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
25 int sysctl_tcp_syn_retries __read_mostly = TCP_SYN_RETRIES;
26 int sysctl_tcp_synack_retries __read_mostly = TCP_SYNACK_RETRIES;
27 int sysctl_tcp_keepalive_time __read_mostly = TCP_KEEPALIVE_TIME;
28 int sysctl_tcp_keepalive_probes __read_mostly = TCP_KEEPALIVE_PROBES;
29 int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
30 int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
31 int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
32 int sysctl_tcp_orphan_retries __read_mostly;
33 int sysctl_tcp_thin_linear_timeouts __read_mostly;
34
tcp_write_err(struct sock * sk)35 static void tcp_write_err(struct sock *sk)
36 {
37 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
38 sk->sk_error_report(sk);
39
40 tcp_done(sk);
41 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
42 }
43
44 /* Do not allow orphaned sockets to eat all our resources.
45 * This is direct violation of TCP specs, but it is required
46 * to prevent DoS attacks. It is called when a retransmission timeout
47 * or zero probe timeout occurs on orphaned socket.
48 *
49 * Also close if our net namespace is exiting; in that case there is no
50 * hope of ever communicating again since all netns interfaces are already
51 * down (or about to be down), and we need to release our dst references,
52 * which have been moved to the netns loopback interface, so the namespace
53 * can finish exiting. This condition is only possible if we are a kernel
54 * socket, as those do not hold references to the namespace.
55 *
56 * Criteria is still not confirmed experimentally and may change.
57 * We kill the socket, if:
58 * 1. If number of orphaned sockets exceeds an administratively configured
59 * limit.
60 * 2. If we have strong memory pressure.
61 * 3. If our net namespace is exiting.
62 */
tcp_out_of_resources(struct sock * sk,bool do_reset)63 static int tcp_out_of_resources(struct sock *sk, bool do_reset)
64 {
65 struct tcp_sock *tp = tcp_sk(sk);
66 int shift = 0;
67
68 /* If peer does not open window for long time, or did not transmit
69 * anything for long time, penalize it. */
70 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
71 shift++;
72
73 /* If some dubious ICMP arrived, penalize even more. */
74 if (sk->sk_err_soft)
75 shift++;
76
77 if (tcp_check_oom(sk, shift)) {
78 /* Catch exceptional cases, when connection requires reset.
79 * 1. Last segment was sent recently. */
80 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
81 /* 2. Window is closed. */
82 (!tp->snd_wnd && !tp->packets_out))
83 do_reset = true;
84 if (do_reset)
85 tcp_send_active_reset(sk, GFP_ATOMIC);
86 tcp_done(sk);
87 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
88 return 1;
89 }
90
91 if (!check_net(sock_net(sk))) {
92 /* Not possible to send reset; just close */
93 tcp_done(sk);
94 return 1;
95 }
96
97 return 0;
98 }
99
100 /* Calculate maximal number or retries on an orphaned socket. */
tcp_orphan_retries(struct sock * sk,int alive)101 static int tcp_orphan_retries(struct sock *sk, int alive)
102 {
103 int retries = sysctl_tcp_orphan_retries; /* May be zero. */
104
105 /* We know from an ICMP that something is wrong. */
106 if (sk->sk_err_soft && !alive)
107 retries = 0;
108
109 /* However, if socket sent something recently, select some safe
110 * number of retries. 8 corresponds to >100 seconds with minimal
111 * RTO of 200msec. */
112 if (retries == 0 && alive)
113 retries = 8;
114 return retries;
115 }
116
tcp_mtu_probing(struct inet_connection_sock * icsk,struct sock * sk)117 static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
118 {
119 /* Black hole detection */
120 if (sysctl_tcp_mtu_probing) {
121 if (!icsk->icsk_mtup.enabled) {
122 icsk->icsk_mtup.enabled = 1;
123 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
124 } else {
125 struct tcp_sock *tp = tcp_sk(sk);
126 int mss;
127
128 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
129 mss = min(sysctl_tcp_base_mss, mss);
130 mss = max(mss, 68 - tp->tcp_header_len);
131 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
132 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
133 }
134 }
135 }
136
137 /* This function calculates a "timeout" which is equivalent to the timeout of a
138 * TCP connection after "boundary" unsuccessful, exponentially backed-off
139 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
140 * syn_set flag is set.
