1 // SPDX-License-Identifier: GPL-2.0-only
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Implementation of the Transmission Control Protocol(TCP).
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Mark Evans, <evansmp@uhura.aston.ac.uk>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
14 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
15 * Linus Torvalds, <torvalds@cs.helsinki.fi>
16 * Alan Cox, <gw4pts@gw4pts.ampr.org>
17 * Matthew Dillon, <dillon@apollo.west.oic.com>
18 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
19 * Jorge Cwik, <jorge@laser.satlink.net>
22 #include <linux/module.h>
23 #include <linux/gfp.h>
26 static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
28 struct inet_connection_sock *icsk = inet_csk(sk);
29 u32 elapsed, start_ts;
32 start_ts = tcp_sk(sk)->retrans_stamp;
33 if (!icsk->icsk_user_timeout)
34 return icsk->icsk_rto;
35 elapsed = tcp_time_stamp(tcp_sk(sk)) - start_ts;
36 remaining = icsk->icsk_user_timeout - elapsed;
38 return 1; /* user timeout has passed; fire ASAP */
40 return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(remaining));
44 * tcp_write_err() - close socket and save error info
45 * @sk: The socket the error has appeared on.
47 * Returns: Nothing (void)
50 static void tcp_write_err(struct sock *sk)
52 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
53 sk->sk_error_report(sk);
55 tcp_write_queue_purge(sk);
57 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
61 * tcp_out_of_resources() - Close socket if out of resources
62 * @sk: pointer to current socket
63 * @do_reset: send a last packet with reset flag
65 * Do not allow orphaned sockets to eat all our resources.
66 * This is direct violation of TCP specs, but it is required
67 * to prevent DoS attacks. It is called when a retransmission timeout
68 * or zero probe timeout occurs on orphaned socket.
70 * Also close if our net namespace is exiting; in that case there is no
71 * hope of ever communicating again since all netns interfaces are already
72 * down (or about to be down), and we need to release our dst references,
73 * which have been moved to the netns loopback interface, so the namespace
74 * can finish exiting. This condition is only possible if we are a kernel
75 * socket, as those do not hold references to the namespace.
77 * Criteria is still not confirmed experimentally and may change.
78 * We kill the socket, if:
79 * 1. If number of orphaned sockets exceeds an administratively configured
81 * 2. If we have strong memory pressure.
82 * 3. If our net namespace is exiting.
84 static int tcp_out_of_resources(struct sock *sk, bool do_reset)
86 struct tcp_sock *tp = tcp_sk(sk);
89 /* If peer does not open window for long time, or did not transmit
90 * anything for long time, penalize it. */
91 if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
94 /* If some dubious ICMP arrived, penalize even more. */
98 if (tcp_check_oom(sk, shift)) {
99 /* Catch exceptional cases, when connection requires reset.
100 * 1. Last segment was sent recently. */
101 if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
102 /* 2. Window is closed. */
103 (!tp->snd_wnd && !tp->packets_out))
106 tcp_send_active_reset(sk, GFP_ATOMIC);
108 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
112 if (!check_net(sock_net(sk))) {
113 /* Not possible to send reset; just close */
122 * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket
123 * @sk: Pointer to the current socket.
124 * @alive: bool, socket alive state
126 static int tcp_orphan_retries(struct sock *sk, bool alive)
128 int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */
130 /* We know from an ICMP that something is wrong. */
131 if (sk->sk_err_soft && !alive)
134 /* However, if socket sent something recently, select some safe
135 * number of retries. 8 corresponds to >100 seconds with minimal
137 if (retries == 0 && alive)
142 static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
144 const struct net *net = sock_net(sk);
147 /* Black hole detection */
148 if (!net->ipv4.sysctl_tcp_mtu_probing)
151 if (!icsk->icsk_mtup.enabled) {
152 icsk->icsk_mtup.enabled = 1;
153 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
155 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
156 mss = min(net->ipv4.sysctl_tcp_base_mss, mss);
157 mss = max(mss, net->ipv4.sysctl_tcp_mtu_probe_floor);
158 mss = max(mss, net->ipv4.sysctl_tcp_min_snd_mss);
159 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
161 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
164 static unsigned int tcp_model_timeout(struct sock *sk,
165 unsigned int boundary,
166 unsigned int rto_base)
168 unsigned int linear_backoff_thresh, timeout;
170 linear_backoff_thresh = ilog2(TCP_RTO_MAX / rto_base);
171 if (boundary <= linear_backoff_thresh)
172 timeout = ((2 << boundary) - 1) * rto_base;
174 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
175 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
176 return jiffies_to_msecs(timeout);
179 * retransmits_timed_out() - returns true if this connection has timed out
180 * @sk: The current socket
181 * @boundary: max number of retransmissions
182 * @timeout: A custom timeout value.
