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 "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/datachecksum_state.h"
22 : #include "postmaster/postmaster.h"
23 : #include "replication/logicallauncher.h"
24 : #include "replication/logicalworker.h"
25 : #include "storage/ipc.h"
26 : #include "storage/latch.h"
27 : #include "storage/lwlock.h"
28 : #include "storage/pmsignal.h"
29 : #include "storage/proc.h"
30 : #include "storage/procarray.h"
31 : #include "storage/procsignal.h"
32 : #include "storage/shmem.h"
33 : #include "storage/subsystems.h"
34 : #include "tcop/tcopprot.h"
35 : #include "utils/ascii.h"
36 : #include "utils/memutils.h"
37 : #include "utils/ps_status.h"
38 : #include "utils/timeout.h"
39 : #include "utils/wait_event.h"
40 :
41 : /*
42 : * The postmaster's list of registered background workers, in private memory.
43 : */
44 : dlist_head BackgroundWorkerList = DLIST_STATIC_INIT(BackgroundWorkerList);
45 :
46 : /*
47 : * BackgroundWorkerSlots exist in shared memory and can be accessed (via
48 : * the BackgroundWorkerArray) by both the postmaster and by regular backends.
49 : * However, the postmaster cannot take locks, even spinlocks, because this
50 : * might allow it to crash or become wedged if shared memory gets corrupted.
51 : * Such an outcome is intolerable. Therefore, we need a lockless protocol
52 : * for coordinating access to this data.
53 : *
54 : * The 'in_use' flag is used to hand off responsibility for the slot between
55 : * the postmaster and the rest of the system. When 'in_use' is false,
56 : * the postmaster will ignore the slot entirely, except for the 'in_use' flag
57 : * itself, which it may read. In this state, regular backends may modify the
58 : * slot. Once a backend sets 'in_use' to true, the slot becomes the
59 : * responsibility of the postmaster. Regular backends may no longer modify it,
60 : * but the postmaster may examine it. Thus, a backend initializing a slot
61 : * must fully initialize the slot - and insert a write memory barrier - before
62 : * marking it as in use.
63 : *
64 : * As an exception, however, even when the slot is in use, regular backends
65 : * may set the 'terminate' flag for a slot, telling the postmaster not
66 : * to restart it. Once the background worker is no longer running, the slot
67 : * will be released for reuse.
68 : *
69 : * In addition to coordinating with the postmaster, backends modifying this
70 : * data structure must coordinate with each other. Since they can take locks,
71 : * this is straightforward: any backend wishing to manipulate a slot must
72 : * take BackgroundWorkerLock in exclusive mode. Backends wishing to read
73 : * data that might get concurrently modified by other backends should take
74 : * this lock in shared mode. No matter what, backends reading this data
75 : * structure must be able to tolerate concurrent modifications by the
76 : * postmaster.
77 : */
78 : typedef struct BackgroundWorkerSlot
79 : {
80 : bool in_use;
81 : bool terminate;
82 : pid_t pid; /* InvalidPid = not started yet; 0 = dead */
83 : uint64 generation; /* incremented when slot is recycled */
84 : BackgroundWorker worker;
85 : } BackgroundWorkerSlot;
86 :
87 : /*
88 : * In order to limit the total number of parallel workers (according to
89 : * max_parallel_workers GUC), we maintain the number of active parallel
90 : * workers. Since the postmaster cannot take locks, two variables are used for
91 : * this purpose: the number of registered parallel workers (modified by the
92 : * backends, protected by BackgroundWorkerLock) and the number of terminated
93 : * parallel workers (modified only by the postmaster, lockless). The active
94 : * number of parallel workers is the number of registered workers minus the
95 : * terminated ones. These counters can of course overflow, but it's not
96 : * important here since the subtraction will still give the right number.
97 : */
98 : typedef struct BackgroundWorkerArray
99 : {
100 : int total_slots;
101 : uint32 parallel_register_count;
102 : uint32 parallel_terminate_count;
103 : BackgroundWorkerSlot slot[FLEXIBLE_ARRAY_MEMBER];
104 : } BackgroundWorkerArray;
105 :
106 : struct BackgroundWorkerHandle
107 : {
108 : int slot;
109 : uint64 generation;
110 : };
111 :
112 : static BackgroundWorkerArray *BackgroundWorkerData;
113 :
114 : static void BackgroundWorkerShmemRequest(void *arg);
115 : static void BackgroundWorkerShmemInit(void *arg);
116 :
117 : const ShmemCallbacks BackgroundWorkerShmemCallbacks = {
118 : .request_fn = BackgroundWorkerShmemRequest,
119 : .init_fn = BackgroundWorkerShmemInit,
120 : };
121 :
122 : /*
123 : * List of internal background worker entry points. We need this for
124 : * reasons explained in LookupBackgroundWorkerFunction(), below.
125 : */
126 : static const struct
127 : {
128 : const char *fn_name;
129 : bgworker_main_type fn_addr;
130 : } InternalBGWorkers[] =
131 :
132 : {
133 : {
134 : .fn_name = "ApplyLauncherMain",
135 : .fn_addr = ApplyLauncherMain
136 : },
137 : {
138 : .fn_name = "ApplyWorkerMain",
139 : .fn_addr = ApplyWorkerMain
140 : },
141 : {
142 : .fn_name = "ParallelApplyWorkerMain",
143 : .fn_addr = ParallelApplyWorkerMain
144 : },
145 : {
146 : .fn_name = "ParallelWorkerMain",
147 : .fn_addr = ParallelWorkerMain
148 : },
149 : {
150 : .fn_name = "SequenceSyncWorkerMain",
151 : .fn_addr = SequenceSyncWorkerMain
152 : },
153 : {
154 : .fn_name = "TableSyncWorkerMain",
155 : .fn_addr = TableSyncWorkerMain
156 : },
157 : {
158 : .fn_name = "DataChecksumsWorkerLauncherMain",
159 : .fn_addr = DataChecksumsWorkerLauncherMain
160 : },
161 : {
162 : .fn_name = "DataChecksumsWorkerMain",
163 : .fn_addr = DataChecksumsWorkerMain
164 : }
165 : };
166 :
167 : /* Private functions. */
168 : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
169 :
170 :
171 : /*
172 : * Register shared memory needed for background workers.
173 : */
174 : static void
175 1227 : BackgroundWorkerShmemRequest(void *arg)
176 : {
177 : Size size;
178 :
179 : /* Array of workers is variably sized. */
180 1227 : size = offsetof(BackgroundWorkerArray, slot);
181 1227 : size = add_size(size, mul_size(max_worker_processes,
182 : sizeof(BackgroundWorkerSlot)));
183 1227 : ShmemRequestStruct(.name = "Background Worker Data",
184 : .size = size,
185 : .ptr = (void **) &BackgroundWorkerData,
186 : );
187 1227 : }
188 :
189 : /*
190 : * Initialize shared memory for background workers.
191 : */
192 : static void
193 1224 : BackgroundWorkerShmemInit(void *arg)
194 : {
195 : dlist_iter iter;
196 1224 : int slotno = 0;
197 :
198 1224 : BackgroundWorkerData->total_slots = max_worker_processes;
199 1224 : BackgroundWorkerData->parallel_register_count = 0;
200 1224 : BackgroundWorkerData->parallel_terminate_count = 0;
201 :
202 : /*
203 : * Copy contents of worker list into shared memory. Record the shared
204 : * memory slot assigned to each worker. This ensures a 1-to-1
205 : * correspondence between the postmaster's private list and the array in
206 : * shared memory.
207 : */
208 2145 : dlist_foreach(iter, &BackgroundWorkerList)
209 : {
210 921 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
211 : RegisteredBgWorker *rw;
212 :
213 921 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
214 : Assert(slotno < max_worker_processes);
215 921 : slot->in_use = true;
216 921 : slot->terminate = false;
217 921 : slot->pid = InvalidPid;
218 921 : slot->generation = 0;
219 921 : rw->rw_shmem_slot = slotno;
220 921 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
221 921 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
222 921 : ++slotno;
223 : }
224 :
225 : /*
226 : * Mark any remaining slots as not in use.
227 : */
228 10086 : while (slotno < max_worker_processes)
229 : {
230 8862 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
231 :
232 8862 : slot->in_use = false;
233 8862 : ++slotno;
234 : }
235 1224 : }
236 :
237 : /*
238 : * Search the postmaster's backend-private list of RegisteredBgWorker objects
239 : * for the one that maps to the given slot number.
240 : */
241 : static RegisteredBgWorker *
242 5514 : FindRegisteredWorkerBySlotNumber(int slotno)
243 : {
244 : dlist_iter iter;
245 :
246 13551 : dlist_foreach(iter, &BackgroundWorkerList)
247 : {
248 : RegisteredBgWorker *rw;
249 :
250 10869 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
251 10869 : if (rw->rw_shmem_slot == slotno)
252 2832 : return rw;
253 : }
254 :
255 2682 : return NULL;
256 : }
257 :
258 : /*
259 : * Notice changes to shared memory made by other backends.
260 : * Accept new worker requests only if allow_new_workers is true.
261 : *
262 : * This code runs in the postmaster, so we must be very careful not to assume
263 : * that shared memory contents are sane. Otherwise, a rogue backend could
264 : * take out the postmaster.
265 : */
266 : void
267 1656 : BackgroundWorkerStateChange(bool allow_new_workers)
268 : {
269 : int slotno;
270 :
271 : /*
272 : * The total number of slots stored in shared memory should match our
273 : * notion of max_worker_processes. If it does not, something is very
274 : * wrong. Further down, we always refer to this value as
275 : * max_worker_processes, in case shared memory gets corrupted while we're
276 : * looping.
277 : */
278 1656 : if (max_worker_processes != BackgroundWorkerData->total_slots)
279 : {
280 0 : ereport(LOG,
281 : (errmsg("inconsistent background worker state (\"max_worker_processes\"=%d, total slots=%d)",
282 : max_worker_processes,
283 : BackgroundWorkerData->total_slots)));
284 0 : return;
285 : }
286 :
287 : /*
288 : * Iterate through slots, looking for newly-registered workers or workers
289 : * who must die.
290 : */
291 15048 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
292 : {
293 13392 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
294 : RegisteredBgWorker *rw;
295 :
296 13392 : if (!slot->in_use)
297 7878 : continue;
298 :
299 : /*
300 : * Make sure we don't see the in_use flag before the updated slot
301 : * contents.
302 : */
303 5514 : pg_read_barrier();
304 :
305 : /* See whether we already know about this worker. */
306 5514 : rw = FindRegisteredWorkerBySlotNumber(slotno);
307 5514 : if (rw != NULL)
308 : {
309 : /*
310 : * In general, the worker data can't change after it's initially
311 : * registered. However, someone can set the terminate flag.
312 : */
313 2832 : if (slot->terminate && !rw->rw_terminate)
314 : {
315 12 : rw->rw_terminate = true;
316 12 : if (rw->rw_pid != 0)
317 12 : kill(rw->rw_pid, SIGTERM);
318 : else
319 : {
320 : /* Report never-started, now-terminated worker as dead. */
321 0 : ReportBackgroundWorkerPID(rw);
322 : }
323 : }
324 2832 : continue;
325 : }
326 :
327 : /*
328 : * If we aren't allowing new workers, then immediately mark it for
329 : * termination; the next stanza will take care of cleaning it up.
330 : * Doing this ensures that any process waiting for the worker will get
331 : * awoken, even though the worker will never be allowed to run.
332 : */
333 2682 : if (!allow_new_workers)
334 4 : slot->terminate = true;
335 :
336 : /*
337 : * If the worker is marked for termination, we don't need to add it to
338 : * the registered workers list; we can just free the slot. However, if
339 : * bgw_notify_pid is set, the process that registered the worker may
340 : * need to know that we've processed the terminate request, so be sure
341 : * to signal it.
342 : */
343 2682 : if (slot->terminate)
344 4 : {
345 : int notify_pid;
346 :
347 : /*
348 : * We need a memory barrier here to make sure that the load of
349 : * bgw_notify_pid and the update of parallel_terminate_count
350 : * complete before the store to in_use.
351 : */
352 4 : notify_pid = slot->worker.bgw_notify_pid;
353 4 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
354 0 : BackgroundWorkerData->parallel_terminate_count++;
355 4 : slot->pid = 0;
356 :
357 4 : pg_memory_barrier();
358 4 : slot->in_use = false;
359 :
360 4 : if (notify_pid != 0)
361 4 : kill(notify_pid, SIGUSR1);
362 :
363 4 : continue;
364 : }
365 :
366 : /*
367 : * Copy the registration data into the registered workers list.
368 : */
369 2678 : rw = MemoryContextAllocExtended(PostmasterContext,
370 : sizeof(RegisteredBgWorker),
371 : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
372 2678 : if (rw == NULL)
373 : {
374 0 : ereport(LOG,
375 : (errcode(ERRCODE_OUT_OF_MEMORY),
376 : errmsg("out of memory")));
377 0 : return;
378 : }
379 :
380 : /*
381 : * Copy strings in a paranoid way. If shared memory is corrupted, the
382 : * source data might not even be NUL-terminated.
383 : */
384 2678 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
385 2678 : slot->worker.bgw_name, BGW_MAXLEN);
386 2678 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
387 2678 : slot->worker.bgw_type, BGW_MAXLEN);
388 2678 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
389 2678 : slot->worker.bgw_library_name, MAXPGPATH);
390 2678 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
391 2678 : slot->worker.bgw_function_name, BGW_MAXLEN);
392 :
393 : /*
394 : * Copy various fixed-size fields.
395 : *
396 : * flags, start_time, and restart_time are examined by the postmaster,
397 : * but nothing too bad will happen if they are corrupted. The
398 : * remaining fields will only be examined by the child process. It
399 : * might crash, but we won't.
400 : */
401 2678 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
402 2678 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
403 2678 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
404 2678 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
405 2678 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
406 :
407 : /*
408 : * Copy the PID to be notified about state changes, but only if the
409 : * postmaster knows about a backend with that PID. It isn't an error
410 : * if the postmaster doesn't know about the PID, because the backend
411 : * that requested the worker could have died (or been killed) just
412 : * after doing so. Nonetheless, at least until we get some experience
413 : * with how this plays out in the wild, log a message at a relative
414 : * high debug level.
415 : */
416 2678 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
417 2678 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
418 : {
419 0 : elog(DEBUG1, "worker notification PID %d is not valid",
420 : (int) rw->rw_worker.bgw_notify_pid);
421 0 : rw->rw_worker.bgw_notify_pid = 0;
422 : }
423 :
424 : /* Initialize postmaster bookkeeping. */
425 2678 : rw->rw_pid = 0;
426 2678 : rw->rw_crashed_at = 0;
427 2678 : rw->rw_shmem_slot = slotno;
428 2678 : rw->rw_terminate = false;
429 :
430 : /* Log it! */
431 2678 : ereport(DEBUG1,
432 : (errmsg_internal("registering background worker \"%s\"",
433 : rw->rw_worker.bgw_name)));
434 :
435 2678 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
436 : }
437 : }
438 :
439 : /*
440 : * Forget about a background worker that's no longer needed.
441 : *
442 : * NOTE: The entry is unlinked from BackgroundWorkerList. If the caller is
443 : * iterating through it, better use a mutable iterator!
444 : *
445 : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
446 : *
447 : * This function must be invoked only in the postmaster.
448 : */
449 : void
450 2663 : ForgetBackgroundWorker(RegisteredBgWorker *rw)
451 : {
452 : BackgroundWorkerSlot *slot;
453 :
454 : Assert(rw->rw_shmem_slot < max_worker_processes);
455 2663 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
456 : Assert(slot->in_use);
457 :
458 : /*
459 : * We need a memory barrier here to make sure that the update of
460 : * parallel_terminate_count completes before the store to in_use.
461 : */
462 2663 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
463 2013 : BackgroundWorkerData->parallel_terminate_count++;
464 :
465 2663 : pg_memory_barrier();
466 2663 : slot->in_use = false;
467 :
468 2663 : ereport(DEBUG1,
469 : (errmsg_internal("unregistering background worker \"%s\"",
470 : rw->rw_worker.bgw_name)));
471 :
472 2663 : dlist_delete(&rw->rw_lnode);
473 2663 : pfree(rw);
474 2663 : }
475 :
476 : /*
477 : * Report the PID of a newly-launched background worker in shared memory.
478 : *
479 : * This function should only be called from the postmaster.
480 : */
481 : void
482 3490 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
483 : {
484 : BackgroundWorkerSlot *slot;
485 :
486 : Assert(rw->rw_shmem_slot < max_worker_processes);
487 3490 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
488 3490 : slot->pid = rw->rw_pid;
489 :
490 3490 : if (rw->rw_worker.bgw_notify_pid != 0)
491 2678 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
492 3490 : }
493 :
494 : /*
495 : * Report that the PID of a background worker is now zero because a
496 : * previously-running background worker has exited.
497 : *
498 : * NOTE: The entry may be unlinked from BackgroundWorkerList. If the caller
499 : * is iterating through it, better use a mutable iterator!
500 : *
501 : * This function should only be called from the postmaster.
502 : */
503 : void
504 3165 : ReportBackgroundWorkerExit(RegisteredBgWorker *rw)
505 : {
506 : BackgroundWorkerSlot *slot;
507 : int notify_pid;
508 :
509 : Assert(rw->rw_shmem_slot < max_worker_processes);
510 3165 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
511 3165 : slot->pid = rw->rw_pid;
512 3165 : notify_pid = rw->rw_worker.bgw_notify_pid;
513 :
514 : /*
515 : * If this worker is slated for deregistration, do that before notifying
516 : * the process which started it. Otherwise, if that process tries to
517 : * reuse the slot immediately, it might not be available yet. In theory
518 : * that could happen anyway if the process checks slot->pid at just the
519 : * wrong moment, but this makes the window narrower.
520 : */
521 3165 : if (rw->rw_terminate ||
522 778 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
523 2663 : ForgetBackgroundWorker(rw);
524 :
525 3165 : if (notify_pid != 0)
526 2621 : kill(notify_pid, SIGUSR1);
527 3165 : }
528 :
529 : /*
530 : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
531 : *
532 : * This function should only be called from the postmaster.
533 : */
534 : void
535 337 : BackgroundWorkerStopNotifications(pid_t pid)
536 : {
537 : dlist_iter iter;
538 :
539 1086 : dlist_foreach(iter, &BackgroundWorkerList)
540 : {
541 : RegisteredBgWorker *rw;
542 :
543 749 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
544 749 : if (rw->rw_worker.bgw_notify_pid == pid)
545 44 : rw->rw_worker.bgw_notify_pid = 0;
546 : }
547 337 : }
548 :
549 : /*
550 : * Cancel any not-yet-started worker requests that have waiting processes.
551 : *
552 : * This is called during a normal ("smart" or "fast") database shutdown.
553 : * After this point, no new background workers will be started, so anything
554 : * that might be waiting for them needs to be kicked off its wait. We do
555 : * that by canceling the bgworker registration entirely, which is perhaps
556 : * overkill, but since we're shutting down it does not matter whether the
557 : * registration record sticks around.
558 : *
559 : * This function should only be called from the postmaster.
560 : */
561 : void
562 619 : ForgetUnstartedBackgroundWorkers(void)
563 : {
564 : dlist_mutable_iter iter;
565 :
566 1222 : dlist_foreach_modify(iter, &BackgroundWorkerList)
567 : {
568 : RegisteredBgWorker *rw;
569 : BackgroundWorkerSlot *slot;
570 :
571 603 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
572 : Assert(rw->rw_shmem_slot < max_worker_processes);
573 603 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
574 :
575 : /* If it's not yet started, and there's someone waiting ... */
576 603 : if (slot->pid == InvalidPid &&
577 56 : rw->rw_worker.bgw_notify_pid != 0)
578 : {
579 : /* ... then zap it, and notify the waiter */
580 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
581 :
582 0 : ForgetBackgroundWorker(rw);
583 0 : if (notify_pid != 0)
584 0 : kill(notify_pid, SIGUSR1);
585 : }
586 : }
587 619 : }
588 :
589 : /*
590 : * Reset background worker crash state.
591 : *
592 : * We assume that, after a crash-and-restart cycle, background workers without
593 : * the never-restart flag should be restarted immediately, instead of waiting
594 : * for bgw_restart_time to elapse. On the other hand, workers with that flag
595 : * should be forgotten immediately, since we won't ever restart them.
596 : *
597 : * This function should only be called from the postmaster.
598 : */
599 : void
600 5 : ResetBackgroundWorkerCrashTimes(void)
601 : {
602 : dlist_mutable_iter iter;
603 :
604 10 : dlist_foreach_modify(iter, &BackgroundWorkerList)
605 : {
606 : RegisteredBgWorker *rw;
607 :
608 5 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
609 :
610 5 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
611 : {
612 : /*
613 : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
614 : * the crash, so forget about them. (If we wait until after the
615 : * crash to forget about them, and they are parallel workers,
616 : * parallel_terminate_count will get incremented after we've
617 : * already zeroed parallel_register_count, which would be bad.)
618 : */
619 0 : ForgetBackgroundWorker(rw);
620 : }
621 : else
622 : {
623 : /*
624 : * The accounting which we do via parallel_register_count and
625 : * parallel_terminate_count would get messed up if a worker marked
626 : * parallel could survive a crash and restart cycle. All such
627 : * workers should be marked BGW_NEVER_RESTART, and thus control
628 : * should never reach this branch.
629 : */
630 : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
631 :
632 : /*
633 : * Allow this worker to be restarted immediately after we finish
634 : * resetting.
635 : */
636 5 : rw->rw_crashed_at = 0;
637 5 : rw->rw_pid = 0;
638 :
639 : /*
640 : * If there was anyone waiting for it, they're history.
641 : */
642 5 : rw->rw_worker.bgw_notify_pid = 0;
643 : }
644 : }
645 5 : }
646 :
647 : /*
648 : * Complain about the BackgroundWorker definition using error level elevel.
649 : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
650 : * which case we won't return at all in the not-OK case).
651 : */
652 : static bool
653 3484 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
654 : {
655 : /* sanity check for flags */
656 :
657 : /*
658 : * We used to support workers not connected to shared memory, but don't
659 : * anymore. Thus this is a required flag now. We're not removing the flag
660 : * for compatibility reasons and because the flag still provides some
661 : * signal when reading code.
662 : */
663 3484 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
664 : {
665 0 : ereport(elevel,
666 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
667 : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
668 : worker->bgw_name)));
669 0 : return false;
670 : }
671 :
672 3484 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
673 : {
674 3475 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
675 : {
676 0 : ereport(elevel,
677 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
678 : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
679 : worker->bgw_name)));
680 0 : return false;
681 : }
682 :
683 : /* XXX other checks? */
684 : }
685 :
686 : /* Interruptible workers require a database connection */
687 3484 : if ((worker->bgw_flags & BGWORKER_INTERRUPTIBLE) &&
688 4 : !(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
689 : {
690 0 : ereport(elevel,
691 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
692 : errmsg("background worker \"%s\": cannot make background workers interruptible without database access",
693 : worker->bgw_name)));
694 0 : return false;
695 : }
696 :
697 3484 : if ((worker->bgw_restart_time < 0 &&
698 2556 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
699 3484 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
700 : {
701 0 : ereport(elevel,
702 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
703 : errmsg("background worker \"%s\": invalid restart interval",
704 : worker->bgw_name)));
705 0 : return false;
706 : }
707 :
708 : /*
709 : * Parallel workers may not be configured for restart, because the
710 : * parallel_register_count/parallel_terminate_count accounting can't
711 : * handle parallel workers lasting through a crash-and-restart cycle.
712 : */
713 3484 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
714 928 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
715 : {
716 0 : ereport(elevel,
717 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
718 : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
719 : worker->bgw_name)));
720 0 : return false;
721 : }
722 :
723 : /*
724 : * If bgw_type is not filled in, use bgw_name.
725 : */
726 3484 : if (strcmp(worker->bgw_type, "") == 0)
727 0 : strcpy(worker->bgw_type, worker->bgw_name);
728 :
729 3484 : return true;
730 : }
731 :
732 : /*
733 : * Main entry point for background worker processes.
734 : */
735 : void
736 3170 : BackgroundWorkerMain(const void *startup_data, size_t startup_data_len)
737 : {
738 : sigjmp_buf local_sigjmp_buf;
739 : BackgroundWorker *worker;
740 : bgworker_main_type entrypt;
741 :
742 3170 : if (startup_data == NULL)
743 0 : elog(FATAL, "unable to find bgworker entry");
744 : Assert(startup_data_len == sizeof(BackgroundWorker));
745 3170 : worker = MemoryContextAlloc(TopMemoryContext, sizeof(BackgroundWorker));
746 3170 : memcpy(worker, startup_data, sizeof(BackgroundWorker));
747 :
748 : /*
749 : * Now that we're done reading the startup data, release postmaster's
750 : * working memory context.
751 : */
752 3170 : if (PostmasterContext)
753 : {
754 3170 : MemoryContextDelete(PostmasterContext);
755 3170 : PostmasterContext = NULL;
756 : }
757 :
758 3170 : MyBgworkerEntry = worker;
759 3170 : init_ps_display(worker->bgw_name);
760 :
761 : Assert(GetProcessingMode() == InitProcessing);
762 :
763 : /* Apply PostAuthDelay */
764 3170 : if (PostAuthDelay > 0)
765 0 : pg_usleep(PostAuthDelay * 1000000L);
766 :
767 : /*
768 : * Set up signal handlers.
769 : */
770 3170 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
771 : {
772 : /*
773 : * SIGINT is used to signal canceling the current action
774 : */
775 3161 : pqsignal(SIGINT, StatementCancelHandler);
776 3161 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
777 3161 : pqsignal(SIGFPE, FloatExceptionHandler);
778 :
779 : /* XXX Any other handlers needed here? */
780 : }
781 : else
782 : {
783 9 : pqsignal(SIGINT, SIG_IGN);
784 9 : pqsignal(SIGUSR1, SIG_IGN);
785 9 : pqsignal(SIGFPE, SIG_IGN);
786 : }
787 3170 : pqsignal(SIGTERM, die);
788 : /* SIGQUIT handler was already set up by InitPostmasterChild */
789 3170 : pqsignal(SIGHUP, SIG_IGN);
790 :
791 3170 : InitializeTimeouts(); /* establishes SIGALRM handler */
792 :
793 3170 : pqsignal(SIGPIPE, SIG_IGN);
794 3170 : pqsignal(SIGUSR2, SIG_IGN);
795 3170 : pqsignal(SIGCHLD, SIG_DFL);
796 :
797 : /*
798 : * If an exception is encountered, processing resumes here.
799 : *
800 : * We just need to clean up, report the error, and go away.
