1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22#include <linux/module.h>
23#include <linux/gfp.h>
24#include <net/tcp.h>
25
26static u32 tcp_clamp_rto_to_user_timeout(const struct sock *sk)
27{
28 struct inet_connection_sock *icsk = inet_csk(sk);
29 u32 elapsed, start_ts;
30 s32 remaining;
31
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;
37 if (remaining <= 0)
38 return 1;
39
40 return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(remaining));
41}
42
43
44
45
46
47
48
49
50static void tcp_write_err(struct sock *sk)
51{
52 sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
53 sk->sk_error_report(sk);
54
55 tcp_write_queue_purge(sk);
56 tcp_done(sk);
57 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONTIMEOUT);
58}
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84static int tcp_out_of_resources(struct sock *sk, bool do_reset)
85{
86 struct tcp_sock *tp = tcp_sk(sk);
87 int shift = 0;
88
89
90
91 if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)
92 shift++;
93
94
95 if (sk->sk_err_soft)
96 shift++;
97
98 if (tcp_check_oom(sk, shift)) {
99
100
101 if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||
102
103 (!tp->snd_wnd && !tp->packets_out))
104 do_reset = true;
105 if (do_reset)
106 tcp_send_active_reset(sk, GFP_ATOMIC);
107 tcp_done(sk);
108 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTONMEMORY);
109 return 1;
110 }
111
112 if (!check_net(sock_net(sk))) {
113
114 tcp_done(sk);
115 return 1;
116 }
117
118 return 0;
119}
120
121
122
123
124
125
126static int tcp_orphan_retries(struct sock *sk, bool alive)
127{
128 int retries = sock_net(sk)->ipv4.sysctl_tcp_orphan_retries;
129
130
131 if (sk->sk_err_soft && !alive)
132 retries = 0;
133
134
135
136
137 if (retries == 0 && alive)
138 retries = 8;
139 return retries;
140}
141
142static void tcp_mtu_probing(struct inet_connection_sock *icsk, struct sock *sk)
143{
144 const struct net *net = sock_net(sk);
145 int mss;
146
147
148 if (!net->ipv4.sysctl_tcp_mtu_probing)
149 return;
150
151 if (!icsk->icsk_mtup.enabled) {
152 icsk->icsk_mtup.enabled = 1;
153 icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;
154 } else {
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);
160 }
161 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);
162}
163
164static unsigned int tcp_model_timeout(struct sock *sk,
165 unsigned int boundary,
166 unsigned int rto_base)
167{
168 unsigned int linear_backoff_thresh, timeout;
169
170 linear_backoff_thresh = ilog2(TCP_RTO_MAX / rto_base);
171 if (boundary <= linear_backoff_thresh)
172 timeout = ((2 << boundary) - 1) * rto_base;
173 else
174 timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +
175 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
176 return jiffies_to_msecs(timeout);
177}
178
179
180
181
182
183
184
185
186
187
188
189
190
191static bool retransmits_timed_out(struct sock *sk,
192 unsigned int boundary,
193 unsigned int timeout)
194{
195 unsigned int start_ts;
196
197 if (!inet_csk(sk)->icsk_retransmits)
198 return false;
199
200 start_ts = tcp_sk(sk)->retrans_stamp;
201 if (likely(timeout == 0))
202 timeout = tcp_model_timeout(sk, boundary, TCP_RTO_MIN);
203
204 return (s32)(tcp_time_stamp(tcp_sk(sk)) - start_ts - timeout) >= 0;
205}
206
207
208static int tcp_write_timeout(struct sock *sk)
209{
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;
214 int retry_until;
215
216 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
217 if (icsk->icsk_retransmits) {
218 dst_negative_advice(sk);
219 } else {
220 sk_rethink_txhash(sk);
221 }
222 retry_until = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_syn_retries;
223 expired = icsk->icsk_retransmits >= retry_until;
224 } else {
225 if (retransmits_timed_out(sk, net->ipv4.sysctl_tcp_retries1, 0)) {
226
227 tcp_mtu_probing(icsk, sk);
228
229 dst_negative_advice(sk);
230 } else {
231 sk_rethink_txhash(sk);
232 }
233
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;
237
238 retry_until = tcp_orphan_retries(sk, alive);
239 do_reset = alive ||
240 !retransmits_timed_out(sk, retry_until, 0);
241
242 if (tcp_out_of_resources(sk, do_reset))
243 return 1;
244 }
245 expired = retransmits_timed_out(sk, retry_until,
246 icsk->icsk_user_timeout);
247 }
248 tcp_fastopen_active_detect_blackhole(sk, expired);
249
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);
254
255 if (expired) {
256
257 tcp_write_err(sk);
258 return 1;
259 }
260
261 return 0;
262}
263
264
265void tcp_delack_timer_handler(struct sock *sk)
266{
267 struct inet_connection_sock *icsk = inet_csk(sk);
268
269 sk_mem_reclaim_partial(sk);
270
271 if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||
272 !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
273 goto out;
274
275 if (time_after(icsk->icsk_ack.timeout, jiffies)) {
276 sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);
277 goto out;
278 }
279 icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
280
281 if (inet_csk_ack_scheduled(sk)) {
282 if (!inet_csk_in_pingpong_mode(sk)) {
283
284 icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1, icsk->icsk_rto);
285 } else {
286
287
288
289 inet_csk_exit_pingpong_mode(sk);
290 icsk->icsk_ack.ato = TCP_ATO_MIN;
291 }
292 tcp_mstamp_refresh(tcp_sk(sk));
293 tcp_send_ack(sk);
294 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKS);
295 }
296
297out:
298 if (tcp_under_memory_pressure(sk))
299 sk_mem_reclaim(sk);
300}
301
302
303
304
305
306
307
308
309
310
311
312static void tcp_delack_timer(struct timer_list *t)
313{
314 struct inet_connection_sock *icsk =
315 from_timer(icsk, t, icsk_delack_timer);
316 struct sock *sk = &icsk->icsk_inet.sk;
317
318 bh_lock_sock(sk);
319 if (!sock_owned_by_user(sk)) {
320 tcp_delack_timer_handler(sk);
321 } else {
322 icsk->icsk_ack.blocked = 1;
323 __NET_INC_STATS(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
324
325 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags))
326 sock_hold(sk);
327 }
328 bh_unlock_sock(sk);
329 sock_put(sk);
330}
331
332static void tcp_probe_timer(struct sock *sk)
333{
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);
337 int max_probes;
338
339 if (tp->packets_out || !skb) {
340 icsk->icsk_probes_out = 0;
341 return;
342 }
343
344
345
346
347
348
349
350
351
352 if (icsk->icsk_user_timeout) {
353 u32 elapsed = tcp_model_timeout(sk, icsk->icsk_probes_out,
354 tcp_probe0_base(sk));
355
356 if (elapsed >= icsk->icsk_user_timeout)
357 goto abort;
358 }
359
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;
363
364 max_probes = tcp_orphan_retries(sk, alive);
365 if (!alive && icsk->icsk_backoff >= max_probes)
366 goto abort;
367 if (tcp_out_of_resources(sk, true))
368 return;
369 }
370
371 if (icsk->icsk_probes_out >= max_probes) {
372abort: tcp_write_err(sk);
373 } else {
374
375 tcp_send_probe0(sk);
376 }
377}
378
379
380
381
382
383static void tcp_fastopen_synack_timer(struct sock *sk)
384{
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;
388 struct tcp_sock *tp = tcp_sk(sk);
389 struct request_sock *req;
390
391 req = tcp_sk(sk)->fastopen_rsk;
392 req->rsk_ops->syn_ack_timeout(req);
393
394 if (req->num_timeout >= max_retries) {
395 tcp_write_err(sk);
396 return;
397 }
398
399 if (icsk->icsk_retransmits == 1)
400 tcp_enter_loss(sk);
401
402
403
404
405
406 inet_rtx_syn_ack(sk, req);
407 req->num_timeout++;
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);
413}
414
415
416
417
418
419
420
421
422
423
424
425
426
427void tcp_retransmit_timer(struct sock *sk)
428{
429 struct tcp_sock *tp = tcp_sk(sk);
430 struct net *net = sock_net(sk);
431 struct inet_connection_sock *icsk = inet_csk(sk);
432
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
438
439
440 return;
441 }
442 if (!tp->packets_out || WARN_ON_ONCE(tcp_rtx_queue_empty(sk)))
443 return;
444
445 tp->tlp_high_seq = 0;
446
447 if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&
448 !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {
449
450
451
452
453
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",
457 &inet->inet_daddr,
458 ntohs(inet->inet_dport),
459 inet->inet_num,
460 tp->snd_una, tp->snd_nxt);
461 }
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",
465 &sk->sk_v6_daddr,
466 ntohs(inet->inet_dport),
467 inet->inet_num,
468 tp->snd_una, tp->snd_nxt);
469 }
470#endif
471 if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) {
472 tcp_write_err(sk);
473 goto out;
474 }
475 tcp_enter_loss(sk);
476 tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1);
477 __sk_dst_reset(sk);
478 goto out_reset_timer;
479 }
480
481 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEOUTS);
482 if (tcp_write_timeout(sk))
483 goto out;
484
485 if (icsk->icsk_retransmits == 0) {
486 int mib_idx = 0;
487
488 if (icsk->icsk_ca_state == TCP_CA_Recovery) {
489 if (tcp_is_sack(tp))
490 mib_idx = LINUX_MIB_TCPSACKRECOVERYFAIL;
491 else
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) ||
496 tp->sacked_out) {
497 if (tcp_is_sack(tp))
498 mib_idx = LINUX_MIB_TCPSACKFAILURES;
499 else
500 mib_idx = LINUX_MIB_TCPRENOFAILURES;
501 }
502 if (mib_idx)
503 __NET_INC_STATS(sock_net(sk), mib_idx);
504 }
505
506 tcp_enter_loss(sk);
507
508 icsk->icsk_retransmits++;
509 if (tcp_retransmit_skb(sk, tcp_rtx_queue_head(sk), 1) > 0) {
510
511
512
513 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
514 TCP_RESOURCE_PROBE_INTERVAL,
515 TCP_RTO_MAX);
516 goto out;
517 }
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534 icsk->icsk_backoff++;
535
536out_reset_timer:
537
538
539
540
541
542
543
544
545
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);
552 } else {
553
554 icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
555 }
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))
559 __sk_dst_reset(sk);
560
561out:;
562}
563
564
565
566void 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 tcp_mstamp_refresh(tcp_sk(sk));
581 event = icsk->icsk_pending;
582
583 switch (event) {
584 case ICSK_TIME_REO_TIMEOUT:
585 tcp_rack_reo_timeout(sk);
586 break;
587 case ICSK_TIME_LOSS_PROBE:
588 tcp_send_loss_probe(sk);
589 break;
590 case ICSK_TIME_RETRANS:
591 icsk->icsk_pending = 0;
592 tcp_retransmit_timer(sk);
593 break;
594 case ICSK_TIME_PROBE0:
595 icsk->icsk_pending = 0;
596 tcp_probe_timer(sk);
597 break;
598 }
599
600out:
601 sk_mem_reclaim(sk);
602}
603
604static void tcp_write_timer(struct timer_list *t)
605{
606 struct inet_connection_sock *icsk =
607 from_timer(icsk, t, icsk_retransmit_timer);
608 struct sock *sk = &icsk->icsk_inet.sk;
609
610 bh_lock_sock(sk);
611 if (!sock_owned_by_user(sk)) {
612 tcp_write_timer_handler(sk);
613 } else {
614
615 if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags))
616 sock_hold(sk);
617 }
618 bh_unlock_sock(sk);
619 sock_put(sk);
620}
621
622void tcp_syn_ack_timeout(const struct request_sock *req)
623{
624 struct net *net = read_pnet(&inet_rsk(req)->ireq_net);
625
626 __NET_INC_STATS(net, LINUX_MIB_TCPTIMEOUTS);
627}
628EXPORT_SYMBOL(tcp_syn_ack_timeout);
629
630void tcp_set_keepalive(struct sock *sk, int val)
631{
632 if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))
633 return;
634
635 if (val && !sock_flag(sk, SOCK_KEEPOPEN))
636 inet_csk_reset_keepalive_timer(sk, keepalive_time_when(tcp_sk(sk)));
637 else if (!val)
638 inet_csk_delete_keepalive_timer(sk);
639}
640EXPORT_SYMBOL_GPL(tcp_set_keepalive);
641
642
643static void tcp_keepalive_timer (struct timer_list *t)
644{
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);
648 u32 elapsed;
649
650
651 bh_lock_sock(sk);
652 if (sock_owned_by_user(sk)) {
653
654 inet_csk_reset_keepalive_timer (sk, HZ/20);
655 goto out;
656 }
657
658 if (sk->sk_state == TCP_LISTEN) {
659 pr_err("Hmm... keepalive on a LISTEN ???\n");
660 goto out;
661 }
662
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;
667
668 if (tmo > 0) {
669 tcp_time_wait(sk, TCP_FIN_WAIT2, tmo);
670 goto out;
671 }
672 }
673 tcp_send_active_reset(sk, GFP_ATOMIC);
674 goto death;
675 }
676
677 if (!sock_flag(sk, SOCK_KEEPOPEN) ||
678 ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))
679 goto out;
680
681 elapsed = keepalive_time_when(tp);
682
683
684 if (tp->packets_out || !tcp_write_queue_empty(sk))
685 goto resched;
686
687 elapsed = keepalive_time_elapsed(tp);
688
689 if (elapsed >= keepalive_time_when(tp)) {
690
691
692
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);
699 tcp_write_err(sk);
700 goto out;
701 }
702 if (tcp_write_wakeup(sk, LINUX_MIB_TCPKEEPALIVE) <= 0) {
703 icsk->icsk_probes_out++;
704 elapsed = keepalive_intvl_when(tp);
705 } else {
706
707
708
709 elapsed = TCP_RESOURCE_PROBE_INTERVAL;
710 }
711 } else {
712
713 elapsed = keepalive_time_when(tp) - elapsed;
714 }
715
716 sk_mem_reclaim(sk);
717
718resched:
719 inet_csk_reset_keepalive_timer (sk, elapsed);
720 goto out;
721
722death:
723 tcp_done(sk);
724
725out:
726 bh_unlock_sock(sk);
727 sock_put(sk);
728}
729
730static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
731{
732 struct tcp_sock *tp = container_of(timer, struct tcp_sock, compressed_ack_timer);
733 struct sock *sk = (struct sock *)tp;
734
735 bh_lock_sock(sk);
736 if (!sock_owned_by_user(sk)) {
737 if (tp->compressed_ack > TCP_FASTRETRANS_THRESH)
738 tcp_send_ack(sk);
739 } else {
740 if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
741 &sk->sk_tsq_flags))
742 sock_hold(sk);
743 }
744 bh_unlock_sock(sk);
745
746 sock_put(sk);
747
748 return HRTIMER_NORESTART;
749}
750
751void tcp_init_xmit_timers(struct sock *sk)
752{
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;
758
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;
762}
763