183 * If set to 0 the default timeout is calculated and used.
184 * Using TCP_RTO_MIN and the number of unsuccessful retransmits.
186 * The default "timeout" value this function can calculate and use
187 * is equivalent to the timeout of a TCP Connection
188 * after "boundary" unsuccessful, exponentially backed-off
189 * retransmissions with an initial RTO of TCP_RTO_MIN.
191 static bool retransmits_timed_out(struct sock *sk,
192 unsigned int boundary,
193 unsigned int timeout)
195 unsigned int start_ts;
197 if (!inet_csk(sk)->icsk_retransmits)
200 start_ts = tcp_sk(sk)->retrans_stamp;
201 if (likely(timeout == 0)) {
202 unsigned int rto_base = TCP_RTO_MIN;
204 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
205 rto_base = tcp_timeout_init(sk);
206 timeout = tcp_model_timeout(sk, boundary, rto_base);
209 return (s32)(tcp_time_stamp(tcp_sk(sk)) - start_ts - timeout) >= 0;
212 /* A write timeout has occurred. Process the after effects. */
213 static int tcp_write_timeout(struct sock *sk)
215 struct inet_connection_sock *icsk = inet_csk(sk);
216 struct tcp_sock *tp = tcp_sk(sk);
217 struct net *net = sock_net(sk);
218 bool expired = false, do_reset;
221 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
222 if (icsk->icsk_retransmits) {
223 dst_negative_advice(sk);
225 sk_rethink_txhash(sk);
227 retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
228 expired = icsk->icsk_retransmits >= retry_until;
230 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0)) {
231 /* Black hole detection */
232 tcp_mtu_probing(icsk, sk);
234 dst_negative_advice(sk);
236 sk_rethink_txhash(sk);
239 retry_until = net->ipv4.sysctl_tcp_retries2;
240 if (sock_flag(sk, SOCK_DEAD)) {
241 const bool alive = icsk->icsk_rto < TCP_RTO_MAX;
243 retry_until = tcp_orphan_retries(sk, alive);
245 !retransmits_timed_out(sk, retry_until, 0);
247 if (tcp_out_of_resources(sk, do_reset))
252 expired = retransmits_timed_out(sk, retry_until,
253 icsk->icsk_user_timeout);
254 tcp_fastopen_active_detect_blackhole(sk, expired);
256 if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RTO_CB_FLAG))
257 tcp_call_bpf_3arg(sk, BPF_SOCK_OPS_RTO_CB,
258 icsk->icsk_retransmits,
259 icsk->icsk_rto, (int)expired);
262 /* Has it gone just too far? */
270 /* Called with BH disabled */
271 void tcp_delack_timer_handler(struct sock *sk)
273 struct inet_connection_sock *icsk = inet_csk(sk);
275 sk_mem_reclaim_partial(sk);
277 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
278 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
281 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
282 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
285 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
287 if (inet_csk_ack_scheduled(sk)) {
288 if (!inet_csk_in_pingpong_mode(sk)) {
289 /* Delayed ACK missed: inflate ATO. */
290 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
292 /* Delayed ACK missed: leave pingpong mode and
295 inet_csk_exit_pingpong_mode(sk);
296 icsk->icsk_ack.ato = TCP_ATO_MIN;
298 tcp_mstamp_refresh(tcp_sk(sk));
300 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
304 if (tcp_under_memory_pressure(sk))
310 * tcp_delack_timer() - The TCP delayed ACK timeout handler
311 * @data: Pointer to the current socket. (gets casted to struct sock *)
313 * This function gets (indirectly) called when the kernel timer for a TCP packet
314 * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work.
316 * Returns: Nothing (void)
318 static void tcp_delack_timer(struct timer_list *t)
320 struct inet_connection_sock *icsk =
321 from_timer(icsk, t, icsk_delack_timer);
322 struct sock *sk = &icsk->icsk_inet.sk;
325 if (!sock_owned_by_user(sk)) {
326 tcp_delack_timer_handler(sk);
328 icsk->icsk_ack.blocked = 1;
329 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
330 /* deleguate our work to tcp_release_cb() */
331 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags))
338 static void tcp_probe_timer(struct sock *sk)
340 struct inet_connection_sock *icsk = inet_csk(sk);
341 struct sk_buff *skb = tcp_send_head(sk);
342 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 if (icsk->icsk_user_timeout) {
359 u32 elapsed = tcp_model_timeout(sk, icsk->icsk_probes_out,
360 tcp_probe0_base(sk));
362 if (elapsed >= icsk->icsk_user_timeout)
366 max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2;
367 if (sock_flag(sk, SOCK_DEAD)) {
368 const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
370 max_probes = tcp_orphan_retries(sk, alive);
371 if (!alive && icsk->icsk_backoff >= max_probes)
373 if (tcp_out_of_resources(sk, true))
377 if (icsk->icsk_probes_out >= max_probes) {
378 abort: tcp_write_err(sk);
380 /* Only send another probe if we didn't close things up. */
386 * Timer for Fast Open socket to retransmit SYNACK. Note that the
387 * sk here is the child socket, not the parent (listener) socket.
