Line data Source code
1 : /*--------------------------------------------------------------------
2 : * bgworker.c
3 : * POSTGRES pluggable background workers implementation
4 : *
5 : * Portions Copyright (c) 1996-2024, 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 :
138 : /* Private functions. */
139 : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
140 :
141 :
142 : /*
143 : * Calculate shared memory needed.
144 : */
145 : Size
146 5240 : BackgroundWorkerShmemSize(void)
147 : {
148 : Size size;
149 :
150 : /* Array of workers is variably sized. */
151 5240 : size = offsetof(BackgroundWorkerArray, slot);
152 5240 : size = add_size(size, mul_size(max_worker_processes,
153 : sizeof(BackgroundWorkerSlot)));
154 :
155 5240 : return size;
156 : }
157 :
158 : /*
159 : * Initialize shared memory.
160 : */
161 : void
162 1828 : BackgroundWorkerShmemInit(void)
163 : {
164 : bool found;
165 :
166 1828 : BackgroundWorkerData = ShmemInitStruct("Background Worker Data",
167 : BackgroundWorkerShmemSize(),
168 : &found);
169 1828 : if (!IsUnderPostmaster)
170 : {
171 : dlist_iter iter;
172 1828 : int slotno = 0;
173 :
174 1828 : BackgroundWorkerData->total_slots = max_worker_processes;
175 1828 : BackgroundWorkerData->parallel_register_count = 0;
176 1828 : BackgroundWorkerData->parallel_terminate_count = 0;
177 :
178 : /*
179 : * Copy contents of worker list into shared memory. Record the shared
180 : * memory slot assigned to each worker. This ensures a 1-to-1
181 : * correspondence between the postmaster's private list and the array
182 : * in shared memory.
183 : */
184 3266 : dlist_foreach(iter, &BackgroundWorkerList)
185 : {
186 1438 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
187 : RegisteredBgWorker *rw;
188 :
189 1438 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
190 : Assert(slotno < max_worker_processes);
191 1438 : slot->in_use = true;
192 1438 : slot->terminate = false;
193 1438 : slot->pid = InvalidPid;
194 1438 : slot->generation = 0;
195 1438 : rw->rw_shmem_slot = slotno;
196 1438 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
197 1438 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
198 1438 : ++slotno;
199 : }
200 :
201 : /*
202 : * Mark any remaining slots as not in use.
203 : */
204 15038 : while (slotno < max_worker_processes)
205 : {
206 13210 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
207 :
208 13210 : slot->in_use = false;
209 13210 : ++slotno;
210 : }
211 : }
212 : else
213 : Assert(found);
214 1828 : }
215 :
216 : /*
217 : * Search the postmaster's backend-private list of RegisteredBgWorker objects
218 : * for the one that maps to the given slot number.
219 : */
220 : static RegisteredBgWorker *
221 6606 : FindRegisteredWorkerBySlotNumber(int slotno)
222 : {
223 : dlist_iter iter;
224 :
225 15924 : dlist_foreach(iter, &BackgroundWorkerList)
226 : {
227 : RegisteredBgWorker *rw;
228 :
229 12440 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
230 12440 : if (rw->rw_shmem_slot == slotno)
231 3122 : return rw;
232 : }
233 :
234 3484 : return NULL;
235 : }
236 :
237 : /*
238 : * Notice changes to shared memory made by other backends.
239 : * Accept new worker requests only if allow_new_workers is true.
240 : *
241 : * This code runs in the postmaster, so we must be very careful not to assume
242 : * that shared memory contents are sane. Otherwise, a rogue backend could
243 : * take out the postmaster.
244 : */
245 : void
246 1998 : BackgroundWorkerStateChange(bool allow_new_workers)
247 : {
248 : int slotno;
249 :
250 : /*
251 : * The total number of slots stored in shared memory should match our
252 : * notion of max_worker_processes. If it does not, something is very
253 : * wrong. Further down, we always refer to this value as
254 : * max_worker_processes, in case shared memory gets corrupted while we're
255 : * looping.
256 : */
257 1998 : if (max_worker_processes != BackgroundWorkerData->total_slots)
258 : {
259 0 : ereport(LOG,
260 : (errmsg("inconsistent background worker state (\"max_worker_processes\"=%d, total slots=%d)",
261 : max_worker_processes,
262 : BackgroundWorkerData->total_slots)));
263 0 : return;
264 : }
265 :
266 : /*
267 : * Iterate through slots, looking for newly-registered workers or workers
268 : * who must die.
269 : */
270 18270 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
271 : {
272 16272 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
273 : RegisteredBgWorker *rw;
274 :
275 16272 : if (!slot->in_use)
276 9666 : continue;
277 :
278 : /*
279 : * Make sure we don't see the in_use flag before the updated slot
280 : * contents.
281 : */
282 6606 : pg_read_barrier();
283 :
284 : /* See whether we already know about this worker. */
285 6606 : rw = FindRegisteredWorkerBySlotNumber(slotno);
286 6606 : if (rw != NULL)
287 : {
288 : /*
289 : * In general, the worker data can't change after it's initially
290 : * registered. However, someone can set the terminate flag.
291 : */
292 3122 : if (slot->terminate && !rw->rw_terminate)
293 : {
294 12 : rw->rw_terminate = true;
295 12 : if (rw->rw_pid != 0)
296 12 : kill(rw->rw_pid, SIGTERM);
297 : else
298 : {
299 : /* Report never-started, now-terminated worker as dead. */
300 0 : ReportBackgroundWorkerPID(rw);
301 : }
302 : }
303 3122 : continue;
304 : }
305 :
306 : /*
307 : * If we aren't allowing new workers, then immediately mark it for
308 : * termination; the next stanza will take care of cleaning it up.
309 : * Doing this ensures that any process waiting for the worker will get
310 : * awoken, even though the worker will never be allowed to run.
311 : */
312 3484 : if (!allow_new_workers)
313 0 : slot->terminate = true;
314 :
315 : /*
316 : * If the worker is marked for termination, we don't need to add it to
317 : * the registered workers list; we can just free the slot. However, if
318 : * bgw_notify_pid is set, the process that registered the worker may
319 : * need to know that we've processed the terminate request, so be sure
320 : * to signal it.
321 : */
322 3484 : if (slot->terminate)
323 : {
324 : int notify_pid;
325 :
326 : /*
327 : * We need a memory barrier here to make sure that the load of
328 : * bgw_notify_pid and the update of parallel_terminate_count
329 : * complete before the store to in_use.
330 : */
331 0 : notify_pid = slot->worker.bgw_notify_pid;
332 0 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
333 0 : BackgroundWorkerData->parallel_terminate_count++;
334 0 : slot->pid = 0;
335 :
336 0 : pg_memory_barrier();
337 0 : slot->in_use = false;
338 :
339 0 : if (notify_pid != 0)
340 0 : kill(notify_pid, SIGUSR1);
341 :
342 0 : continue;
343 : }
344 :
345 : /*
346 : * Copy the registration data into the registered workers list.
347 : */
348 3484 : rw = MemoryContextAllocExtended(PostmasterContext,
349 : sizeof(RegisteredBgWorker),
350 : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
351 3484 : if (rw == NULL)
352 : {
353 0 : ereport(LOG,
354 : (errcode(ERRCODE_OUT_OF_MEMORY),
355 : errmsg("out of memory")));
356 0 : return;
357 : }
358 :
359 : /*
360 : * Copy strings in a paranoid way. If shared memory is corrupted, the
361 : * source data might not even be NUL-terminated.
362 : */
363 3484 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
364 3484 : slot->worker.bgw_name, BGW_MAXLEN);
365 3484 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
366 3484 : slot->worker.bgw_type, BGW_MAXLEN);
367 3484 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
368 3484 : slot->worker.bgw_library_name, MAXPGPATH);
369 3484 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
370 3484 : slot->worker.bgw_function_name, BGW_MAXLEN);
371 :
372 : /*
373 : * Copy various fixed-size fields.
374 : *
375 : * flags, start_time, and restart_time are examined by the postmaster,
376 : * but nothing too bad will happen if they are corrupted. The
377 : * remaining fields will only be examined by the child process. It
378 : * might crash, but we won't.
379 : */
380 3484 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
381 3484 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
382 3484 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
383 3484 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
384 3484 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
385 :
386 : /*
387 : * Copy the PID to be notified about state changes, but only if the
388 : * postmaster knows about a backend with that PID. It isn't an error
389 : * if the postmaster doesn't know about the PID, because the backend
390 : * that requested the worker could have died (or been killed) just
391 : * after doing so. Nonetheless, at least until we get some experience
392 : * with how this plays out in the wild, log a message at a relative
393 : * high debug level.
394 : */
395 3484 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
396 3484 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
397 : {
398 0 : elog(DEBUG1, "worker notification PID %d is not valid",
399 : (int) rw->rw_worker.bgw_notify_pid);
400 0 : rw->rw_worker.bgw_notify_pid = 0;
401 : }
402 :
403 : /* Initialize postmaster bookkeeping. */
404 3484 : rw->rw_pid = 0;
405 3484 : rw->rw_crashed_at = 0;
406 3484 : rw->rw_shmem_slot = slotno;
407 3484 : rw->rw_terminate = false;
408 :
409 : /* Log it! */
410 3484 : ereport(DEBUG1,
411 : (errmsg_internal("registering background worker \"%s\"",
412 : rw->rw_worker.bgw_name)));
413 :
414 3484 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
415 : }
416 : }
417 :
418 : /*
419 : * Forget about a background worker that's no longer needed.
420 : *
421 : * NOTE: The entry is unlinked from BackgroundWorkerList. If the caller is
422 : * iterating through it, better use a mutable iterator!
423 : *
424 : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
425 : *
426 : * This function must be invoked only in the postmaster.
427 : */
428 : void
429 3456 : ForgetBackgroundWorker(RegisteredBgWorker *rw)
430 : {
431 : BackgroundWorkerSlot *slot;
432 :
433 : Assert(rw->rw_shmem_slot < max_worker_processes);
434 3456 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
435 : Assert(slot->in_use);
436 :
437 : /*
438 : * We need a memory barrier here to make sure that the update of
439 : * parallel_terminate_count completes before the store to in_use.
440 : */
441 3456 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
442 2680 : BackgroundWorkerData->parallel_terminate_count++;
443 :
444 3456 : pg_memory_barrier();
445 3456 : slot->in_use = false;
446 :
447 3456 : ereport(DEBUG1,
448 : (errmsg_internal("unregistering background worker \"%s\"",
449 : rw->rw_worker.bgw_name)));
450 :
451 3456 : dlist_delete(&rw->rw_lnode);
452 3456 : pfree(rw);
453 3456 : }
454 :
455 : /*
456 : * Report the PID of a newly-launched background worker in shared memory.
457 : *
458 : * This function should only be called from the postmaster.
459 : */
460 : void
461 4720 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
462 : {
463 : BackgroundWorkerSlot *slot;
464 :
465 : Assert(rw->rw_shmem_slot < max_worker_processes);
466 4720 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
467 4720 : slot->pid = rw->rw_pid;
468 :
469 4720 : if (rw->rw_worker.bgw_notify_pid != 0)
470 3484 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
471 4720 : }
472 :
473 : /*
474 : * Report that the PID of a background worker is now zero because a
475 : * previously-running background worker has exited.
476 : *
477 : * NOTE: The entry may be unlinked from BackgroundWorkerList. If the caller
478 : * is iterating through it, better use a mutable iterator!
479 : *
480 : * This function should only be called from the postmaster.
481 : */
482 : void
483 4164 : ReportBackgroundWorkerExit(RegisteredBgWorker *rw)
484 : {
485 : BackgroundWorkerSlot *slot;
486 : int notify_pid;
487 :
488 : Assert(rw->rw_shmem_slot < max_worker_processes);
489 4164 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
490 4164 : slot->pid = rw->rw_pid;
491 4164 : notify_pid = rw->rw_worker.bgw_notify_pid;
492 :
493 : /*
494 : * If this worker is slated for deregistration, do that before notifying
495 : * the process which started it. Otherwise, if that process tries to
496 : * reuse the slot immediately, it might not be available yet. In theory
497 : * that could happen anyway if the process checks slot->pid at just the
498 : * wrong moment, but this makes the window narrower.
499 : */
500 4164 : if (rw->rw_terminate ||
501 1060 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
502 3456 : ForgetBackgroundWorker(rw);
503 :
504 4164 : if (notify_pid != 0)
505 3400 : kill(notify_pid, SIGUSR1);
506 4164 : }
507 :
508 : /*
509 : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
510 : *
511 : * This function should only be called from the postmaster.
512 : */
513 : void
514 446 : BackgroundWorkerStopNotifications(pid_t pid)
515 : {
516 : dlist_iter iter;
517 :
518 1330 : dlist_foreach(iter, &BackgroundWorkerList)
519 : {
520 : RegisteredBgWorker *rw;
521 :
522 884 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
523 884 : if (rw->rw_worker.bgw_notify_pid == pid)
524 60 : rw->rw_worker.bgw_notify_pid = 0;
525 : }
526 446 : }
527 :
528 : /*
529 : * Cancel any not-yet-started worker requests that have waiting processes.
530 : *
531 : * This is called during a normal ("smart" or "fast") database shutdown.
532 : * After this point, no new background workers will be started, so anything
533 : * that might be waiting for them needs to be kicked off its wait. We do
534 : * that by canceling the bgworker registration entirely, which is perhaps
535 : * overkill, but since we're shutting down it does not matter whether the
536 : * registration record sticks around.
537 : *
538 : * This function should only be called from the postmaster.
539 : */
540 : void
541 862 : ForgetUnstartedBackgroundWorkers(void)
542 : {
543 : dlist_mutable_iter iter;
544 :
545 1742 : dlist_foreach_modify(iter, &BackgroundWorkerList)
546 : {
547 : RegisteredBgWorker *rw;
548 : BackgroundWorkerSlot *slot;
549 :
550 880 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
551 : Assert(rw->rw_shmem_slot < max_worker_processes);
552 880 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
553 :
554 : /* If it's not yet started, and there's someone waiting ... */
555 880 : if (slot->pid == InvalidPid &&
556 100 : rw->rw_worker.bgw_notify_pid != 0)
557 : {
558 : /* ... then zap it, and notify the waiter */
559 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
560 :
561 0 : ForgetBackgroundWorker(rw);
562 0 : if (notify_pid != 0)
563 0 : kill(notify_pid, SIGUSR1);
564 : }
565 : }
566 862 : }
567 :
568 : /*
569 : * Reset background worker crash state.
570 : *
571 : * We assume that, after a crash-and-restart cycle, background workers without
572 : * the never-restart flag should be restarted immediately, instead of waiting
573 : * for bgw_restart_time to elapse. On the other hand, workers with that flag
574 : * should be forgotten immediately, since we won't ever restart them.
575 : *
576 : * This function should only be called from the postmaster.
577 : */
578 : void
579 8 : ResetBackgroundWorkerCrashTimes(void)
580 : {
581 : dlist_mutable_iter iter;
582 :
583 16 : dlist_foreach_modify(iter, &BackgroundWorkerList)
584 : {
585 : RegisteredBgWorker *rw;
586 :
587 8 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
588 :
589 8 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
590 : {
591 : /*
592 : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
593 : * the crash, so forget about them. (If we wait until after the
594 : * crash to forget about them, and they are parallel workers,
595 : * parallel_terminate_count will get incremented after we've
596 : * already zeroed parallel_register_count, which would be bad.)
597 : */
598 0 : ForgetBackgroundWorker(rw);
599 : }
600 : else
601 : {
602 : /*
603 : * The accounting which we do via parallel_register_count and
604 : * parallel_terminate_count would get messed up if a worker marked
605 : * parallel could survive a crash and restart cycle. All such
606 : * workers should be marked BGW_NEVER_RESTART, and thus control
607 : * should never reach this branch.
608 : */
609 : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
610 :
611 : /*
612 : * Allow this worker to be restarted immediately after we finish
613 : * resetting.
614 : */
615 8 : rw->rw_crashed_at = 0;
616 :
617 : /*
618 : * If there was anyone waiting for it, they're history.
619 : */
620 8 : rw->rw_worker.bgw_notify_pid = 0;
621 : }
622 : }
623 8 : }
624 :
625 : /*
626 : * Complain about the BackgroundWorker definition using error level elevel.
627 : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
628 : * which case we won't return at all in the not-OK case).
629 : */
630 : static bool
631 4802 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
632 : {
633 : /* sanity check for flags */
634 :
635 : /*
636 : * We used to support workers not connected to shared memory, but don't
637 : * anymore. Thus this is a required flag now. We're not removing the flag
638 : * for compatibility reasons and because the flag still provides some
639 : * signal when reading code.
640 : */
641 4802 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
642 : {
643 0 : ereport(elevel,
644 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
645 : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
646 : worker->bgw_name)));
647 0 : return false;
648 : }
649 :
650 4802 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
651 : {
652 4784 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
653 : {
654 0 : ereport(elevel,
655 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
656 : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
657 : worker->bgw_name)));
658 0 : return false;
659 : }
660 :
661 : /* XXX other checks? */
662 : }
663 :
664 4802 : if ((worker->bgw_restart_time < 0 &&
665 3368 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
666 4802 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
667 : {
668 0 : ereport(elevel,
669 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
670 : errmsg("background worker \"%s\": invalid restart interval",
671 : worker->bgw_name)));
672 0 : return false;
673 : }
674 :
675 : /*
676 : * Parallel workers may not be configured for restart, because the
677 : * parallel_register_count/parallel_terminate_count accounting can't
678 : * handle parallel workers lasting through a crash-and-restart cycle.
679 : */
680 4802 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
681 1434 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
682 : {
683 0 : ereport(elevel,
684 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
685 : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
686 : worker->bgw_name)));
687 0 : return false;
688 : }
689 :
690 : /*
691 : * If bgw_type is not filled in, use bgw_name.
692 : */
693 4802 : if (strcmp(worker->bgw_type, "") == 0)
694 0 : strcpy(worker->bgw_type, worker->bgw_name);
695 :
696 4802 : return true;
697 : }
698 :
699 : /*
700 : * Standard SIGTERM handler for background workers
701 : */
702 : static void
703 0 : bgworker_die(SIGNAL_ARGS)
704 : {
705 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
706 :
707 0 : ereport(FATAL,
708 : (errcode(ERRCODE_ADMIN_SHUTDOWN),
709 : errmsg("terminating background worker \"%s\" due to administrator command",
710 : MyBgworkerEntry->bgw_type)));
711 : }
712 :
713 : /*
714 : * Main entry point for background worker processes.
715 : */
716 : void
717 4170 : BackgroundWorkerMain(char *startup_data, size_t startup_data_len)
718 : {
719 : sigjmp_buf local_sigjmp_buf;
720 : BackgroundWorker *worker;
721 : bgworker_main_type entrypt;
722 :
723 4170 : if (startup_data == NULL)
724 0 : elog(FATAL, "unable to find bgworker entry");
725 : Assert(startup_data_len == sizeof(BackgroundWorker));
726 4170 : worker = MemoryContextAlloc(TopMemoryContext, sizeof(BackgroundWorker));
727 4170 : memcpy(worker, startup_data, sizeof(BackgroundWorker));
728 :
729 : /*
730 : * Now that we're done reading the startup data, release postmaster's
731 : * working memory context.
732 : */
733 4170 : if (PostmasterContext)
734 : {
735 4170 : MemoryContextDelete(PostmasterContext);
736 4170 : PostmasterContext = NULL;
737 : }
738 :
739 4170 : MyBgworkerEntry = worker;
740 4170 : MyBackendType = B_BG_WORKER;
741 4170 : init_ps_display(worker->bgw_name);
742 :
743 : Assert(GetProcessingMode() == InitProcessing);
744 :
745 : /* Apply PostAuthDelay */
746 4170 : if (PostAuthDelay > 0)
747 0 : pg_usleep(PostAuthDelay * 1000000L);
748 :
749 : /*
750 : * Set up signal handlers.
751 : */
752 4170 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
753 : {
754 : /*
755 : * SIGINT is used to signal canceling the current action
756 : */
757 4152 : pqsignal(SIGINT, StatementCancelHandler);
758 4152 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
759 4152 : pqsignal(SIGFPE, FloatExceptionHandler);
760 :
761 : /* XXX Any other handlers needed here? */
762 : }
763 : else
764 : {
765 18 : pqsignal(SIGINT, SIG_IGN);
766 18 : pqsignal(SIGUSR1, SIG_IGN);
767 18 : pqsignal(SIGFPE, SIG_IGN);
768 : }
769 4170 : pqsignal(SIGTERM, bgworker_die);
770 : /* SIGQUIT handler was already set up by InitPostmasterChild */
771 4170 : pqsignal(SIGHUP, SIG_IGN);
772 :
773 4170 : InitializeTimeouts(); /* establishes SIGALRM handler */
774 :
775 4170 : pqsignal(SIGPIPE, SIG_IGN);
776 4170 : pqsignal(SIGUSR2, SIG_IGN);
777 4170 : pqsignal(SIGCHLD, SIG_DFL);
778 :
779 : /*
780 : * If an exception is encountered, processing resumes here.
781 : *
782 : * We just need to clean up, report the error, and go away.
783 : */
784 4170 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
785 : {
786 : /* Since not using PG_TRY, must reset error stack by hand */
787 176 : error_context_stack = NULL;
788 :
789 : /* Prevent interrupts while cleaning up */
790 176 : HOLD_INTERRUPTS();
791 :
792 : /*
793 : * sigsetjmp will have blocked all signals, but we may need to accept
794 : * signals while communicating with our parallel leader. Once we've
795 : * done HOLD_INTERRUPTS() it should be safe to unblock signals.
796 : */
797 176 : BackgroundWorkerUnblockSignals();
798 :
799 : /* Report the error to the parallel leader and the server log */
800 176 : EmitErrorReport();
801 :
802 : /*
803 : * Do we need more cleanup here? For shmem-connected bgworkers, we
804 : * will call InitProcess below, which will install ProcKill as exit
805 : * callback. That will take care of releasing locks, etc.
806 : */
807 :
808 : /* and go away */
809 176 : proc_exit(1);
810 : }
811 :
812 : /* We can now handle ereport(ERROR) */
813 4170 : PG_exception_stack = &local_sigjmp_buf;
814 :
815 : /*
816 : * Create a per-backend PGPROC struct in shared memory. We must do this
817 : * before we can use LWLocks or access any shared memory.
818 : */
819 4170 : InitProcess();
820 :
821 : /*
822 : * Early initialization.
823 : */
824 4170 : BaseInit();
825 :
826 : /*
827 : * Look up the entry point function, loading its library if necessary.
828 : */
829 8340 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
830 4170 : worker->bgw_function_name);
831 :
832 : /*
833 : * Note that in normal processes, we would call InitPostgres here. For a
834 : * worker, however, we don't know what database to connect to, yet; so we
835 : * need to wait until the user code does it via
836 : * BackgroundWorkerInitializeConnection().
837 : */
838 :
839 : /*
840 : * Now invoke the user-defined worker code
841 : */
842 4170 : entrypt(worker->bgw_main_arg);
843 :
844 : /* ... and if it returns, we're done */
845 2674 : proc_exit(0);
846 : }
847 :
848 : /*
849 : * Connect background worker to a database.
850 : */
851 : void
852 712 : BackgroundWorkerInitializeConnection(const char *dbname, const char *username, uint32 flags)
853 : {
854 712 : BackgroundWorker *worker = MyBgworkerEntry;
855 712 : bits32 init_flags = 0; /* never honor session_preload_libraries */
856 :
857 : /* ignore datallowconn? */
858 712 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
859 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
860 : /* ignore rolcanlogin? */
861 712 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
862 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
863 :
864 : /* XXX is this the right errcode? */
865 712 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
866 0 : ereport(FATAL,
867 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
868 : errmsg("database connection requirement not indicated during registration")));
869 :
870 712 : InitPostgres(dbname, InvalidOid, /* database to connect to */
871 : username, InvalidOid, /* role to connect as */
872 : init_flags,
873 : NULL); /* no out_dbname */
874 :
875 : /* it had better not gotten out of "init" mode yet */
876 712 : if (!IsInitProcessingMode())
877 0 : ereport(ERROR,
878 : (errmsg("invalid processing mode in background worker")));
879 712 : SetProcessingMode(NormalProcessing);
880 712 : }
881 :
882 : /*
883 : * Connect background worker to a database using OIDs.
884 : */
885 : void
886 3440 : BackgroundWorkerInitializeConnectionByOid(Oid dboid, Oid useroid, uint32 flags)
887 : {
888 3440 : BackgroundWorker *worker = MyBgworkerEntry;
889 3440 : bits32 init_flags = 0; /* never honor session_preload_libraries */
890 :
891 : /* ignore datallowconn? */
892 3440 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
893 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
894 : /* ignore rolcanlogin? */
895 3440 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
896 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
897 :
898 : /* XXX is this the right errcode? */
899 3440 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
900 0 : ereport(FATAL,
901 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
902 : errmsg("database connection requirement not indicated during registration")));
903 :
904 3440 : InitPostgres(NULL, dboid, /* database to connect to */
905 : NULL, useroid, /* role to connect as */
906 : init_flags,
907 : NULL); /* no out_dbname */
908 :
909 : /* it had better not gotten out of "init" mode yet */
910 3426 : if (!IsInitProcessingMode())
911 0 : ereport(ERROR,
912 : (errmsg("invalid processing mode in background worker")));
913 3426 : SetProcessingMode(NormalProcessing);
914 3426 : }
915 :
916 : /*
917 : * Block/unblock signals in a background worker
918 : */
919 : void
920 0 : BackgroundWorkerBlockSignals(void)
921 : {
922 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
923 0 : }
924 :
925 : void
926 4346 : BackgroundWorkerUnblockSignals(void)
927 : {
928 4346 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
929 4346 : }
930 :
931 : /*
932 : * Register a new static background worker.
933 : *
934 : * This can only be called directly from postmaster or in the _PG_init
935 : * function of a module library that's loaded by shared_preload_libraries;
936 : * otherwise it will have no effect.
937 : */
938 : void
939 1440 : RegisterBackgroundWorker(BackgroundWorker *worker)
940 : {
941 : RegisteredBgWorker *rw;
942 : static int numworkers = 0;
943 :
944 : /*
945 : * Static background workers can only be registered in the postmaster
946 : * process.
947 : */
948 1440 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
949 : {
950 : /*
951 : * In EXEC_BACKEND or single-user mode, we process
952 : * shared_preload_libraries in backend processes too. We cannot
953 : * register static background workers at that stage, but many
954 : * libraries' _PG_init() functions don't distinguish whether they're
955 : * being loaded in the postmaster or in a backend, they just check
956 : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
957 : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
958 : * the background workers should already have been registered when the
959 : * library was loaded in postmaster.
960 : */
961 0 : if (process_shared_preload_libraries_in_progress)
962 0 : return;
963 0 : ereport(LOG,
964 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
965 : errmsg("background worker \"%s\": must be registered in \"shared_preload_libraries\"",
966 : worker->bgw_name)));
967 0 : return;
968 : }
969 :
970 : /*
971 : * Cannot register static background workers after calling
972 : * BackgroundWorkerShmemInit().
973 : */
974 1440 : if (BackgroundWorkerData != NULL)
975 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
976 : worker->bgw_name);
977 :
978 1440 : ereport(DEBUG1,
979 : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
980 :
981 1440 : if (!SanityCheckBackgroundWorker(worker, LOG))
982 0 : return;
983 :
984 1440 : if (worker->bgw_notify_pid != 0)
985 : {
986 0 : ereport(LOG,
987 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
988 : errmsg("background worker \"%s\": only dynamic background workers can request notification",
989 : worker->bgw_name)));
990 0 : return;
991 : }
992 :
993 : /*
994 : * Enforce maximum number of workers. Note this is overly restrictive: we
995 : * could allow more non-shmem-connected workers, because these don't count
996 : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
997 : * important to relax this restriction.
998 : */
999 1440 : if (++numworkers > max_worker_processes)
1000 : {
1001 0 : ereport(LOG,
1002 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
1003 : errmsg("too many background workers"),
1004 : errdetail_plural("Up to %d background worker can be registered with the current settings.",
1005 : "Up to %d background workers can be registered with the current settings.",
1006 : max_worker_processes,
1007 : max_worker_processes),
1008 : errhint("Consider increasing the configuration parameter \"%s\".", "max_worker_processes")));
1009 0 : return;
1010 : }
1011 :
1012 : /*
1013 : * Copy the registration data into the registered workers list.
1014 : */
1015 1440 : rw = MemoryContextAllocExtended(PostmasterContext,
1016 : sizeof(RegisteredBgWorker),
1017 : MCXT_ALLOC_NO_OOM);
1018 1440 : if (rw == NULL)
1019 : {
1020 0 : ereport(LOG,
1021 : (errcode(ERRCODE_OUT_OF_MEMORY),
1022 : errmsg("out of memory")));
1023 0 : return;
1024 : }
1025 :
1026 1440 : rw->rw_worker = *worker;
1027 1440 : rw->rw_pid = 0;
1028 1440 : rw->rw_crashed_at = 0;
1029 1440 : rw->rw_terminate = false;
1030 :
1031 1440 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
1032 : }
1033 :
1034 : /*
1035 : * Register a new background worker from a regular backend.
1036 : *
1037 : * Returns true on success and false on failure. Failure typically indicates
1038 : * that no background worker slots are currently available.
1039 : *
1040 : * If handle != NULL, we'll set *handle to a pointer that can subsequently
1041 : * be used as an argument to GetBackgroundWorkerPid(). The caller can
1042 : * free this pointer using pfree(), if desired.
1043 : */
1044 : bool
1045 3362 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
1046 : BackgroundWorkerHandle **handle)
1047 : {
1048 : int slotno;
1049 3362 : bool success = false;
1050 : bool parallel;
1051 3362 : uint64 generation = 0;
1052 :
1053 : /*
1054 : * We can't register dynamic background workers from the postmaster. If
1055 : * this is a standalone backend, we're the only process and can't start
1056 : * any more. In a multi-process environment, it might be theoretically
1057 : * possible, but we don't currently support it due to locking
1058 : * considerations; see comments on the BackgroundWorkerSlot data
1059 : * structure.
1060 : */
1061 3362 : if (!IsUnderPostmaster)
1062 0 : return false;
1063 :
1064 3362 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1065 0 : return false;
1066 :
1067 3362 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1068 :
1069 3362 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1070 :
1071 : /*
1072 : * If this is a parallel worker, check whether there are already too many
1073 : * parallel workers; if so, don't register another one. Our view of
1074 : * parallel_terminate_count may be slightly stale, but that doesn't really
1075 : * matter: we would have gotten the same result if we'd arrived here
1076 : * slightly earlier anyway. There's no help for it, either, since the
1077 : * postmaster must not take locks; a memory barrier wouldn't guarantee
1078 : * anything useful.
1079 : */
1080 3362 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1081 2710 : BackgroundWorkerData->parallel_terminate_count) >=
1082 : max_parallel_workers)
1083 : {
1084 : Assert(BackgroundWorkerData->parallel_register_count -
1085 : BackgroundWorkerData->parallel_terminate_count <=
1086 : MAX_PARALLEL_WORKER_LIMIT);
1087 20 : LWLockRelease(BackgroundWorkerLock);
1088 20 : return false;
1089 : }
1090 :
1091 : /*
1092 : * Look for an unused slot. If we find one, grab it.
1093 : */
1094 10390 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1095 : {
1096 10380 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1097 :
1098 10380 : if (!slot->in_use)
1099 : {
1100 3332 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1101 3332 : slot->pid = InvalidPid; /* indicates not started yet */
1102 3332 : slot->generation++;
1103 3332 : slot->terminate = false;
1104 3332 : generation = slot->generation;
1105 3332 : if (parallel)
1106 2680 : BackgroundWorkerData->parallel_register_count++;
1107 :
1108 : /*
1109 : * Make sure postmaster doesn't see the slot as in use before it
1110 : * sees the new contents.
1111 : */
1112 3332 : pg_write_barrier();
1113 :
1114 3332 : slot->in_use = true;
1115 3332 : success = true;
1116 3332 : break;
1117 : }
1118 : }
1119 :
1120 3342 : LWLockRelease(BackgroundWorkerLock);
1121 :
1122 : /* If we found a slot, tell the postmaster to notice the change. */
1123 3342 : if (success)
1124 3332 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1125 :
1126 : /*
1127 : * If we found a slot and the user has provided a handle, initialize it.
1128 : */
1129 3342 : if (success && handle)
1130 : {
1131 3332 : *handle = palloc(sizeof(BackgroundWorkerHandle));
1132 3332 : (*handle)->slot = slotno;
1133 3332 : (*handle)->generation = generation;
1134 : }
1135 :
1136 3342 : return success;
1137 : }
1138 :
1139 : /*
1140 : * Get the PID of a dynamically-registered background worker.
1141 : *
1142 : * If the worker is determined to be running, the return value will be
1143 : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1144 : * postmaster has not yet attempted to start the worker, the return value will
1145 : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1146 : *
1147 : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1148 : * (because it is configured for automatic restart and exited non-zero),
1149 : * or that the worker is permanently stopped (because it exited with exit
1150 : * code 0, or was not configured for automatic restart), or even that the
1151 : * worker was unregistered without ever starting (either because startup
1152 : * failed and the worker is not configured for automatic restart, or because
1153 : * TerminateBackgroundWorker was used before the worker was successfully
1154 : * started).
1155 : */
1156 : BgwHandleStatus
1157 14378142 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1158 : {
1159 : BackgroundWorkerSlot *slot;
1160 : pid_t pid;
1161 :
1162 : Assert(handle->slot < max_worker_processes);
1163 14378142 : slot = &BackgroundWorkerData->slot[handle->slot];
1164 :
1165 : /*
1166 : * We could probably arrange to synchronize access to data using memory
1167 : * barriers only, but for now, let's just keep it simple and grab the
1168 : * lock. It seems unlikely that there will be enough traffic here to
1169 : * result in meaningful contention.
1170 : */
1171 14378142 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1172 :
1173 : /*
1174 : * The generation number can't be concurrently changed while we hold the
1175 : * lock. The pid, which is updated by the postmaster, can change at any
1176 : * time, but we assume such changes are atomic. So the value we read
1177 : * won't be garbage, but it might be out of date by the time the caller
1178 : * examines it (but that's unavoidable anyway).
1179 : *
1180 : * The in_use flag could be in the process of changing from true to false,
1181 : * but if it is already false then it can't change further.
1182 : */
1183 14378142 : if (handle->generation != slot->generation || !slot->in_use)
1184 2682 : pid = 0;
1185 : else
1186 14375460 : pid = slot->pid;
1187 :
1188 : /* All done. */
1189 14378142 : LWLockRelease(BackgroundWorkerLock);
1190 :
1191 14378142 : if (pid == 0)
1192 2682 : return BGWH_STOPPED;
1193 14375460 : else if (pid == InvalidPid)
1194 4989574 : return BGWH_NOT_YET_STARTED;
1195 9385886 : *pidp = pid;
1196 9385886 : return BGWH_STARTED;
1197 : }
1198 :
1199 : /*
1200 : * Wait for a background worker to start up.
1201 : *
1202 : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1203 : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1204 : * returned. However, if the postmaster has died, we give up and return
1205 : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1206 : * take place.
1207 : *
1208 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1209 : * else we will not be awoken promptly when the worker's state changes.
1210 : */
1211 : BgwHandleStatus
1212 36 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1213 : {
1214 : BgwHandleStatus status;
1215 : int rc;
1216 :
1217 : for (;;)
1218 22 : {
1219 : pid_t pid;
1220 :
1221 36 : CHECK_FOR_INTERRUPTS();
1222 :
1223 36 : status = GetBackgroundWorkerPid(handle, &pid);
1224 36 : if (status == BGWH_STARTED)
1225 14 : *pidp = pid;
1226 36 : if (status != BGWH_NOT_YET_STARTED)
1227 14 : break;
1228 :
1229 22 : rc = WaitLatch(MyLatch,
1230 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1231 : WAIT_EVENT_BGWORKER_STARTUP);
1232 :
1233 22 : if (rc & WL_POSTMASTER_DEATH)
1234 : {
1235 0 : status = BGWH_POSTMASTER_DIED;
1236 0 : break;
1237 : }
1238 :
1239 22 : ResetLatch(MyLatch);
1240 : }
1241 :
1242 14 : return status;
1243 : }
1244 :
1245 : /*
1246 : * Wait for a background worker to stop.
1247 : *
1248 : * If the worker hasn't yet started, or is running, we wait for it to stop
1249 : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1250 : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1251 : * notifies us when a worker's state changes.
1252 : *
1253 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1254 : * else we will not be awoken promptly when the worker's state changes.
1255 : */
1256 : BgwHandleStatus
1257 5122 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1258 : {
1259 : BgwHandleStatus status;
1260 : int rc;
1261 :
1262 : for (;;)
1263 2440 : {
1264 : pid_t pid;
1265 :
1266 5122 : CHECK_FOR_INTERRUPTS();
1267 :
1268 5122 : status = GetBackgroundWorkerPid(handle, &pid);
1269 5122 : if (status == BGWH_STOPPED)
1270 2682 : break;
1271 :
1272 2440 : rc = WaitLatch(MyLatch,
1273 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1274 : WAIT_EVENT_BGWORKER_SHUTDOWN);
1275 :
1276 2440 : if (rc & WL_POSTMASTER_DEATH)
1277 : {
1278 0 : status = BGWH_POSTMASTER_DIED;
1279 0 : break;
1280 : }
1281 :
1282 2440 : ResetLatch(MyLatch);
1283 : }
1284 :
1285 2682 : return status;
1286 : }
1287 :
1288 : /*
1289 : * Instruct the postmaster to terminate a background worker.
1290 : *
1291 : * Note that it's safe to do this without regard to whether the worker is
1292 : * still running, or even if the worker may already have exited and been
1293 : * unregistered.
1294 : */
1295 : void
1296 12 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1297 : {
1298 : BackgroundWorkerSlot *slot;
1299 12 : bool signal_postmaster = false;
1300 :
1301 : Assert(handle->slot < max_worker_processes);
1302 12 : slot = &BackgroundWorkerData->slot[handle->slot];
1303 :
1304 : /* Set terminate flag in shared memory, unless slot has been reused. */
1305 12 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1306 12 : if (handle->generation == slot->generation)
1307 : {
1308 12 : slot->terminate = true;
1309 12 : signal_postmaster = true;
1310 : }
1311 12 : LWLockRelease(BackgroundWorkerLock);
1312 :
1313 : /* Make sure the postmaster notices the change to shared memory. */
1314 12 : if (signal_postmaster)
1315 12 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1316 12 : }
1317 :
1318 : /*
1319 : * Look up (and possibly load) a bgworker entry point function.
1320 : *
1321 : * For functions contained in the core code, we use library name "postgres"
1322 : * and consult the InternalBGWorkers array. External functions are
1323 : * looked up, and loaded if necessary, using load_external_function().
1324 : *
1325 : * The point of this is to pass function names as strings across process
1326 : * boundaries. We can't pass actual function addresses because of the
1327 : * possibility that the function has been loaded at a different address
1328 : * in a different process. This is obviously a hazard for functions in
1329 : * loadable libraries, but it can happen even for functions in the core code
1330 : * on platforms using EXEC_BACKEND (e.g., Windows).
1331 : *
1332 : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1333 : * in favor of applying load_external_function() for core functions too;
1334 : * but that raises portability issues that are not worth addressing now.
1335 : */
1336 : static bgworker_main_type
1337 4170 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1338 : {
1339 : /*
1340 : * If the function is to be loaded from postgres itself, search the
1341 : * InternalBGWorkers array.
1342 : */
1343 4170 : if (strcmp(libraryname, "postgres") == 0)
1344 : {
1345 : int i;
1346 :
1347 7056 : for (i = 0; i < lengthof(InternalBGWorkers); i++)
1348 : {
1349 7056 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1350 4144 : return InternalBGWorkers[i].fn_addr;
1351 : }
1352 :
1353 : /* We can only reach this by programming error. */
1354 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1355 : }
1356 :
1357 : /* Otherwise load from external library. */
1358 26 : return (bgworker_main_type)
1359 26 : load_external_function(libraryname, funcname, true, NULL);
1360 : }
1361 :
1362 : /*
1363 : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1364 : * not a valid background worker.
1365 : *
1366 : * The return value is in static memory belonging to this function, so it has
1367 : * to be used before calling this function again. This is so that the caller
1368 : * doesn't have to worry about the background worker locking protocol.
1369 : */
1370 : const char *
1371 1530 : GetBackgroundWorkerTypeByPid(pid_t pid)
1372 : {
1373 : int slotno;
1374 1530 : bool found = false;
1375 : static char result[BGW_MAXLEN];
1376 :
1377 1530 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1378 :
1379 1718 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1380 : {
1381 1718 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1382 :
1383 1718 : if (slot->pid > 0 && slot->pid == pid)
1384 : {
1385 1530 : strcpy(result, slot->worker.bgw_type);
1386 1530 : found = true;
1387 1530 : break;
1388 : }
1389 : }
1390 :
1391 1530 : LWLockRelease(BackgroundWorkerLock);
1392 :
1393 1530 : if (!found)
1394 0 : return NULL;
1395 :
1396 1530 : return result;
1397 : }
|