Line data Source code
1 : /*--------------------------------------------------------------------
2 : * bgworker.c
3 : * POSTGRES pluggable background workers implementation
4 : *
5 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
6 : *
7 : * IDENTIFICATION
8 : * src/backend/postmaster/bgworker.c
9 : *
10 : *-------------------------------------------------------------------------
11 : */
12 :
13 : #include "postgres.h"
14 :
15 : #include "access/parallel.h"
16 : #include "libpq/pqsignal.h"
17 : #include "miscadmin.h"
18 : #include "pgstat.h"
19 : #include "port/atomics.h"
20 : #include "postmaster/bgworker_internals.h"
21 : #include "postmaster/postmaster.h"
22 : #include "replication/logicallauncher.h"
23 : #include "replication/logicalworker.h"
24 : #include "storage/ipc.h"
25 : #include "storage/latch.h"
26 : #include "storage/lwlock.h"
27 : #include "storage/pmsignal.h"
28 : #include "storage/proc.h"
29 : #include "storage/procsignal.h"
30 : #include "storage/shmem.h"
31 : #include "tcop/tcopprot.h"
32 : #include "utils/ascii.h"
33 : #include "utils/memutils.h"
34 : #include "utils/ps_status.h"
35 : #include "utils/timeout.h"
36 :
37 : /*
38 : * The postmaster's list of registered background workers, in private memory.
39 : */
40 : dlist_head BackgroundWorkerList = DLIST_STATIC_INIT(BackgroundWorkerList);
41 :
42 : /*
43 : * BackgroundWorkerSlots exist in shared memory and can be accessed (via
44 : * the BackgroundWorkerArray) by both the postmaster and by regular backends.
45 : * However, the postmaster cannot take locks, even spinlocks, because this
46 : * might allow it to crash or become wedged if shared memory gets corrupted.
47 : * Such an outcome is intolerable. Therefore, we need a lockless protocol
48 : * for coordinating access to this data.
49 : *
50 : * The 'in_use' flag is used to hand off responsibility for the slot between
51 : * the postmaster and the rest of the system. When 'in_use' is false,
52 : * the postmaster will ignore the slot entirely, except for the 'in_use' flag
53 : * itself, which it may read. In this state, regular backends may modify the
54 : * slot. Once a backend sets 'in_use' to true, the slot becomes the
55 : * responsibility of the postmaster. Regular backends may no longer modify it,
56 : * but the postmaster may examine it. Thus, a backend initializing a slot
57 : * must fully initialize the slot - and insert a write memory barrier - before
58 : * marking it as in use.
59 : *
60 : * As an exception, however, even when the slot is in use, regular backends
61 : * may set the 'terminate' flag for a slot, telling the postmaster not
62 : * to restart it. Once the background worker is no longer running, the slot
63 : * will be released for reuse.
64 : *
65 : * In addition to coordinating with the postmaster, backends modifying this
66 : * data structure must coordinate with each other. Since they can take locks,
67 : * this is straightforward: any backend wishing to manipulate a slot must
68 : * take BackgroundWorkerLock in exclusive mode. Backends wishing to read
69 : * data that might get concurrently modified by other backends should take
70 : * this lock in shared mode. No matter what, backends reading this data
71 : * structure must be able to tolerate concurrent modifications by the
72 : * postmaster.
73 : */
74 : typedef struct BackgroundWorkerSlot
75 : {
76 : bool in_use;
77 : bool terminate;
78 : pid_t pid; /* InvalidPid = not started yet; 0 = dead */
79 : uint64 generation; /* incremented when slot is recycled */
80 : BackgroundWorker worker;
81 : } BackgroundWorkerSlot;
82 :
83 : /*
84 : * In order to limit the total number of parallel workers (according to
85 : * max_parallel_workers GUC), we maintain the number of active parallel
86 : * workers. Since the postmaster cannot take locks, two variables are used for
87 : * this purpose: the number of registered parallel workers (modified by the
88 : * backends, protected by BackgroundWorkerLock) and the number of terminated
89 : * parallel workers (modified only by the postmaster, lockless). The active
90 : * number of parallel workers is the number of registered workers minus the
91 : * terminated ones. These counters can of course overflow, but it's not
92 : * important here since the subtraction will still give the right number.
93 : */
94 : typedef struct BackgroundWorkerArray
95 : {
96 : int total_slots;
97 : uint32 parallel_register_count;
98 : uint32 parallel_terminate_count;
99 : BackgroundWorkerSlot slot[FLEXIBLE_ARRAY_MEMBER];
100 : } BackgroundWorkerArray;
101 :
102 : struct BackgroundWorkerHandle
103 : {
104 : int slot;
105 : uint64 generation;
106 : };
107 :
108 : static BackgroundWorkerArray *BackgroundWorkerData;
109 :
110 : /*
111 : * List of internal background worker entry points. We need this for
112 : * reasons explained in LookupBackgroundWorkerFunction(), below.
113 : */
114 : static const struct
115 : {
116 : const char *fn_name;
117 : bgworker_main_type fn_addr;
118 : } InternalBGWorkers[] =
119 :
120 : {
121 : {
122 : "ParallelWorkerMain", ParallelWorkerMain
123 : },
124 : {
125 : "ApplyLauncherMain", ApplyLauncherMain
126 : },
127 : {
128 : "ApplyWorkerMain", ApplyWorkerMain
129 : },
130 : {
131 : "ParallelApplyWorkerMain", ParallelApplyWorkerMain
132 : },
133 : {
134 : "TableSyncWorkerMain", TableSyncWorkerMain
135 : },
136 : {
137 : "SequenceSyncWorkerMain", SequenceSyncWorkerMain
138 : }
139 : };
140 :
141 : /* Private functions. */
142 : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
143 :
144 :
145 : /*
146 : * Calculate shared memory needed.
147 : */
148 : Size
149 6300 : BackgroundWorkerShmemSize(void)
150 : {
151 : Size size;
152 :
153 : /* Array of workers is variably sized. */
154 6300 : size = offsetof(BackgroundWorkerArray, slot);
155 6300 : size = add_size(size, mul_size(max_worker_processes,
156 : sizeof(BackgroundWorkerSlot)));
157 :
158 6300 : return size;
159 : }
160 :
161 : /*
162 : * Initialize shared memory.
163 : */
164 : void
165 2200 : BackgroundWorkerShmemInit(void)
166 : {
167 : bool found;
168 :
169 2200 : BackgroundWorkerData = ShmemInitStruct("Background Worker Data",
170 : BackgroundWorkerShmemSize(),
171 : &found);
172 2200 : if (!IsUnderPostmaster)
173 : {
174 : dlist_iter iter;
175 2200 : int slotno = 0;
176 :
177 2200 : BackgroundWorkerData->total_slots = max_worker_processes;
178 2200 : BackgroundWorkerData->parallel_register_count = 0;
179 2200 : BackgroundWorkerData->parallel_terminate_count = 0;
180 :
181 : /*
182 : * Copy contents of worker list into shared memory. Record the shared
183 : * memory slot assigned to each worker. This ensures a 1-to-1
184 : * correspondence between the postmaster's private list and the array
185 : * in shared memory.
186 : */
187 3866 : dlist_foreach(iter, &BackgroundWorkerList)
188 : {
189 1666 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
190 : RegisteredBgWorker *rw;
191 :
192 1666 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
193 : Assert(slotno < max_worker_processes);
194 1666 : slot->in_use = true;
195 1666 : slot->terminate = false;
196 1666 : slot->pid = InvalidPid;
197 1666 : slot->generation = 0;
198 1666 : rw->rw_shmem_slot = slotno;
199 1666 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
200 1666 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
201 1666 : ++slotno;
202 : }
203 :
204 : /*
205 : * Mark any remaining slots as not in use.
206 : */
207 18130 : while (slotno < max_worker_processes)
208 : {
209 15930 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
210 :
211 15930 : slot->in_use = false;
212 15930 : ++slotno;
213 : }
214 : }
215 : else
216 : Assert(found);
217 2200 : }
218 :
219 : /*
220 : * Search the postmaster's backend-private list of RegisteredBgWorker objects
221 : * for the one that maps to the given slot number.
222 : */
223 : static RegisteredBgWorker *
224 8360 : FindRegisteredWorkerBySlotNumber(int slotno)
225 : {
226 : dlist_iter iter;
227 :
228 20768 : dlist_foreach(iter, &BackgroundWorkerList)
229 : {
230 : RegisteredBgWorker *rw;
231 :
232 16914 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
233 16914 : if (rw->rw_shmem_slot == slotno)
234 4506 : return rw;
235 : }
236 :
237 3854 : return NULL;
238 : }
239 :
240 : /*
241 : * Notice changes to shared memory made by other backends.
242 : * Accept new worker requests only if allow_new_workers is true.
243 : *
244 : * This code runs in the postmaster, so we must be very careful not to assume
245 : * that shared memory contents are sane. Otherwise, a rogue backend could
246 : * take out the postmaster.
247 : */
248 : void
249 2472 : BackgroundWorkerStateChange(bool allow_new_workers)
250 : {
251 : int slotno;
252 :
253 : /*
254 : * The total number of slots stored in shared memory should match our
255 : * notion of max_worker_processes. If it does not, something is very
256 : * wrong. Further down, we always refer to this value as
257 : * max_worker_processes, in case shared memory gets corrupted while we're
258 : * looping.
259 : */
260 2472 : if (max_worker_processes != BackgroundWorkerData->total_slots)
261 : {
262 0 : ereport(LOG,
263 : (errmsg("inconsistent background worker state (\"max_worker_processes\"=%d, total slots=%d)",
264 : max_worker_processes,
265 : BackgroundWorkerData->total_slots)));
266 0 : return;
267 : }
268 :
269 : /*
270 : * Iterate through slots, looking for newly-registered workers or workers
271 : * who must die.
272 : */
273 22536 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
274 : {
275 20064 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
276 : RegisteredBgWorker *rw;
277 :
278 20064 : if (!slot->in_use)
279 11704 : continue;
280 :
281 : /*
282 : * Make sure we don't see the in_use flag before the updated slot
283 : * contents.
284 : */
285 8360 : pg_read_barrier();
286 :
287 : /* See whether we already know about this worker. */
288 8360 : rw = FindRegisteredWorkerBySlotNumber(slotno);
289 8360 : if (rw != NULL)
290 : {
291 : /*
292 : * In general, the worker data can't change after it's initially
293 : * registered. However, someone can set the terminate flag.
294 : */
295 4506 : if (slot->terminate && !rw->rw_terminate)
296 : {
297 12 : rw->rw_terminate = true;
298 12 : if (rw->rw_pid != 0)
299 12 : kill(rw->rw_pid, SIGTERM);
300 : else
301 : {
302 : /* Report never-started, now-terminated worker as dead. */
303 0 : ReportBackgroundWorkerPID(rw);
304 : }
305 : }
306 4506 : continue;
307 : }
308 :
309 : /*
310 : * If we aren't allowing new workers, then immediately mark it for
311 : * termination; the next stanza will take care of cleaning it up.
312 : * Doing this ensures that any process waiting for the worker will get
313 : * awoken, even though the worker will never be allowed to run.
314 : */
315 3854 : if (!allow_new_workers)
316 0 : slot->terminate = true;
317 :
318 : /*
319 : * If the worker is marked for termination, we don't need to add it to
320 : * the registered workers list; we can just free the slot. However, if
321 : * bgw_notify_pid is set, the process that registered the worker may
322 : * need to know that we've processed the terminate request, so be sure
323 : * to signal it.
324 : */
325 3854 : if (slot->terminate)
326 0 : {
327 : int notify_pid;
328 :
329 : /*
330 : * We need a memory barrier here to make sure that the load of
331 : * bgw_notify_pid and the update of parallel_terminate_count
332 : * complete before the store to in_use.
333 : */
334 0 : notify_pid = slot->worker.bgw_notify_pid;
335 0 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
336 0 : BackgroundWorkerData->parallel_terminate_count++;
337 0 : slot->pid = 0;
338 :
339 0 : pg_memory_barrier();
340 0 : slot->in_use = false;
341 :
342 0 : if (notify_pid != 0)
343 0 : kill(notify_pid, SIGUSR1);
344 :
345 0 : continue;
346 : }
347 :
348 : /*
349 : * Copy the registration data into the registered workers list.
350 : */
351 3854 : rw = MemoryContextAllocExtended(PostmasterContext,
352 : sizeof(RegisteredBgWorker),
353 : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
354 3854 : if (rw == NULL)
355 : {
356 0 : ereport(LOG,
357 : (errcode(ERRCODE_OUT_OF_MEMORY),
358 : errmsg("out of memory")));
359 0 : return;
360 : }
361 :
362 : /*
363 : * Copy strings in a paranoid way. If shared memory is corrupted, the
364 : * source data might not even be NUL-terminated.
365 : */
366 3854 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
367 3854 : slot->worker.bgw_name, BGW_MAXLEN);
368 3854 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
369 3854 : slot->worker.bgw_type, BGW_MAXLEN);
370 3854 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
371 3854 : slot->worker.bgw_library_name, MAXPGPATH);
372 3854 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
373 3854 : slot->worker.bgw_function_name, BGW_MAXLEN);
374 :
375 : /*
376 : * Copy various fixed-size fields.
377 : *
378 : * flags, start_time, and restart_time are examined by the postmaster,
379 : * but nothing too bad will happen if they are corrupted. The
380 : * remaining fields will only be examined by the child process. It
381 : * might crash, but we won't.
382 : */
383 3854 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
384 3854 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
385 3854 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
386 3854 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
387 3854 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
388 :
389 : /*
390 : * Copy the PID to be notified about state changes, but only if the
391 : * postmaster knows about a backend with that PID. It isn't an error
392 : * if the postmaster doesn't know about the PID, because the backend
393 : * that requested the worker could have died (or been killed) just
394 : * after doing so. Nonetheless, at least until we get some experience
395 : * with how this plays out in the wild, log a message at a relative
396 : * high debug level.
397 : */
398 3854 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
399 3854 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
400 : {
401 0 : elog(DEBUG1, "worker notification PID %d is not valid",
402 : (int) rw->rw_worker.bgw_notify_pid);
403 0 : rw->rw_worker.bgw_notify_pid = 0;
404 : }
405 :
406 : /* Initialize postmaster bookkeeping. */
407 3854 : rw->rw_pid = 0;
408 3854 : rw->rw_crashed_at = 0;
409 3854 : rw->rw_shmem_slot = slotno;
410 3854 : rw->rw_terminate = false;
411 :
412 : /* Log it! */
413 3854 : ereport(DEBUG1,
414 : (errmsg_internal("registering background worker \"%s\"",
415 : rw->rw_worker.bgw_name)));
416 :
417 3854 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
418 : }
419 : }
420 :
421 : /*
422 : * Forget about a background worker that's no longer needed.
423 : *
424 : * NOTE: The entry is unlinked from BackgroundWorkerList. If the caller is
425 : * iterating through it, better use a mutable iterator!
426 : *
427 : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
428 : *
429 : * This function must be invoked only in the postmaster.
430 : */
431 : void
432 3804 : ForgetBackgroundWorker(RegisteredBgWorker *rw)
433 : {
434 : BackgroundWorkerSlot *slot;
435 :
436 : Assert(rw->rw_shmem_slot < max_worker_processes);
437 3804 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
438 : Assert(slot->in_use);
439 :
440 : /*
441 : * We need a memory barrier here to make sure that the update of
442 : * parallel_terminate_count completes before the store to in_use.
443 : */
444 3804 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
445 2752 : BackgroundWorkerData->parallel_terminate_count++;
446 :
447 3804 : pg_memory_barrier();
448 3804 : slot->in_use = false;
449 :
450 3804 : ereport(DEBUG1,
451 : (errmsg_internal("unregistering background worker \"%s\"",
452 : rw->rw_worker.bgw_name)));
453 :
454 3804 : dlist_delete(&rw->rw_lnode);
455 3804 : pfree(rw);
456 3804 : }
457 :
458 : /*
459 : * Report the PID of a newly-launched background worker in shared memory.
460 : *
461 : * This function should only be called from the postmaster.
462 : */
463 : void
464 5310 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
465 : {
466 : BackgroundWorkerSlot *slot;
467 :
468 : Assert(rw->rw_shmem_slot < max_worker_processes);
469 5310 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
470 5310 : slot->pid = rw->rw_pid;
471 :
472 5310 : if (rw->rw_worker.bgw_notify_pid != 0)
473 3854 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
474 5310 : }
475 :
476 : /*
477 : * Report that the PID of a background worker is now zero because a
478 : * previously-running background worker has exited.
479 : *
480 : * NOTE: The entry may be unlinked from BackgroundWorkerList. If the caller
481 : * is iterating through it, better use a mutable iterator!
482 : *
483 : * This function should only be called from the postmaster.
484 : */
485 : void
486 4664 : ReportBackgroundWorkerExit(RegisteredBgWorker *rw)
487 : {
488 : BackgroundWorkerSlot *slot;
489 : int notify_pid;
490 :
491 : Assert(rw->rw_shmem_slot < max_worker_processes);
492 4664 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
493 4664 : slot->pid = rw->rw_pid;
494 4664 : notify_pid = rw->rw_worker.bgw_notify_pid;
495 :
496 : /*
497 : * If this worker is slated for deregistration, do that before notifying
498 : * the process which started it. Otherwise, if that process tries to
499 : * reuse the slot immediately, it might not be available yet. In theory
500 : * that could happen anyway if the process checks slot->pid at just the
501 : * wrong moment, but this makes the window narrower.
502 : */
503 4664 : if (rw->rw_terminate ||
504 1308 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
505 3804 : ForgetBackgroundWorker(rw);
506 :
507 4664 : if (notify_pid != 0)
508 3762 : kill(notify_pid, SIGUSR1);
509 4664 : }
510 :
511 : /*
512 : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
513 : *
514 : * This function should only be called from the postmaster.
515 : */
516 : void
517 498 : BackgroundWorkerStopNotifications(pid_t pid)
518 : {
519 : dlist_iter iter;
520 :
521 1588 : dlist_foreach(iter, &BackgroundWorkerList)
522 : {
523 : RegisteredBgWorker *rw;
524 :
525 1090 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
526 1090 : if (rw->rw_worker.bgw_notify_pid == pid)
527 44 : rw->rw_worker.bgw_notify_pid = 0;
528 : }
529 498 : }
530 :
531 : /*
532 : * Cancel any not-yet-started worker requests that have waiting processes.
533 : *
534 : * This is called during a normal ("smart" or "fast") database shutdown.
535 : * After this point, no new background workers will be started, so anything
536 : * that might be waiting for them needs to be kicked off its wait. We do
537 : * that by canceling the bgworker registration entirely, which is perhaps
538 : * overkill, but since we're shutting down it does not matter whether the
539 : * registration record sticks around.
540 : *
541 : * This function should only be called from the postmaster.
542 : */
543 : void
544 1076 : ForgetUnstartedBackgroundWorkers(void)
545 : {
546 : dlist_mutable_iter iter;
547 :
548 2140 : dlist_foreach_modify(iter, &BackgroundWorkerList)
549 : {
550 : RegisteredBgWorker *rw;
551 : BackgroundWorkerSlot *slot;
552 :
553 1064 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
554 : Assert(rw->rw_shmem_slot < max_worker_processes);
555 1064 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
556 :
557 : /* If it's not yet started, and there's someone waiting ... */
558 1064 : if (slot->pid == InvalidPid &&
559 110 : rw->rw_worker.bgw_notify_pid != 0)
560 : {
561 : /* ... then zap it, and notify the waiter */
562 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
563 :
564 0 : ForgetBackgroundWorker(rw);
565 0 : if (notify_pid != 0)
566 0 : kill(notify_pid, SIGUSR1);
567 : }
568 : }
569 1076 : }
570 :
571 : /*
572 : * Reset background worker crash state.
573 : *
574 : * We assume that, after a crash-and-restart cycle, background workers without
575 : * the never-restart flag should be restarted immediately, instead of waiting
576 : * for bgw_restart_time to elapse. On the other hand, workers with that flag
577 : * should be forgotten immediately, since we won't ever restart them.
578 : *
579 : * This function should only be called from the postmaster.
580 : */
581 : void
582 10 : ResetBackgroundWorkerCrashTimes(void)
583 : {
584 : dlist_mutable_iter iter;
585 :
586 20 : dlist_foreach_modify(iter, &BackgroundWorkerList)
587 : {
588 : RegisteredBgWorker *rw;
589 :
590 10 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
591 :
592 10 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
593 : {
594 : /*
595 : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
596 : * the crash, so forget about them. (If we wait until after the
597 : * crash to forget about them, and they are parallel workers,
598 : * parallel_terminate_count will get incremented after we've
599 : * already zeroed parallel_register_count, which would be bad.)
600 : */
601 0 : ForgetBackgroundWorker(rw);
602 : }
603 : else
604 : {
605 : /*
606 : * The accounting which we do via parallel_register_count and
607 : * parallel_terminate_count would get messed up if a worker marked
608 : * parallel could survive a crash and restart cycle. All such
609 : * workers should be marked BGW_NEVER_RESTART, and thus control
610 : * should never reach this branch.
611 : */
612 : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
613 :
614 : /*
615 : * Allow this worker to be restarted immediately after we finish
616 : * resetting.
617 : */
618 10 : rw->rw_crashed_at = 0;
619 10 : rw->rw_pid = 0;
620 :
621 : /*
622 : * If there was anyone waiting for it, they're history.
623 : */
624 10 : rw->rw_worker.bgw_notify_pid = 0;
625 : }
626 : }
627 10 : }
628 :
629 : /*
630 : * Complain about the BackgroundWorker definition using error level elevel.
631 : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
632 : * which case we won't return at all in the not-OK case).
633 : */
634 : static bool
635 5302 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
636 : {
637 : /* sanity check for flags */
638 :
639 : /*
640 : * We used to support workers not connected to shared memory, but don't
641 : * anymore. Thus this is a required flag now. We're not removing the flag
642 : * for compatibility reasons and because the flag still provides some
643 : * signal when reading code.
644 : */
645 5302 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
646 : {
647 0 : ereport(elevel,
648 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
649 : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
650 : worker->bgw_name)));
651 0 : return false;
652 : }
653 :
654 5302 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
655 : {
656 5284 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
657 : {
658 0 : ereport(elevel,
659 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
660 : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
661 : worker->bgw_name)));
662 0 : return false;
663 : }
664 :
665 : /* XXX other checks? */
666 : }
667 :
668 5302 : if ((worker->bgw_restart_time < 0 &&
669 3638 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
670 5302 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
671 : {
672 0 : ereport(elevel,
673 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
674 : errmsg("background worker \"%s\": invalid restart interval",
675 : worker->bgw_name)));
676 0 : return false;
677 : }
678 :
679 : /*
680 : * Parallel workers may not be configured for restart, because the
681 : * parallel_register_count/parallel_terminate_count accounting can't
682 : * handle parallel workers lasting through a crash-and-restart cycle.
683 : */
684 5302 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
685 1664 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
686 : {
687 0 : ereport(elevel,
688 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
689 : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
690 : worker->bgw_name)));
691 0 : return false;
692 : }
693 :
694 : /*
695 : * If bgw_type is not filled in, use bgw_name.
696 : */
697 5302 : if (strcmp(worker->bgw_type, "") == 0)
698 0 : strcpy(worker->bgw_type, worker->bgw_name);
699 :
700 5302 : return true;
701 : }
702 :
703 : /*
704 : * Standard SIGTERM handler for background workers
705 : */
706 : static void
707 0 : bgworker_die(SIGNAL_ARGS)
708 : {
709 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
710 :
711 0 : ereport(FATAL,
712 : (errcode(ERRCODE_ADMIN_SHUTDOWN),
713 : errmsg("terminating background worker \"%s\" due to administrator command",
714 : MyBgworkerEntry->bgw_type)));
715 : }
716 :
717 : /*
718 : * Main entry point for background worker processes.
719 : */
720 : void
721 4690 : BackgroundWorkerMain(const void *startup_data, size_t startup_data_len)
722 : {
723 : sigjmp_buf local_sigjmp_buf;
724 : BackgroundWorker *worker;
725 : bgworker_main_type entrypt;
726 :
727 4690 : if (startup_data == NULL)
728 0 : elog(FATAL, "unable to find bgworker entry");
729 : Assert(startup_data_len == sizeof(BackgroundWorker));
730 4690 : worker = MemoryContextAlloc(TopMemoryContext, sizeof(BackgroundWorker));
731 4690 : memcpy(worker, startup_data, sizeof(BackgroundWorker));
732 :
733 : /*
734 : * Now that we're done reading the startup data, release postmaster's
735 : * working memory context.
736 : */
737 4690 : if (PostmasterContext)
738 : {
739 4690 : MemoryContextDelete(PostmasterContext);
740 4690 : PostmasterContext = NULL;
741 : }
742 :
743 4690 : MyBgworkerEntry = worker;
744 4690 : MyBackendType = B_BG_WORKER;
745 4690 : init_ps_display(worker->bgw_name);
746 :
747 : Assert(GetProcessingMode() == InitProcessing);
748 :
749 : /* Apply PostAuthDelay */
750 4690 : if (PostAuthDelay > 0)
751 0 : pg_usleep(PostAuthDelay * 1000000L);
752 :
753 : /*
754 : * Set up signal handlers.
755 : */
756 4690 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
757 : {
758 : /*
759 : * SIGINT is used to signal canceling the current action
760 : */
761 4672 : pqsignal(SIGINT, StatementCancelHandler);
762 4672 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
763 4672 : pqsignal(SIGFPE, FloatExceptionHandler);
764 :
765 : /* XXX Any other handlers needed here? */
766 : }
767 : else
768 : {
769 18 : pqsignal(SIGINT, SIG_IGN);
770 18 : pqsignal(SIGUSR1, SIG_IGN);
771 18 : pqsignal(SIGFPE, SIG_IGN);
772 : }
773 4690 : pqsignal(SIGTERM, bgworker_die);
774 : /* SIGQUIT handler was already set up by InitPostmasterChild */
775 4690 : pqsignal(SIGHUP, SIG_IGN);
776 :
777 4690 : InitializeTimeouts(); /* establishes SIGALRM handler */
778 :
779 4690 : pqsignal(SIGPIPE, SIG_IGN);
780 4690 : pqsignal(SIGUSR2, SIG_IGN);
781 4690 : pqsignal(SIGCHLD, SIG_DFL);
782 :
783 : /*
784 : * If an exception is encountered, processing resumes here.
785 : *
786 : * We just need to clean up, report the error, and go away.
787 : */
788 4690 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
789 : {
790 : /* Since not using PG_TRY, must reset error stack by hand */
791 252 : error_context_stack = NULL;
792 :
793 : /* Prevent interrupts while cleaning up */
794 252 : HOLD_INTERRUPTS();
795 :
796 : /*
797 : * sigsetjmp will have blocked all signals, but we may need to accept
798 : * signals while communicating with our parallel leader. Once we've
799 : * done HOLD_INTERRUPTS() it should be safe to unblock signals.
800 : */
801 252 : BackgroundWorkerUnblockSignals();
802 :
803 : /* Report the error to the parallel leader and the server log */
804 252 : EmitErrorReport();
805 :
806 : /*
807 : * Do we need more cleanup here? For shmem-connected bgworkers, we
808 : * will call InitProcess below, which will install ProcKill as exit
809 : * callback. That will take care of releasing locks, etc.
810 : */
811 :
812 : /* and go away */
813 252 : proc_exit(1);
814 : }
815 :
816 : /* We can now handle ereport(ERROR) */
817 4690 : PG_exception_stack = &local_sigjmp_buf;
818 :
819 : /*
820 : * Create a per-backend PGPROC struct in shared memory. We must do this
821 : * before we can use LWLocks or access any shared memory.
822 : */
823 4690 : InitProcess();
824 :
825 : /*
826 : * Early initialization.
827 : */
828 4690 : BaseInit();
829 :
830 : /*
831 : * Look up the entry point function, loading its library if necessary.
832 : */
833 9380 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
834 4690 : worker->bgw_function_name);
835 :
836 : /*
837 : * Note that in normal processes, we would call InitPostgres here. For a
838 : * worker, however, we don't know what database to connect to, yet; so we
839 : * need to wait until the user code does it via
840 : * BackgroundWorkerInitializeConnection().
841 : */
842 :
843 : /*
844 : * Now invoke the user-defined worker code
845 : */
846 4690 : entrypt(worker->bgw_main_arg);
847 :
848 : /* ... and if it returns, we're done */
849 2746 : proc_exit(0);
850 : }
851 :
852 : /*
853 : * Connect background worker to a database.
854 : */
855 : void
856 866 : BackgroundWorkerInitializeConnection(const char *dbname, const char *username, uint32 flags)
857 : {
858 866 : BackgroundWorker *worker = MyBgworkerEntry;
859 866 : bits32 init_flags = 0; /* never honor session_preload_libraries */
860 :
861 : /* ignore datallowconn and ACL_CONNECT? */
862 866 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
863 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
864 : /* ignore rolcanlogin? */
865 866 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
866 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
867 :
868 : /* XXX is this the right errcode? */
869 866 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
870 0 : ereport(FATAL,
871 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
872 : errmsg("database connection requirement not indicated during registration")));
873 :
874 866 : InitPostgres(dbname, InvalidOid, /* database to connect to */
875 : username, InvalidOid, /* role to connect as */
876 : init_flags,
877 : NULL); /* no out_dbname */
878 :
879 : /* it had better not gotten out of "init" mode yet */
880 866 : if (!IsInitProcessingMode())
881 0 : ereport(ERROR,
882 : (errmsg("invalid processing mode in background worker")));
883 866 : SetProcessingMode(NormalProcessing);
884 866 : }
885 :
886 : /*
887 : * Connect background worker to a database using OIDs.
888 : */
889 : void
890 3806 : BackgroundWorkerInitializeConnectionByOid(Oid dboid, Oid useroid, uint32 flags)
891 : {
892 3806 : BackgroundWorker *worker = MyBgworkerEntry;
893 3806 : bits32 init_flags = 0; /* never honor session_preload_libraries */
894 :
895 : /* ignore datallowconn and ACL_CONNECT? */
896 3806 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
897 2752 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
898 : /* ignore rolcanlogin? */
899 3806 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
900 2752 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
901 :
902 : /* XXX is this the right errcode? */
903 3806 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
904 0 : ereport(FATAL,
905 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
906 : errmsg("database connection requirement not indicated during registration")));
907 :
908 3806 : InitPostgres(NULL, dboid, /* database to connect to */
909 : NULL, useroid, /* role to connect as */
910 : init_flags,
911 : NULL); /* no out_dbname */
912 :
913 : /* it had better not gotten out of "init" mode yet */
914 3794 : if (!IsInitProcessingMode())
915 0 : ereport(ERROR,
916 : (errmsg("invalid processing mode in background worker")));
917 3794 : SetProcessingMode(NormalProcessing);
918 3794 : }
919 :
920 : /*
921 : * Block/unblock signals in a background worker
922 : */
923 : void
924 0 : BackgroundWorkerBlockSignals(void)
925 : {
926 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
927 0 : }
928 :
929 : void
930 4942 : BackgroundWorkerUnblockSignals(void)
931 : {
932 4942 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
933 4942 : }
934 :
935 : /*
936 : * Register a new static background worker.
937 : *
938 : * This can only be called directly from postmaster or in the _PG_init
939 : * function of a module library that's loaded by shared_preload_libraries;
940 : * otherwise it will have no effect.
941 : */
942 : void
943 1670 : RegisterBackgroundWorker(BackgroundWorker *worker)
944 : {
945 : RegisteredBgWorker *rw;
946 : static int numworkers = 0;
947 :
948 : /*
949 : * Static background workers can only be registered in the postmaster
950 : * process.
951 : */
952 1670 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
953 : {
954 : /*
955 : * In EXEC_BACKEND or single-user mode, we process
956 : * shared_preload_libraries in backend processes too. We cannot
957 : * register static background workers at that stage, but many
958 : * libraries' _PG_init() functions don't distinguish whether they're
959 : * being loaded in the postmaster or in a backend, they just check
960 : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
961 : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
962 : * the background workers should already have been registered when the
963 : * library was loaded in postmaster.
964 : */
965 0 : if (process_shared_preload_libraries_in_progress)
966 0 : return;
967 0 : ereport(LOG,
968 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
969 : errmsg("background worker \"%s\": must be registered in \"shared_preload_libraries\"",
970 : worker->bgw_name)));
971 0 : return;
972 : }
973 :
974 : /*
975 : * Cannot register static background workers after calling
976 : * BackgroundWorkerShmemInit().
977 : */
978 1670 : if (BackgroundWorkerData != NULL)
979 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
980 : worker->bgw_name);
981 :
982 1670 : ereport(DEBUG1,
983 : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
984 :
985 1670 : if (!SanityCheckBackgroundWorker(worker, LOG))
986 0 : return;
987 :
988 1670 : if (worker->bgw_notify_pid != 0)
989 : {
990 0 : ereport(LOG,
991 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
992 : errmsg("background worker \"%s\": only dynamic background workers can request notification",
993 : worker->bgw_name)));
994 0 : return;
995 : }
996 :
997 : /*
998 : * Enforce maximum number of workers. Note this is overly restrictive: we
999 : * could allow more non-shmem-connected workers, because these don't count
1000 : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
1001 : * important to relax this restriction.
1002 : */
1003 1670 : if (++numworkers > max_worker_processes)
1004 : {
1005 0 : ereport(LOG,
1006 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
1007 : errmsg("too many background workers"),
1008 : errdetail_plural("Up to %d background worker can be registered with the current settings.",
1009 : "Up to %d background workers can be registered with the current settings.",
1010 : max_worker_processes,
1011 : max_worker_processes),
1012 : errhint("Consider increasing the configuration parameter \"%s\".", "max_worker_processes")));
1013 0 : return;
1014 : }
1015 :
1016 : /*
1017 : * Copy the registration data into the registered workers list.
1018 : */
1019 1670 : rw = MemoryContextAllocExtended(PostmasterContext,
1020 : sizeof(RegisteredBgWorker),
1021 : MCXT_ALLOC_NO_OOM);
1022 1670 : if (rw == NULL)
1023 : {
1024 0 : ereport(LOG,
1025 : (errcode(ERRCODE_OUT_OF_MEMORY),
1026 : errmsg("out of memory")));
1027 0 : return;
1028 : }
1029 :
1030 1670 : rw->rw_worker = *worker;
1031 1670 : rw->rw_pid = 0;
1032 1670 : rw->rw_crashed_at = 0;
1033 1670 : rw->rw_terminate = false;
1034 :
1035 1670 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
1036 : }
1037 :
1038 : /*
1039 : * Register a new background worker from a regular backend.
1040 : *
1041 : * Returns true on success and false on failure. Failure typically indicates
1042 : * that no background worker slots are currently available.
1043 : *
1044 : * If handle != NULL, we'll set *handle to a pointer that can subsequently
1045 : * be used as an argument to GetBackgroundWorkerPid(). The caller can
1046 : * free this pointer using pfree(), if desired.
1047 : */
1048 : bool
1049 3632 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
1050 : BackgroundWorkerHandle **handle)
1051 : {
1052 : int slotno;
1053 3632 : bool success = false;
1054 : bool parallel;
1055 3632 : uint64 generation = 0;
1056 :
1057 : /*
1058 : * We can't register dynamic background workers from the postmaster. If
1059 : * this is a standalone backend, we're the only process and can't start
1060 : * any more. In a multi-process environment, it might be theoretically
1061 : * possible, but we don't currently support it due to locking
1062 : * considerations; see comments on the BackgroundWorkerSlot data
1063 : * structure.
1064 : */
1065 3632 : if (!IsUnderPostmaster)
1066 0 : return false;
1067 :
1068 3632 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1069 0 : return false;
1070 :
1071 3632 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1072 :
1073 3632 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1074 :
1075 : /*
1076 : * If this is a parallel worker, check whether there are already too many
1077 : * parallel workers; if so, don't register another one. Our view of
1078 : * parallel_terminate_count may be slightly stale, but that doesn't really
1079 : * matter: we would have gotten the same result if we'd arrived here
1080 : * slightly earlier anyway. There's no help for it, either, since the
1081 : * postmaster must not take locks; a memory barrier wouldn't guarantee
1082 : * anything useful.
1083 : */
1084 3632 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1085 2780 : BackgroundWorkerData->parallel_terminate_count) >=
1086 : max_parallel_workers)
1087 : {
1088 : Assert(BackgroundWorkerData->parallel_register_count -
1089 : BackgroundWorkerData->parallel_terminate_count <=
1090 : MAX_PARALLEL_WORKER_LIMIT);
1091 20 : LWLockRelease(BackgroundWorkerLock);
1092 20 : return false;
1093 : }
1094 :
1095 : /*
1096 : * Look for an unused slot. If we find one, grab it.
1097 : */
1098 11428 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1099 : {
1100 11420 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1101 :
1102 11420 : if (!slot->in_use)
1103 : {
1104 3604 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1105 3604 : slot->pid = InvalidPid; /* indicates not started yet */
1106 3604 : slot->generation++;
1107 3604 : slot->terminate = false;
1108 3604 : generation = slot->generation;
1109 3604 : if (parallel)
1110 2752 : BackgroundWorkerData->parallel_register_count++;
1111 :
1112 : /*
1113 : * Make sure postmaster doesn't see the slot as in use before it
1114 : * sees the new contents.
1115 : */
1116 3604 : pg_write_barrier();
1117 :
1118 3604 : slot->in_use = true;
1119 3604 : success = true;
1120 3604 : break;
1121 : }
1122 : }
1123 :
1124 3612 : LWLockRelease(BackgroundWorkerLock);
1125 :
1126 : /* If we found a slot, tell the postmaster to notice the change. */
1127 3612 : if (success)
1128 3604 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1129 :
1130 : /*
1131 : * If we found a slot and the user has provided a handle, initialize it.
1132 : */
1133 3612 : if (success && handle)
1134 : {
1135 3604 : *handle = palloc(sizeof(BackgroundWorkerHandle));
1136 3604 : (*handle)->slot = slotno;
1137 3604 : (*handle)->generation = generation;
1138 : }
1139 :
1140 3612 : return success;
1141 : }
1142 :
1143 : /*
1144 : * Get the PID of a dynamically-registered background worker.
1145 : *
1146 : * If the worker is determined to be running, the return value will be
1147 : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1148 : * postmaster has not yet attempted to start the worker, the return value will
1149 : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1150 : *
1151 : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1152 : * (because it is configured for automatic restart and exited non-zero),
1153 : * or that the worker is permanently stopped (because it exited with exit
1154 : * code 0, or was not configured for automatic restart), or even that the
1155 : * worker was unregistered without ever starting (either because startup
1156 : * failed and the worker is not configured for automatic restart, or because
1157 : * TerminateBackgroundWorker was used before the worker was successfully
1158 : * started).
1159 : */
1160 : BgwHandleStatus
1161 22040552 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1162 : {
1163 : BackgroundWorkerSlot *slot;
1164 : pid_t pid;
1165 :
1166 : Assert(handle->slot < max_worker_processes);
1167 22040552 : slot = &BackgroundWorkerData->slot[handle->slot];
1168 :
1169 : /*
1170 : * We could probably arrange to synchronize access to data using memory
1171 : * barriers only, but for now, let's just keep it simple and grab the
1172 : * lock. It seems unlikely that there will be enough traffic here to
1173 : * result in meaningful contention.
1174 : */
1175 22040552 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1176 :
1177 : /*
1178 : * The generation number can't be concurrently changed while we hold the
1179 : * lock. The pid, which is updated by the postmaster, can change at any
1180 : * time, but we assume such changes are atomic. So the value we read
1181 : * won't be garbage, but it might be out of date by the time the caller
1182 : * examines it (but that's unavoidable anyway).
1183 : *
1184 : * The in_use flag could be in the process of changing from true to false,
1185 : * but if it is already false then it can't change further.
1186 : */
1187 22040552 : if (handle->generation != slot->generation || !slot->in_use)
1188 2754 : pid = 0;
1189 : else
1190 22037798 : pid = slot->pid;
1191 :
1192 : /* All done. */
1193 22040552 : LWLockRelease(BackgroundWorkerLock);
1194 :
1195 22040552 : if (pid == 0)
1196 2754 : return BGWH_STOPPED;
1197 22037798 : else if (pid == InvalidPid)
1198 1403000 : return BGWH_NOT_YET_STARTED;
1199 20634798 : *pidp = pid;
1200 20634798 : return BGWH_STARTED;
1201 : }
1202 :
1203 : /*
1204 : * Wait for a background worker to start up.
1205 : *
1206 : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1207 : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1208 : * returned. However, if the postmaster has died, we give up and return
1209 : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1210 : * take place.
1211 : *
1212 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1213 : * else we will not be awoken promptly when the worker's state changes.
1214 : */
1215 : BgwHandleStatus
1216 12 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1217 : {
1218 : BgwHandleStatus status;
1219 : int rc;
1220 :
1221 : for (;;)
1222 4 : {
1223 : pid_t pid;
1224 :
1225 16 : CHECK_FOR_INTERRUPTS();
1226 :
1227 16 : status = GetBackgroundWorkerPid(handle, &pid);
1228 16 : if (status == BGWH_STARTED)
1229 12 : *pidp = pid;
1230 16 : if (status != BGWH_NOT_YET_STARTED)
1231 12 : break;
1232 :
1233 4 : rc = WaitLatch(MyLatch,
1234 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1235 : WAIT_EVENT_BGWORKER_STARTUP);
1236 :
1237 4 : if (rc & WL_POSTMASTER_DEATH)
1238 : {
1239 0 : status = BGWH_POSTMASTER_DIED;
1240 0 : break;
1241 : }
1242 :
1243 4 : ResetLatch(MyLatch);
1244 : }
1245 :
1246 12 : return status;
1247 : }
1248 :
1249 : /*
1250 : * Wait for a background worker to stop.
1251 : *
1252 : * If the worker hasn't yet started, or is running, we wait for it to stop
1253 : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1254 : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1255 : * notifies us when a worker's state changes.
1256 : *
1257 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1258 : * else we will not be awoken promptly when the worker's state changes.
1259 : */
1260 : BgwHandleStatus
1261 2754 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1262 : {
1263 : BgwHandleStatus status;
1264 : int rc;
1265 :
1266 : for (;;)
1267 3840 : {
1268 : pid_t pid;
1269 :
1270 6594 : CHECK_FOR_INTERRUPTS();
1271 :
1272 6594 : status = GetBackgroundWorkerPid(handle, &pid);
1273 6594 : if (status == BGWH_STOPPED)
1274 2754 : break;
1275 :
1276 3840 : rc = WaitLatch(MyLatch,
1277 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1278 : WAIT_EVENT_BGWORKER_SHUTDOWN);
1279 :
1280 3840 : if (rc & WL_POSTMASTER_DEATH)
1281 : {
1282 0 : status = BGWH_POSTMASTER_DIED;
1283 0 : break;
1284 : }
1285 :
1286 3840 : ResetLatch(MyLatch);
1287 : }
1288 :
1289 2754 : return status;
1290 : }
1291 :
1292 : /*
1293 : * Instruct the postmaster to terminate a background worker.
1294 : *
1295 : * Note that it's safe to do this without regard to whether the worker is
1296 : * still running, or even if the worker may already have exited and been
1297 : * unregistered.
1298 : */
1299 : void
1300 12 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1301 : {
1302 : BackgroundWorkerSlot *slot;
1303 12 : bool signal_postmaster = false;
1304 :
1305 : Assert(handle->slot < max_worker_processes);
1306 12 : slot = &BackgroundWorkerData->slot[handle->slot];
1307 :
1308 : /* Set terminate flag in shared memory, unless slot has been reused. */
1309 12 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1310 12 : if (handle->generation == slot->generation)
1311 : {
1312 12 : slot->terminate = true;
1313 12 : signal_postmaster = true;
1314 : }
1315 12 : LWLockRelease(BackgroundWorkerLock);
1316 :
1317 : /* Make sure the postmaster notices the change to shared memory. */
1318 12 : if (signal_postmaster)
1319 12 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1320 12 : }
1321 :
1322 : /*
1323 : * Look up (and possibly load) a bgworker entry point function.
1324 : *
1325 : * For functions contained in the core code, we use library name "postgres"
1326 : * and consult the InternalBGWorkers array. External functions are
1327 : * looked up, and loaded if necessary, using load_external_function().
1328 : *
1329 : * The point of this is to pass function names as strings across process
1330 : * boundaries. We can't pass actual function addresses because of the
1331 : * possibility that the function has been loaded at a different address
1332 : * in a different process. This is obviously a hazard for functions in
1333 : * loadable libraries, but it can happen even for functions in the core code
1334 : * on platforms using EXEC_BACKEND (e.g., Windows).
1335 : *
1336 : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1337 : * in favor of applying load_external_function() for core functions too;
1338 : * but that raises portability issues that are not worth addressing now.
1339 : */
1340 : static bgworker_main_type
1341 4690 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1342 : {
1343 : /*
1344 : * If the function is to be loaded from postgres itself, search the
1345 : * InternalBGWorkers array.
1346 : */
1347 4690 : if (strcmp(libraryname, "postgres") == 0)
1348 : {
1349 : int i;
1350 :
1351 8496 : for (i = 0; i < lengthof(InternalBGWorkers); i++)
1352 : {
1353 8496 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1354 4664 : return InternalBGWorkers[i].fn_addr;
1355 : }
1356 :
1357 : /* We can only reach this by programming error. */
1358 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1359 : }
1360 :
1361 : /* Otherwise load from external library. */
1362 26 : return (bgworker_main_type)
1363 26 : load_external_function(libraryname, funcname, true, NULL);
1364 : }
1365 :
1366 : /*
1367 : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1368 : * not a valid background worker.
1369 : *
1370 : * The return value is in static memory belonging to this function, so it has
1371 : * to be used before calling this function again. This is so that the caller
1372 : * doesn't have to worry about the background worker locking protocol.
1373 : */
1374 : const char *
1375 2030 : GetBackgroundWorkerTypeByPid(pid_t pid)
1376 : {
1377 : int slotno;
1378 2030 : bool found = false;
1379 : static char result[BGW_MAXLEN];
1380 :
1381 2030 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1382 :
1383 2240 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1384 : {
1385 2240 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1386 :
1387 2240 : if (slot->pid > 0 && slot->pid == pid)
1388 : {
1389 2030 : strcpy(result, slot->worker.bgw_type);
1390 2030 : found = true;
1391 2030 : break;
1392 : }
1393 : }
1394 :
1395 2030 : LWLockRelease(BackgroundWorkerLock);
1396 :
1397 2030 : if (!found)
1398 0 : return NULL;
1399 :
1400 2030 : return result;
1401 : }
|
