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_thin_linear_timeouts __read_mostly;
26
27 /**
28 * tcp_write_err() - close socket and save error info
29 * @sk: The socket the error has appeared on.
30 *
31 * Returns: Nothing (void)
32 */
33
tcp_write_err(struct sock * sk)34 static void tcp_write_err(struct sock *sk)
35 {
36 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
37 sk->sk_error_report(sk);
38
39 tcp_done(sk);
40 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
41 }
42
43 /**
44 * tcp_out_of_resources() - Close socket if out of resources
45 * @sk: pointer to current socket
46 * @do_reset: send a last packet with reset flag
47 *
48 * Do not allow orphaned sockets to eat all our resources.
49 * This is direct violation of TCP specs, but it is required
50 * to prevent DoS attacks. It is called when a retransmission timeout
51 * or zero probe timeout occurs on orphaned socket.
52 *
53 * Also close if our net namespace is exiting; in that case there is no
54 * hope of ever communicating again since all netns interfaces are already
55 * down (or about to be down), and we need to release our dst references,
56 * which have been moved to the netns loopback interface, so the namespace
57 * can finish exiting. This condition is only possible if we are a kernel
58 * socket, as those do not hold references to the namespace.
59 *
60 * Criteria is still not confirmed experimentally and may change.
61 * We kill the socket, if:
62 * 1. If number of orphaned sockets exceeds an administratively configured
63 * limit.
64 * 2. If we have strong memory pressure.
65 * 3. If our net namespace is exiting.
66 */
tcp_out_of_resources(struct sock * sk,bool do_reset)67 static int tcp_out_of_resources(struct sock *sk, bool do_reset)
68 {
69 struct tcp_sock *tp = tcp_sk(sk);
70 int shift = 0;
71
72 /* If peer does not open window for long time, or did not transmit
73 * anything for long time, penalize it. */
74 if ((s32)(tcp_time_stamp - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
75 shift++;
76
77 /* If some dubious ICMP arrived, penalize even more. */
78 if (sk->sk_err_soft)
79 shift++;
80
81 if (tcp_check_oom(sk, shift)) {
82 /* Catch exceptional cases, when connection requires reset.
83 * 1. Last segment was sent recently. */
84 if ((s32)(tcp_time_stamp - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
85 /* 2. Window is closed. */
86 (!tp->snd_wnd && !tp->packets_out))
87 do_reset = true;
88 if (do_reset)
89 tcp_send_active_reset(sk, GFP_ATOMIC);
90 tcp_done(sk);
91 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
92 return 1;
93 }
94
95 if (!check_net(sock_net(sk))) {
96 /* Not possible to send reset; just close */
97 tcp_done(sk);
98 return 1;
99 }
100
101 return 0;
102 }
103
104 /**
105 * tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket
106 * @sk: Pointer to the current socket.
107 * @alive: bool, socket alive state
108 */
tcp_orphan_retries(struct sock * sk,bool alive)109 static int tcp_orphan_retries(struct sock *sk, bool alive)
110 {
111 int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries; /* May be zero. */
112
113 /* We know from an ICMP that something is wrong. */
114 if (sk->sk_err_soft && !alive)
115 retries = 0;
116
117 /* However, if socket sent something recently, select some safe
118 * number of retries. 8 corresponds to >100 seconds with minimal
119 * RTO of 200msec. */
120 if (retries == 0 && alive)
121 retries = 8;
122 return retries;
123 }
124
tcp_mtu_probing(struct inet_connection_sock * icsk,struct sock * sk)125 static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
126 {
127 struct net *net = sock_net(sk);
128
129 /* Black hole detection */
130 if (net->ipv4.sysctl_tcp_mtu_probing) {
131 if (!icsk->icsk_mtup.enabled) {
132 icsk->icsk_mtup.enabled = 1;
133 icsk->icsk_mtup.probe_timestamp = tcp_time_stamp;
134 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
135 } else {
136 struct net *net = sock_net(sk);
137 struct tcp_sock *tp = tcp_sk(sk);
138 int mss;
139
140 mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;
141 mss = min(net->ipv4.sysctl_tcp_base_mss, mss);
142 mss = max(mss, 68 - tp->tcp_header_len);
143 icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
144 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
145 }
146 }
147 }
148
149
150 /**
151 * retransmits_timed_out() - returns true if this connection has timed out
152 * @sk: The current socket
153 * @boundary: max number of retransmissions
154 * @timeout: A custom timeout value.
155 * If set to 0 the default timeout is calculated and used.
156 * Using TCP_RTO_MIN and the number of unsuccessful retransmits.
157 * @syn_set: true if the SYN Bit was set.
158 *
159 * The default "timeout" value this function can calculate and use
160 * is equivalent to the timeout of a TCP Connection
161 * after "boundary" unsuccessful, exponentially backed-off
162 * retransmissions with an initial RTO of TCP_RTO_MIN or TCP_TIMEOUT_INIT if
163 * syn_set flag is set.
164 *
165 */
retransmits_timed_out(struct sock * sk,unsigned int boundary,unsigned int timeout,bool syn_set)166 static bool retransmits_timed_out(struct sock *sk,
167 unsigned int boundary,
168 unsigned int timeout,
169 bool syn_set)
170 {
171 unsigned int linear_backoff_thresh, start_ts;
172 unsigned int rto_base = syn_set ? TCP_TIMEOUT_INIT : TCP_RTO_MIN;
173
174 if (!inet_csk(sk)->icsk_retransmits)
175 return false;
176
177 start_ts = tcp_sk(sk)->retrans_stamp;
178 if (unlikely(!start_ts))
179 start_ts = tcp_skb_timestamp(tcp_write_queue_head(sk));
180
181 if (likely(timeout == 0)) {
182 linear_backoff_thresh = ilog2(TCP_RTO_MAX/rto_base);
183
184 if (boundary <= linear_backoff_thresh)
185 timeout = ((2 << boundary) - 1) * rto_base;
186 else
187 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
188 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
189 }
190 return (tcp_time_stamp - start_ts) >= timeout;
191 }
192
193 /* A write timeout has occurred. Process the after effects. */
tcp_write_timeout(struct sock * sk)194 static int tcp_write_timeout(struct sock *sk)
195 {
196 struct inet_connection_sock *icsk = inet_csk(sk);
197 struct tcp_sock *tp = tcp_sk(sk);
198 struct net *net = sock_net(sk);
199 int retry_until;
200 bool do_reset, syn_set = false;
201
202 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
203 if (icsk->icsk_retransmits) {
204 dst_negative_advice(sk);
205 if (tp->syn_fastopen || tp->syn_data)
206 tcp_fastopen_cache_set(sk, 0, NULL, true, 0);
207 if (tp->syn_data && icsk->icsk_retransmits == 1)
208 NET_INC_STATS(sock_net(sk),
209 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
210 } else if (!tp->syn_data && !tp->syn_fastopen) {
211 sk_rethink_txhash(sk);
212 }
213 retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
214 syn_set = true;
215 } else {
216 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0, 0)) {
217 /* Some middle-boxes may black-hole Fast Open _after_
218 * the handshake. Therefore we conservatively disable
219 * Fast Open on this path on recurring timeouts with
220 * few or zero bytes acked after Fast Open.
221 */
222 if (tp->syn_data_acked &&
223 tp->bytes_acked <= tp->rx_opt.mss_clamp) {
224 tcp_fastopen_cache_set(sk, 0, NULL, true, 0);
225 if (icsk->icsk_retransmits == net->ipv4.sysctl_tcp_retries1)
226 NET_INC_STATS(sock_net(sk),
227 LINUX_MIB_TCPFASTOPENACTIVEFAIL);
228 }
229 /* Black hole detection */
230 tcp_mtu_probing(icsk, sk);
231
232 dst_negative_advice(sk);
233 } else {
234 sk_rethink_txhash(sk);
235 }
236
237 retry_until = net->ipv4.sysctl_tcp_retries2;
238 if (sock_flag(sk, SOCK_DEAD)) {
239 const bool alive = icsk->icsk_rto < TCP_RTO_MAX;
240
241 retry_until = tcp_orphan_retries(sk, alive);
242 do_reset = alive ||
243 !retransmits_timed_out(sk, retry_until, 0, 0);
244
245 if (tcp_out_of_resources(sk, do_reset))
246 return 1;
247 }
248 }
249
250 if (retransmits_timed_out(sk, retry_until,
251 syn_set ? 0 : icsk->icsk_user_timeout, syn_set)) {
252 /* Has it gone just too far? */
253 tcp_write_err(sk);
254 return 1;
255 }
256 return 0;
257 }
258
259 /* Called with BH disabled */
tcp_delack_timer_handler(struct sock * sk)260 void tcp_delack_timer_handler(struct sock *sk)
261 {
262 struct tcp_sock *tp = tcp_sk(sk);
263 struct inet_connection_sock *icsk = inet_csk(sk);
264
265 sk_mem_reclaim_partial(sk);
266
267 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
268 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
269 goto out;
270
271 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
272 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
273 goto out;
274 }
275 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
276
277 if (!skb_queue_empty(&tp->ucopy.prequeue)) {
278 struct sk_buff *skb;
279
280 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPSCHEDULERFAILED);
281
282 while ((skb = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
283 sk_backlog_rcv(sk, skb);
284
285 tp->ucopy.memory = 0;
286 }
287
288 if (inet_csk_ack_scheduled(sk)) {
289 if (!icsk->icsk_ack.pingpong) {
290 /* Delayed ACK missed: inflate ATO. */
291 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
292 } else {
293 /* Delayed ACK missed: leave pingpong mode and
294 * deflate ATO.
295 */
296 icsk->icsk_ack.pingpong = 0;
297 icsk->icsk_ack.ato = TCP_ATO_MIN;
298 }
299 tcp_send_ack(sk);
300 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
301 }
302
303 out:
304 if (tcp_under_memory_pressure(sk))
305 sk_mem_reclaim(sk);
306 }
307
308
309 /**
310 * tcp_delack_timer() - The TCP delayed ACK timeout handler
311 * @data: Pointer to the current socket. (gets casted to struct sock *)
312 *
313 * This function gets (indirectly) called when the kernel timer for a TCP packet
314 * of this socket expires. Calls tcp_delack_timer_handler() to do the actual work.
315 *
316 * Returns: Nothing (void)
317 */
tcp_delack_timer(unsigned long data)318 static void tcp_delack_timer(unsigned long data)
319 {
320 struct sock *sk = (struct sock *)data;
321
322 bh_lock_sock(sk);
323 if (!sock_owned_by_user(sk)) {
324 tcp_delack_timer_handler(sk);
325 } else {
326 inet_csk(sk)->icsk_ack.blocked = 1;
327 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
328 /* deleguate our work to tcp_release_cb() */
329 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
330 sock_hold(sk);
331 }
332 bh_unlock_sock(sk);
333 sock_put(sk);
334 }
335
tcp_probe_timer(struct sock * sk)336 static void tcp_probe_timer(struct sock *sk)
337 {
338 struct inet_connection_sock *icsk = inet_csk(sk);
339 struct tcp_sock *tp = tcp_sk(sk);
340 int max_probes;
341 u32 start_ts;
342
343 if (tp->packets_out || !tcp_send_head(sk)) {
344 icsk->icsk_probes_out = 0;
345 return;
346 }
347
348 /* RFC 1122 4.2.2.17 requires the sender to stay open indefinitely as
349 * long as the receiver continues to respond probes. We support this by
350 * default and reset icsk_probes_out with incoming ACKs. But if the
351 * socket is orphaned or the user specifies TCP_USER_TIMEOUT, we
352 * kill the socket when the retry count and the time exceeds the
353 * corresponding system limit. We also implement similar policy when
354 * we use RTO to probe window in tcp_retransmit_timer().
355 */
356 start_ts = tcp_skb_timestamp(tcp_send_head(sk));
357 if (!start_ts)
358 skb_mstamp_get(&tcp_send_head(sk)->skb_mstamp);
359 else if (icsk->icsk_user_timeout &&
360 (s32)(tcp_time_stamp - start_ts) > icsk->icsk_user_timeout)
361 goto abort;
362
363 max_probes = sock_net(sk)->ipv4.sysctl_tcp_retries2;
364 if (sock_flag(sk, SOCK_DEAD)) {
365 const bool alive = inet_csk_rto_backoff(icsk, TCP_RTO_MAX) < TCP_RTO_MAX;
366
367 max_probes = tcp_orphan_retries(sk, alive);
368 if (!alive && icsk->icsk_backoff >= max_probes)
369 goto abort;
370 if (tcp_out_of_resources(sk, true))
371 return;
372 }
373
374 if (icsk->icsk_probes_out > max_probes) {
375 abort: tcp_write_err(sk);
376 } else {
377 /* Only send another probe if we didn't close things up. */
378 tcp_send_probe0(sk);
379 }
380 }
381
382 /*
383 * Timer for Fast Open socket to retransmit SYNACK. Note that the
384 * sk here is the child socket, not the parent (listener) socket.
385 */
tcp_fastopen_synack_timer(struct sock * sk)386 static void tcp_fastopen_synack_timer(struct sock *sk)
387 {
388 struct inet_connection_sock *icsk = inet_csk(sk);
389 int max_retries = icsk->icsk_syn_retries ? :
390 sock_net(sk)->ipv4.sysctl_tcp_synack_retries + 1; /* add one more retry for fastopen */
391 struct request_sock *req;
392
393 req = tcp_sk(sk)->fastopen_rsk;
394 req->rsk_ops->syn_ack_timeout(req);
395
396 if (req->num_timeout >= max_retries) {
397 tcp_write_err(sk);
398 return;
399 }
400 /* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error
401 * returned from rtx_syn_ack() to make it more persistent like
402 * regular retransmit because if the child socket has been accepted
403 * it's not good to give up too easily.
404 */
405 inet_rtx_syn_ack(sk, req);
406 req->num_timeout++;
407 icsk->icsk_retransmits++;
408 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
409 TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
410 }
411
412
413 /**
414 * tcp_retransmit_timer() - The TCP retransmit timeout handler
415 * @sk: Pointer to the current socket.
416 *
417 * This function gets called when the kernel timer for a TCP packet
418 * of this socket expires.
419 *
420 * It handles retransmission, timer adjustment and other necesarry measures.
421 *
422 * Returns: Nothing (void)
423 */
tcp_retransmit_timer(struct sock * sk)424 void tcp_retransmit_timer(struct sock *sk)
425 {
426 struct tcp_sock *tp = tcp_sk(sk);
427 struct net *net = sock_net(sk);
428 struct inet_connection_sock *icsk = inet_csk(sk);
429
430 if (tp->fastopen_rsk) {
431 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&
432 sk->sk_state != TCP_FIN_WAIT1);
433 tcp_fastopen_synack_timer(sk);
434 /* Before we receive ACK to our SYN-ACK don't retransmit
435 * anything else (e.g., data or FIN segments).
436 */
437 return;
438 }
439 if (!tp->packets_out)
440 goto out;
441
442 WARN_ON(tcp_write_queue_empty(sk));
443
444 tp->tlp_high_seq = 0;
445
446 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
447 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
448 /* Receiver dastardly shrinks window. Our retransmits
449 * become zero probes, but we should not timeout this
450 * connection. If the socket is an orphan, time it out,
451 * we cannot allow such beasts to hang infinitely.
452 */
453 struct inet_sock *inet = inet_sk(sk);
454 if (sk->sk_family == AF_INET) {
455 net_dbg_ratelimited("Peer %pI4:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
456 &inet->inet_daddr,
457 ntohs(inet->inet_dport),
458 inet->inet_num,
459 tp->snd_una, tp->snd_nxt);
460 }
461 #if IS_ENABLED(CONFIG_IPV6)
462 else if (sk->sk_family == AF_INET6) {
463 net_dbg_ratelimited("Peer %pI6:%u/%u unexpectedly shrunk window %u:%u (repaired)\n",
464 &sk->sk_v6_daddr,
465 ntohs(inet->inet_dport),
466 inet->inet_num,
467 tp->snd_una, tp->snd_nxt);
468 }
469 #endif
470 if (tcp_time_stamp - tp->rcv_tstamp > TCP_RTO_MAX) {
471 tcp_write_err(sk);
472 goto out;
473 }
474 tcp_enter_loss(sk);
475 tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1);
476 __sk_dst_reset(sk);
477 goto out_reset_timer;
478 }
479
480 if (tcp_write_timeout(sk))
481 goto out;
482
483 if (icsk->icsk_retransmits == 0) {
484 int mib_idx;
485
486 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
487 if (tcp_is_sack(tp))
488 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
489 else
490 mib_idx = LINUX_MIB_TCPRENORECOVERYFAIL;
491 } else if (icsk->icsk_ca_state == TCP_CA_Loss) {
492 mib_idx = LINUX_MIB_TCPLOSSFAILURES;
493 } else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||
494 tp->sacked_out) {
495 if (tcp_is_sack(tp))
496 mib_idx = LINUX_MIB_TCPSACKFAILURES;
497 else
498 mib_idx = LINUX_MIB_TCPRENOFAILURES;
499 } else {
500 mib_idx = LINUX_MIB_TCPTIMEOUTS;
501 }
502 __NET_INC_STATS(sock_net(sk), mib_idx);
503 }
504
505 tcp_enter_loss(sk);
506
507 if (tcp_retransmit_skb(sk, tcp_write_queue_head(sk), 1) > 0) {
508 /* Retransmission failed because of local congestion,
509 * do not backoff.
510 */
511 if (!icsk->icsk_retransmits)
512 icsk->icsk_retransmits = 1;
513 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
514 min(icsk->icsk_rto, TCP_RESOURCE_PROBE_INTERVAL),
515 TCP_RTO_MAX);
516 goto out;
517 }
518
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.
529 *
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!
533 */
534 icsk->icsk_backoff++;
535 icsk->icsk_retransmits++;
536
537 out_reset_timer:
538 /* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
539 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
540 * might be increased if the stream oscillates between thin and thick,
541 * thus the old value might already be too high compared to the value
542 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
543 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
544 * exponential backoff behaviour to avoid continue hammering
545 * linear-timeout retransmissions into a black hole
546 */
547 if (sk->sk_state == TCP_ESTABLISHED &&
548 (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
549 tcp_stream_is_thin(tp) &&
550 icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
551 icsk->icsk_backoff = 0;
552 icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
553 } else {
554 /* Use normal (exponential) backoff */
555 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
556 }
557 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
558 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1 + 1, 0, 0))
559 __sk_dst_reset(sk);
560
561 out:;
562 }
563
564 /* Called with bottom-half processing disabled.
565 Called by tcp_write_timer() */
tcp_write_timer_handler(struct sock * sk)566 void tcp_write_timer_handler(struct sock *sk)
567 {
568 struct inet_connection_sock *icsk = inet_csk(sk);
569 int event;
570
571 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
572 !icsk->icsk_pending)
573 goto out;
574
575 if (time_after(icsk->icsk_timeout, jiffies)) {
576 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);
577 goto out;
578 }
579
580 event = icsk->icsk_pending;
581
582 switch (event) {
583 case ICSK_TIME_EARLY_RETRANS:
584 tcp_resume_early_retransmit(sk);
585 break;
586 case ICSK_TIME_LOSS_PROBE:
587 tcp_send_loss_probe(sk);
588 break;
589 case ICSK_TIME_RETRANS:
590 icsk->icsk_pending = 0;
591 tcp_retransmit_timer(sk);
592 break;
593 case ICSK_TIME_PROBE0:
594 icsk->icsk_pending = 0;
595 tcp_probe_timer(sk);
596 break;
597 }
598
599 out:
600 sk_mem_reclaim(sk);
601 }
602
tcp_write_timer(unsigned long data)603 static void tcp_write_timer(unsigned long data)
604 {
605 struct sock *sk = (struct sock *)data;
606
607 bh_lock_sock(sk);
608 if (!sock_owned_by_user(sk)) {
609 tcp_write_timer_handler(sk);
610 } else {
611 /* delegate our work to tcp_release_cb() */
612 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &tcp_sk(sk)->tsq_flags))
613 sock_hold(sk);
614 }
615 bh_unlock_sock(sk);
616 sock_put(sk);
617 }
618
tcp_syn_ack_timeout(const struct request_sock * req)619 void tcp_syn_ack_timeout(const struct request_sock *req)
620 {
621 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
622
623 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
624 }
625 EXPORT_SYMBOL(tcp_syn_ack_timeout);
626
tcp_set_keepalive(struct sock * sk,int val)627 void tcp_set_keepalive(struct sock *sk, int val)
628 {
629 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
630 return;
631
632 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
633 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
634 else if (!val)
635 inet_csk_delete_keepalive_timer(sk);
636 }
637
638
tcp_keepalive_timer(unsigned long data)639 static void tcp_keepalive_timer (unsigned long data)
640 {
641 struct sock *sk = (struct sock *) data;
642 struct inet_connection_sock *icsk = inet_csk(sk);
643 struct tcp_sock *tp = tcp_sk(sk);
644 u32 elapsed;
645
646 /* Only process if socket is not in use. */
647 bh_lock_sock(sk);
648 if (sock_owned_by_user(sk)) {
649 /* Try again later. */
650 inet_csk_reset_keepalive_timer (sk, HZ/20);
651 goto out;
652 }
653
654 if (sk->sk_state == TCP_LISTEN) {
655 pr_err("Hmm... keepalive on a LISTEN ???\n");
656 goto out;
657 }
658
659 if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {
660 if (tp->linger2 >= 0) {
661 const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;
662
663 if (tmo > 0) {
664 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
665 goto out;
666 }
667 }
668 tcp_send_active_reset(sk, GFP_ATOMIC);
669 goto death;
670 }
671
672 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
673 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
674 goto out;
675
676 elapsed = keepalive_time_when(tp);
677
678 /* It is alive without keepalive 8) */
679 if (tp->packets_out || tcp_send_head(sk))
680 goto resched;
681
682 elapsed = keepalive_time_elapsed(tp);
683
684 if (elapsed >= keepalive_time_when(tp)) {
685 /* If the TCP_USER_TIMEOUT option is enabled, use that
686 * to determine when to timeout instead.
687 */
688 if ((icsk->icsk_user_timeout != 0 &&
689 elapsed >= icsk->icsk_user_timeout &&
690 icsk->icsk_probes_out > 0) ||
691 (icsk->icsk_user_timeout == 0 &&
692 icsk->icsk_probes_out >= keepalive_probes(tp))) {
693 tcp_send_active_reset(sk, GFP_ATOMIC);
694 tcp_write_err(sk);
695 goto out;
696 }
697 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
698 icsk->icsk_probes_out++;
699 elapsed = keepalive_intvl_when(tp);
700 } else {
701 /* If keepalive was lost due to local congestion,
702 * try harder.
703 */
704 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
705 }
706 } else {
707 /* It is tp->rcv_tstamp + keepalive_time_when(tp) */
708 elapsed = keepalive_time_when(tp) - elapsed;
709 }
710
711 sk_mem_reclaim(sk);
712
713 resched:
714 inet_csk_reset_keepalive_timer (sk, elapsed);
715 goto out;
716
717 death:
718 tcp_done(sk);
719
720 out:
721 bh_unlock_sock(sk);
722 sock_put(sk);
723 }
724
tcp_init_xmit_timers(struct sock * sk)725 void tcp_init_xmit_timers(struct sock *sk)
726 {
727 inet_csk_init_xmit_timers(sk, &tcp_write_timer, &tcp_delack_timer,
728 &tcp_keepalive_timer);
729 }
730