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
, 68 - tcp_sk(sk
)->tcp_header_len
);
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 timeout
= tcp_model_timeout(sk
, boundary
, TCP_RTO_MIN
);
204 return (s32
)(tcp_time_stamp(tcp_sk(sk
)) - start_ts
- timeout
) >= 0;
207 /* A write timeout has occurred. Process the after effects. */
208 static int tcp_write_timeout(struct sock
*sk
)
210 struct inet_connection_sock
*icsk
= inet_csk(sk
);
211 struct tcp_sock
*tp
= tcp_sk(sk
);
212 struct net
*net
= sock_net(sk
);
213 bool expired
, do_reset
;
216 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
217 if (icsk
->icsk_retransmits
) {
218 dst_negative_advice(sk
);
220 sk_rethink_txhash(sk
);
222 retry_until
= icsk
->icsk_syn_retries
? : net
->ipv4
.sysctl_tcp_syn_retries
;
223 expired
= icsk
->icsk_retransmits
>= retry_until
;
225 if (retransmits_timed_out(sk
, net
->ipv4
.sysctl_tcp_retries1
, 0)) {
226 /* Black hole detection */
227 tcp_mtu_probing(icsk
, sk
);
229 dst_negative_advice(sk
);
231 sk_rethink_txhash(sk
);
234 retry_until
= net
->ipv4
.sysctl_tcp_retries2
;
235 if (sock_flag(sk
, SOCK_DEAD
)) {
236 const bool alive
= icsk
->icsk_rto
< TCP_RTO_MAX
;
238 retry_until
= tcp_orphan_retries(sk
, alive
);
240 !retransmits_timed_out(sk
, retry_until
, 0);
242 if (tcp_out_of_resources(sk
, do_reset
))
245 expired
= retransmits_timed_out(sk
, retry_until
,
246 icsk
->icsk_user_timeout
);
248 tcp_fastopen_active_detect_blackhole(sk
, expired
);
250 if (BPF_SOCK_OPS_TEST_FLAG(tp
, BPF_SOCK_OPS_RTO_CB_FLAG
))
251 tcp_call_bpf_3arg(sk
, BPF_SOCK_OPS_RTO_CB
,
252 icsk
->icsk_retransmits
,
253 icsk
->icsk_rto
, (int)expired
);
256 /* Has it gone just too far? */
264 /* Called with BH disabled */
265 void tcp_delack_timer_handler(struct sock
*sk
)
267 struct inet_connection_sock
*icsk
= inet_csk(sk
);
269 sk_mem_reclaim_partial(sk
);
271 if (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)) ||
272 !(icsk
->icsk_ack
.pending
& ICSK_ACK_TIMER
))
275 if (time_after(icsk
->icsk_ack
.timeout
, jiffies
)) {
276 sk_reset_timer(sk
, &icsk
->icsk_delack_timer
, icsk
->icsk_ack
.timeout
);
279 icsk
->icsk_ack
.pending
&= ~ICSK_ACK_TIMER
;
281 if (inet_csk_ack_scheduled(sk
)) {
282 if (!inet_csk_in_pingpong_mode(sk
)) {
283 /* Delayed ACK missed: inflate ATO. */
284 icsk
->icsk_ack
.ato
= min(icsk
->icsk_ack
.ato
<< 1, icsk
->icsk_rto
);
286 /* Delayed ACK missed: leave pingpong mode and
289 inet_csk_exit_pingpong_mode(sk
);
290 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
292 tcp_mstamp_refresh(tcp_sk(sk
));
294 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_DELAYEDACKS
);
298 if (tcp_under_memory_pressure(sk
))
304 * tcp_delack_timer() - The TCP delayed ACK timeout handler
305 * @data: Pointer to the current socket. (gets casted to struct sock *)
307 * This function gets (indirectly) called when the kernel timer for a TCP packet
308 * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work.
310 * Returns: Nothing (void)
312 static void tcp_delack_timer(struct timer_list
*t
)
314 struct inet_connection_sock
*icsk
=
315 from_timer(icsk
, t
, icsk_delack_timer
);
316 struct sock
*sk
= &icsk
->icsk_inet
.sk
;
319 if (!sock_owned_by_user(sk
)) {
320 tcp_delack_timer_handler(sk
);
322 icsk
->icsk_ack
.blocked
= 1;
323 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_DELAYEDACKLOCKED
);
324 /* deleguate our work to tcp_release_cb() */
325 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED
, &sk
->sk_tsq_flags
))
332 static void tcp_probe_timer(struct sock
*sk
)
334 struct inet_connection_sock
*icsk
= inet_csk(sk
);
335 struct sk_buff
*skb
= tcp_send_head(sk
);
336 struct tcp_sock
*tp
= tcp_sk(sk
);
339 if (tp
->packets_out
|| !skb
) {
340 icsk
->icsk_probes_out
= 0;
344 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
345 * long as the receiver continues to respond probes. We support this by
346 * default and reset icsk_probes_out with incoming ACKs. But if the
347 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
348 * kill the socket when the retry count and the time exceeds the
349 * corresponding system limit. We also implement similar policy when
350 * we use RTO to probe window in tcp_retransmit_timer().
352 if (icsk
->icsk_user_timeout
) {
353 u32 elapsed
= tcp_model_timeout(sk
, icsk
->icsk_probes_out
,
354 tcp_probe0_base(sk
));
356 if (elapsed
>= icsk
->icsk_user_timeout
)
360 max_probes
= sock_net(sk
)->ipv4
.sysctl_tcp_retries2
;
361 if (sock_flag(sk
, SOCK_DEAD
)) {
362 const bool alive
= inet_csk_rto_backoff(icsk
, TCP_RTO_MAX
) < TCP_RTO_MAX
;
364 max_probes
= tcp_orphan_retries(sk
, alive
);
365 if (!alive
&& icsk
->icsk_backoff
>= max_probes
)
367 if (tcp_out_of_resources(sk
, true))
371 if (icsk
->icsk_probes_out
>= max_probes
) {
372 abort
: tcp_write_err(sk
);
374 /* Only send another probe if we didn't close things up. */
380 * Timer for Fast Open socket to retransmit SYNACK. Note that the
381 * sk here is the child socket, not the parent (listener) socket.
383 static void tcp_fastopen_synack_timer(struct sock
*sk
)
385 struct inet_connection_sock
*icsk
= inet_csk(sk
);
386 int max_retries
= icsk
->icsk_syn_retries
? :
387 sock_net(sk
)->ipv4
.sysctl_tcp_synack_retries
+ 1; /* add one more retry for fastopen */
388 struct tcp_sock
*tp
= tcp_sk(sk
);
389 struct request_sock
*req
;
391 req
= tcp_sk(sk
)->fastopen_rsk
;
392 req
->rsk_ops
->syn_ack_timeout(req
);
394 if (req
->num_timeout
>= max_retries
) {
398 /* Lower cwnd after certain SYNACK timeout like tcp_init_transfer() */
399 if (icsk
->icsk_retransmits
== 1)
401 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
402 * returned from rtx_syn_ack() to make it more persistent like
403 * regular retransmit because if the child socket has been accepted
404 * it's not good to give up too easily.
406 inet_rtx_syn_ack(sk
, req
);
408 icsk
->icsk_retransmits
++;
409 if (!tp
->retrans_stamp
)
410 tp
->retrans_stamp
= tcp_time_stamp(tp
);
411 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
412 TCP_TIMEOUT_INIT
<< req
->num_timeout
, TCP_RTO_MAX
);
417 * tcp_retransmit_timer() - The TCP retransmit timeout handler
418 * @sk: Pointer to the current socket.
420 * This function gets called when the kernel timer for a TCP packet
421 * of this socket expires.
423 * It handles retransmission, timer adjustment and other necesarry measures.
425 * Returns: Nothing (void)
427 void tcp_retransmit_timer(struct sock
*sk
)
429 struct tcp_sock
*tp
= tcp_sk(sk
);
430 struct net
*net
= sock_net(sk
);
431 struct inet_connection_sock
*icsk
= inet_csk(sk
);
433 if (tp
->fastopen_rsk
) {
434 WARN_ON_ONCE(sk
->sk_state
!= TCP_SYN_RECV
&&
435 sk
->sk_state
!= TCP_FIN_WAIT1
);
436 tcp_fastopen_synack_timer(sk
);
437 /* Before we receive ACK to our SYN-ACK don't retransmit
438 * anything else (e.g., data or FIN segments).
442 if (!tp
->packets_out
|| WARN_ON_ONCE(tcp_rtx_queue_empty(sk
)))
445 tp
->tlp_high_seq
= 0;
447 if (!tp
->snd_wnd
&& !sock_flag(sk
, SOCK_DEAD
) &&
448 !((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
))) {
449 /* Receiver dastardly shrinks window. Our retransmits
450 * become zero probes, but we should not timeout this
451 * connection. If the socket is an orphan, time it out,
452 * we cannot allow such beasts to hang infinitely.
454 struct inet_sock
*inet
= inet_sk(sk
);
455 if (sk
->sk_family
== AF_INET
) {
456 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
458 ntohs(inet
->inet_dport
),
460 tp
->snd_una
, tp
->snd_nxt
);
462 #if IS_ENABLED(CONFIG_IPV6)
463 else if (sk
->sk_family
== AF_INET6
) {
464 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
466 ntohs(inet
->inet_dport
),
468 tp
->snd_una
, tp
->snd_nxt
);
471 if (tcp_jiffies32
- tp
->rcv_tstamp
> TCP_RTO_MAX
) {
476 tcp_retransmit_skb(sk
, tcp_rtx_queue_head(sk
), 1);
478 goto out_reset_timer
;
481 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPTIMEOUTS
);
482 if (tcp_write_timeout(sk
))
485 if (icsk
->icsk_retransmits
== 0) {
488 if (icsk
->icsk_ca_state
== TCP_CA_Recovery
) {
490 mib_idx
= LINUX_MIB_TCPSACKRECOVERYFAIL
;
492 mib_idx
= LINUX_MIB_TCPRENORECOVERYFAIL
;
493 } else if (icsk
->icsk_ca_state
== TCP_CA_Loss
) {
494 mib_idx
= LINUX_MIB_TCPLOSSFAILURES
;
495 } else if ((icsk
->icsk_ca_state
== TCP_CA_Disorder
) ||
498 mib_idx
= LINUX_MIB_TCPSACKFAILURES
;
500 mib_idx
= LINUX_MIB_TCPRENOFAILURES
;
503 __NET_INC_STATS(sock_net(sk
), mib_idx
);
508 icsk
->icsk_retransmits
++;
509 if (tcp_retransmit_skb(sk
, tcp_rtx_queue_head(sk
), 1) > 0) {
510 /* Retransmission failed because of local congestion,
511 * Let senders fight for local resources conservatively.
513 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
514 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
++;
537 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
538 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
539 * might be increased if the stream oscillates between thin and thick,
540 * thus the old value might already be too high compared to the value
541 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
542 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
543 * exponential backoff behaviour to avoid continue hammering
544 * linear-timeout retransmissions into a black hole
546 if (sk
->sk_state
== TCP_ESTABLISHED
&&
547 (tp
->thin_lto
|| net
->ipv4
.sysctl_tcp_thin_linear_timeouts
) &&
548 tcp_stream_is_thin(tp
) &&
549 icsk
->icsk_retransmits
<= TCP_THIN_LINEAR_RETRIES
) {
550 icsk
->icsk_backoff
= 0;
551 icsk
->icsk_rto
= min(__tcp_set_rto(tp
), TCP_RTO_MAX
);
553 /* Use normal (exponential) backoff */
554 icsk
->icsk_rto
= min(icsk
->icsk_rto
<< 1, TCP_RTO_MAX
);
556 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
557 tcp_clamp_rto_to_user_timeout(sk
), TCP_RTO_MAX
);
558 if (retransmits_timed_out(sk
, net
->ipv4
.sysctl_tcp_retries1
+ 1, 0))
564 /* Called with bottom-half processing disabled.
565 Called by tcp_write_timer() */
566 void tcp_write_timer_handler(struct sock
*sk
)
568 struct inet_connection_sock
*icsk
= inet_csk(sk
);
571 if (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)) ||
575 if (time_after(icsk
->icsk_timeout
, jiffies
)) {
576 sk_reset_timer(sk
, &icsk
->icsk_retransmit_timer
, icsk
->icsk_timeout
);
580 tcp_mstamp_refresh(tcp_sk(sk
));
581 event
= icsk
->icsk_pending
;
584 case ICSK_TIME_REO_TIMEOUT
:
585 tcp_rack_reo_timeout(sk
);
587 case ICSK_TIME_LOSS_PROBE
:
588 tcp_send_loss_probe(sk
);
590 case ICSK_TIME_RETRANS
:
591 icsk
->icsk_pending
= 0;
592 tcp_retransmit_timer(sk
);
594 case ICSK_TIME_PROBE0
:
595 icsk
->icsk_pending
= 0;
604 static void tcp_write_timer(struct timer_list
*t
)
606 struct inet_connection_sock
*icsk
=
607 from_timer(icsk
, t
, icsk_retransmit_timer
);
608 struct sock
*sk
= &icsk
->icsk_inet
.sk
;
611 if (!sock_owned_by_user(sk
)) {
612 tcp_write_timer_handler(sk
);
614 /* delegate our work to tcp_release_cb() */
615 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED
, &sk
->sk_tsq_flags
))
622 void tcp_syn_ack_timeout(const struct request_sock
*req
)
624 struct net
*net
= read_pnet(&inet_rsk(req
)->ireq_net
);
626 __NET_INC_STATS(net
, LINUX_MIB_TCPTIMEOUTS
);
628 EXPORT_SYMBOL(tcp_syn_ack_timeout
);
630 void tcp_set_keepalive(struct sock
*sk
, int val
)
632 if ((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
))
635 if (val
&& !sock_flag(sk
, SOCK_KEEPOPEN
))
636 inet_csk_reset_keepalive_timer(sk
, keepalive_time_when(tcp_sk(sk
)));
638 inet_csk_delete_keepalive_timer(sk
);
640 EXPORT_SYMBOL_GPL(tcp_set_keepalive
);
643 static void tcp_keepalive_timer (struct timer_list
*t
)
645 struct sock
*sk
= from_timer(sk
, t
, sk_timer
);
646 struct inet_connection_sock
*icsk
= inet_csk(sk
);
647 struct tcp_sock
*tp
= tcp_sk(sk
);
650 /* Only process if socket is not in use. */
652 if (sock_owned_by_user(sk
)) {
653 /* Try again later. */
654 inet_csk_reset_keepalive_timer (sk
, HZ
/20);
658 if (sk
->sk_state
== TCP_LISTEN
) {
659 pr_err("Hmm... keepalive on a LISTEN ???\n");
663 tcp_mstamp_refresh(tp
);
664 if (sk
->sk_state
== TCP_FIN_WAIT2
&& sock_flag(sk
, SOCK_DEAD
)) {
665 if (tp
->linger2
>= 0) {
666 const int tmo
= tcp_fin_time(sk
) - TCP_TIMEWAIT_LEN
;
669 tcp_time_wait(sk
, TCP_FIN_WAIT2
, tmo
);
673 tcp_send_active_reset(sk
, GFP_ATOMIC
);
677 if (!sock_flag(sk
, SOCK_KEEPOPEN
) ||
678 ((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)))
681 elapsed
= keepalive_time_when(tp
);
683 /* It is alive without keepalive 8) */
684 if (tp
->packets_out
|| !tcp_write_queue_empty(sk
))
687 elapsed
= keepalive_time_elapsed(tp
);
689 if (elapsed
>= keepalive_time_when(tp
)) {
690 /* If the TCP_USER_TIMEOUT option is enabled, use that
691 * to determine when to timeout instead.
693 if ((icsk
->icsk_user_timeout
!= 0 &&
694 elapsed
>= msecs_to_jiffies(icsk
->icsk_user_timeout
) &&
695 icsk
->icsk_probes_out
> 0) ||
696 (icsk
->icsk_user_timeout
== 0 &&
697 icsk
->icsk_probes_out
>= keepalive_probes(tp
))) {
698 tcp_send_active_reset(sk
, GFP_ATOMIC
);
702 if (tcp_write_wakeup(sk
, LINUX_MIB_TCPKEEPALIVE
) <= 0) {
703 icsk
->icsk_probes_out
++;
704 elapsed
= keepalive_intvl_when(tp
);
706 /* If keepalive was lost due to local congestion,
709 elapsed
= TCP_RESOURCE_PROBE_INTERVAL
;
712 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
713 elapsed
= keepalive_time_when(tp
) - elapsed
;
719 inet_csk_reset_keepalive_timer (sk
, elapsed
);
730 static enum hrtimer_restart
tcp_compressed_ack_kick(struct hrtimer
*timer
)
732 struct tcp_sock
*tp
= container_of(timer
, struct tcp_sock
, compressed_ack_timer
);
733 struct sock
*sk
= (struct sock
*)tp
;
736 if (!sock_owned_by_user(sk
)) {
737 if (tp
->compressed_ack
> TCP_FASTRETRANS_THRESH
)
740 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED
,
748 return HRTIMER_NORESTART
;
751 void tcp_init_xmit_timers(struct sock
*sk
)
753 inet_csk_init_xmit_timers(sk
, &tcp_write_timer
, &tcp_delack_timer
,
754 &tcp_keepalive_timer
);
755 hrtimer_init(&tcp_sk(sk
)->pacing_timer
, CLOCK_MONOTONIC
,
756 HRTIMER_MODE_ABS_PINNED_SOFT
);
757 tcp_sk(sk
)->pacing_timer
.function
= tcp_pace_kick
;
759 hrtimer_init(&tcp_sk(sk
)->compressed_ack_timer
, CLOCK_MONOTONIC
,
760 HRTIMER_MODE_REL_PINNED_SOFT
);
761 tcp_sk(sk
)->compressed_ack_timer
.function
= tcp_compressed_ack_kick
;