141 */
retransmits_timed_out(struct sock * sk,unsigned int boundary,unsigned int timeout,bool syn_set)142 static bool retransmits_timed_out(struct sock *sk,
143 unsigned int boundary,
144 unsigned int timeout,
145 bool syn_set)
146 {
147 unsigned int linear_backoff_thresh, start_ts;
148 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
149
150 if (!inet_csk(sk)->icsk_retransmits)
151 return false;
152
153 start_ts = tcp_sk(sk)->retrans_stamp;
154 if (unlikely(!start_ts))
155 start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk));
156
157 if (likely(timeout == 0)) {
158 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
159
160 if (boundary <= linear_backoff_thresh)
161 timeout = ((2 << boundary) - 1) * rto_base;
162 else
163 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
164 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
165 }
166 return (tcp_time_stamp - start_ts) >= timeout;
167 }
168
169 /* A write timeout has occurred. Process the after effects. */
tcp_write_timeout(struct sock * sk)170 static int tcp_write_timeout(struct sock *sk)
171 {
172 struct inet_connection_sock *icsk = inet_csk(sk);
173 struct tcp_sock *tp = tcp_sk(sk);
174 int retry_until;
175 bool do_reset, syn_set = false;
176
177 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
178 if (icsk->icsk_retransmits) {
179 dst_negative_advice(sk);
180 if (tp->syn_fastopen || tp->syn_data)
181 tcp_fastopen_cache_set(sk, 0, NULL, true);
182 if (tp->syn_data)
183 NET_INC_STATS_BH(sock_net(sk),
184 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
185 }
186 retry_until = icsk->icsk_syn_retries ? : sysctl_tcp_syn_retries;
187 syn_set = true;
188 } else {
189 if (retransmits_timed_out(sk, sysctl_tcp_retries1, 0, 0)) {
190 /* Black hole detection */
191 tcp_mtu_probing(icsk, sk);
192
193 dst_negative_advice(sk);
194 }
195
196 retry_until = sysctl_tcp_retries2;
197 if (sock_flag(sk, SOCK_DEAD)) {
198 const int alive = icsk->icsk_rto < TCP_RTO_MAX;
199
200 retry_until = tcp_orphan_retries(sk, alive);
201 do_reset = alive ||
202 !retransmits_timed_out(sk, retry_until, 0, 0);
203
204 if (tcp_out_of_resources(sk, do_reset))
205 return 1;
206 }
207 }
208
209 if (retransmits_timed_out(sk, retry_until,
210 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
211 /* Has it gone just too far? */
212 tcp_write_err(sk);
213 return 1;
214 }
215 return 0;
216 }
217
tcp_delack_timer_handler(struct sock * sk)218 void tcp_delack_timer_handler(struct sock *sk)
219 {
220 struct tcp_sock *tp = tcp_sk(sk);
221 struct inet_connection_sock *icsk = inet_csk(sk);
222
223 sk_mem_reclaim_partial(sk);
224
225 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
226 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
227 goto out;
228
229 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
230 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
231 goto out;
232 }
233 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
234
235 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
236 struct sk_buff *skb;
237
238 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
239
240 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
241 sk_backlog_rcv(sk, skb);
242
243 tp->ucopy.memory = 0;
244 }
245
246 if (inet_csk_ack_scheduled(sk)) {
247 if (!icsk->icsk_ack.pingpong) {
248 /* Delayed ACK missed: inflate ATO. */
249 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
250 } else {
251 /* Delayed ACK missed: leave pingpong mode and
252 * deflate ATO.
253 */
254 icsk->icsk_ack.pingpong = 0;
255 icsk->icsk_ack.ato = TCP_ATO_MIN;
256 }
257 tcp_send_ack(sk);
258 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
259 }
260
261 out:
262 if (sk_under_memory_pressure(sk))
263 sk_mem_reclaim(sk);
264 }
265
tcp_delack_timer(unsigned long data)266 static void tcp_delack_timer(unsigned long data)
267 {
268 struct sock *sk = (struct sock *)data;
269
270 bh_lock_sock(sk);
271 if (!sock_owned_by_user(sk)) {
272 tcp_delack_timer_handler(sk);
273 } else {
274 inet_csk(sk)->icsk_ack.blocked = 1;
275 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
276 /* deleguate our work to tcp_release_cb() */
277 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
278 sock_hold(sk);
279 }
280 bh_unlock_sock(sk);
281 sock_put(sk);
282 }
283
tcp_probe_timer(struct sock * sk)284 static void tcp_probe_timer(struct sock *sk)
285 {
286 struct inet_connection_sock *icsk = inet_csk(sk);
287 struct tcp_sock *tp = tcp_sk(sk);
288 int max_probes;
289 u32 start_ts;
290
291 if (tp->packets_out || !tcp_send_head(sk)) {
292 icsk->icsk_probes_out = 0;
293 return;
294 }
295
296 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
297 * long as the receiver continues to respond probes. We support this by
298 * default and reset icsk_probes_out with incoming ACKs. But if the
299 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
300 * kill the socket when the retry count and the time exceeds the
301 * corresponding system limit. We also implement similar policy when
302 * we use RTO to probe window in tcp_retransmit_timer().
303 */
304 start_ts = tcp_skb_timestamp(tcp_send_head(sk));
305 if (!start_ts)
306 skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp);
307 else if (icsk->icsk_user_timeout &&
308 (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout)
309 goto abort;
310
311 max_probes = sysctl_tcp_retries2;
312 if (sock_flag(sk, SOCK_DEAD)) {
313 const int alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
314
315 max_probes = tcp_orphan_retries(sk, alive);
316 if (!alive && icsk->icsk_backoff >= max_probes)
317 goto abort;
318 if (tcp_out_of_resources(sk, true))
319 return;
320 }
321
322 if (icsk->icsk_probes_out > max_probes) {
323 abort: tcp_write_err(sk);
324 } else {
325 /* Only send another probe if we didn't close things up. */
326 tcp_send_probe0(sk);
327 }
328 }
329
330 /*
331 * Timer for Fast Open socket to retransmit SYNACK. Note that the
332 * sk here is the child socket, not the parent (listener) socket.
333 */
tcp_fastopen_synack_timer(struct sock * sk)334 static void tcp_fastopen_synack_timer(struct sock *sk)
335 {
336 struct inet_connection_sock *icsk = inet_csk(sk);
337 int max_retries = icsk->icsk_syn_retries ? :
338 sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
339 struct request_sock *req;
340
341 req = tcp_sk(sk)->fastopen_rsk;
342 req->rsk_ops->syn_ack_timeout(sk, req);
343
344 if (req->num_timeout >= max_retries) {
345 tcp_write_err(sk);
346 return;
347 }
348 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
349 * returned from rtx_syn_ack() to make it more persistent like
350 * regular retransmit because if the child socket has been accepted
351 * it's not good to give up too easily.
352 */
353 inet_rtx_syn_ack(sk, req);
354 req->num_timeout++;
355 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
356 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
357 }
358
359 /*
360 * The TCP retransmit timer.
361 */
362
tcp_retransmit_timer(struct sock * sk)363 void tcp_retransmit_timer(struct sock *sk)
364 {
365 struct tcp_sock *tp = tcp_sk(sk);
366 struct inet_connection_sock *icsk = inet_csk(sk);
367
368 if (tp->fastopen_rsk) {
369 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
370 sk->sk_state != TCP_FIN_WAIT1);
371 tcp_fastopen_synack_timer(sk);
372 /* Before we receive ACK to our SYN-ACK don't retransmit
373 * anything else (e.g., data or FIN segments).
374 */
375 return;
376 }
377 if (!tp->packets_out)
378 goto out;
379
380 WARN_ON(tcp_write_queue_empty(sk));
381
382 tp->tlp_high_seq = 0;
383
384 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
385 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
386 /* Receiver dastardly shrinks window. Our retransmits
387 * become zero probes, but we should not timeout this
388 * connection. If the socket is an orphan, time it out,
389 * we cannot allow such beasts to hang infinitely.
390 */
391 struct inet_sock *inet = inet_sk(sk);
392 if (sk->sk_family == AF_INET) {
393 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
394 &inet->inet_daddr,
395 ntohs(inet->inet_dport), inet->inet_num,
396 tp->snd_una, tp->snd_nxt);
397 }
398 #if IS_ENABLED(CONFIG_IPV6)
399 else if (sk->sk_family == AF_INET6) {
400 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n"),
401 &sk->sk_v6_daddr,
402 ntohs(inet->inet_dport), inet->inet_num,
403 tp->snd_una, tp->snd_nxt);
404 }
405 #endif
406 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
407 tcp_write_err(sk);
408 goto out;
409 }
410 tcp_enter_loss(sk);
411 tcp_retransmit_skb(sk, tcp_write_queue_head(sk));
412 __sk_dst_reset(sk);
413 goto out_reset_timer;
414 }
415
416 if (tcp_write_timeout(sk))
417 goto out;
418
419 if (icsk->icsk_retransmits == 0) {
420 int mib_idx;
421
422 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
423 if (tcp_is_sack(tp))
424 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
425 else
426 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
427 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
428 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
429 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
430 tp->sacked_out) {
431 if (tcp_is_sack(tp))
432 mib_idx = LINUX_MIB_TCPSACKFAILURES;
433 else
434 mib_idx = LINUX_MIB_TCPRENOFAILURES;
435 } else {
436 mib_idx = LINUX_MIB_TCPTIMEOUTS;
437 }
438 NET_INC_STATS_BH(sock_net(sk), mib_idx);
439 }
440
441 tcp_enter_loss(sk);
442
443 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk)) > 0) {
444 /* Retransmission failed because of local congestion,
445 * do not backoff.
446 */
447 if (!icsk->icsk_retransmits)
448 icsk->icsk_retransmits = 1;
449 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
450 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
451 TCP_RTO_MAX);
452 goto out;
453 }
454
455 /* Increase the timeout each time we retransmit. Note that
456 * we do not increase the rtt estimate. rto is initialized
457 * from rtt, but increases here. Jacobson (SIGCOMM 88) suggests
458 * that doubling rto each time is the least we can get away with.
459 * In KA9Q, Karn uses this for the first few times, and then
460 * goes to quadratic. netBSD doubles, but only goes up to *64,
461 * and clamps at 1 to 64 sec afterwards. Note that 120 sec is
462 * defined in the protocol as the maximum possible RTT. I guess
463 * we'll have to use something other than TCP to talk to the
464 * University of Mars.
465 *
466 * PAWS allows us longer timeouts and large windows, so once
467 * implemented ftp to mars will work nicely. We will have to fix
468 * the 120 second clamps though!
469 */
470 icsk->icsk_backoff++;
471 icsk->icsk_retransmits++;
472
473 out_reset_timer:
474 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
475 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
476 * might be increased if the stream oscillates between thin and thick,
477 * thus the old value might already be too high compared to the value
478 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
479 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
480 * exponential backoff behaviour to avoid continue hammering
481 * linear-timeout retransmissions into a black hole
482 */
483 if (sk->sk_state == TCP_ESTABLISHED &&
484 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
485 tcp_stream_is_thin(tp) &&
486 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
487 icsk->icsk_backoff = 0;
488 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
489 } else {
490 /* Use normal (exponential) backoff */
491 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
492 }
493 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
494 if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1, 0, 0))
495 __sk_dst_reset(sk);
496
497 out:;
498 }
499
tcp_write_timer_handler(struct sock * sk)500 void tcp_write_timer_handler(struct sock *sk)
501 {
502 struct inet_connection_sock *icsk = inet_csk(sk);
503 int event;
504
505 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
506 !icsk->icsk_pending)
507 goto out;
508
509 if (time_after(icsk->icsk_timeout, jiffies)) {
510 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
511 goto out;
512 }
513
514 event = icsk->icsk_pending;
515
516 switch (event) {
517 case ICSK_TIME_EARLY_RETRANS:
518 tcp_resume_early_retransmit(sk);
519 break;
520 case ICSK_TIME_LOSS_PROBE:
521 tcp_send_loss_probe(sk);
522 break;
523 case ICSK_TIME_RETRANS:
524 icsk->icsk_pending = 0;
525 tcp_retransmit_timer(sk);
526 break;
527 case ICSK_TIME_PROBE0:
528 icsk->icsk_pending = 0;
529 tcp_probe_timer(sk);
530 break;
531 }
532
533 out:
534 sk_mem_reclaim(sk);
535 }
536
tcp_write_timer(unsigned long data)537 static void tcp_write_timer(unsigned long data)
538 {
539 struct sock *sk = (struct sock *)data;
540
541 bh_lock_sock(sk);
542 if (!sock_owned_by_user(sk)) {
543 tcp_write_timer_handler(sk);
544 } else {
545 /* deleguate our work to tcp_release_cb() */
546 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
547 sock_hold(sk);
548 }
549 bh_unlock_sock(sk);
550 sock_put(sk);
551 }
552
553 /*
554 * Timer for listening sockets
555 */
556
tcp_synack_timer(struct sock * sk)557 static void tcp_synack_timer(struct sock *sk)
558 {
559 inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL,
560 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
561 }
562
tcp_syn_ack_timeout(struct sock * sk,struct request_sock * req)563 void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
564 {
565 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
566 }
567 EXPORT_SYMBOL(tcp_syn_ack_timeout);
568
tcp_set_keepalive(struct sock * sk,int val)569 void tcp_set_keepalive(struct sock *sk, int val)
570 {
571 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
572 return;
573
574 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
575 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
576 else if (!val)
577 inet_csk_delete_keepalive_timer(sk);
578 }
579
580
tcp_keepalive_timer(unsigned long data)581 static void tcp_keepalive_timer (unsigned long data)
582 {
583 struct sock *sk = (struct sock *) data;
584 struct inet_connection_sock *icsk = inet_csk(sk);
585 struct tcp_sock *tp = tcp_sk(sk);
586 u32 elapsed;
587
588 /* Only process if socket is not in use. */
589 bh_lock_sock(sk);
590 if (sock_owned_by_user(sk)) {
591 /* Try again later. */
592 inet_csk_reset_keepalive_timer (sk, HZ/20);
593 goto out;
594 }
595
596 if (sk->sk_state == TCP_LISTEN) {
597 tcp_synack_timer(sk);
598 goto out;
599 }
600
601 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
602 if (tp->linger2 >= 0) {
603 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
604
605 if (tmo > 0) {
606 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
607 goto out;
608 }
609 }
610 tcp_send_active_reset(sk, GFP_ATOMIC);
611 goto death;
612 }
613
614 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
615 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
616 goto out;
617
618 elapsed = keepalive_time_when(tp);
619
620 /* It is alive without keepalive 8) */
621 if (tp->packets_out || tcp_send_head(sk))
622 goto resched;
623
624 elapsed = keepalive_time_elapsed(tp);
625
626 if (elapsed >= keepalive_time_when(tp)) {
627 /* If the TCP_USER_TIMEOUT option is enabled, use that
628 * to determine when to timeout instead.
629 */
630 if ((icsk->icsk_user_timeout != 0 &&
631 elapsed >= icsk->icsk_user_timeout &&
632 icsk->icsk_probes_out > 0) ||
633 (icsk->icsk_user_timeout == 0 &&
634 icsk->icsk_probes_out >= keepalive_probes(tp))) {
635 tcp_send_active_reset(sk, GFP_ATOMIC);
636 tcp_write_err(sk);
637 goto out;
638 }
639 if (tcp_write_wakeup(sk) <= 0) {
640 icsk->icsk_probes_out++;
641 elapsed = keepalive_intvl_when(tp);
642 } else {
643 /* If keepalive was lost due to local congestion,
644 * try harder.
645 */
646 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
647 }
648 } else {
649 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
650 elapsed = keepalive_time_when(tp) - elapsed;
651 }
652
653 sk_mem_reclaim(sk);
654
655 resched:
656 inet_csk_reset_keepalive_timer (sk, elapsed);
657 goto out;
658
659 death:
660 tcp_done(sk);
661
662 out:
663 bh_unlock_sock(sk);
664 sock_put(sk);
665 }
666
tcp_init_xmit_timers(struct sock * sk)667 void tcp_init_xmit_timers(struct sock *sk)
668 {
669 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
670 &tcp_keepalive_timer);
671 }
672 EXPORT_SYMBOL(tcp_init_xmit_timers);
673