1/*
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.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * Authors: Ross Biro
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>
19 */
20
21#include <linux/module.h>
22#include <linux/gfp.h>
23#include <net/tcp.h>
24
25static u32 tcp_retransmit_stamp(const struct sock *sk)
26{
27 u32 start_ts = tcp_sk(sk)->retrans_stamp;
28
29 if (unlikely(!start_ts)) {
30 struct sk_buff *head = tcp_rtx_queue_head(sk);
31
32 if (!head)
33 return 0;
34 start_ts = tcp_skb_timestamp(head);
35 }
36 return start_ts;
37}
38
39static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
40{
41 struct inet_connection_sock *icsk = inet_csk(sk);
42 u32 elapsed, start_ts;
43
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 */
50 else
51 return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(icsk->icsk_user_timeout - elapsed));
52}
53
54/**
55 * tcp_write_err() - close socket and save error info
56 * @sk: The socket the error has appeared on.
57 *
58 * Returns: Nothing (void)
59 */
60
61static void tcp_write_err(struct sock *sk)
62{
63 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
64 sk->sk_error_report(sk);
65
66 tcp_write_queue_purge(sk);
67 tcp_done(sk);
68 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
69}
70
71/**
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
75 *
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.
80 *
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.
87 *
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
91 * limit.
92 * 2. If we have strong memory pressure.
93 * 3. If our net namespace is exiting.
94 */
95static int tcp_out_of_resources(struct sock *sk, bool do_reset)
96{
97 struct tcp_sock *tp = tcp_sk(sk);
98 int shift = 0;
99
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)
103 shift++;
104
105 /* If some dubious ICMP arrived, penalize even more. */
106 if (sk->sk_err_soft)
107 shift++;
108
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))
115 do_reset = true;
116 if (do_reset)
117 tcp_send_active_reset(sk, GFP_ATOMIC);
118 tcp_done(sk);
119 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
120 return 1;
121 }
122
123 if (!check_net(sock_net(sk))) {
124 /* Not possible to send reset; just close */
125 tcp_done(sk);
126 return 1;
127 }
128
129 return 0;
130}
131
132/**
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
136 */
137static int tcp_orphan_retries(struct sock *sk, bool alive)
138{
139 int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */
140
141 /* We know from an ICMP that something is wrong. */
142 if (sk->sk_err_soft && !alive)
143 retries = 0;
144
145 /* However, if socket sent something recently, select some safe
146 * number of retries. 8 corresponds to >100 seconds with minimal
147 * RTO of 200msec. */
148 if (retries == 0 && alive)
149 retries = 8;
150 return retries;
151}
152
153static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
154{
155 const struct net *net = sock_net(sk);
156 int mss;
157
158 /* Black hole detection */
159 if (!net->ipv4.sysctl_tcp_mtu_probing)
160 return;
161
162 if (!icsk->icsk_mtup.enabled) {
163 icsk->icsk_mtup.enabled = 1;
164 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
165 } else {
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);
170 }
171 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
172}
173
174
175/**
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.
182 *
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.
187 */
188static bool retransmits_timed_out(struct sock *sk,
189 unsigned int boundary,
190 unsigned int timeout)
191{
192 const unsigned int rto_base = TCP_RTO_MIN;
193 unsigned int linear_backoff_thresh, start_ts;
194
195 if (!inet_csk(sk)->icsk_retransmits)
196 return false;
197
198 start_ts = tcp_retransmit_stamp(sk);
199 if (!start_ts)
200 return false;
201
202 if (likely(timeout == 0)) {
203 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
204
205 if (boundary <= linear_backoff_thresh)
206 timeout = ((2 << boundary) - 1) * rto_base;
207 else
208 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
209 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
210 timeout = jiffies_to_msecs(timeout);
211 }
212 return (tcp_time_stamp(tcp_sk(sk)) - start_ts) >= timeout;
213}
214
215/* A write timeout has occurred. Process the after effects. */
216static int tcp_write_timeout(struct sock *sk)
217{
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;
222 int retry_until;
223
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);
229 }
230 retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
231 expired = icsk->icsk_retransmits >= retry_until;
232 } else {
233 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0)) {
234 /* Black hole detection */
235 tcp_mtu_probing(icsk, sk);
236
237 dst_negative_advice(sk);
238 } else {
239 sk_rethink_txhash(sk);
240 }
241
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;
245
246 retry_until = tcp_orphan_retries(sk, alive);
247 do_reset = alive ||
248 !retransmits_timed_out(sk, retry_until, 0);
249
250 if (tcp_out_of_resources(sk, do_reset))
251 return 1;
252 }
253 expired = retransmits_timed_out(sk, retry_until,
254 icsk->icsk_user_timeout);
255 }
256 tcp_fastopen_active_detect_blackhole(sk, expired);
257
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);
262
263 if (expired) {
264 /* Has it gone just too far? */
265 tcp_write_err(sk);
266 return 1;
267 }
268
269 return 0;
270}
271
272/* Called with BH disabled */
273void tcp_delack_timer_handler(struct sock *sk)
274{
275 struct inet_connection_sock *icsk = inet_csk(sk);
276
277 sk_mem_reclaim_partial(sk);
278
279 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
280 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
281 goto out;
282
283 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
284 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
285 goto out;
286 }
287 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
288
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);
293 } else {
294 /* Delayed ACK missed: leave pingpong mode and
295 * deflate ATO.
296 */
297 icsk->icsk_ack.pingpong = 0;
298 icsk->icsk_ack.ato = TCP_ATO_MIN;
299 }
300 tcp_mstamp_refresh(tcp_sk(sk));
301 tcp_send_ack(sk);
302 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
303 }
304
305out:
306 if (tcp_under_memory_pressure(sk))
307 sk_mem_reclaim(sk);
308}
309
310
311/**
312 * tcp_delack_timer() - The TCP delayed ACK timeout handler
313 * @data: Pointer to the current socket. (gets casted to struct sock *)
314 *
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.
317 *
318 * Returns: Nothing (void)
319 */
320static void tcp_delack_timer(struct timer_list *t)
321{
322 struct inet_connection_sock *icsk =
323 from_timer(icsk, t, icsk_delack_timer);
324 struct sock *sk = &icsk->icsk_inet.sk;
325
326 bh_lock_sock(sk);
327 if (!sock_owned_by_user(sk)) {
328 tcp_delack_timer_handler(sk);
329 } else {
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))
334 sock_hold(sk);
335 }
336 bh_unlock_sock(sk);
337 sock_put(sk);
338}
339
340static void tcp_probe_timer(struct sock *sk)
341{
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);
345 int max_probes;
346 u32 start_ts;
347
348 if (tp->packets_out || !skb) {
349 icsk->icsk_probes_out = 0;
350 return;
351 }
352
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().
360 */
361 start_ts = tcp_skb_timestamp(skb);
362 if (!start_ts)
363 skb->skb_mstamp = tp->tcp_mstamp;
364 else if (icsk->icsk_user_timeout &&
365 (s32)(tcp_time_stamp(tp) - start_ts) > icsk->icsk_user_timeout)
366 goto abort;
367
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;
371
372 max_probes = tcp_orphan_retries(sk, alive);
373 if (!alive && icsk->icsk_backoff >= max_probes)
374 goto abort;
375 if (tcp_out_of_resources(sk, true))
376 return;
377 }
378
379 if (icsk->icsk_probes_out > max_probes) {
380abort: tcp_write_err(sk);
381 } else {
382 /* Only send another probe if we didn't close things up. */
383 tcp_send_probe0(sk);
384 }
385}
386
387/*
388 * Timer for Fast Open socket to retransmit SYNACK. Note that the
389 * sk here is the child socket, not the parent (listener) socket.
390 */
391static void tcp_fastopen_synack_timer(struct sock *sk)
392{
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;
397
398 req = tcp_sk(sk)->fastopen_rsk;
399 req->rsk_ops->syn_ack_timeout(req);
400
401 if (req->num_timeout >= max_retries) {
402 tcp_write_err(sk);
403 return;
404 }
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.
409 */
410 inet_rtx_syn_ack(sk, req);
411 req->num_timeout++;
412 icsk->icsk_retransmits++;
413 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
414 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
415}
416
417
418/**
419 * tcp_retransmit_timer() - The TCP retransmit timeout handler
420 * @sk: Pointer to the current socket.
421 *
422 * This function gets called when the kernel timer for a TCP packet
423 * of this socket expires.
424 *
425 * It handles retransmission, timer adjustment and other necesarry measures.
426 *
427 * Returns: Nothing (void)
428 */
429void tcp_retransmit_timer(struct sock *sk)
430{
431 struct tcp_sock *tp = tcp_sk(sk);
432 struct net *net = sock_net(sk);
433 struct inet_connection_sock *icsk = inet_csk(sk);
434
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).
441 */
442 return;
443 }
444 if (!tp->packets_out)
445 goto out;
446
447 WARN_ON(tcp_rtx_queue_empty(sk));
448
449 tp->tlp_high_seq = 0;
450
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.
457 */
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",
461 &inet->inet_daddr,
462 ntohs(inet->inet_dport),
463 inet->inet_num,
464 tp->snd_una, tp->snd_nxt);
465 }
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",
469 &sk->sk_v6_daddr,
470 ntohs(inet->inet_dport),
471 inet->inet_num,
472 tp->snd_una, tp->snd_nxt);
473 }
474#endif
475 if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) {
476 tcp_write_err(sk);
477 goto out;
478 }
479 tcp_enter_loss(sk);
480 tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1);
481 __sk_dst_reset(sk);
482 goto out_reset_timer;
483 }
484
485 if (tcp_write_timeout(sk))
486 goto out;
487
488 if (icsk->icsk_retransmits == 0) {
489 int mib_idx;
490
491 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
492 if (tcp_is_sack(tp))
493 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
494 else
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) ||
499 tp->sacked_out) {
500 if (tcp_is_sack(tp))
501 mib_idx = LINUX_MIB_TCPSACKFAILURES;
502 else
503 mib_idx = LINUX_MIB_TCPRENOFAILURES;
504 } else {
505 mib_idx = LINUX_MIB_TCPTIMEOUTS;
506 }
507 __NET_INC_STATS(sock_net(sk), mib_idx);
508 }
509
510 tcp_enter_loss(sk);
511
512 if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) {
513 /* Retransmission failed because of local congestion,
514 * do not backoff.
515 */
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),
520 TCP_RTO_MAX);
521 goto out;
522 }
523
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.
534 *
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!
538 */
539 icsk->icsk_backoff++;
540 icsk->icsk_retransmits++;
541
542out_reset_timer:
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
551 */
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);
558 } else {
559 /* Use normal (exponential) backoff */
560 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
561 }
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))
565 __sk_dst_reset(sk);
566
567out:;
568}
569
570/* Called with bottom-half processing disabled.
571 Called by tcp_write_timer() */
572void tcp_write_timer_handler(struct sock *sk)
573{
574 struct inet_connection_sock *icsk = inet_csk(sk);
575 int event;
576
577 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
578 !icsk->icsk_pending)
579 goto out;
580
581 if (time_after(icsk->icsk_timeout, jiffies)) {
582 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
583 goto out;
584 }
585
586 tcp_mstamp_refresh(tcp_sk(sk));
587 event = icsk->icsk_pending;
588
589 switch (event) {
590 case ICSK_TIME_REO_TIMEOUT:
591 tcp_rack_reo_timeout(sk);
592 break;
593 case ICSK_TIME_LOSS_PROBE:
594 tcp_send_loss_probe(sk);
595 break;
596 case ICSK_TIME_RETRANS:
597 icsk->icsk_pending = 0;
598 tcp_retransmit_timer(sk);
599 break;
600 case ICSK_TIME_PROBE0:
601 icsk->icsk_pending = 0;
602 tcp_probe_timer(sk);
603 break;
604 }
605
606out:
607 sk_mem_reclaim(sk);
608}
609
610static void tcp_write_timer(struct timer_list *t)
611{
612 struct inet_connection_sock *icsk =
613 from_timer(icsk, t, icsk_retransmit_timer);
614 struct sock *sk = &icsk->icsk_inet.sk;
615
616 bh_lock_sock(sk);
617 if (!sock_owned_by_user(sk)) {
618 tcp_write_timer_handler(sk);
619 } else {
620 /* delegate our work to tcp_release_cb() */
621 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags))
622 sock_hold(sk);
623 }
624 bh_unlock_sock(sk);
625 sock_put(sk);
626}
627
628void tcp_syn_ack_timeout(const struct request_sock *req)
629{
630 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
631
632 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
633}
634EXPORT_SYMBOL(tcp_syn_ack_timeout);
635
636void tcp_set_keepalive(struct sock *sk, int val)
637{
638 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
639 return;
640
641 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
642 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
643 else if (!val)
644 inet_csk_delete_keepalive_timer(sk);
645}
646EXPORT_SYMBOL_GPL(tcp_set_keepalive);
647
648
649static void tcp_keepalive_timer (struct timer_list *t)
650{
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);
654 u32 elapsed;
655
656 /* Only process if socket is not in use. */
657 bh_lock_sock(sk);
658 if (sock_owned_by_user(sk)) {
659 /* Try again later. */
660 inet_csk_reset_keepalive_timer (sk, HZ/20);
661 goto out;
662 }
663
664 if (sk->sk_state == TCP_LISTEN) {
665 pr_err("Hmm... keepalive on a LISTEN ???\n");
666 goto out;
667 }
668
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;
673
674 if (tmo > 0) {
675 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
676 goto out;
677 }
678 }
679 tcp_send_active_reset(sk, GFP_ATOMIC);
680 goto death;
681 }
682
683 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
684 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
685 goto out;
686
687 elapsed = keepalive_time_when(tp);
688
689 /* It is alive without keepalive 8) */
690 if (tp->packets_out || !tcp_write_queue_empty(sk))
691 goto resched;
692
693 elapsed = keepalive_time_elapsed(tp);
694
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.
698 */
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);
705 tcp_write_err(sk);
706 goto out;
707 }
708 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
709 icsk->icsk_probes_out++;
710 elapsed = keepalive_intvl_when(tp);
711 } else {
712 /* If keepalive was lost due to local congestion,
713 * try harder.
714 */
715 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
716 }
717 } else {
718 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
719 elapsed = keepalive_time_when(tp) - elapsed;
720 }
721
722 sk_mem_reclaim(sk);
723
724resched:
725 inet_csk_reset_keepalive_timer (sk, elapsed);
726 goto out;
727
728death:
729 tcp_done(sk);
730
731out:
732 bh_unlock_sock(sk);
733 sock_put(sk);
734}
735
736static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
737{
738 struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer);
739 struct sock *sk = (struct sock *)tp;
740
741 bh_lock_sock(sk);
742 if (!sock_owned_by_user(sk)) {
743 if (tp->compressed_ack)
744 tcp_send_ack(sk);
745 } else {
746 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
747 &sk->sk_tsq_flags))
748 sock_hold(sk);
749 }
750 bh_unlock_sock(sk);
751
752 sock_put(sk);
753
754 return HRTIMER_NORESTART;
755}
756
757void tcp_init_xmit_timers(struct sock *sk)
758{
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;
764
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;
768}
769