Line data Source code
1 : /*--------------------------------------------------------------------
2 : * bgworker.c
3 : * POSTGRES pluggable background workers implementation
4 : *
5 : * Portions Copyright (c) 1996-2026, 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 "commands/repack.h"
17 : #include "libpq/pqsignal.h"
18 : #include "miscadmin.h"
19 : #include "pgstat.h"
20 : #include "port/atomics.h"
21 : #include "postmaster/bgworker_internals.h"
22 : #include "postmaster/datachecksum_state.h"
23 : #include "postmaster/postmaster.h"
24 : #include "replication/logicallauncher.h"
25 : #include "replication/logicalworker.h"
26 : #include "storage/ipc.h"
27 : #include "storage/latch.h"
28 : #include "storage/lwlock.h"
29 : #include "storage/pmsignal.h"
30 : #include "storage/proc.h"
31 : #include "storage/procarray.h"
32 : #include "storage/procsignal.h"
33 : #include "storage/shmem.h"
34 : #include "storage/subsystems.h"
35 : #include "tcop/tcopprot.h"
36 : #include "utils/ascii.h"
37 : #include "utils/memutils.h"
38 : #include "utils/ps_status.h"
39 : #include "utils/timeout.h"
40 : #include "utils/wait_event.h"
41 :
42 : /*
43 : * The postmaster's list of registered background workers, in private memory.
44 : */
45 : dlist_head BackgroundWorkerList = DLIST_STATIC_INIT(BackgroundWorkerList);
46 :
47 : /*
48 : * BackgroundWorkerSlots exist in shared memory and can be accessed (via
49 : * the BackgroundWorkerArray) by both the postmaster and by regular backends.
50 : * However, the postmaster cannot take locks, even spinlocks, because this
51 : * might allow it to crash or become wedged if shared memory gets corrupted.
52 : * Such an outcome is intolerable. Therefore, we need a lockless protocol
53 : * for coordinating access to this data.
54 : *
55 : * The 'in_use' flag is used to hand off responsibility for the slot between
56 : * the postmaster and the rest of the system. When 'in_use' is false,
57 : * the postmaster will ignore the slot entirely, except for the 'in_use' flag
58 : * itself, which it may read. In this state, regular backends may modify the
59 : * slot. Once a backend sets 'in_use' to true, the slot becomes the
60 : * responsibility of the postmaster. Regular backends may no longer modify it,
61 : * but the postmaster may examine it. Thus, a backend initializing a slot
62 : * must fully initialize the slot - and insert a write memory barrier - before
63 : * marking it as in use.
64 : *
65 : * As an exception, however, even when the slot is in use, regular backends
66 : * may set the 'terminate' flag for a slot, telling the postmaster not
67 : * to restart it. Once the background worker is no longer running, the slot
68 : * will be released for reuse.
69 : *
70 : * In addition to coordinating with the postmaster, backends modifying this
71 : * data structure must coordinate with each other. Since they can take locks,
72 : * this is straightforward: any backend wishing to manipulate a slot must
73 : * take BackgroundWorkerLock in exclusive mode. Backends wishing to read
74 : * data that might get concurrently modified by other backends should take
75 : * this lock in shared mode. No matter what, backends reading this data
76 : * structure must be able to tolerate concurrent modifications by the
77 : * postmaster.
78 : */
79 : typedef struct BackgroundWorkerSlot
80 : {
81 : bool in_use;
82 : bool terminate;
83 : pid_t pid; /* InvalidPid = not started yet; 0 = dead */
84 : uint64 generation; /* incremented when slot is recycled */
85 : BackgroundWorker worker;
86 : } BackgroundWorkerSlot;
87 :
88 : /*
89 : * In order to limit the total number of parallel workers (according to
90 : * max_parallel_workers GUC), we maintain the number of active parallel
91 : * workers. Since the postmaster cannot take locks, two variables are used for
92 : * this purpose: the number of registered parallel workers (modified by the
93 : * backends, protected by BackgroundWorkerLock) and the number of terminated
94 : * parallel workers (modified only by the postmaster, lockless). The active
95 : * number of parallel workers is the number of registered workers minus the
96 : * terminated ones. These counters can of course overflow, but it's not
97 : * important here since the subtraction will still give the right number.
98 : */
99 : typedef struct BackgroundWorkerArray
100 : {
101 : int total_slots;
102 : uint32 parallel_register_count;
103 : uint32 parallel_terminate_count;
104 : BackgroundWorkerSlot slot[FLEXIBLE_ARRAY_MEMBER];
105 : } BackgroundWorkerArray;
106 :
107 : struct BackgroundWorkerHandle
108 : {
109 : int slot;
110 : uint64 generation;
111 : };
112 :
113 : static BackgroundWorkerArray *BackgroundWorkerData;
114 :
115 : static void BackgroundWorkerShmemRequest(void *arg);
116 : static void BackgroundWorkerShmemInit(void *arg);
117 :
118 : const ShmemCallbacks BackgroundWorkerShmemCallbacks = {
119 : .request_fn = BackgroundWorkerShmemRequest,
120 : .init_fn = BackgroundWorkerShmemInit,
121 : };
122 :
123 : /*
124 : * List of internal background worker entry points. We need this for
125 : * reasons explained in LookupBackgroundWorkerFunction(), below.
126 : */
127 : static const struct
128 : {
129 : const char *fn_name;
130 : bgworker_main_type fn_addr;
131 : } InternalBGWorkers[] =
132 :
133 : {
134 : {
135 : .fn_name = "ApplyLauncherMain",
136 : .fn_addr = ApplyLauncherMain
137 : },
138 : {
139 : .fn_name = "ApplyWorkerMain",
140 : .fn_addr = ApplyWorkerMain
141 : },
142 : {
143 : .fn_name = "ParallelApplyWorkerMain",
144 : .fn_addr = ParallelApplyWorkerMain
145 : },
146 : {
147 : .fn_name = "ParallelWorkerMain",
148 : .fn_addr = ParallelWorkerMain
149 : },
150 : {
151 : .fn_name = "RepackWorkerMain",
152 : .fn_addr = RepackWorkerMain
153 : },
154 : {
155 : .fn_name = "SequenceSyncWorkerMain",
156 : .fn_addr = SequenceSyncWorkerMain
157 : },
158 : {
159 : .fn_name = "TableSyncWorkerMain",
160 : .fn_addr = TableSyncWorkerMain
161 : },
162 : {
163 : .fn_name = "DataChecksumsWorkerLauncherMain",
164 : .fn_addr = DataChecksumsWorkerLauncherMain
165 : },
166 : {
167 : .fn_name = "DataChecksumsWorkerMain",
168 : .fn_addr = DataChecksumsWorkerMain
169 : }
170 : };
171 :
172 : /* Private functions. */
173 : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
174 :
175 :
176 : /*
177 : * Register shared memory needed for background workers.
178 : */
179 : static void
180 1234 : BackgroundWorkerShmemRequest(void *arg)
181 : {
182 : Size size;
183 :
184 : /* Array of workers is variably sized. */
185 1234 : size = offsetof(BackgroundWorkerArray, slot);
186 1234 : size = add_size(size, mul_size(max_worker_processes,
187 : sizeof(BackgroundWorkerSlot)));
188 1234 : ShmemRequestStruct(.name = "Background Worker Data",
189 : .size = size,
190 : .ptr = (void **) &BackgroundWorkerData,
191 : );
192 1234 : }
193 :
194 : /*
195 : * Initialize shared memory for background workers.
196 : */
197 : static void
198 1231 : BackgroundWorkerShmemInit(void *arg)
199 : {
200 : dlist_iter iter;
201 1231 : int slotno = 0;
202 :
203 1231 : BackgroundWorkerData->total_slots = max_worker_processes;
204 1231 : BackgroundWorkerData->parallel_register_count = 0;
205 1231 : BackgroundWorkerData->parallel_terminate_count = 0;
206 :
207 : /*
208 : * Copy contents of worker list into shared memory. Record the shared
209 : * memory slot assigned to each worker. This ensures a 1-to-1
210 : * correspondence between the postmaster's private list and the array in
211 : * shared memory.
212 : */
213 2159 : dlist_foreach(iter, &BackgroundWorkerList)
214 : {
215 928 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
216 : RegisteredBgWorker *rw;
217 :
218 928 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
219 : Assert(slotno < max_worker_processes);
220 928 : slot->in_use = true;
221 928 : slot->terminate = false;
222 928 : slot->pid = InvalidPid;
223 928 : slot->generation = 0;
224 928 : rw->rw_shmem_slot = slotno;
225 928 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
226 928 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
227 928 : ++slotno;
228 : }
229 :
230 : /*
231 : * Mark any remaining slots as not in use.
232 : */
233 10144 : while (slotno < max_worker_processes)
234 : {
235 8913 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
236 :
237 8913 : slot->in_use = false;
238 8913 : ++slotno;
239 : }
240 1231 : }
241 :
242 : /*
243 : * Search the postmaster's backend-private list of RegisteredBgWorker objects
244 : * for the one that maps to the given slot number.
245 : */
246 : static RegisteredBgWorker *
247 5641 : FindRegisteredWorkerBySlotNumber(int slotno)
248 : {
249 : dlist_iter iter;
250 :
251 13787 : dlist_foreach(iter, &BackgroundWorkerList)
252 : {
253 : RegisteredBgWorker *rw;
254 :
255 11087 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
256 11087 : if (rw->rw_shmem_slot == slotno)
257 2941 : return rw;
258 : }
259 :
260 2700 : return NULL;
261 : }
262 :
263 : /*
264 : * Notice changes to shared memory made by other backends.
265 : * Accept new worker requests only if allow_new_workers is true.
266 : *
267 : * This code runs in the postmaster, so we must be very careful not to assume
268 : * that shared memory contents are sane. Otherwise, a rogue backend could
269 : * take out the postmaster.
270 : */
271 : void
272 1702 : BackgroundWorkerStateChange(bool allow_new_workers)
273 : {
274 : int slotno;
275 :
276 : /*
277 : * The total number of slots stored in shared memory should match our
278 : * notion of max_worker_processes. If it does not, something is very
279 : * wrong. Further down, we always refer to this value as
280 : * max_worker_processes, in case shared memory gets corrupted while we're
281 : * looping.
282 : */
283 1702 : if (max_worker_processes != BackgroundWorkerData->total_slots)
284 : {
285 0 : ereport(LOG,
286 : (errmsg("inconsistent background worker state (\"max_worker_processes\"=%d, total slots=%d)",
287 : max_worker_processes,
288 : BackgroundWorkerData->total_slots)));
289 0 : return;
290 : }
291 :
292 : /*
293 : * Iterate through slots, looking for newly-registered workers or workers
294 : * who must die.
295 : */
296 15472 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
297 : {
298 13770 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
299 : RegisteredBgWorker *rw;
300 :
301 13770 : if (!slot->in_use)
302 8129 : continue;
303 :
304 : /*
305 : * Make sure we don't see the in_use flag before the updated slot
306 : * contents.
307 : */
308 5641 : pg_read_barrier();
309 :
310 : /* See whether we already know about this worker. */
311 5641 : rw = FindRegisteredWorkerBySlotNumber(slotno);
312 5641 : if (rw != NULL)
313 : {
314 : /*
315 : * In general, the worker data can't change after it's initially
316 : * registered. However, someone can set the terminate flag.
317 : */
318 2941 : if (slot->terminate && !rw->rw_terminate)
319 : {
320 14 : rw->rw_terminate = true;
321 14 : if (rw->rw_pid != 0)
322 14 : kill(rw->rw_pid, SIGTERM);
323 : else
324 : {
325 : /* Report never-started, now-terminated worker as dead. */
326 0 : ReportBackgroundWorkerPID(rw);
327 : }
328 : }
329 2941 : continue;
330 : }
331 :
332 : /*
333 : * If we aren't allowing new workers, then immediately mark it for
334 : * termination; the next stanza will take care of cleaning it up.
335 : * Doing this ensures that any process waiting for the worker will get
336 : * awoken, even though the worker will never be allowed to run.
337 : */
338 2700 : if (!allow_new_workers)
339 8 : slot->terminate = true;
340 :
341 : /*
342 : * If the worker is marked for termination, we don't need to add it to
343 : * the registered workers list; we can just free the slot. However, if
344 : * bgw_notify_pid is set, the process that registered the worker may
345 : * need to know that we've processed the terminate request, so be sure
346 : * to signal it.
347 : */
348 2700 : if (slot->terminate)
349 8 : {
350 : int notify_pid;
351 :
352 : /*
353 : * We need a memory barrier here to make sure that the load of
354 : * bgw_notify_pid and the update of parallel_terminate_count
355 : * complete before the store to in_use.
356 : */
357 8 : notify_pid = slot->worker.bgw_notify_pid;
358 8 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
359 0 : BackgroundWorkerData->parallel_terminate_count++;
360 8 : slot->pid = 0;
361 :
362 8 : pg_memory_barrier();
363 8 : slot->in_use = false;
364 :
365 8 : if (notify_pid != 0)
366 8 : kill(notify_pid, SIGUSR1);
367 :
368 8 : continue;
369 : }
370 :
371 : /*
372 : * Copy the registration data into the registered workers list.
373 : */
374 2692 : rw = MemoryContextAllocExtended(PostmasterContext,
375 : sizeof(RegisteredBgWorker),
376 : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
377 2692 : if (rw == NULL)
378 : {
379 0 : ereport(LOG,
380 : (errcode(ERRCODE_OUT_OF_MEMORY),
381 : errmsg("out of memory")));
382 0 : return;
383 : }
384 :
385 : /*
386 : * Copy strings in a paranoid way. If shared memory is corrupted, the
387 : * source data might not even be NUL-terminated.
388 : */
389 2692 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
390 2692 : slot->worker.bgw_name, BGW_MAXLEN);
391 2692 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
392 2692 : slot->worker.bgw_type, BGW_MAXLEN);
393 2692 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
394 2692 : slot->worker.bgw_library_name, MAXPGPATH);
395 2692 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
396 2692 : slot->worker.bgw_function_name, BGW_MAXLEN);
397 :
398 : /*
399 : * Copy various fixed-size fields.
400 : *
401 : * flags, start_time, and restart_time are examined by the postmaster,
402 : * but nothing too bad will happen if they are corrupted. The
403 : * remaining fields will only be examined by the child process. It
404 : * might crash, but we won't.
405 : */
406 2692 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
407 2692 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
408 2692 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
409 2692 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
410 2692 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
411 :
412 : /*
413 : * Copy the PID to be notified about state changes, but only if the
414 : * postmaster knows about a backend with that PID. It isn't an error
415 : * if the postmaster doesn't know about the PID, because the backend
416 : * that requested the worker could have died (or been killed) just
417 : * after doing so. Nonetheless, at least until we get some experience
418 : * with how this plays out in the wild, log a message at a relative
419 : * high debug level.
420 : */
421 2692 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
422 2692 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
423 : {
424 0 : elog(DEBUG1, "worker notification PID %d is not valid",
425 : (int) rw->rw_worker.bgw_notify_pid);
426 0 : rw->rw_worker.bgw_notify_pid = 0;
427 : }
428 :
429 : /* Initialize postmaster bookkeeping. */
430 2692 : rw->rw_pid = 0;
431 2692 : rw->rw_crashed_at = 0;
432 2692 : rw->rw_shmem_slot = slotno;
433 2692 : rw->rw_terminate = false;
434 :
435 : /* Log it! */
436 2692 : ereport(DEBUG1,
437 : (errmsg_internal("registering background worker \"%s\"",
438 : rw->rw_worker.bgw_name)));
439 :
440 2692 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
441 : }
442 : }
443 :
444 : /*
445 : * Forget about a background worker that's no longer needed.
446 : *
447 : * NOTE: The entry is unlinked from BackgroundWorkerList. If the caller is
448 : * iterating through it, better use a mutable iterator!
449 : *
450 : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
451 : *
452 : * This function must be invoked only in the postmaster.
453 : */
454 : void
455 2679 : ForgetBackgroundWorker(RegisteredBgWorker *rw)
456 : {
457 : BackgroundWorkerSlot *slot;
458 :
459 : Assert(rw->rw_shmem_slot < max_worker_processes);
460 2679 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
461 : Assert(slot->in_use);
462 :
463 : /*
464 : * We need a memory barrier here to make sure that the update of
465 : * parallel_terminate_count completes before the store to in_use.
466 : */
467 2679 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
468 2008 : BackgroundWorkerData->parallel_terminate_count++;
469 :
470 2679 : pg_memory_barrier();
471 2679 : slot->in_use = false;
472 :
473 2679 : ereport(DEBUG1,
474 : (errmsg_internal("unregistering background worker \"%s\"",
475 : rw->rw_worker.bgw_name)));
476 :
477 2679 : dlist_delete(&rw->rw_lnode);
478 2679 : pfree(rw);
479 2679 : }
480 :
481 : /*
482 : * Report the PID of a newly-launched background worker in shared memory.
483 : *
484 : * This function should only be called from the postmaster.
485 : */
486 : void
487 3510 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
488 : {
489 : BackgroundWorkerSlot *slot;
490 :
491 : Assert(rw->rw_shmem_slot < max_worker_processes);
492 3510 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
493 3510 : slot->pid = rw->rw_pid;
494 :
495 3510 : if (rw->rw_worker.bgw_notify_pid != 0)
496 2692 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
497 3510 : }
498 :
499 : /*
500 : * Report that the PID of a background worker is now zero because a
501 : * previously-running background worker has exited.
502 : *
503 : * NOTE: The entry may be unlinked from BackgroundWorkerList. If the caller
504 : * is iterating through it, better use a mutable iterator!
505 : *
506 : * This function should only be called from the postmaster.
507 : */
508 : void
509 3187 : ReportBackgroundWorkerExit(RegisteredBgWorker *rw)
510 : {
511 : BackgroundWorkerSlot *slot;
512 : int notify_pid;
513 :
514 : Assert(rw->rw_shmem_slot < max_worker_processes);
515 3187 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
516 3187 : slot->pid = rw->rw_pid;
517 3187 : notify_pid = rw->rw_worker.bgw_notify_pid;
518 :
519 : /*
520 : * If this worker is slated for deregistration, do that before notifying
521 : * the process which started it. Otherwise, if that process tries to
522 : * reuse the slot immediately, it might not be available yet. In theory
523 : * that could happen anyway if the process checks slot->pid at just the
524 : * wrong moment, but this makes the window narrower.
525 : */
526 3187 : if (rw->rw_terminate ||
527 804 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
528 2679 : ForgetBackgroundWorker(rw);
529 :
530 3187 : if (notify_pid != 0)
531 2635 : kill(notify_pid, SIGUSR1);
532 3187 : }
533 :
534 : /*
535 : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
536 : *
537 : * This function should only be called from the postmaster.
538 : */
539 : void
540 344 : BackgroundWorkerStopNotifications(pid_t pid)
541 : {
542 : dlist_iter iter;
543 :
544 1079 : dlist_foreach(iter, &BackgroundWorkerList)
545 : {
546 : RegisteredBgWorker *rw;
547 :
548 735 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
549 735 : if (rw->rw_worker.bgw_notify_pid == pid)
550 46 : rw->rw_worker.bgw_notify_pid = 0;
551 : }
552 344 : }
553 :
554 : /*
555 : * Cancel any not-yet-started worker requests that have waiting processes.
556 : *
557 : * This is called during a normal ("smart" or "fast") database shutdown.
558 : * After this point, no new background workers will be started, so anything
559 : * that might be waiting for them needs to be kicked off its wait. We do
560 : * that by canceling the bgworker registration entirely, which is perhaps
561 : * overkill, but since we're shutting down it does not matter whether the
562 : * registration record sticks around.
563 : *
564 : * This function should only be called from the postmaster.
565 : */
566 : void
567 626 : ForgetUnstartedBackgroundWorkers(void)
568 : {
569 : dlist_mutable_iter iter;
570 :
571 1245 : dlist_foreach_modify(iter, &BackgroundWorkerList)
572 : {
573 : RegisteredBgWorker *rw;
574 : BackgroundWorkerSlot *slot;
575 :
576 619 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
577 : Assert(rw->rw_shmem_slot < max_worker_processes);
578 619 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
579 :
580 : /* If it's not yet started, and there's someone waiting ... */
581 619 : if (slot->pid == InvalidPid &&
582 57 : rw->rw_worker.bgw_notify_pid != 0)
583 : {
584 : /* ... then zap it, and notify the waiter */
585 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
586 :
587 0 : ForgetBackgroundWorker(rw);
588 0 : if (notify_pid != 0)
589 0 : kill(notify_pid, SIGUSR1);
590 : }
591 : }
592 626 : }
593 :
594 : /*
595 : * Reset background worker crash state.
596 : *
597 : * We assume that, after a crash-and-restart cycle, background workers without
598 : * the never-restart flag should be restarted immediately, instead of waiting
599 : * for bgw_restart_time to elapse. On the other hand, workers with that flag
600 : * should be forgotten immediately, since we won't ever restart them.
601 : *
602 : * This function should only be called from the postmaster.
603 : */
604 : void
605 5 : ResetBackgroundWorkerCrashTimes(void)
606 : {
607 : dlist_mutable_iter iter;
608 :
609 10 : dlist_foreach_modify(iter, &BackgroundWorkerList)
610 : {
611 : RegisteredBgWorker *rw;
612 :
613 5 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
614 :
615 5 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
616 : {
617 : /*
618 : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
619 : * the crash, so forget about them. (If we wait until after the
620 : * crash to forget about them, and they are parallel workers,
621 : * parallel_terminate_count will get incremented after we've
622 : * already zeroed parallel_register_count, which would be bad.)
623 : */
624 0 : ForgetBackgroundWorker(rw);
625 : }
626 : else
627 : {
628 : /*
629 : * The accounting which we do via parallel_register_count and
630 : * parallel_terminate_count would get messed up if a worker marked
631 : * parallel could survive a crash and restart cycle. All such
632 : * workers should be marked BGW_NEVER_RESTART, and thus control
633 : * should never reach this branch.
634 : */
635 : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
636 :
637 : /*
638 : * Allow this worker to be restarted immediately after we finish
639 : * resetting.
640 : */
641 5 : rw->rw_crashed_at = 0;
642 5 : rw->rw_pid = 0;
643 :
644 : /*
645 : * If there was anyone waiting for it, they're history.
646 : */
647 5 : rw->rw_worker.bgw_notify_pid = 0;
648 : }
649 : }
650 5 : }
651 :
652 : /*
653 : * Complain about the BackgroundWorker definition using error level elevel.
654 : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
655 : * which case we won't return at all in the not-OK case).
656 : */
657 : static bool
658 3502 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
659 : {
660 : /* sanity check for flags */
661 :
662 : /*
663 : * We used to support workers not connected to shared memory, but don't
664 : * anymore. Thus this is a required flag now. We're not removing the flag
665 : * for compatibility reasons and because the flag still provides some
666 : * signal when reading code.
667 : */
668 3502 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
669 : {
670 0 : ereport(elevel,
671 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
672 : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
673 : worker->bgw_name)));
674 0 : return false;
675 : }
676 :
677 3502 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
678 : {
679 3493 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
680 : {
681 0 : ereport(elevel,
682 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
683 : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
684 : worker->bgw_name)));
685 0 : return false;
686 : }
687 :
688 : /* XXX other checks? */
689 : }
690 :
691 : /* Interruptible workers require a database connection */
692 3502 : if ((worker->bgw_flags & BGWORKER_INTERRUPTIBLE) &&
693 4 : !(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
694 : {
695 0 : ereport(elevel,
696 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
697 : errmsg("background worker \"%s\": cannot make background workers interruptible without database access",
698 : worker->bgw_name)));
699 0 : return false;
700 : }
701 :
702 3502 : if ((worker->bgw_restart_time < 0 &&
703 2567 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
704 3502 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
705 : {
706 0 : ereport(elevel,
707 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
708 : errmsg("background worker \"%s\": invalid restart interval",
709 : worker->bgw_name)));
710 0 : return false;
711 : }
712 :
713 : /*
714 : * Parallel workers may not be configured for restart, because the
715 : * parallel_register_count/parallel_terminate_count accounting can't
716 : * handle parallel workers lasting through a crash-and-restart cycle.
717 : */
718 3502 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
719 935 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
720 : {
721 0 : ereport(elevel,
722 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
723 : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
724 : worker->bgw_name)));
725 0 : return false;
726 : }
727 :
728 : /*
729 : * If bgw_type is not filled in, use bgw_name.
730 : */
731 3502 : if (strcmp(worker->bgw_type, "") == 0)
732 0 : strcpy(worker->bgw_type, worker->bgw_name);
733 :
734 3502 : return true;
735 : }
736 :
737 : /*
738 : * Main entry point for background worker processes.
739 : */
740 : void
741 3194 : BackgroundWorkerMain(const void *startup_data, size_t startup_data_len)
742 : {
743 : sigjmp_buf local_sigjmp_buf;
744 : BackgroundWorker *worker;
745 : bgworker_main_type entrypt;
746 :
747 3194 : if (startup_data == NULL)
748 0 : elog(FATAL, "unable to find bgworker entry");
749 : Assert(startup_data_len == sizeof(BackgroundWorker));
750 3194 : worker = MemoryContextAlloc(TopMemoryContext, sizeof(BackgroundWorker));
751 3194 : memcpy(worker, startup_data, sizeof(BackgroundWorker));
752 :
753 : /*
754 : * Now that we're done reading the startup data, release postmaster's
755 : * working memory context.
756 : */
757 3194 : if (PostmasterContext)
758 : {
759 3194 : MemoryContextDelete(PostmasterContext);
760 3194 : PostmasterContext = NULL;
761 : }
762 :
763 3194 : MyBgworkerEntry = worker;
764 3194 : init_ps_display(worker->bgw_name);
765 :
766 : Assert(GetProcessingMode() == InitProcessing);
767 :
768 : /* Apply PostAuthDelay */
769 3194 : if (PostAuthDelay > 0)
770 0 : pg_usleep(PostAuthDelay * 1000000L);
771 :
772 : /*
773 : * Set up signal handlers.
774 : */
775 3194 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
776 : {
777 : /*
778 : * SIGINT is used to signal canceling the current action
779 : */
780 3185 : pqsignal(SIGINT, StatementCancelHandler);
781 3185 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
782 3185 : pqsignal(SIGFPE, FloatExceptionHandler);
783 :
784 : /* XXX Any other handlers needed here? */
785 : }
786 : else
787 : {
788 9 : pqsignal(SIGINT, SIG_IGN);
789 9 : pqsignal(SIGUSR1, SIG_IGN);
790 9 : pqsignal(SIGFPE, SIG_IGN);
791 : }
792 3194 : pqsignal(SIGTERM, die);
793 : /* SIGQUIT handler was already set up by InitPostmasterChild */
794 3194 : pqsignal(SIGHUP, SIG_IGN);
795 :
796 3194 : InitializeTimeouts(); /* establishes SIGALRM handler */
797 :
798 3194 : pqsignal(SIGPIPE, SIG_IGN);
799 3194 : pqsignal(SIGUSR2, SIG_IGN);
800 3194 : pqsignal(SIGCHLD, SIG_DFL);
801 :
802 : /*
803 : * If an exception is encountered, processing resumes here.
804 : *
805 : * We just need to clean up, report the error, and go away.
806 : */
807 3194 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
808 : {
809 : /* Since not using PG_TRY, must reset error stack by hand */
810 181 : error_context_stack = NULL;
811 :
812 : /* Prevent interrupts while cleaning up */
813 181 : HOLD_INTERRUPTS();
814 :
815 : /*
816 : * sigsetjmp will have blocked all signals, but we may need to accept
817 : * signals while communicating with our parallel leader. Once we've
818 : * done HOLD_INTERRUPTS() it should be safe to unblock signals.
819 : */
820 181 : BackgroundWorkerUnblockSignals();
821 :
822 : /* Report the error to the parallel leader and the server log */
823 181 : EmitErrorReport();
824 :
825 : /*
826 : * Do we need more cleanup here? For shmem-connected bgworkers, we
827 : * will call InitProcess below, which will install ProcKill as exit
828 : * callback. That will take care of releasing locks, etc.
829 : */
830 :
831 : /* and go away */
832 181 : proc_exit(1);
833 : }
834 :
835 : /* We can now handle ereport(ERROR) */
836 3194 : PG_exception_stack = &local_sigjmp_buf;
837 :
838 : /*
839 : * Create a per-backend PGPROC struct in shared memory. We must do this
840 : * before we can use LWLocks or access any shared memory.
841 : */
842 3194 : InitProcess();
843 :
844 : /*
845 : * Early initialization.
846 : */
847 3194 : BaseInit();
848 :
849 : /*
850 : * Look up the entry point function, loading its library if necessary.
851 : */
852 6388 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
853 3194 : worker->bgw_function_name);
854 :
855 : /*
856 : * Note that in normal processes, we would call InitPostgres here. For a
857 : * worker, however, we don't know what database to connect to, yet; so we
858 : * need to wait until the user code does it via
859 : * BackgroundWorkerInitializeConnection().
860 : */
861 :
862 : /*
863 : * Now invoke the user-defined worker code
864 : */
865 3194 : entrypt(worker->bgw_main_arg);
866 :
867 : /* ... and if it returns, we're done */
868 2037 : proc_exit(0);
869 : }
870 :
871 : /*
872 : * Connect background worker to a database.
873 : */
874 : void
875 511 : BackgroundWorkerInitializeConnection(const char *dbname, const char *username, uint32 flags)
876 : {
877 511 : BackgroundWorker *worker = MyBgworkerEntry;
878 511 : uint32 init_flags = 0; /* never honor session_preload_libraries */
879 :
880 : /* ignore datallowconn and ACL_CONNECT? */
881 511 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
882 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
883 : /* ignore rolcanlogin? */
884 511 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
885 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
886 :
887 : /* XXX is this the right errcode? */
888 511 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
889 0 : ereport(FATAL,
890 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
891 : errmsg("database connection requirement not indicated during registration")));
892 :
893 511 : InitPostgres(dbname, InvalidOid, /* database to connect to */
894 : username, InvalidOid, /* role to connect as */
895 : init_flags,
896 : NULL); /* no out_dbname */
897 :
898 : /* it had better not gotten out of "init" mode yet */
899 511 : if (!IsInitProcessingMode())
900 0 : ereport(ERROR,
901 : (errmsg("invalid processing mode in background worker")));
902 511 : SetProcessingMode(NormalProcessing);
903 511 : }
904 :
905 : /*
906 : * Connect background worker to a database using OIDs.
907 : */
908 : void
909 2674 : BackgroundWorkerInitializeConnectionByOid(Oid dboid, Oid useroid, uint32 flags)
910 : {
911 2674 : BackgroundWorker *worker = MyBgworkerEntry;
912 2674 : uint32 init_flags = 0; /* never honor session_preload_libraries */
913 :
914 : /* ignore datallowconn and ACL_CONNECT? */
915 2674 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
916 2028 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
917 : /* ignore rolcanlogin? */
918 2674 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
919 2008 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
920 :
921 : /* XXX is this the right errcode? */
922 2674 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
923 0 : ereport(FATAL,
924 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
925 : errmsg("database connection requirement not indicated during registration")));
926 :
927 2674 : InitPostgres(NULL, dboid, /* database to connect to */
928 : NULL, useroid, /* role to connect as */
929 : init_flags,
930 : NULL); /* no out_dbname */
931 :
932 : /* it had better not gotten out of "init" mode yet */
933 2670 : if (!IsInitProcessingMode())
934 0 : ereport(ERROR,
935 : (errmsg("invalid processing mode in background worker")));
936 2670 : SetProcessingMode(NormalProcessing);
937 2670 : }
938 :
939 : /*
940 : * Block/unblock signals in a background worker
941 : */
942 : void
943 0 : BackgroundWorkerBlockSignals(void)
944 : {
945 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
946 0 : }
947 :
948 : void
949 3375 : BackgroundWorkerUnblockSignals(void)
950 : {
951 3375 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
952 3375 : }
953 :
954 : /*
955 : * Register a new static background worker.
956 : *
957 : * This can only be called directly from postmaster or in the _PG_init
958 : * function of a module library that's loaded by shared_preload_libraries;
959 : * otherwise it will have no effect.
960 : */
961 : void
962 938 : RegisterBackgroundWorker(BackgroundWorker *worker)
963 : {
964 : RegisteredBgWorker *rw;
965 : static int numworkers = 0;
966 :
967 : /*
968 : * Static background workers can only be registered in the postmaster
969 : * process.
970 : */
971 938 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
972 : {
973 : /*
974 : * In EXEC_BACKEND or single-user mode, we process
975 : * shared_preload_libraries in backend processes too. We cannot
976 : * register static background workers at that stage, but many
977 : * libraries' _PG_init() functions don't distinguish whether they're
978 : * being loaded in the postmaster or in a backend, they just check
979 : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
980 : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
981 : * the background workers should already have been registered when the
982 : * library was loaded in postmaster.
983 : */
984 0 : if (process_shared_preload_libraries_in_progress)
985 0 : return;
986 0 : ereport(LOG,
987 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
988 : errmsg("background worker \"%s\": must be registered in \"shared_preload_libraries\"",
989 : worker->bgw_name)));
990 0 : return;
991 : }
992 :
993 : /*
994 : * Cannot register static background workers after calling
995 : * BackgroundWorkerShmemInit().
996 : */
997 938 : if (BackgroundWorkerData != NULL)
998 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
999 : worker->bgw_name);
1000 :
1001 938 : ereport(DEBUG1,
1002 : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
1003 :
1004 938 : if (!SanityCheckBackgroundWorker(worker, LOG))
1005 0 : return;
1006 :
1007 938 : if (worker->bgw_notify_pid != 0)
1008 : {
1009 0 : ereport(LOG,
1010 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1011 : errmsg("background worker \"%s\": only dynamic background workers can request notification",
1012 : worker->bgw_name)));
1013 0 : return;
1014 : }
1015 :
1016 : /*
1017 : * Enforce maximum number of workers. Note this is overly restrictive: we
1018 : * could allow more non-shmem-connected workers, because these don't count
1019 : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
1020 : * important to relax this restriction.
1021 : */
1022 938 : if (++numworkers > max_worker_processes)
1023 : {
1024 0 : ereport(LOG,
1025 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
1026 : errmsg("too many background workers"),
1027 : errdetail_plural("Up to %d background worker can be registered with the current settings.",
1028 : "Up to %d background workers can be registered with the current settings.",
1029 : max_worker_processes,
1030 : max_worker_processes),
1031 : errhint("Consider increasing the configuration parameter \"%s\".", "max_worker_processes")));
1032 0 : return;
1033 : }
1034 :
1035 : /*
1036 : * Copy the registration data into the registered workers list.
1037 : */
1038 938 : rw = MemoryContextAllocExtended(PostmasterContext,
1039 : sizeof(RegisteredBgWorker),
1040 : MCXT_ALLOC_NO_OOM);
1041 938 : if (rw == NULL)
1042 : {
1043 0 : ereport(LOG,
1044 : (errcode(ERRCODE_OUT_OF_MEMORY),
1045 : errmsg("out of memory")));
1046 0 : return;
1047 : }
1048 :
1049 938 : rw->rw_worker = *worker;
1050 938 : rw->rw_pid = 0;
1051 938 : rw->rw_crashed_at = 0;
1052 938 : rw->rw_terminate = false;
1053 :
1054 938 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
1055 : }
1056 :
1057 : /*
1058 : * Register a new background worker from a regular backend.
1059 : *
1060 : * Returns true on success and false on failure. Failure typically indicates
1061 : * that no background worker slots are currently available.
1062 : *
1063 : * If handle != NULL, we'll set *handle to a pointer that can subsequently
1064 : * be used as an argument to GetBackgroundWorkerPid(). The caller can
1065 : * free this pointer using pfree(), if desired.
1066 : */
1067 : bool
1068 2564 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
1069 : BackgroundWorkerHandle **handle)
1070 : {
1071 : int slotno;
1072 2564 : bool success = false;
1073 : bool parallel;
1074 2564 : uint64 generation = 0;
1075 :
1076 : /*
1077 : * We can't register dynamic background workers from the postmaster. If
1078 : * this is a standalone backend, we're the only process and can't start
1079 : * any more. In a multi-process environment, it might be theoretically
1080 : * possible, but we don't currently support it due to locking
1081 : * considerations; see comments on the BackgroundWorkerSlot data
1082 : * structure.
1083 : */
1084 2564 : if (!IsUnderPostmaster)
1085 0 : return false;
1086 :
1087 2564 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1088 0 : return false;
1089 :
1090 2564 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1091 :
1092 2564 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1093 :
1094 : /*
1095 : * If this is a parallel worker, check whether there are already too many
1096 : * parallel workers; if so, don't register another one. Our view of
1097 : * parallel_terminate_count may be slightly stale, but that doesn't really
1098 : * matter: we would have gotten the same result if we'd arrived here
1099 : * slightly earlier anyway. There's no help for it, either, since the
1100 : * postmaster must not take locks; a memory barrier wouldn't guarantee
1101 : * anything useful.
1102 : */
1103 2564 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1104 2030 : BackgroundWorkerData->parallel_terminate_count) >=
1105 : max_parallel_workers)
1106 : {
1107 : Assert(BackgroundWorkerData->parallel_register_count -
1108 : BackgroundWorkerData->parallel_terminate_count <=
1109 : MAX_PARALLEL_WORKER_LIMIT);
1110 13 : LWLockRelease(BackgroundWorkerLock);
1111 13 : return false;
1112 : }
1113 :
1114 : /*
1115 : * Look for an unused slot. If we find one, grab it.
1116 : */
1117 8072 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1118 : {
1119 8063 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1120 :
1121 8063 : if (!slot->in_use)
1122 : {
1123 2542 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1124 2542 : slot->pid = InvalidPid; /* indicates not started yet */
1125 2542 : slot->generation++;
1126 2542 : slot->terminate = false;
1127 2542 : generation = slot->generation;
1128 2542 : if (parallel)
1129 2008 : BackgroundWorkerData->parallel_register_count++;
1130 :
1131 : /*
1132 : * Make sure postmaster doesn't see the slot as in use before it
1133 : * sees the new contents.
1134 : */
1135 2542 : pg_write_barrier();
1136 :
1137 2542 : slot->in_use = true;
1138 2542 : success = true;
1139 2542 : break;
1140 : }
1141 : }
1142 :
1143 2551 : LWLockRelease(BackgroundWorkerLock);
1144 :
1145 : /* If we found a slot, tell the postmaster to notice the change. */
1146 2551 : if (success)
1147 2542 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1148 :
1149 : /*
1150 : * If we found a slot and the user has provided a handle, initialize it.
1151 : */
1152 2551 : if (success && handle)
1153 : {
1154 2542 : *handle = palloc_object(BackgroundWorkerHandle);
1155 2542 : (*handle)->slot = slotno;
1156 2542 : (*handle)->generation = generation;
1157 : }
1158 :
1159 2551 : return success;
1160 : }
1161 :
1162 : /*
1163 : * Get the PID of a dynamically-registered background worker.
1164 : *
1165 : * If the worker is determined to be running, the return value will be
1166 : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1167 : * postmaster has not yet attempted to start the worker, the return value will
1168 : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1169 : *
1170 : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1171 : * (because it is configured for automatic restart and exited non-zero),
1172 : * or that the worker is permanently stopped (because it exited with exit
1173 : * code 0, or was not configured for automatic restart), or even that the
1174 : * worker was unregistered without ever starting (either because startup
1175 : * failed and the worker is not configured for automatic restart, or because
1176 : * TerminateBackgroundWorker was used before the worker was successfully
1177 : * started).
1178 : */
1179 : BgwHandleStatus
1180 10894622 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1181 : {
1182 : BackgroundWorkerSlot *slot;
1183 : pid_t pid;
1184 :
1185 : Assert(handle->slot < max_worker_processes);
1186 10894622 : slot = &BackgroundWorkerData->slot[handle->slot];
1187 :
1188 : /*
1189 : * We could probably arrange to synchronize access to data using memory
1190 : * barriers only, but for now, let's just keep it simple and grab the
1191 : * lock. It seems unlikely that there will be enough traffic here to
1192 : * result in meaningful contention.
1193 : */
1194 10894622 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1195 :
1196 : /*
1197 : * The generation number can't be concurrently changed while we hold the
1198 : * lock. The pid, which is updated by the postmaster, can change at any
1199 : * time, but we assume such changes are atomic. So the value we read
1200 : * won't be garbage, but it might be out of date by the time the caller
1201 : * examines it (but that's unavoidable anyway).
1202 : *
1203 : * The in_use flag could be in the process of changing from true to false,
1204 : * but if it is already false then it can't change further.
1205 : */
1206 10894622 : if (handle->generation != slot->generation || !slot->in_use)
1207 2030 : pid = 0;
1208 : else
1209 10892592 : pid = slot->pid;
1210 :
1211 : /* All done. */
1212 10894622 : LWLockRelease(BackgroundWorkerLock);
1213 :
1214 10894622 : if (pid == 0)
1215 2030 : return BGWH_STOPPED;
1216 10892592 : else if (pid == InvalidPid)
1217 809319 : return BGWH_NOT_YET_STARTED;
1218 10083273 : *pidp = pid;
1219 10083273 : return BGWH_STARTED;
1220 : }
1221 :
1222 : /*
1223 : * Wait for a background worker to start up.
1224 : *
1225 : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1226 : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1227 : * returned. However, if the postmaster has died, we give up and return
1228 : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1229 : * take place.
1230 : *
1231 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1232 : * else we will not be awoken promptly when the worker's state changes.
1233 : */
1234 : BgwHandleStatus
1235 32 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1236 : {
1237 : BgwHandleStatus status;
1238 : int rc;
1239 :
1240 : for (;;)
1241 23 : {
1242 : pid_t pid;
1243 :
1244 55 : CHECK_FOR_INTERRUPTS();
1245 :
1246 55 : status = GetBackgroundWorkerPid(handle, &pid);
1247 55 : if (status == BGWH_STARTED)
1248 32 : *pidp = pid;
1249 55 : if (status != BGWH_NOT_YET_STARTED)
1250 32 : break;
1251 :
1252 23 : rc = WaitLatch(MyLatch,
1253 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1254 : WAIT_EVENT_BGWORKER_STARTUP);
1255 :
1256 23 : if (rc & WL_POSTMASTER_DEATH)
1257 : {
1258 0 : status = BGWH_POSTMASTER_DIED;
1259 0 : break;
1260 : }
1261 :
1262 23 : ResetLatch(MyLatch);
1263 : }
1264 :
1265 32 : return status;
1266 : }
1267 :
1268 : /*
1269 : * Wait for a background worker to stop.
1270 : *
1271 : * If the worker hasn't yet started, or is running, we wait for it to stop
1272 : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1273 : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1274 : * notifies us when a worker's state changes.
1275 : *
1276 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1277 : * else we will not be awoken promptly when the worker's state changes.
1278 : */
1279 : BgwHandleStatus
1280 2031 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1281 : {
1282 : BgwHandleStatus status;
1283 : int rc;
1284 :
1285 : for (;;)
1286 2390 : {
1287 : pid_t pid;
1288 :
1289 4421 : CHECK_FOR_INTERRUPTS();
1290 :
1291 4420 : status = GetBackgroundWorkerPid(handle, &pid);
1292 4420 : if (status == BGWH_STOPPED)
1293 2030 : break;
1294 :
1295 2390 : rc = WaitLatch(MyLatch,
1296 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1297 : WAIT_EVENT_BGWORKER_SHUTDOWN);
1298 :
1299 2390 : if (rc & WL_POSTMASTER_DEATH)
1300 : {
1301 0 : status = BGWH_POSTMASTER_DIED;
1302 0 : break;
1303 : }
1304 :
1305 2390 : ResetLatch(MyLatch);
1306 : }
1307 :
1308 2030 : return status;
1309 : }
1310 :
1311 : /*
1312 : * Instruct the postmaster to terminate a background worker.
1313 : *
1314 : * Note that it's safe to do this without regard to whether the worker is
1315 : * still running, or even if the worker may already have exited and been
1316 : * unregistered.
1317 : */
1318 : void
1319 10 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1320 : {
1321 : BackgroundWorkerSlot *slot;
1322 10 : bool signal_postmaster = false;
1323 :
1324 : Assert(handle->slot < max_worker_processes);
1325 10 : slot = &BackgroundWorkerData->slot[handle->slot];
1326 :
1327 : /* Set terminate flag in shared memory, unless slot has been reused. */
1328 10 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1329 10 : if (handle->generation == slot->generation)
1330 : {
1331 10 : slot->terminate = true;
1332 10 : signal_postmaster = true;
1333 : }
1334 10 : LWLockRelease(BackgroundWorkerLock);
1335 :
1336 : /* Make sure the postmaster notices the change to shared memory. */
1337 10 : if (signal_postmaster)
1338 10 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1339 10 : }
1340 :
1341 : /*
1342 : * Look up (and possibly load) a bgworker entry point function.
1343 : *
1344 : * For functions contained in the core code, we use library name "postgres"
1345 : * and consult the InternalBGWorkers array. External functions are
1346 : * looked up, and loaded if necessary, using load_external_function().
1347 : *
1348 : * The point of this is to pass function names as strings across process
1349 : * boundaries. We can't pass actual function addresses because of the
1350 : * possibility that the function has been loaded at a different address
1351 : * in a different process. This is obviously a hazard for functions in
1352 : * loadable libraries, but it can happen even for functions in the core code
1353 : * on platforms using EXEC_BACKEND (e.g., Windows).
1354 : *
1355 : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1356 : * in favor of applying load_external_function() for core functions too;
1357 : * but that raises portability issues that are not worth addressing now.
1358 : */
1359 : static bgworker_main_type
1360 3194 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1361 : {
1362 : /*
1363 : * If the function is to be loaded from postgres itself, search the
1364 : * InternalBGWorkers array.
1365 : */
1366 3194 : if (strcmp(libraryname, "postgres") == 0)
1367 : {
1368 : int i;
1369 :
1370 11212 : for (i = 0; i < lengthof(InternalBGWorkers); i++)
1371 : {
1372 11212 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1373 3176 : return InternalBGWorkers[i].fn_addr;
1374 : }
1375 :
1376 : /* We can only reach this by programming error. */
1377 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1378 : }
1379 :
1380 : /* Otherwise load from external library. */
1381 18 : return (bgworker_main_type)
1382 18 : load_external_function(libraryname, funcname, true, NULL);
1383 : }
1384 :
1385 : /*
1386 : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1387 : * not a valid background worker.
1388 : *
1389 : * The return value is in static memory belonging to this function, so it has
1390 : * to be used before calling this function again. This is so that the caller
1391 : * doesn't have to worry about the background worker locking protocol.
1392 : */
1393 : const char *
1394 781 : GetBackgroundWorkerTypeByPid(pid_t pid)
1395 : {
1396 : int slotno;
1397 781 : bool found = false;
1398 : static char result[BGW_MAXLEN];
1399 :
1400 781 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1401 :
1402 923 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1403 : {
1404 923 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1405 :
1406 923 : if (slot->pid > 0 && slot->pid == pid)
1407 : {
1408 781 : strcpy(result, slot->worker.bgw_type);
1409 781 : found = true;
1410 781 : break;
1411 : }
1412 : }
1413 :
1414 781 : LWLockRelease(BackgroundWorkerLock);
1415 :
1416 781 : if (!found)
1417 0 : return NULL;
1418 :
1419 781 : return result;
1420 : }
1421 :
1422 : /*
1423 : * Terminate all background workers connected to the given database, if the
1424 : * workers can be interrupted.
1425 : */
1426 : void
1427 7 : TerminateBackgroundWorkersForDatabase(Oid databaseId)
1428 : {
1429 7 : bool signal_postmaster = false;
1430 :
1431 7 : elog(DEBUG1, "attempting worker termination for database %u",
1432 : databaseId);
1433 :
1434 7 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1435 :
1436 : /*
1437 : * Iterate through slots, looking for workers connected to the given
1438 : * database.
1439 : */
1440 63 : for (int slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1441 : {
1442 56 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1443 :
1444 56 : if (slot->in_use &&
1445 14 : (slot->worker.bgw_flags & BGWORKER_INTERRUPTIBLE))
1446 : {
1447 4 : PGPROC *proc = BackendPidGetProc(slot->pid);
1448 :
1449 4 : if (proc && proc->databaseId == databaseId)
1450 : {
1451 4 : slot->terminate = true;
1452 4 : signal_postmaster = true;
1453 :
1454 4 : elog(DEBUG1, "termination requested for worker (PID %d) on database %u",
1455 : (int) slot->pid, databaseId);
1456 : }
1457 : }
1458 : }
1459 :
1460 7 : LWLockRelease(BackgroundWorkerLock);
1461 :
1462 : /* Make sure the postmaster notices the change to shared memory. */
1463 7 : if (signal_postmaster)
1464 4 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1465 7 : }
|
