer(struct timer_list *t)
319 struct inet_connection_sock *icsk =
320 from_timer(icsk, t, icsk_delack_timer);
321 struct sock *sk = &icsk->icsk_inet.sk;
324 if (!sock_owned_by_user(sk)) {
325 tcp_delack_timer_handler(sk);
327 icsk->icsk_ack.blocked = 1;
328 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
329 /* deleguate our work to tcp_release_cb() */
330 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags))
337 static void tcp_probe_timer(struct sock *sk)
339 struct inet_connection_sock *icsk = inet_csk(sk);
340 struct sk_buff *skb = tcp_send_head(sk);
341 struct tcp_sock *tp = tcp_sk(sk);
345 if (tp->packets_out || !skb) {
346 icsk->icsk_probes_out = 0;
350 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
351 * long as the receiver continues to respond probes. We support this by
352 * default and reset icsk_probes_out with incoming ACKs. But if the
353 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
354 * kill the socket when the retry count and the time exceeds the
355 * corresponding system limit. We also implement similar policy when
356 * we use RTO to probe window in tcp_retransmit_timer().
358 start_ts = tcp_skb_timestamp(skb);
360 skb->skb_mstamp = tp->tcp_mstamp;
361 else if (icsk->icsk_user_timeout &&
362 (s32)(tcp_time_stamp(tp) - start_ts) > icsk->icsk_user_timeout)
365 max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2;
366 if (sock_flag(sk, SOCK_DEAD)) {
367 const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
369 max_probes = tcp_orphan_retries(sk, alive);
370 if (!alive && icsk->icsk_backoff >= max_probes)
372 if (tcp_out_of_resources(sk, true))
376 if (icsk->icsk_probes_out >= max_probes) {
377 abort: tcp_write_err(sk);
379 /* Only send another probe if we didn't close things up. */
385 * Timer for Fast Open socket to retransmit SYNACK. Note that the
386 * sk here is the child socket, not the parent (listener) socket.
388 static void tcp_fastopen_synack_timer(struct sock *sk)
390 struct inet_connection_sock *icsk = inet_csk(sk);
391 int max_retries = icsk->icsk_syn_retries ? :
392 sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
393 struct request_sock *req;
395 req = tcp_sk(sk)->fastopen_rsk;
396 req->rsk_ops->syn_ack_timeout(req);
398 if (req->num_timeout >= max_retries) {
402 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
403 * returned from rtx_syn_ack() to make it more persistent like
404 * regular retransmit because if the child socket has been accepted
405 * it's not good to give up too easily.
407 inet_rtx_syn_ack(sk, req);
409 icsk->icsk_retransmits++;
410 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
411 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
416 * tcp_retransmit_timer() - The TCP retransmit timeout handler
417 * @sk: Pointer to the current socket.
419 * This function gets called when the kernel timer for a TCP packet
420 * of this socket expires.
422 * It handles retransmission, timer adjustment and other necesarry measures.
424 * Returns: Nothing (void)
426 void tcp_retransmit_timer(struct sock *sk)
428 struct tcp_sock *tp = tcp_sk(sk);
429 struct net *net = sock_net(sk);
430 struct inet_connection_sock *icsk = inet_csk(sk);
432 if (tp->fastopen_rsk) {
433 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
434 sk->sk_state != TCP_FIN_WAIT1);
435 tcp_fastopen_synack_timer(sk);
436 /* Before we receive ACK to our SYN-ACK don't retransmit
437 * anything else (e.g., data or FIN segments).
441 if (!tp->packets_out || WARN_ON_ONCE(tcp_rtx_queue_empty(sk)))
444 tp->tlp_high_seq = 0;
446 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
447 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
448 /* Receiver dastardly shrinks window. Our retransmits
449 * become zero probes, but we should not timeout this
450 * connection. If the socket is an orphan, time it out,
451 * we cannot allow such beasts to hang infinitely.
453 struct inet_sock *inet = inet_sk(sk);
454 if (sk->sk_family == AF_INET) {
455 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
457 ntohs(inet->inet_dport),
459 tp->snd_una, tp->snd_nxt);
461 #if IS_ENABLED(CONFIG_IPV6)
462 else if (sk->sk_family == AF_INET6) {
463 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
465 ntohs(inet->inet_dport),
467 tp->snd_una, tp->snd_nxt);
470 if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) {
475 tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1);
477 goto out_reset_timer;
480 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
481 if (tcp_write_timeout(sk))
484 if (icsk->icsk_retransmits == 0) {
487 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
489 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
491 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
492 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
493 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
494 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
497 mib_idx = LINUX_MIB_TCPSACKFAILURES;
499 mib_idx = LINUX_MIB_TCPRENOFAILURES;
502 __NET_INC_STATS(sock_net(sk), mib_idx);
507 if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) {
508 /* Retransmission failed because of local congestion,
511 if (!icsk->icsk_retransmits)
512 icsk->icsk_retransmits = 1;
513 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
514 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
519 /* Increase the timeout each time we retransmit. Note that
520 * we do not increase the rtt estimate. rto is initialized
521 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
522 * that doubling rto each time is the least we can get away with.
523 * In KA9Q, Karn uses this for the first few times, and then
524 * goes to quadratic. netBSD doubles, but only goes up to *64,
525 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
526 * defined in the protocol as the maximum possible RTT. I guess
527 * we'll have to use something other than TCP to talk to the
528 * University of Mars.
530 * PAWS allows us longer timeouts and large windows, so once
531 * implemented ftp to mars will work nicely. We will have to fix
532 * the 120 second clamps though!
534 icsk->icsk_backoff++;
535 icsk->icsk_retransmits++;
538 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
539 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
540 * might be increased if the stream oscillates between thin and thick,
541 * thus the old value might already be too high compared to the value
542 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
543 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
544 * exponential backoff behaviour to avoid continue hammering
545 * linear-timeout retransmissions into a black hole
547 if (sk->sk_state == TCP_ESTABLISHED &&
548 (tp->thin_lto || net->ipv4.sysctl_tcp_thin_linear_timeouts) &&
549 tcp_stream_is_thin(tp) &&
550 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
551 icsk->icsk_backoff = 0;
552 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
554 /* Use normal (exponential) backoff */
555 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
557 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
558 tcp_clamp_rto_to_user_timeout(sk), TCP_RTO_MAX);
559 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0))
565 /* Called with bottom-half processing disabled.
566 Called by tcp_write_timer() */
567 void tcp_write_timer_handler(struct sock *sk)
569 struct inet_connection_sock *icsk = inet_csk(sk);
572 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
576 if (time_after(icsk->icsk_timeout, jiffies)) {
577 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
581 tcp_mstamp_refresh(tcp_sk(sk));
582 event = icsk->icsk_pending;
585 case ICSK_TIME_REO_TIMEOUT:
586 tcp_rack_reo_timeout(sk);
588 case ICSK_TIME_LOSS_PROBE:
589 tcp_send_loss_probe(sk);
591 case ICSK_TIME_RETRANS:
592 icsk->icsk_pending = 0;
593 tcp_retransmit_timer(sk);
595 case ICSK_TIME_PROBE0:
596 icsk->icsk_pending = 0;
605 static void tcp_write_timer(struct timer_list *t)
607 struct inet_connection_sock *icsk =
608 from_timer(icsk, t, icsk_retransmit_timer);
609 struct sock *sk = &icsk->icsk_inet.sk;
612 if (!sock_owned_by_user(sk)) {
613 tcp_write_timer_handler(sk);
615 /* delegate our work to tcp_release_cb() */
616 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags))
623 void tcp_syn_ack_timeout(const struct request_sock *req)
625 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
627 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
629 EXPORT_SYMBOL(tcp_syn_ack_timeout);
631 void tcp_set_keepalive(struct sock *sk, int val)
633 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
636 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
637 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
639 inet_csk_delete_keepalive_timer(sk);
641 EXPORT_SYMBOL_GPL(tcp_set_keepalive);
644 static void tcp_keepalive_timer (struct timer_list *t)
646 struct sock *sk = from_timer(sk, t, sk_timer);
647 struct inet_connection_sock *icsk = inet_csk(sk);
648 struct tcp_sock *tp = tcp_sk(sk);
651 /* Only process if socket is not in use. */
653 if (sock_owned_by_user(sk)) {
654 /* Try again later. */
655 inet_csk_reset_keepalive_timer (sk, HZ/20);
659 if (sk->sk_state == TCP_LISTEN) {
660 pr_err("Hmm... keepalive on a LISTEN ???\n");
664 tcp_mstamp_refresh(tp);
665 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
666 if (tp->linger2 >= 0) {
667 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
670 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
674 tcp_send_active_reset(sk, GFP_ATOMIC);
678 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
679 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
682 elapsed = keepalive_time_when(tp);
684 /* It is alive without keepalive 8) */
685 if (tp->packets_out || !tcp_write_queue_empty(sk))
688 elapsed = keepalive_time_elapsed(tp);
690 if (elapsed >= keepalive_time_when(tp)) {
691 /* If the TCP_USER_TIMEOUT option is enabled, use that
692 * to determine when to timeout instead.
694 if ((icsk->icsk_user_timeout != 0 &&
695 elapsed >= msecs_to_jiffies(icsk->icsk_user_timeout) &&
696 icsk->icsk_probes_out > 0) ||
697 (icsk->icsk_user_timeout == 0 &&
698 icsk->icsk_probes_out >= keepalive_probes(tp))) {
699 tcp_send_active_reset(sk, GFP_ATOMIC);
703 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
704 icsk->icsk_probes_out++;
705 elapsed = keepalive_intvl_when(tp);
707 /* If keepalive was lost due to local congestion,
710 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
713 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
714 elapsed = keepalive_time_when(tp) - elapsed;
720 inet_csk_reset_keepalive_timer (sk, elapsed);
731 static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
733 struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer);
734 struct sock *sk = (struct sock *)tp;
737 if (!sock_owned_by_user(sk)) {
738 if (tp->compressed_ack > TCP_FASTRETRANS_THRESH)
741 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
749 return HRTIMER_NORESTART;
752 void tcp_init_xmit_timers(struct sock *sk)
754 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
755 &tcp_keepalive_timer);
756 hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC,
757 HRTIMER_MODE_ABS_PINNED_SOFT);
758 tcp_sk(sk)->pacing_timer.function = tcp_pace_kick;
760 hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC,
761 HRTIMER_MODE_REL_PINNED_SOFT);
762 tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick;