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).
22 #include <linux/module.h>
23 #include <linux/gfp.h>
25 #include <net/rstreason.h>
27 static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
29 const struct inet_connection_sock *icsk = inet_csk(sk);
30 const struct tcp_sock *tp = tcp_sk(sk);
31 u32 elapsed, user_timeout;
34 user_timeout = READ_ONCE(icsk->icsk_user_timeout);
36 return icsk->icsk_rto;
38 elapsed = tcp_time_stamp_ts(tp) - tp->retrans_stamp;
40 elapsed /= USEC_PER_MSEC;
42 remaining = user_timeout - elapsed;
44 return 1; /* user timeout has passed; fire ASAP */
46 return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(remaining));
49 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when)
51 const struct inet_connection_sock *icsk = inet_csk(sk);
52 u32 remaining, user_timeout;
55 user_timeout = READ_ONCE(icsk->icsk_user_timeout);
56 if (!user_timeout || !icsk->icsk_probes_tstamp)
59 elapsed = tcp_jiffies32 - icsk->icsk_probes_tstamp;
60 if (unlikely(elapsed < 0))
62 remaining = msecs_to_jiffies(user_timeout) - elapsed;
63 remaining = max_t(u32, remaining, TCP_TIMEOUT_MIN);
65 return min_t(u32, remaining, when);
69 * tcp_write_err() - close socket and save error info
70 * @sk: The socket the error has appeared on.
72 * Returns: Nothing (void)
75 static void tcp_write_err(struct sock *sk)
77 tcp_done_with_error(sk, READ_ONCE(sk->sk_err_soft) ? : ETIMEDOUT);
78 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
82 * tcp_out_of_resources() - Close socket if out of resources
83 * @sk: pointer to current socket
84 * @do_reset: send a last packet with reset flag
86 * Do not allow orphaned sockets to eat all our resources.
87 * This is direct violation of TCP specs, but it is required
88 * to prevent DoS attacks. It is called when a retransmission timeout
89 * or zero probe timeout occurs on orphaned socket.
91 * Also close if our net namespace is exiting; in that case there is no
92 * hope of ever communicating again since all netns interfaces are already
93 * down (or about to be down), and we need to release our dst references,
94 * which have been moved to the netns loopback interface, so the namespace
95 * can finish exiting. This condition is only possible if we are a kernel
96 * socket, as those do not hold references to the namespace.
98 * Criteria is still not confirmed experimentally and may change.
99 * We kill the socket, if:
100 * 1. If number of orphaned sockets exceeds an administratively configured
102 * 2. If we have strong memory pressure.
103 * 3. If our net namespace is exiting.
105 static int tcp_out_of_resources(struct sock *sk, bool do_reset)
107 struct tcp_sock *tp = tcp_sk(sk);
110 /* If peer does not open window for long time, or did not transmit
111 * anything for long time, penalize it. */
112 if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
115 /* If some dubious ICMP arrived, penalize even more. */
116 if (READ_ONCE(sk->sk_err_soft))
119 if (tcp_check_oom(sk, shift)) {
120 /* Catch exceptional cases, when connection requires reset.
121 * 1. Last segment was sent recently. */
122 if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
123 /* 2. Window is closed. */
124 (!tp->snd_wnd && !tp->packets_out))
127 tcp_send_active_reset(sk, GFP_ATOMIC,
128 SK_RST_REASON_NOT_SPECIFIED);
130 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
134 if (!check_net(sock_net(sk))) {
135 /* Not possible to send reset; just close */
144 * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket
145 * @sk: Pointer to the current socket.
146 * @alive: bool, socket alive state
148 static int tcp_orphan_retries(struct sock *sk, bool alive)
150 int retries = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_orphan_retries); /* May be zero. */
152 /* We know from an ICMP that something is wrong. */
153 if (READ_ONCE(sk->sk_err_soft) && !alive)
156 /* However, if socket sent something recently, select some safe
157 * number of retries. 8 corresponds to >100 seconds with minimal
159 if (retries == 0 && alive)
164 static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
166 const struct net *net = sock_net(sk);
169 /* Black hole detection */
170 if (!READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing))
173 if (!icsk->icsk_mtup.enabled) {
174 icsk->icsk_mtup.enabled = 1;
175 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
177 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
178 mss = min(READ_ONCE(net->ipv4.sysctl_tcp_base_mss), mss);
179 mss = max(mss, READ_ONCE(net->ipv4.sysctl_tcp_mtu_probe_floor));
180 mss = max(mss, READ_ONCE(net->ipv4.sysctl_tcp_min_snd_mss));
181 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
183 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
186 static unsigned int tcp_model_timeout(struct sock *sk,
187 unsigned int boundary,
188 unsigned int rto_base)
190 unsigned int linear_backoff_thresh, timeout;
192 linear_backoff_thresh = ilog2(TCP_RTO_MAX / rto_base);
193 if (boundary <= linear_backoff_thresh)
194 timeout = ((2 << boundary) - 1) * rto_base;
196 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
197 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
198 return jiffies_to_msecs(timeout);
201 * retransmits_timed_out() - returns true if this connection has timed out
202 * @sk: The current socket
203 * @boundary: max number of retransmissions
204 * @timeout: A custom timeout value.
205 * If set to 0 the default timeout is calculated and used.
206 * Using TCP_RTO_MIN and the number of unsuccessful retransmits.
208 * The default "timeout" value this function can calculate and use
209 * is equivalent to the timeout of a TCP Connection
210 * after "boundary" unsuccessful, exponentially backed-off
211 * retransmissions with an initial RTO of TCP_RTO_MIN.
213 static bool retransmits_timed_out(struct sock *sk,
214 unsigned int boundary,
215 unsigned int timeout)
217 struct tcp_sock *tp = tcp_sk(sk);
218 unsigned int start_ts, delta;
220 if (!inet_csk(sk)->icsk_retransmits)
223 start_ts = tp->retrans_stamp;
224 if (likely(timeout == 0)) {
225 unsigned int rto_base = TCP_RTO_MIN;
227 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))
228 rto_base = tcp_timeout_init(sk);
229 timeout = tcp_model_timeout(sk, boundary, rto_base);
232 if (tp->tcp_usec_ts) {
233 /* delta maybe off up to a jiffy due to timer granularity. */
234 delta = tp->tcp_mstamp - start_ts + jiffies_to_usecs(1);
235 return (s32)(delta - timeout * USEC_PER_MSEC) >= 0;
237 return (s32)(tcp_time_stamp_ts(tp) - start_ts - timeout) >= 0;
240 /* A write timeout has occurred. Process the after effects. */
241 static int tcp_write_timeout(struct sock *sk)
243 struct inet_connection_sock *icsk = inet_csk(sk);
244 struct tcp_sock *tp = tcp_sk(sk);
245 struct net *net = sock_net(sk);
246 bool expired = false, do_reset;
247 int retry_until, max_retransmits;
249 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
250 if (icsk->icsk_retransmits)
251 __dst_negative_advice(sk);
252 /* Paired with WRITE_ONCE() in tcp_sock_set_syncnt() */
253 retry_until = READ_ONCE(icsk->icsk_syn_retries) ? :
254 READ_ONCE(net->ipv4.sysctl_tcp_syn_retries);
256 max_retransmits = retry_until;
257 if (sk->sk_state == TCP_SYN_SENT)
258 max_retransmits += READ_ONCE(net->ipv4.sysctl_tcp_syn_linear_timeouts);
260 expired = icsk->icsk_retransmits >= max_retransmits;
262 if (retransmits_timed_out(sk, READ_ONCE(net->ipv4.sysctl_tcp_retries1), 0)) {
263 /* Black hole detection */
264 tcp_mtu_probing(icsk, sk);
266 __dst_negative_advice(sk);
269 retry_until = READ_ONCE(net->ipv4.sysctl_tcp_retries2);
270 if (sock_flag(sk, SOCK_DEAD)) {
271 const bool alive = icsk->icsk_rto < TCP_RTO_MAX;
273 retry_until = tcp_orphan_retries(sk, alive);
275 !retransmits_timed_out(sk, retry_until, 0);
277 if (tcp_out_of_resources(sk, do_reset))
282 expired = retransmits_timed_out(sk, retry_until,
283 READ_ONCE(icsk->icsk_user_timeout));
284 tcp_fastopen_active_detect_blackhole(sk, expired);
286 if (BPF_SOCK_OPS_TEST_FLAG(tp, BPF_SOCK_OPS_RTO_CB_FLAG))
287 tcp_call_bpf_3arg(sk, BPF_SOCK_OPS_RTO_CB,
288 icsk->icsk_retransmits,
289 icsk->icsk_rto, (int)expired);
292 /* Has it gone just too far? */
297 if (sk_rethink_txhash(sk)) {
298 tp->timeout_rehash++;
299 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTREHASH);
305 /* Called with BH disabled */
306 void tcp_delack_timer_handler(struct sock *sk)
308 struct inet_connection_sock *icsk = inet_csk(sk);
309 struct tcp_sock *tp = tcp_sk(sk);
311 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
314 /* Handling the sack compression case */
315 if (tp->compressed_ack) {
316 tcp_mstamp_refresh(tp);
317 tcp_sack_compress_send_ack(sk);
321 if (!(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
324 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
325 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
328 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
330 if (inet_csk_ack_scheduled(sk)) {
331 if (!inet_csk_in_pingpong_mode(sk)) {
332 /* Delayed ACK missed: inflate ATO. */
333 icsk->icsk_ack.ato = min_t(u32, icsk->icsk_ack.ato << 1, icsk->icsk_rto);
335 /* Delayed ACK missed: leave pingpong mode and
338 inet_csk_exit_pingpong_mode(sk);
339 icsk->icsk_ack.ato = TCP_ATO_MIN;
341 tcp_mstamp_refresh(tp);
343 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
349 * tcp_delack_timer() - The TCP delayed ACK timeout handler
350 * @t: Pointer to the timer. (gets casted to struct sock *)
352 * This function gets (indirectly) called when the kernel timer for a TCP packet
353 * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work.
355 * Returns: Nothing (void)
357 static void tcp_delack_timer(struct timer_list *t)
359 struct inet_connection_sock *icsk =
360 from_timer(icsk, t, icsk_delack_timer);
361 struct sock *sk = &icsk->icsk_inet.sk;
364 if (!sock_owned_by_user(sk)) {
365 tcp_delack_timer_handler(sk);
367 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
368 /* deleguate our work to tcp_release_cb() */
369 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags))
376 static void tcp_probe_timer(struct sock *sk)
378 struct inet_connection_sock *icsk = inet_csk(sk);
379 struct sk_buff *skb = tcp_send_head(sk);
380 struct tcp_sock *tp = tcp_sk(sk);
383 if (tp->packets_out || !skb) {
384 icsk->icsk_probes_out = 0;
385 icsk->icsk_probes_tstamp = 0;
389 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
390 * long as the receiver continues to respond probes. We support this by
391 * default and reset icsk_probes_out with incoming ACKs. But if the
392 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
393 * kill the socket when the retry count and the time exceeds the
394 * corresponding system limit. We also implement similar policy when
395 * we use RTO to probe window in tcp_retransmit_timer().
397 if (!icsk->icsk_probes_tstamp) {
398 icsk->icsk_probes_tstamp = tcp_jiffies32;
400 u32 user_timeout = READ_ONCE(icsk->icsk_user_timeout);
403 (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >=
404 msecs_to_jiffies(user_timeout))
407 max_probes = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retries2);
408 if (sock_flag(sk, SOCK_DEAD)) {
409 const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
411 max_probes = tcp_orphan_retries(sk, alive);
412 if (!alive && icsk->icsk_backoff >= max_probes)
414 if (tcp_out_of_resources(sk, true))
418 if (icsk->icsk_probes_out >= max_probes) {
419 abort: tcp_write_err(sk);
421 /* Only send another probe if we didn't close things up. */
426 static void tcp_update_rto_stats(struct sock *sk)
428 struct inet_connection_sock *icsk = inet_csk(sk);
429 struct tcp_sock *tp = tcp_sk(sk);
431 if (!icsk->icsk_retransmits) {
432 tp->total_rto_recoveries++;
433 tp->rto_stamp = tcp_time_stamp_ms(tp);
435 icsk->icsk_retransmits++;
440 * Timer for Fast Open socket to retransmit SYNACK. Note that the
441 * sk here is the child socket, not the parent (listener) socket.
443 static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req)
445 struct inet_connection_sock *icsk = inet_csk(sk);
446 struct tcp_sock *tp = tcp_sk(sk);
449 req->rsk_ops->syn_ack_timeout(req);
451 /* Add one more retry for fastopen.
452 * Paired with WRITE_ONCE() in tcp_sock_set_syncnt()
454 max_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
455 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_synack_retries) + 1;
457 if (req->num_timeout >= max_retries) {
461 /* Lower cwnd after certain SYNACK timeout like tcp_init_transfer() */
462 if (icsk->icsk_retransmits == 1)
464 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
465 * returned from rtx_syn_ack() to make it more persistent like
466 * regular retransmit because if the child socket has been accepted
467 * it's not good to give up too easily.
469 inet_rtx_syn_ack(sk, req);
471 tcp_update_rto_stats(sk);
472 if (!tp->retrans_stamp)
473 tp->retrans_stamp = tcp_time_stamp_ts(tp);
474 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
475 req->timeout << req->num_timeout, TCP_RTO_MAX);
478 static bool tcp_rtx_probe0_timed_out(const struct sock *sk,
479 const struct sk_buff *skb,
482 const struct inet_connection_sock *icsk = inet_csk(sk);
483 u32 user_timeout = READ_ONCE(icsk->icsk_user_timeout);
484 const struct tcp_sock *tp = tcp_sk(sk);
485 int timeout = TCP_RTO_MAX * 2;
489 /* If user application specified a TCP_USER_TIMEOUT,
490 * it does not want win 0 packets to 'reset the timer'
491 * while retransmits are not making progress.
493 if (rtx_delta > user_timeout)
495 timeout = min_t(u32, timeout, msecs_to_jiffies(user_timeout));
497 /* Note: timer interrupt might have been delayed by at least one jiffy,
498 * and tp->rcv_tstamp might very well have been written recently.
499 * rcv_delta can thus be negative.
501 rcv_delta = icsk->icsk_timeout - tp->rcv_tstamp;
502 if (rcv_delta <= timeout)
505 return msecs_to_jiffies(rtx_delta) > timeout;
509 * tcp_retransmit_timer() - The TCP retransmit timeout handler
510 * @sk: Pointer to the current socket.
512 * This function gets called when the kernel timer for a TCP packet
513 * of this socket expires.
515 * It handles retransmission, timer adjustment and other necessary measures.
517 * Returns: Nothing (void)
519 void tcp_retransmit_timer(struct sock *sk)
521 struct tcp_sock *tp = tcp_sk(sk);
522 struct net *net = sock_net(sk);
523 struct inet_connection_sock *icsk = inet_csk(sk);
524 struct request_sock *req;
527 req = rcu_dereference_protected(tp->fastopen_rsk,
528 lockdep_sock_is_held(sk));
530 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
531 sk->sk_state != TCP_FIN_WAIT1);
532 tcp_fastopen_synack_timer(sk, req);
533 /* Before we receive ACK to our SYN-ACK don't retransmit
534 * anything else (e.g., data or FIN segments).
539 if (!tp->packets_out)
542 skb = tcp_rtx_queue_head(sk);
543 if (WARN_ON_ONCE(!skb))
546 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
547 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
548 /* Receiver dastardly shrinks window. Our retransmits
549 * become zero probes, but we should not timeout this
550 * connection. If the socket is an orphan, time it out,
551 * we cannot allow such beasts to hang infinitely.
553 struct inet_sock *inet = inet_sk(sk);
556 rtx_delta = tcp_time_stamp_ts(tp) - (tp->retrans_stamp ?:
557 tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));
559 rtx_delta /= USEC_PER_MSEC;
561 if (sk->sk_family == AF_INET) {
562 net_dbg_ratelimited("Probing zero-window on %pI4:%u/%u, seq=%u:%u, recv %ums ago, lasting %ums\n",
563 &inet->inet_daddr, ntohs(inet->inet_dport),
564 inet->inet_num, tp->snd_una, tp->snd_nxt,
565 jiffies_to_msecs(jiffies - tp->rcv_tstamp),
568 #if IS_ENABLED(CONFIG_IPV6)
569 else if (sk->sk_family == AF_INET6) {
570 net_dbg_ratelimited("Probing zero-window on %pI6:%u/%u, seq=%u:%u, recv %ums ago, lasting %ums\n",
571 &sk->sk_v6_daddr, ntohs(inet->inet_dport),
572 inet->inet_num, tp->snd_una, tp->snd_nxt,
573 jiffies_to_msecs(jiffies - tp->rcv_tstamp),
577 if (tcp_rtx_probe0_timed_out(sk, skb, rtx_delta)) {
582 tcp_retransmit_skb(sk, skb, 1);
584 goto out_reset_timer;
587 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
588 if (tcp_write_timeout(sk))
591 if (icsk->icsk_retransmits == 0) {
594 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
596 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
598 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
599 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
600 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
601 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
604 mib_idx = LINUX_MIB_TCPSACKFAILURES;
606 mib_idx = LINUX_MIB_TCPRENOFAILURES;
609 __NET_INC_STATS(sock_net(sk), mib_idx);
614 tcp_update_rto_stats(sk);
615 if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) {
616 /* Retransmission failed because of local congestion,
617 * Let senders fight for local resources conservatively.
619 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
620 TCP_RESOURCE_PROBE_INTERVAL,
625 /* Increase the timeout each time we retransmit. Note that
626 * we do not increase the rtt estimate. rto is initialized
627 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
628 * that doubling rto each time is the least we can get away with.
629 * In KA9Q, Karn uses this for the first few times, and then
630 * goes to quadratic. netBSD doubles, but only goes up to *64,
631 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
632 * defined in the protocol as the maximum possible RTT. I guess
633 * we'll have to use something other than TCP to talk to the
634 * University of Mars.
636 * PAWS allows us longer timeouts and large windows, so once
637 * implemented ftp to mars will work nicely. We will have to fix
638 * the 120 second clamps though!
642 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
643 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
644 * might be increased if the stream oscillates between thin and thick,
645 * thus the old value might already be too high compared to the value
646 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
647 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
648 * exponential backoff behaviour to avoid continue hammering
649 * linear-timeout retransmissions into a black hole
651 if (sk->sk_state == TCP_ESTABLISHED &&
652 (tp->thin_lto || READ_ONCE(net->ipv4.sysctl_tcp_thin_linear_timeouts)) &&
653 tcp_stream_is_thin(tp) &&
654 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
655 icsk->icsk_backoff = 0;
656 icsk->icsk_rto = clamp(__tcp_set_rto(tp),
659 } else if (sk->sk_state != TCP_SYN_SENT ||
661 READ_ONCE(net->ipv4.sysctl_tcp_syn_linear_timeouts)) {
662 /* Use normal (exponential) backoff unless linear timeouts are
665 icsk->icsk_backoff++;
666 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
668 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
669 tcp_clamp_rto_to_user_timeout(sk), TCP_RTO_MAX);
670 if (retransmits_timed_out(sk, READ_ONCE(net->ipv4.sysctl_tcp_retries1) + 1, 0))
676 /* Called with bottom-half processing disabled.
677 Called by tcp_write_timer() */
678 void tcp_write_timer_handler(struct sock *sk)
680 struct inet_connection_sock *icsk = inet_csk(sk);
683 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
687 if (time_after(icsk->icsk_timeout, jiffies)) {
688 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
692 tcp_mstamp_refresh(tcp_sk(sk));
693 event = icsk->icsk_pending;
696 case ICSK_TIME_REO_TIMEOUT:
697 tcp_rack_reo_timeout(sk);
699 case ICSK_TIME_LOSS_PROBE:
700 tcp_send_loss_probe(sk);
702 case ICSK_TIME_RETRANS:
703 icsk->icsk_pending = 0;
704 tcp_retransmit_timer(sk);
706 case ICSK_TIME_PROBE0:
707 icsk->icsk_pending = 0;
713 static void tcp_write_timer(struct timer_list *t)
715 struct inet_connection_sock *icsk =
716 from_timer(icsk, t, icsk_retransmit_timer);
717 struct sock *sk = &icsk->icsk_inet.sk;
720 if (!sock_owned_by_user(sk)) {
721 tcp_write_timer_handler(sk);
723 /* delegate our work to tcp_release_cb() */
724 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags))
731 void tcp_syn_ack_timeout(const struct request_sock *req)
733 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
735 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
737 EXPORT_SYMBOL(tcp_syn_ack_timeout);
739 void tcp_set_keepalive(struct sock *sk, int val)
741 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
744 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
745 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
747 inet_csk_delete_keepalive_timer(sk);
749 EXPORT_SYMBOL_GPL(tcp_set_keepalive);
752 static void tcp_keepalive_timer (struct timer_list *t)
754 struct sock *sk = from_timer(sk, t, sk_timer);
755 struct inet_connection_sock *icsk = inet_csk(sk);
756 struct tcp_sock *tp = tcp_sk(sk);
759 /* Only process if socket is not in use. */
761 if (sock_owned_by_user(sk)) {
762 /* Try again later. */
763 inet_csk_reset_keepalive_timer (sk, HZ/20);
767 if (sk->sk_state == TCP_LISTEN) {
768 pr_err("Hmm... keepalive on a LISTEN ???\n");
772 tcp_mstamp_refresh(tp);
773 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
774 if (READ_ONCE(tp->linger2) >= 0) {
775 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
778 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
782 tcp_send_active_reset(sk, GFP_ATOMIC, SK_RST_REASON_NOT_SPECIFIED);
786 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
787 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
790 elapsed = keepalive_time_when(tp);
792 /* It is alive without keepalive 8) */
793 if (tp->packets_out || !tcp_write_queue_empty(sk))
796 elapsed = keepalive_time_elapsed(tp);
798 if (elapsed >= keepalive_time_when(tp)) {
799 u32 user_timeout = READ_ONCE(icsk->icsk_user_timeout);
801 /* If the TCP_USER_TIMEOUT option is enabled, use that
802 * to determine when to timeout instead.
804 if ((user_timeout != 0 &&
805 elapsed >= msecs_to_jiffies(user_timeout) &&
806 icsk->icsk_probes_out > 0) ||
807 (user_timeout == 0 &&
808 icsk->icsk_probes_out >= keepalive_probes(tp))) {
809 tcp_send_active_reset(sk, GFP_ATOMIC,
810 SK_RST_REASON_NOT_SPECIFIED);
814 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
815 icsk->icsk_probes_out++;
816 elapsed = keepalive_intvl_when(tp);
818 /* If keepalive was lost due to local congestion,
821 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
824 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
825 elapsed = keepalive_time_when(tp) - elapsed;
829 inet_csk_reset_keepalive_timer (sk, elapsed);
840 static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
842 struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer);
843 struct sock *sk = (struct sock *)tp;
846 if (!sock_owned_by_user(sk)) {
847 if (tp->compressed_ack) {
848 /* Since we have to send one ack finally,
849 * subtract one from tp->compressed_ack to keep
850 * LINUX_MIB_TCPACKCOMPRESSED accurate.
852 tp->compressed_ack--;
856 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
864 return HRTIMER_NORESTART;
867 void tcp_init_xmit_timers(struct sock *sk)
869 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
870 &tcp_keepalive_timer);
871 hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC,
872 HRTIMER_MODE_ABS_PINNED_SOFT);
873 tcp_sk(sk)->pacing_timer.function = tcp_pace_kick;
875 hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC,
876 HRTIMER_MODE_REL_PINNED_SOFT);
877 tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick;