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_retransmit_stamp(const struct sock
*sk
)
27 u32 start_ts
= tcp_sk(sk
)->retrans_stamp
;
29 if (unlikely(!start_ts
)) {
30 struct sk_buff
*head
= tcp_rtx_queue_head(sk
);
34 start_ts
= tcp_skb_timestamp(head
);
39 static u32
tcp_clamp_rto_to_user_timeout(const struct sock
*sk
)
41 struct inet_connection_sock
*icsk
= inet_csk(sk
);
42 u32 elapsed
, start_ts
;
44 start_ts
= tcp_retransmit_stamp(sk
);
45 if (!icsk
->icsk_user_timeout
|| !start_ts
)
46 return icsk
->icsk_rto
;
47 elapsed
= tcp_time_stamp(tcp_sk(sk
)) - start_ts
;
48 if (elapsed
>= icsk
->icsk_user_timeout
)
49 return 1; /* user timeout has passed; fire ASAP */
51 return min_t(u32
, icsk
->icsk_rto
, msecs_to_jiffies(icsk
->icsk_user_timeout
- elapsed
));
55 * tcp_write_err() - close socket and save error info
56 * @sk: The socket the error has appeared on.
58 * Returns: Nothing (void)
61 static void tcp_write_err(struct sock
*sk
)
63 sk
->sk_err
= sk
->sk_err_soft
? : ETIMEDOUT
;
64 sk
->sk_error_report(sk
);
66 tcp_write_queue_purge(sk
);
68 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPABORTONTIMEOUT
);
72 * tcp_out_of_resources() - Close socket if out of resources
73 * @sk: pointer to current socket
74 * @do_reset: send a last packet with reset flag
76 * Do not allow orphaned sockets to eat all our resources.
77 * This is direct violation of TCP specs, but it is required
78 * to prevent DoS attacks. It is called when a retransmission timeout
79 * or zero probe timeout occurs on orphaned socket.
81 * Also close if our net namespace is exiting; in that case there is no
82 * hope of ever communicating again since all netns interfaces are already
83 * down (or about to be down), and we need to release our dst references,
84 * which have been moved to the netns loopback interface, so the namespace
85 * can finish exiting. This condition is only possible if we are a kernel
86 * socket, as those do not hold references to the namespace.
88 * Criteria is still not confirmed experimentally and may change.
89 * We kill the socket, if:
90 * 1. If number of orphaned sockets exceeds an administratively configured
92 * 2. If we have strong memory pressure.
93 * 3. If our net namespace is exiting.
95 static int tcp_out_of_resources(struct sock
*sk
, bool do_reset
)
97 struct tcp_sock
*tp
= tcp_sk(sk
);
100 /* If peer does not open window for long time, or did not transmit
101 * anything for long time, penalize it. */
102 if ((s32
)(tcp_jiffies32
- tp
->lsndtime
) > 2*TCP_RTO_MAX
|| !do_reset
)
105 /* If some dubious ICMP arrived, penalize even more. */
109 if (tcp_check_oom(sk
, shift
)) {
110 /* Catch exceptional cases, when connection requires reset.
111 * 1. Last segment was sent recently. */
112 if ((s32
)(tcp_jiffies32
- tp
->lsndtime
) <= TCP_TIMEWAIT_LEN
||
113 /* 2. Window is closed. */
114 (!tp
->snd_wnd
&& !tp
->packets_out
))
117 tcp_send_active_reset(sk
, GFP_ATOMIC
);
119 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_TCPABORTONMEMORY
);
123 if (!check_net(sock_net(sk
))) {
124 /* Not possible to send reset; just close */
133 * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket
134 * @sk: Pointer to the current socket.
135 * @alive: bool, socket alive state
137 static int tcp_orphan_retries(struct sock
*sk
, bool alive
)
139 int retries
= sock_net(sk
)->ipv4
.sysctl_tcp_orphan_retries
; /* May be zero. */
141 /* We know from an ICMP that something is wrong. */
142 if (sk
->sk_err_soft
&& !alive
)
145 /* However, if socket sent something recently, select some safe
146 * number of retries. 8 corresponds to >100 seconds with minimal
148 if (retries
== 0 && alive
)
153 static void tcp_mtu_probing(struct inet_connection_sock
*icsk
, struct sock
*sk
)
155 const struct net
*net
= sock_net(sk
);
158 /* Black hole detection */
159 if (!net
->ipv4
.sysctl_tcp_mtu_probing
)
162 if (!icsk
->icsk_mtup
.enabled
) {
163 icsk
->icsk_mtup
.enabled
= 1;
164 icsk
->icsk_mtup
.probe_timestamp
= tcp_jiffies32
;
166 mss
= tcp_mtu_to_mss(sk
, icsk
->icsk_mtup
.search_low
) >> 1;
167 mss
= min(net
->ipv4
.sysctl_tcp_base_mss
, mss
);
168 mss
= max(mss
, 68 - tcp_sk(sk
)->tcp_header_len
);
169 icsk
->icsk_mtup
.search_low
= tcp_mss_to_mtu(sk
, mss
);
171 tcp_sync_mss(sk
, icsk
->icsk_pmtu_cookie
);
176 * retransmits_timed_out() - returns true if this connection has timed out
177 * @sk: The current socket
178 * @boundary: max number of retransmissions
179 * @timeout: A custom timeout value.
180 * If set to 0 the default timeout is calculated and used.
181 * Using TCP_RTO_MIN and the number of unsuccessful retransmits.
183 * The default "timeout" value this function can calculate and use
184 * is equivalent to the timeout of a TCP Connection
185 * after "boundary" unsuccessful, exponentially backed-off
186 * retransmissions with an initial RTO of TCP_RTO_MIN.
188 static bool retransmits_timed_out(struct sock
*sk
,
189 unsigned int boundary
,
190 unsigned int timeout
)
192 const unsigned int rto_base
= TCP_RTO_MIN
;
193 unsigned int linear_backoff_thresh
, start_ts
;
195 if (!inet_csk(sk
)->icsk_retransmits
)
198 start_ts
= tcp_retransmit_stamp(sk
);
202 if (likely(timeout
== 0)) {
203 linear_backoff_thresh
= ilog2(TCP_RTO_MAX
/rto_base
);
205 if (boundary
<= linear_backoff_thresh
)
206 timeout
= ((2 << boundary
) - 1) * rto_base
;
208 timeout
= ((2 << linear_backoff_thresh
) - 1) * rto_base
+
209 (boundary
- linear_backoff_thresh
) * TCP_RTO_MAX
;
210 timeout
= jiffies_to_msecs(timeout
);
212 return (tcp_time_stamp(tcp_sk(sk
)) - start_ts
) >= timeout
;
215 /* A write timeout has occurred. Process the after effects. */
216 static int tcp_write_timeout(struct sock
*sk
)
218 struct inet_connection_sock
*icsk
= inet_csk(sk
);
219 struct tcp_sock
*tp
= tcp_sk(sk
);
220 struct net
*net
= sock_net(sk
);
221 bool expired
, do_reset
;
224 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
225 if (icsk
->icsk_retransmits
) {
226 dst_negative_advice(sk
);
227 } else if (!tp
->syn_data
&& !tp
->syn_fastopen
) {
228 sk_rethink_txhash(sk
);
230 retry_until
= icsk
->icsk_syn_retries
? : net
->ipv4
.sysctl_tcp_syn_retries
;
231 expired
= icsk
->icsk_retransmits
>= retry_until
;
233 if (retransmits_timed_out(sk
, net
->ipv4
.sysctl_tcp_retries1
, 0)) {
234 /* Black hole detection */
235 tcp_mtu_probing(icsk
, sk
);
237 dst_negative_advice(sk
);
239 sk_rethink_txhash(sk
);
242 retry_until
= net
->ipv4
.sysctl_tcp_retries2
;
243 if (sock_flag(sk
, SOCK_DEAD
)) {
244 const bool alive
= icsk
->icsk_rto
< TCP_RTO_MAX
;
246 retry_until
= tcp_orphan_retries(sk
, alive
);
248 !retransmits_timed_out(sk
, retry_until
, 0);
250 if (tcp_out_of_resources(sk
, do_reset
))
253 expired
= retransmits_timed_out(sk
, retry_until
,
254 icsk
->icsk_user_timeout
);
256 tcp_fastopen_active_detect_blackhole(sk
, expired
);
258 if (BPF_SOCK_OPS_TEST_FLAG(tp
, BPF_SOCK_OPS_RTO_CB_FLAG
))
259 tcp_call_bpf_3arg(sk
, BPF_SOCK_OPS_RTO_CB
,
260 icsk
->icsk_retransmits
,
261 icsk
->icsk_rto
, (int)expired
);
264 /* Has it gone just too far? */
272 /* Called with BH disabled */
273 void tcp_delack_timer_handler(struct sock
*sk
)
275 struct inet_connection_sock
*icsk
= inet_csk(sk
);
277 sk_mem_reclaim_partial(sk
);
279 if (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)) ||
280 !(icsk
->icsk_ack
.pending
& ICSK_ACK_TIMER
))
283 if (time_after(icsk
->icsk_ack
.timeout
, jiffies
)) {
284 sk_reset_timer(sk
, &icsk
->icsk_delack_timer
, icsk
->icsk_ack
.timeout
);
287 icsk
->icsk_ack
.pending
&= ~ICSK_ACK_TIMER
;
289 if (inet_csk_ack_scheduled(sk
)) {
290 if (!icsk
->icsk_ack
.pingpong
) {
291 /* Delayed ACK missed: inflate ATO. */
292 icsk
->icsk_ack
.ato
= min(icsk
->icsk_ack
.ato
<< 1, icsk
->icsk_rto
);
294 /* Delayed ACK missed: leave pingpong mode and
297 icsk
->icsk_ack
.pingpong
= 0;
298 icsk
->icsk_ack
.ato
= TCP_ATO_MIN
;
300 tcp_mstamp_refresh(tcp_sk(sk
));
302 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_DELAYEDACKS
);
306 if (tcp_under_memory_pressure(sk
))
312 * tcp_delack_timer() - The TCP delayed ACK timeout handler
313 * @data: Pointer to the current socket. (gets casted to struct sock *)
315 * This function gets (indirectly) called when the kernel timer for a TCP packet
316 * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work.
318 * Returns: Nothing (void)
320 static void tcp_delack_timer(struct timer_list
*t
)
322 struct inet_connection_sock
*icsk
=
323 from_timer(icsk
, t
, icsk_delack_timer
);
324 struct sock
*sk
= &icsk
->icsk_inet
.sk
;
327 if (!sock_owned_by_user(sk
)) {
328 tcp_delack_timer_handler(sk
);
330 icsk
->icsk_ack
.blocked
= 1;
331 __NET_INC_STATS(sock_net(sk
), LINUX_MIB_DELAYEDACKLOCKED
);
332 /* deleguate our work to tcp_release_cb() */
333 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED
, &sk
->sk_tsq_flags
))
340 static void tcp_probe_timer(struct sock
*sk
)
342 struct inet_connection_sock
*icsk
= inet_csk(sk
);
343 struct sk_buff
*skb
= tcp_send_head(sk
);
344 struct tcp_sock
*tp
= tcp_sk(sk
);
348 if (tp
->packets_out
|| !skb
) {
349 icsk
->icsk_probes_out
= 0;
353 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
354 * long as the receiver continues to respond probes. We support this by
355 * default and reset icsk_probes_out with incoming ACKs. But if the
356 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
357 * kill the socket when the retry count and the time exceeds the
358 * corresponding system limit. We also implement similar policy when
359 * we use RTO to probe window in tcp_retransmit_timer().
361 start_ts
= tcp_skb_timestamp(skb
);
363 skb
->skb_mstamp_ns
= tp
->tcp_clock_cache
;
364 else if (icsk
->icsk_user_timeout
&&
365 (s32
)(tcp_time_stamp(tp
) - start_ts
) > icsk
->icsk_user_timeout
)
368 max_probes
= sock_net(sk
)->ipv4
.sysctl_tcp_retries2
;
369 if (sock_flag(sk
, SOCK_DEAD
)) {
370 const bool alive
= inet_csk_rto_backoff(icsk
, TCP_RTO_MAX
) < TCP_RTO_MAX
;
372 max_probes
= tcp_orphan_retries(sk
, alive
);
373 if (!alive
&& icsk
->icsk_backoff
>= max_probes
)
375 if (tcp_out_of_resources(sk
, true))
379 if (icsk
->icsk_probes_out
> max_probes
) {
380 abort
: tcp_write_err(sk
);
382 /* Only send another probe if we didn't close things up. */
388 * Timer for Fast Open socket to retransmit SYNACK. Note that the
389 * sk here is the child socket, not the parent (listener) socket.
391 static void tcp_fastopen_synack_timer(struct sock
*sk
)
393 struct inet_connection_sock
*icsk
= inet_csk(sk
);
394 int max_retries
= icsk
->icsk_syn_retries
? :
395 sock_net(sk
)->ipv4
.sysctl_tcp_synack_retries
+ 1; /* add one more retry for fastopen */
396 struct request_sock
*req
;
398 req
= tcp_sk(sk
)->fastopen_rsk
;
399 req
->rsk_ops
->syn_ack_timeout(req
);
401 if (req
->num_timeout
>= max_retries
) {
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 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
414 TCP_TIMEOUT_INIT
<< req
->num_timeout
, TCP_RTO_MAX
);
419 * tcp_retransmit_timer() - The TCP retransmit timeout handler
420 * @sk: Pointer to the current socket.
422 * This function gets called when the kernel timer for a TCP packet
423 * of this socket expires.
425 * It handles retransmission, timer adjustment and other necesarry measures.
427 * Returns: Nothing (void)
429 void tcp_retransmit_timer(struct sock
*sk
)
431 struct tcp_sock
*tp
= tcp_sk(sk
);
432 struct net
*net
= sock_net(sk
);
433 struct inet_connection_sock
*icsk
= inet_csk(sk
);
435 if (tp
->fastopen_rsk
) {
436 WARN_ON_ONCE(sk
->sk_state
!= TCP_SYN_RECV
&&
437 sk
->sk_state
!= TCP_FIN_WAIT1
);
438 tcp_fastopen_synack_timer(sk
);
439 /* Before we receive ACK to our SYN-ACK don't retransmit
440 * anything else (e.g., data or FIN segments).
444 if (!tp
->packets_out
)
447 WARN_ON(tcp_rtx_queue_empty(sk
));
449 tp
->tlp_high_seq
= 0;
451 if (!tp
->snd_wnd
&& !sock_flag(sk
, SOCK_DEAD
) &&
452 !((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
))) {
453 /* Receiver dastardly shrinks window. Our retransmits
454 * become zero probes, but we should not timeout this
455 * connection. If the socket is an orphan, time it out,
456 * we cannot allow such beasts to hang infinitely.
458 struct inet_sock
*inet
= inet_sk(sk
);
459 if (sk
->sk_family
== AF_INET
) {
460 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
462 ntohs(inet
->inet_dport
),
464 tp
->snd_una
, tp
->snd_nxt
);
466 #if IS_ENABLED(CONFIG_IPV6)
467 else if (sk
->sk_family
== AF_INET6
) {
468 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
470 ntohs(inet
->inet_dport
),
472 tp
->snd_una
, tp
->snd_nxt
);
475 if (tcp_jiffies32
- tp
->rcv_tstamp
> TCP_RTO_MAX
) {
480 tcp_retransmit_skb(sk
, tcp_rtx_queue_head(sk
), 1);
482 goto out_reset_timer
;
485 if (tcp_write_timeout(sk
))
488 if (icsk
->icsk_retransmits
== 0) {
491 if (icsk
->icsk_ca_state
== TCP_CA_Recovery
) {
493 mib_idx
= LINUX_MIB_TCPSACKRECOVERYFAIL
;
495 mib_idx
= LINUX_MIB_TCPRENORECOVERYFAIL
;
496 } else if (icsk
->icsk_ca_state
== TCP_CA_Loss
) {
497 mib_idx
= LINUX_MIB_TCPLOSSFAILURES
;
498 } else if ((icsk
->icsk_ca_state
== TCP_CA_Disorder
) ||
501 mib_idx
= LINUX_MIB_TCPSACKFAILURES
;
503 mib_idx
= LINUX_MIB_TCPRENOFAILURES
;
505 mib_idx
= LINUX_MIB_TCPTIMEOUTS
;
507 __NET_INC_STATS(sock_net(sk
), mib_idx
);
512 if (tcp_retransmit_skb(sk
, tcp_rtx_queue_head(sk
), 1) > 0) {
513 /* Retransmission failed because of local congestion,
516 if (!icsk
->icsk_retransmits
)
517 icsk
->icsk_retransmits
= 1;
518 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
519 min(icsk
->icsk_rto
, TCP_RESOURCE_PROBE_INTERVAL
),
524 /* Increase the timeout each time we retransmit. Note that
525 * we do not increase the rtt estimate. rto is initialized
526 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
527 * that doubling rto each time is the least we can get away with.