389 static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req)
391 struct inet_connection_sock *icsk = inet_csk(sk);
392 int max_retries = icsk->icsk_syn_retries ? :
393 sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
394 struct tcp_sock *tp = tcp_sk(sk);
396 req->rsk_ops->syn_ack_timeout(req);
398 if (req->num_timeout >= max_retries) {
402 /* Lower cwnd after certain SYNACK timeout like tcp_init_transfer() */
403 if (icsk->icsk_retransmits == 1)
405 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
406 * returned from rtx_syn_ack() to make it more persistent like
407 * regular retransmit because if the child socket has been accepted
408 * it's not good to give up too easily.
410 inet_rtx_syn_ack(sk, req);
412 icsk->icsk_retransmits++;
413 if (!tp->retrans_stamp)
414 tp->retrans_stamp = tcp_time_stamp(tp);
415 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
416 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
421 * tcp_retransmit_timer() - The TCP retransmit timeout handler
422 * @sk: Pointer to the current socket.
424 * This function gets called when the kernel timer for a TCP packet
425 * of this socket expires.
427 * It handles retransmission, timer adjustment and other necesarry measures.
429 * Returns: Nothing (void)
431 void tcp_retransmit_timer(struct sock *sk)
433 struct tcp_sock *tp = tcp_sk(sk);
434 struct net *net = sock_net(sk);
435 struct inet_connection_sock *icsk = inet_csk(sk);
436 struct request_sock *req;
439 req = rcu_dereference_protected(tp->fastopen_rsk,
440 lockdep_sock_is_held(sk));
442 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
443 sk->sk_state != TCP_FIN_WAIT1);
444 tcp_fastopen_synack_timer(sk, req);
445 /* Before we receive ACK to our SYN-ACK don't retransmit
446 * anything else (e.g., data or FIN segments).
451 if (!tp->packets_out)
454 skb = tcp_rtx_queue_head(sk);
455 if (WARN_ON_ONCE(!skb))
458 tp->tlp_high_seq = 0;
460 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
461 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
462 /* Receiver dastardly shrinks window. Our retransmits
463 * become zero probes, but we should not timeout this
464 * connection. If the socket is an orphan, time it out,
465 * we cannot allow such beasts to hang infinitely.
467 struct inet_sock *inet = inet_sk(sk);
468 if (sk->sk_family == AF_INET) {
469 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
471 ntohs(inet->inet_dport),
473 tp->snd_una, tp->snd_nxt);
475 #if IS_ENABLED(CONFIG_IPV6)
476 else if (sk->sk_family == AF_INET6) {
477 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
479 ntohs(inet->inet_dport),
481 tp->snd_una, tp->snd_nxt);
484 if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) {
489 tcp_retransmit_skb(sk, skb, 1);
491 goto out_reset_timer;
494 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
495 if (tcp_write_timeout(sk))
498 if (icsk->icsk_retransmits == 0) {
501 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
503 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
505 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
506 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
507 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
508 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
511 mib_idx = LINUX_MIB_TCPSACKFAILURES;
513 mib_idx = LINUX_MIB_TCPRENOFAILURES;
516 __NET_INC_STATS(sock_net(sk), mib_idx);
521 icsk->icsk_retransmits++;
522 if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) {
523 /* Retransmission failed because of local congestion,
524 * Let senders fight for local resources conservatively.
526 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
527 TCP_RESOURCE_PROBE_INTERVAL,
532 /* Increase the timeout each time we retransmit. Note that
533 * we do not increase the rtt estimate. rto is initialized
534 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
535 * that doubling rto each time is the least we can get away with.
536 * In KA9Q, Karn uses this for the first few times, and then
537 * goes to quadratic. netBSD doubles, but only goes up to *64,
538 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
539 * defined in the protocol as the maximum possible RTT. I guess
540 * we'll have to use something other than TCP to talk to the
541 * University of Mars.
543 * PAWS allows us longer timeouts and large windows, so once
544 * implemented ftp to mars will work nicely. We will have to fix
545 * the 120 second clamps though!
547 icsk->icsk_backoff++;
550 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
551 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
552 * might be increased if the stream oscillates between thin and thick,
553 * thus the old value might already be too high compared to the value
554 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
555 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
556 * exponential backoff behaviour to avoid continue hammering
557 * linear-timeout retransmissions into a black hole
559 if (sk->sk_state == TCP_ESTABLISHED &&
560 (tp->thin_lto || net->ipv4.sysctl_tcp_thin_linear_timeouts) &&
561 tcp_stream_is_thin(tp) &&
562 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
563 icsk->icsk_backoff = 0;
564 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
566 /* Use normal (exponential) backoff */
567 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
569 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
570 tcp_clamp_rto_to_user_timeout(sk), TCP_RTO_MAX);
571 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0))
577 /* Called with bottom-half processing disabled.
578 Called by tcp_write_timer() */
579 void tcp_write_timer_handler(struct sock *sk)
581 struct inet_connection_sock *icsk = inet_csk(sk);
584 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
588 if (time_after(icsk->icsk_timeout, jiffies)) {
589 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
593 tcp_mstamp_refresh(tcp_sk(sk));
594 event = icsk->icsk_pending;
597 case ICSK_TIME_REO_TIMEOUT:
598 tcp_rack_reo_timeout(sk);
600 case ICSK_TIME_LOSS_PROBE:
601 tcp_send_loss_probe(sk);
603 case ICSK_TIME_RETRANS:
604 icsk->icsk_pending = 0;
605 tcp_retransmit_timer(sk);
607 case ICSK_TIME_PROBE0:
608 icsk->icsk_pending = 0;
617 static void tcp_write_timer(struct timer_list *t)
619 struct inet_connection_sock *icsk =
620 from_timer(icsk, t, icsk_retransmit_timer);
621 struct sock *sk = &icsk->icsk_inet.sk;
624 if (!sock_owned_by_user(sk)) {
625 tcp_write_timer_handler(sk);
627 /* delegate our work to tcp_release_cb() */
628 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags))
635 void tcp_syn_ack_timeout(const struct request_sock *req)
637 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
639 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
641 EXPORT_SYMBOL(tcp_syn_ack_timeout);
643 void tcp_set_keepalive(struct sock *sk, int val)
645 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
648 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
649 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
651 inet_csk_delete_keepalive_timer(sk);
653 EXPORT_SYMBOL_GPL(tcp_set_keepalive);
656 static void tcp_keepalive_timer (struct timer_list *t)
658 struct sock *sk = from_timer(sk, t, sk_timer);
659 struct inet_connection_sock *icsk = inet_csk(sk);
660 struct tcp_sock *tp = tcp_sk(sk);
663 /* Only process if socket is not in use. */
665 if (sock_owned_by_user(sk)) {
666 /* Try again later. */
667 inet_csk_reset_keepalive_timer (sk, HZ/20);
671 if (sk->sk_state == TCP_LISTEN) {
672 pr_err("Hmm... keepalive on a LISTEN ???\n");
676 tcp_mstamp_refresh(tp);
677 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
678 if (tp->linger2 >= 0) {
679 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
682 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
686 tcp_send_active_reset(sk, GFP_ATOMIC);
690 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
691 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
694 elapsed = keepalive_time_when(tp);
696 /* It is alive without keepalive 8) */
697 if (tp->packets_out || !tcp_write_queue_empty(sk))
700 elapsed = keepalive_time_elapsed(tp);
702 if (elapsed >= keepalive_time_when(tp)) {
703 /* If the TCP_USER_TIMEOUT option is enabled, use that
704 * to determine when to timeout instead.
706 if ((icsk->icsk_user_timeout != 0 &&
707 elapsed >= msecs_to_jiffies(icsk->icsk_user_timeout) &&
708 icsk->icsk_probes_out > 0) ||
709 (icsk->icsk_user_timeout == 0 &&
710 icsk->icsk_probes_out >= keepalive_probes(tp))) {
711 tcp_send_active_reset(sk, GFP_ATOMIC);
715 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
716 icsk->icsk_probes_out++;
717 elapsed = keepalive_intvl_when(tp);
719 /* If keepalive was lost due to local congestion,
722 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
725 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
726 elapsed = keepalive_time_when(tp) - elapsed;
732 inet_csk_reset_keepalive_timer (sk, elapsed);
743 static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
745 struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer);
746 struct sock *sk = (struct sock *)tp;
749 if (!sock_owned_by_user(sk)) {
750 if (tp->compressed_ack > TCP_FASTRETRANS_THRESH)
753 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
761 return HRTIMER_NORESTART;
764 void tcp_init_xmit_timers(struct sock *sk)
766 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
767 &tcp_keepalive_timer);
768 hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC,
769 HRTIMER_MODE_ABS_PINNED_SOFT);
770 tcp_sk(sk)->pacing_timer.function = tcp_pace_kick;
772 hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC,
773 HRTIMER_MODE_REL_PINNED_SOFT);
774 tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick;