1 | /*------------------------------------------------------------------------- |
2 | * |
3 | * timeout.c |
4 | * Routines to multiplex SIGALRM interrupts for multiple timeout reasons. |
5 | * |
6 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
7 | * Portions Copyright (c) 1994, Regents of the University of California |
8 | * |
9 | * |
10 | * IDENTIFICATION |
11 | * src/backend/utils/misc/timeout.c |
12 | * |
13 | *------------------------------------------------------------------------- |
14 | */ |
15 | #include "postgres.h" |
16 | |
17 | #include <sys/time.h> |
18 | |
19 | #include "miscadmin.h" |
20 | #include "storage/proc.h" |
21 | #include "utils/timeout.h" |
22 | #include "utils/timestamp.h" |
23 | |
24 | |
25 | /* Data about any one timeout reason */ |
26 | typedef struct timeout_params |
27 | { |
28 | TimeoutId index; /* identifier of timeout reason */ |
29 | |
30 | /* volatile because it may be changed from the signal handler */ |
31 | volatile bool indicator; /* true if timeout has occurred */ |
32 | |
33 | /* callback function for timeout, or NULL if timeout not registered */ |
34 | timeout_handler_proc timeout_handler; |
35 | |
36 | TimestampTz start_time; /* time that timeout was last activated */ |
37 | TimestampTz fin_time; /* time it is, or was last, due to fire */ |
38 | } timeout_params; |
39 | |
40 | /* |
41 | * List of possible timeout reasons in the order of enum TimeoutId. |
42 | */ |
43 | static timeout_params all_timeouts[MAX_TIMEOUTS]; |
44 | static bool all_timeouts_initialized = false; |
45 | |
46 | /* |
47 | * List of active timeouts ordered by their fin_time and priority. |
48 | * This list is subject to change by the interrupt handler, so it's volatile. |
49 | */ |
50 | static volatile int num_active_timeouts = 0; |
51 | static timeout_params *volatile active_timeouts[MAX_TIMEOUTS]; |
52 | |
53 | /* |
54 | * Flag controlling whether the signal handler is allowed to do anything. |
55 | * We leave this "false" when we're not expecting interrupts, just in case. |
56 | * |
57 | * Note that we don't bother to reset any pending timer interrupt when we |
58 | * disable the signal handler; it's not really worth the cycles to do so, |
59 | * since the probability of the interrupt actually occurring while we have |
60 | * it disabled is low. See comments in schedule_alarm() about that. |
61 | */ |
62 | static volatile sig_atomic_t alarm_enabled = false; |
63 | |
64 | #define disable_alarm() (alarm_enabled = false) |
65 | #define enable_alarm() (alarm_enabled = true) |
66 | |
67 | |
68 | /***************************************************************************** |
69 | * Internal helper functions |
70 | * |
71 | * For all of these, it is caller's responsibility to protect them from |
72 | * interruption by the signal handler. Generally, call disable_alarm() |
73 | * first to prevent interruption, then update state, and last call |
74 | * schedule_alarm(), which will re-enable the signal handler if needed. |
75 | *****************************************************************************/ |
76 | |
77 | /* |
78 | * Find the index of a given timeout reason in the active array. |
79 | * If it's not there, return -1. |
80 | */ |
81 | static int |
82 | find_active_timeout(TimeoutId id) |
83 | { |
84 | int i; |
85 | |
86 | for (i = 0; i < num_active_timeouts; i++) |
87 | { |
88 | if (active_timeouts[i]->index == id) |
89 | return i; |
90 | } |
91 | |
92 | return -1; |
93 | } |
94 | |
95 | /* |
96 | * Insert specified timeout reason into the list of active timeouts |
97 | * at the given index. |
98 | */ |
99 | static void |
100 | insert_timeout(TimeoutId id, int index) |
101 | { |
102 | int i; |
103 | |
104 | if (index < 0 || index > num_active_timeouts) |
105 | elog(FATAL, "timeout index %d out of range 0..%d" , index, |
106 | num_active_timeouts); |
107 | |
108 | for (i = num_active_timeouts - 1; i >= index; i--) |
109 | active_timeouts[i + 1] = active_timeouts[i]; |
110 | |
111 | active_timeouts[index] = &all_timeouts[id]; |
112 | |
113 | num_active_timeouts++; |
114 | } |
115 | |
116 | /* |
117 | * Remove the index'th element from the timeout list. |
118 | */ |
119 | static void |
120 | remove_timeout_index(int index) |
121 | { |
122 | int i; |
123 | |
124 | if (index < 0 || index >= num_active_timeouts) |
125 | elog(FATAL, "timeout index %d out of range 0..%d" , index, |
126 | num_active_timeouts - 1); |
127 | |
128 | for (i = index + 1; i < num_active_timeouts; i++) |
129 | active_timeouts[i - 1] = active_timeouts[i]; |
130 | |
131 | num_active_timeouts--; |
132 | } |
133 | |
134 | /* |
135 | * Enable the specified timeout reason |
136 | */ |
137 | static void |
138 | enable_timeout(TimeoutId id, TimestampTz now, TimestampTz fin_time) |
139 | { |
140 | int i; |
141 | |
142 | /* Assert request is sane */ |
143 | Assert(all_timeouts_initialized); |
144 | Assert(all_timeouts[id].timeout_handler != NULL); |
145 | |
146 | /* |
147 | * If this timeout was already active, momentarily disable it. We |
148 | * interpret the call as a directive to reschedule the timeout. |
149 | */ |
150 | i = find_active_timeout(id); |
151 | if (i >= 0) |
152 | remove_timeout_index(i); |
153 | |
154 | /* |
155 | * Find out the index where to insert the new timeout. We sort by |
156 | * fin_time, and for equal fin_time by priority. |
157 | */ |
158 | for (i = 0; i < num_active_timeouts; i++) |
159 | { |
160 | timeout_params *old_timeout = active_timeouts[i]; |
161 | |
162 | if (fin_time < old_timeout->fin_time) |
163 | break; |
164 | if (fin_time == old_timeout->fin_time && id < old_timeout->index) |
165 | break; |
166 | } |
167 | |
168 | /* |
169 | * Mark the timeout active, and insert it into the active list. |
170 | */ |
171 | all_timeouts[id].indicator = false; |
172 | all_timeouts[id].start_time = now; |
173 | all_timeouts[id].fin_time = fin_time; |
174 | |
175 | insert_timeout(id, i); |
176 | } |
177 | |
178 | /* |
179 | * Schedule alarm for the next active timeout, if any |
180 | * |
181 | * We assume the caller has obtained the current time, or a close-enough |
182 | * approximation. |
183 | */ |
184 | static void |
185 | schedule_alarm(TimestampTz now) |
186 | { |
187 | if (num_active_timeouts > 0) |
188 | { |
189 | struct itimerval timeval; |
190 | long secs; |
191 | int usecs; |
192 | |
193 | MemSet(&timeval, 0, sizeof(struct itimerval)); |
194 | |
195 | /* Get the time remaining till the nearest pending timeout */ |
196 | TimestampDifference(now, active_timeouts[0]->fin_time, |
197 | &secs, &usecs); |
198 | |
199 | /* |
200 | * It's possible that the difference is less than a microsecond; |
201 | * ensure we don't cancel, rather than set, the interrupt. |
202 | */ |
203 | if (secs == 0 && usecs == 0) |
204 | usecs = 1; |
205 | |
206 | timeval.it_value.tv_sec = secs; |
207 | timeval.it_value.tv_usec = usecs; |
208 | |
209 | /* |
210 | * We must enable the signal handler before calling setitimer(); if we |
211 | * did it in the other order, we'd have a race condition wherein the |
212 | * interrupt could occur before we can set alarm_enabled, so that the |
213 | * signal handler would fail to do anything. |
214 | * |
215 | * Because we didn't bother to reset the timer in disable_alarm(), |
216 | * it's possible that a previously-set interrupt will fire between |
217 | * enable_alarm() and setitimer(). This is safe, however. There are |
218 | * two possible outcomes: |
219 | * |
220 | * 1. The signal handler finds nothing to do (because the nearest |
221 | * timeout event is still in the future). It will re-set the timer |
222 | * and return. Then we'll overwrite the timer value with a new one. |
223 | * This will mean that the timer fires a little later than we |
224 | * intended, but only by the amount of time it takes for the signal |
225 | * handler to do nothing useful, which shouldn't be much. |
226 | * |
227 | * 2. The signal handler executes and removes one or more timeout |
228 | * events. When it returns, either the queue is now empty or the |
229 | * frontmost event is later than the one we looked at above. So we'll |
230 | * overwrite the timer value with one that is too soon (plus or minus |
231 | * the signal handler's execution time), causing a useless interrupt |
232 | * to occur. But the handler will then re-set the timer and |
233 | * everything will still work as expected. |
234 | * |
235 | * Since these cases are of very low probability (the window here |
236 | * being quite narrow), it's not worth adding cycles to the mainline |
237 | * code to prevent occasional wasted interrupts. |
238 | */ |
239 | enable_alarm(); |
240 | |
241 | /* Set the alarm timer */ |
242 | if (setitimer(ITIMER_REAL, &timeval, NULL) != 0) |
243 | elog(FATAL, "could not enable SIGALRM timer: %m" ); |
244 | } |
245 | } |
246 | |
247 | |
248 | /***************************************************************************** |
249 | * Signal handler |
250 | *****************************************************************************/ |
251 | |
252 | /* |
253 | * Signal handler for SIGALRM |
254 | * |
255 | * Process any active timeout reasons and then reschedule the interrupt |
256 | * as needed. |
257 | */ |
258 | static void |
259 | handle_sig_alarm(SIGNAL_ARGS) |
260 | { |
261 | int save_errno = errno; |
262 | |
263 | /* |
264 | * Bump the holdoff counter, to make sure nothing we call will process |
265 | * interrupts directly. No timeout handler should do that, but these |
266 | * failures are hard to debug, so better be sure. |
267 | */ |
268 | HOLD_INTERRUPTS(); |
269 | |
270 | /* |
271 | * SIGALRM is always cause for waking anything waiting on the process |
272 | * latch. |
273 | */ |
274 | SetLatch(MyLatch); |
275 | |
276 | /* |
277 | * Fire any pending timeouts, but only if we're enabled to do so. |
278 | */ |
279 | if (alarm_enabled) |
280 | { |
281 | /* |
282 | * Disable alarms, just in case this platform allows signal handlers |
283 | * to interrupt themselves. schedule_alarm() will re-enable if |
284 | * appropriate. |
285 | */ |
286 | disable_alarm(); |
287 | |
288 | if (num_active_timeouts > 0) |
289 | { |
290 | TimestampTz now = GetCurrentTimestamp(); |
291 | |
292 | /* While the first pending timeout has been reached ... */ |
293 | while (num_active_timeouts > 0 && |
294 | now >= active_timeouts[0]->fin_time) |
295 | { |
296 | timeout_params *this_timeout = active_timeouts[0]; |
297 | |
298 | /* Remove it from the active list */ |
299 | remove_timeout_index(0); |
300 | |
301 | /* Mark it as fired */ |
302 | this_timeout->indicator = true; |
303 | |
304 | /* And call its handler function */ |
305 | this_timeout->timeout_handler(); |
306 | |
307 | /* |
308 | * The handler might not take negligible time (CheckDeadLock |
309 | * for instance isn't too cheap), so let's update our idea of |
310 | * "now" after each one. |
311 | */ |
312 | now = GetCurrentTimestamp(); |
313 | } |
314 | |
315 | /* Done firing timeouts, so reschedule next interrupt if any */ |
316 | schedule_alarm(now); |
317 | } |
318 | } |
319 | |
320 | RESUME_INTERRUPTS(); |
321 | |
322 | errno = save_errno; |
323 | } |
324 | |
325 | |
326 | /***************************************************************************** |
327 | * Public API |
328 | *****************************************************************************/ |
329 | |
330 | /* |
331 | * Initialize timeout module. |
332 | * |
333 | * This must be called in every process that wants to use timeouts. |
334 | * |
335 | * If the process was forked from another one that was also using this |
336 | * module, be sure to call this before re-enabling signals; else handlers |
337 | * meant to run in the parent process might get invoked in this one. |
338 | */ |
339 | void |
340 | InitializeTimeouts(void) |
341 | { |
342 | int i; |
343 | |
344 | /* Initialize, or re-initialize, all local state */ |
345 | disable_alarm(); |
346 | |
347 | num_active_timeouts = 0; |
348 | |
349 | for (i = 0; i < MAX_TIMEOUTS; i++) |
350 | { |
351 | all_timeouts[i].index = i; |
352 | all_timeouts[i].indicator = false; |
353 | all_timeouts[i].timeout_handler = NULL; |
354 | all_timeouts[i].start_time = 0; |
355 | all_timeouts[i].fin_time = 0; |
356 | } |
357 | |
358 | all_timeouts_initialized = true; |
359 | |
360 | /* Now establish the signal handler */ |
361 | pqsignal(SIGALRM, handle_sig_alarm); |
362 | } |
363 | |
364 | /* |
365 | * Register a timeout reason |
366 | * |
367 | * For predefined timeouts, this just registers the callback function. |
368 | * |
369 | * For user-defined timeouts, pass id == USER_TIMEOUT; we then allocate and |
370 | * return a timeout ID. |
371 | */ |
372 | TimeoutId |
373 | RegisterTimeout(TimeoutId id, timeout_handler_proc handler) |
374 | { |
375 | Assert(all_timeouts_initialized); |
376 | |
377 | /* There's no need to disable the signal handler here. */ |
378 | |
379 | if (id >= USER_TIMEOUT) |
380 | { |
381 | /* Allocate a user-defined timeout reason */ |
382 | for (id = USER_TIMEOUT; id < MAX_TIMEOUTS; id++) |
383 | if (all_timeouts[id].timeout_handler == NULL) |
384 | break; |
385 | if (id >= MAX_TIMEOUTS) |
386 | ereport(FATAL, |
387 | (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED), |
388 | errmsg("cannot add more timeout reasons" ))); |
389 | } |
390 | |
391 | Assert(all_timeouts[id].timeout_handler == NULL); |
392 | |
393 | all_timeouts[id].timeout_handler = handler; |
394 | |
395 | return id; |
396 | } |
397 | |
398 | /* |
399 | * Reschedule any pending SIGALRM interrupt. |
400 | * |
401 | * This can be used during error recovery in case query cancel resulted in loss |
402 | * of a SIGALRM event (due to longjmp'ing out of handle_sig_alarm before it |
403 | * could do anything). But note it's not necessary if any of the public |
404 | * enable_ or disable_timeout functions are called in the same area, since |
405 | * those all do schedule_alarm() internally if needed. |
406 | */ |
407 | void |
408 | reschedule_timeouts(void) |
409 | { |
410 | /* For flexibility, allow this to be called before we're initialized. */ |
411 | if (!all_timeouts_initialized) |
412 | return; |
413 | |
414 | /* Disable timeout interrupts for safety. */ |
415 | disable_alarm(); |
416 | |
417 | /* Reschedule the interrupt, if any timeouts remain active. */ |
418 | if (num_active_timeouts > 0) |
419 | schedule_alarm(GetCurrentTimestamp()); |
420 | } |
421 | |
422 | /* |
423 | * Enable the specified timeout to fire after the specified delay. |
424 | * |
425 | * Delay is given in milliseconds. |
426 | */ |
427 | void |
428 | enable_timeout_after(TimeoutId id, int delay_ms) |
429 | { |
430 | TimestampTz now; |
431 | TimestampTz fin_time; |
432 | |
433 | /* Disable timeout interrupts for safety. */ |
434 | disable_alarm(); |
435 | |
436 | /* Queue the timeout at the appropriate time. */ |
437 | now = GetCurrentTimestamp(); |
438 | fin_time = TimestampTzPlusMilliseconds(now, delay_ms); |
439 | enable_timeout(id, now, fin_time); |
440 | |
441 | /* Set the timer interrupt. */ |
442 | schedule_alarm(now); |
443 | } |
444 | |
445 | /* |
446 | * Enable the specified timeout to fire at the specified time. |
447 | * |
448 | * This is provided to support cases where there's a reason to calculate |
449 | * the timeout by reference to some point other than "now". If there isn't, |
450 | * use enable_timeout_after(), to avoid calling GetCurrentTimestamp() twice. |
451 | */ |
452 | void |
453 | enable_timeout_at(TimeoutId id, TimestampTz fin_time) |
454 | { |
455 | TimestampTz now; |
456 | |
457 | /* Disable timeout interrupts for safety. */ |
458 | disable_alarm(); |
459 | |
460 | /* Queue the timeout at the appropriate time. */ |
461 | now = GetCurrentTimestamp(); |
462 | enable_timeout(id, now, fin_time); |
463 | |
464 | /* Set the timer interrupt. */ |
465 | schedule_alarm(now); |
466 | } |
467 | |
468 | /* |
469 | * Enable multiple timeouts at once. |
470 | * |
471 | * This works like calling enable_timeout_after() and/or enable_timeout_at() |
472 | * multiple times. Use this to reduce the number of GetCurrentTimestamp() |
473 | * and setitimer() calls needed to establish multiple timeouts. |
474 | */ |
475 | void |
476 | enable_timeouts(const EnableTimeoutParams *timeouts, int count) |
477 | { |
478 | TimestampTz now; |
479 | int i; |
480 | |
481 | /* Disable timeout interrupts for safety. */ |
482 | disable_alarm(); |
483 | |
484 | /* Queue the timeout(s) at the appropriate times. */ |
485 | now = GetCurrentTimestamp(); |
486 | |
487 | for (i = 0; i < count; i++) |
488 | { |
489 | TimeoutId id = timeouts[i].id; |
490 | TimestampTz fin_time; |
491 | |
492 | switch (timeouts[i].type) |
493 | { |
494 | case TMPARAM_AFTER: |
495 | fin_time = TimestampTzPlusMilliseconds(now, |
496 | timeouts[i].delay_ms); |
497 | enable_timeout(id, now, fin_time); |
498 | break; |
499 | |
500 | case TMPARAM_AT: |
501 | enable_timeout(id, now, timeouts[i].fin_time); |
502 | break; |
503 | |
504 | default: |
505 | elog(ERROR, "unrecognized timeout type %d" , |
506 | (int) timeouts[i].type); |
507 | break; |
508 | } |
509 | } |
510 | |
511 | /* Set the timer interrupt. */ |
512 | schedule_alarm(now); |
513 | } |
514 | |
515 | /* |
516 | * Cancel the specified timeout. |
517 | * |
518 | * The timeout's I've-been-fired indicator is reset, |
519 | * unless keep_indicator is true. |
520 | * |
521 | * When a timeout is canceled, any other active timeout remains in force. |
522 | * It's not an error to disable a timeout that is not enabled. |
523 | */ |
524 | void |
525 | disable_timeout(TimeoutId id, bool keep_indicator) |
526 | { |
527 | int i; |
528 | |
529 | /* Assert request is sane */ |
530 | Assert(all_timeouts_initialized); |
531 | Assert(all_timeouts[id].timeout_handler != NULL); |
532 | |
533 | /* Disable timeout interrupts for safety. */ |
534 | disable_alarm(); |
535 | |
536 | /* Find the timeout and remove it from the active list. */ |
537 | i = find_active_timeout(id); |
538 | if (i >= 0) |
539 | remove_timeout_index(i); |
540 | |
541 | /* Mark it inactive, whether it was active or not. */ |
542 | if (!keep_indicator) |
543 | all_timeouts[id].indicator = false; |
544 | |
545 | /* Reschedule the interrupt, if any timeouts remain active. */ |
546 | if (num_active_timeouts > 0) |
547 | schedule_alarm(GetCurrentTimestamp()); |
548 | } |
549 | |
550 | /* |
551 | * Cancel multiple timeouts at once. |
552 | * |
553 | * The timeouts' I've-been-fired indicators are reset, |
554 | * unless timeouts[i].keep_indicator is true. |
555 | * |
556 | * This works like calling disable_timeout() multiple times. |
557 | * Use this to reduce the number of GetCurrentTimestamp() |
558 | * and setitimer() calls needed to cancel multiple timeouts. |
559 | */ |
560 | void |
561 | disable_timeouts(const DisableTimeoutParams *timeouts, int count) |
562 | { |
563 | int i; |
564 | |
565 | Assert(all_timeouts_initialized); |
566 | |
567 | /* Disable timeout interrupts for safety. */ |
568 | disable_alarm(); |
569 | |
570 | /* Cancel the timeout(s). */ |
571 | for (i = 0; i < count; i++) |
572 | { |
573 | TimeoutId id = timeouts[i].id; |
574 | int idx; |
575 | |
576 | Assert(all_timeouts[id].timeout_handler != NULL); |
577 | |
578 | idx = find_active_timeout(id); |
579 | if (idx >= 0) |
580 | remove_timeout_index(idx); |
581 | |
582 | if (!timeouts[i].keep_indicator) |
583 | all_timeouts[id].indicator = false; |
584 | } |
585 | |
586 | /* Reschedule the interrupt, if any timeouts remain active. */ |
587 | if (num_active_timeouts > 0) |
588 | schedule_alarm(GetCurrentTimestamp()); |
589 | } |
590 | |
591 | /* |
592 | * Disable SIGALRM and remove all timeouts from the active list, |
593 | * and optionally reset their timeout indicators. |
594 | */ |
595 | void |
596 | disable_all_timeouts(bool keep_indicators) |
597 | { |
598 | disable_alarm(); |
599 | |
600 | /* |
601 | * Only bother to reset the timer if we think it's active. We could just |
602 | * let the interrupt happen anyway, but it's probably a bit cheaper to do |
603 | * setitimer() than to let the useless interrupt happen. |
604 | */ |
605 | if (num_active_timeouts > 0) |
606 | { |
607 | struct itimerval timeval; |
608 | |
609 | MemSet(&timeval, 0, sizeof(struct itimerval)); |
610 | if (setitimer(ITIMER_REAL, &timeval, NULL) != 0) |
611 | elog(FATAL, "could not disable SIGALRM timer: %m" ); |
612 | } |
613 | |
614 | num_active_timeouts = 0; |
615 | |
616 | if (!keep_indicators) |
617 | { |
618 | int i; |
619 | |
620 | for (i = 0; i < MAX_TIMEOUTS; i++) |
621 | all_timeouts[i].indicator = false; |
622 | } |
623 | } |
624 | |
625 | /* |
626 | * Return the timeout's I've-been-fired indicator |
627 | * |
628 | * If reset_indicator is true, reset the indicator when returning true. |
629 | * To avoid missing timeouts due to race conditions, we are careful not to |
630 | * reset the indicator when returning false. |
631 | */ |
632 | bool |
633 | get_timeout_indicator(TimeoutId id, bool reset_indicator) |
634 | { |
635 | if (all_timeouts[id].indicator) |
636 | { |
637 | if (reset_indicator) |
638 | all_timeouts[id].indicator = false; |
639 | return true; |
640 | } |
641 | return false; |
642 | } |
643 | |
644 | /* |
645 | * Return the time when the timeout was most recently activated |
646 | * |
647 | * Note: will return 0 if timeout has never been activated in this process. |
648 | * However, we do *not* reset the start_time when a timeout occurs, so as |
649 | * not to create a race condition if SIGALRM fires just as some code is |
650 | * about to fetch the value. |
651 | */ |
652 | TimestampTz |
653 | get_timeout_start_time(TimeoutId id) |
654 | { |
655 | return all_timeouts[id].start_time; |
656 | } |
657 | |
658 | /* |
659 | * Return the time when the timeout is, or most recently was, due to fire |
660 | * |
661 | * Note: will return 0 if timeout has never been activated in this process. |
662 | * However, we do *not* reset the fin_time when a timeout occurs, so as |
663 | * not to create a race condition if SIGALRM fires just as some code is |
664 | * about to fetch the value. |
665 | */ |
666 | TimestampTz |
667 | get_timeout_finish_time(TimeoutId id) |
668 | { |
669 | return all_timeouts[id].fin_time; |
670 | } |
671 | |