528 * In KA9Q, Karn uses this for the first few times, and then
529 * goes to quadratic. netBSD doubles, but only goes up to *64,
530 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
531 * defined in the protocol as the maximum possible RTT. I guess
532 * we'll have to use something other than TCP to talk to the
533 * University of Mars.
535 * PAWS allows us longer timeouts and large windows, so once
536 * implemented ftp to mars will work nicely. We will have to fix
537 * the 120 second clamps though!
539 icsk
->icsk_backoff
++;
540 icsk
->icsk_retransmits
++;
543 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
544 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
545 * might be increased if the stream oscillates between thin and thick,
546 * thus the old value might already be too high compared to the value
547 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
548 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
549 * exponential backoff behaviour to avoid continue hammering
550 * linear-timeout retransmissions into a black hole
552 if (sk
->sk_state
== TCP_ESTABLISHED
&&
553 (tp
->thin_lto
|| net
->ipv4
.sysctl_tcp_thin_linear_timeouts
) &&
554 tcp_stream_is_thin(tp
) &&
555 icsk
->icsk_retransmits
<= TCP_THIN_LINEAR_RETRIES
) {
556 icsk
->icsk_backoff
= 0;
557 icsk
->icsk_rto
= min(__tcp_set_rto(tp
), TCP_RTO_MAX
);
559 /* Use normal (exponential) backoff */
560 icsk
->icsk_rto
= min(icsk
->icsk_rto
<< 1, TCP_RTO_MAX
);
562 inet_csk_reset_xmit_timer(sk
, ICSK_TIME_RETRANS
,
563 tcp_clamp_rto_to_user_timeout(sk
), TCP_RTO_MAX
);
564 if (retransmits_timed_out(sk
, net
->ipv4
.sysctl_tcp_retries1
+ 1, 0))
570 /* Called with bottom-half processing disabled.
571 Called by tcp_write_timer() */
572 void tcp_write_timer_handler(struct sock
*sk
)
574 struct inet_connection_sock
*icsk
= inet_csk(sk
);
577 if (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)) ||
581 if (time_after(icsk
->icsk_timeout
, jiffies
)) {
582 sk_reset_timer(sk
, &icsk
->icsk_retransmit_timer
, icsk
->icsk_timeout
);
586 tcp_mstamp_refresh(tcp_sk(sk
));
587 event
= icsk
->icsk_pending
;
590 case ICSK_TIME_REO_TIMEOUT
:
591 tcp_rack_reo_timeout(sk
);
593 case ICSK_TIME_LOSS_PROBE
:
594 tcp_send_loss_probe(sk
);
596 case ICSK_TIME_RETRANS
:
597 icsk
->icsk_pending
= 0;
598 tcp_retransmit_timer(sk
);
600 case ICSK_TIME_PROBE0
:
601 icsk
->icsk_pending
= 0;
610 static void tcp_write_timer(struct timer_list
*t
)
612 struct inet_connection_sock
*icsk
=
613 from_timer(icsk
, t
, icsk_retransmit_timer
);
614 struct sock
*sk
= &icsk
->icsk_inet
.sk
;
617 if (!sock_owned_by_user(sk
)) {
618 tcp_write_timer_handler(sk
);
620 /* delegate our work to tcp_release_cb() */
621 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED
, &sk
->sk_tsq_flags
))
628 void tcp_syn_ack_timeout(const struct request_sock
*req
)
630 struct net
*net
= read_pnet(&inet_rsk(req
)->ireq_net
);
632 __NET_INC_STATS(net
, LINUX_MIB_TCPTIMEOUTS
);
634 EXPORT_SYMBOL(tcp_syn_ack_timeout
);
636 void tcp_set_keepalive(struct sock
*sk
, int val
)
638 if ((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_LISTEN
))
641 if (val
&& !sock_flag(sk
, SOCK_KEEPOPEN
))
642 inet_csk_reset_keepalive_timer(sk
, keepalive_time_when(tcp_sk(sk
)));
644 inet_csk_delete_keepalive_timer(sk
);
646 EXPORT_SYMBOL_GPL(tcp_set_keepalive
);
649 static void tcp_keepalive_timer (struct timer_list
*t
)
651 struct sock
*sk
= from_timer(sk
, t
, sk_timer
);
652 struct inet_connection_sock
*icsk
= inet_csk(sk
);
653 struct tcp_sock
*tp
= tcp_sk(sk
);
656 /* Only process if socket is not in use. */
658 if (sock_owned_by_user(sk
)) {
659 /* Try again later. */
660 inet_csk_reset_keepalive_timer (sk
, HZ
/20);
664 if (sk
->sk_state
== TCP_LISTEN
) {
665 pr_err("Hmm... keepalive on a LISTEN ???\n");
669 tcp_mstamp_refresh(tp
);
670 if (sk
->sk_state
== TCP_FIN_WAIT2
&& sock_flag(sk
, SOCK_DEAD
)) {
671 if (tp
->linger2
>= 0) {
672 const int tmo
= tcp_fin_time(sk
) - TCP_TIMEWAIT_LEN
;
675 tcp_time_wait(sk
, TCP_FIN_WAIT2
, tmo
);
679 tcp_send_active_reset(sk
, GFP_ATOMIC
);
683 if (!sock_flag(sk
, SOCK_KEEPOPEN
) ||
684 ((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)))
687 elapsed
= keepalive_time_when(tp
);
689 /* It is alive without keepalive 8) */
690 if (tp
->packets_out
|| !tcp_write_queue_empty(sk
))
693 elapsed
= keepalive_time_elapsed(tp
);
695 if (elapsed
>= keepalive_time_when(tp
)) {
696 /* If the TCP_USER_TIMEOUT option is enabled, use that
697 * to determine when to timeout instead.
699 if ((icsk
->icsk_user_timeout
!= 0 &&
700 elapsed
>= msecs_to_jiffies(icsk
->icsk_user_timeout
) &&
701 icsk
->icsk_probes_out
> 0) ||
702 (icsk
->icsk_user_timeout
== 0 &&
703 icsk
->icsk_probes_out
>= keepalive_probes(tp
))) {
704 tcp_send_active_reset(sk
, GFP_ATOMIC
);
708 if (tcp_write_wakeup(sk
, LINUX_MIB_TCPKEEPALIVE
) <= 0) {
709 icsk
->icsk_probes_out
++;
710 elapsed
= keepalive_intvl_when(tp
);
712 /* If keepalive was lost due to local congestion,
715 elapsed
= TCP_RESOURCE_PROBE_INTERVAL
;
718 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
719 elapsed
= keepalive_time_when(tp
) - elapsed
;
725 inet_csk_reset_keepalive_timer (sk
, elapsed
);
736 static enum hrtimer_restart
tcp_compressed_ack_kick(struct hrtimer
*timer
)
738 struct tcp_sock
*tp
= container_of(timer
, struct tcp_sock
, compressed_ack_timer
);
739 struct sock
*sk
= (struct sock
*)tp
;
742 if (!sock_owned_by_user(sk
)) {
743 if (tp
->compressed_ack
> TCP_FASTRETRANS_THRESH
)
746 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED
,
754 return HRTIMER_NORESTART
;
757 void tcp_init_xmit_timers(struct sock
*sk
)
759 inet_csk_init_xmit_timers(sk
, &tcp_write_timer
, &tcp_delack_timer
,
760 &tcp_keepalive_timer
);
761 hrtimer_init(&tcp_sk(sk
)->pacing_timer
, CLOCK_MONOTONIC
,
762 HRTIMER_MODE_ABS_PINNED_SOFT
);
763 tcp_sk(sk
)->pacing_timer
.function
= tcp_pace_kick
;
765 hrtimer_init(&tcp_sk(sk
)->compressed_ack_timer
, CLOCK_MONOTONIC
,
766 HRTIMER_MODE_REL_PINNED_SOFT
);
767 tcp_sk(sk
)->compressed_ack_timer
.function
= tcp_compressed_ack_kick
;