801 : */
802 3170 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
803 : {
804 : /* Since not using PG_TRY, must reset error stack by hand */
805 160 : error_context_stack = NULL;
806 :
807 : /* Prevent interrupts while cleaning up */
808 160 : HOLD_INTERRUPTS();
809 :
810 : /*
811 : * sigsetjmp will have blocked all signals, but we may need to accept
812 : * signals while communicating with our parallel leader. Once we've
813 : * done HOLD_INTERRUPTS() it should be safe to unblock signals.
814 : */
815 160 : BackgroundWorkerUnblockSignals();
816 :
817 : /* Report the error to the parallel leader and the server log */
818 160 : EmitErrorReport();
819 :
820 : /*
821 : * Do we need more cleanup here? For shmem-connected bgworkers, we
822 : * will call InitProcess below, which will install ProcKill as exit
823 : * callback. That will take care of releasing locks, etc.
824 : */
825 :
826 : /* and go away */
827 160 : proc_exit(1);
828 : }
829 :
830 : /* We can now handle ereport(ERROR) */
831 3170 : PG_exception_stack = &local_sigjmp_buf;
832 :
833 : /*
834 : * Create a per-backend PGPROC struct in shared memory. We must do this
835 : * before we can use LWLocks or access any shared memory.
836 : */
837 3170 : InitProcess();
838 :
839 : /*
840 : * Early initialization.
841 : */
842 3170 : BaseInit();
843 :
844 : /*
845 : * Look up the entry point function, loading its library if necessary.
846 : */
847 6340 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
848 3170 : worker->bgw_function_name);
849 :
850 : /*
851 : * Note that in normal processes, we would call InitPostgres here. For a
852 : * worker, however, we don't know what database to connect to, yet; so we
853 : * need to wait until the user code does it via
854 : * BackgroundWorkerInitializeConnection().
855 : */
856 :
857 : /*
858 : * Now invoke the user-defined worker code
859 : */
860 3170 : entrypt(worker->bgw_main_arg);
861 :
862 : /* ... and if it returns, we're done */
863 2040 : proc_exit(0);
864 : }
865 :
866 : /*
867 : * Connect background worker to a database.
868 : */
869 : void
870 505 : BackgroundWorkerInitializeConnection(const char *dbname, const char *username, uint32 flags)
871 : {
872 505 : BackgroundWorker *worker = MyBgworkerEntry;
873 505 : uint32 init_flags = 0; /* never honor session_preload_libraries */
874 :
875 : /* ignore datallowconn and ACL_CONNECT? */
876 505 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
877 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
878 : /* ignore rolcanlogin? */
879 505 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
880 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
881 :
882 : /* XXX is this the right errcode? */
883 505 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
884 0 : ereport(FATAL,
885 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
886 : errmsg("database connection requirement not indicated during registration")));
887 :
888 505 : InitPostgres(dbname, InvalidOid, /* database to connect to */
889 : username, InvalidOid, /* role to connect as */
890 : init_flags,
891 : NULL); /* no out_dbname */
892 :
893 : /* it had better not gotten out of "init" mode yet */
894 505 : if (!IsInitProcessingMode())
895 0 : ereport(ERROR,
896 : (errmsg("invalid processing mode in background worker")));
897 505 : SetProcessingMode(NormalProcessing);
898 505 : }
899 :
900 : /*
901 : * Connect background worker to a database using OIDs.
902 : */
903 : void
904 2656 : BackgroundWorkerInitializeConnectionByOid(Oid dboid, Oid useroid, uint32 flags)
905 : {
906 2656 : BackgroundWorker *worker = MyBgworkerEntry;
907 2656 : uint32 init_flags = 0; /* never honor session_preload_libraries */
908 :
909 : /* ignore datallowconn and ACL_CONNECT? */
910 2656 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
911 2033 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
912 : /* ignore rolcanlogin? */
913 2656 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
914 2013 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
915 :
916 : /* XXX is this the right errcode? */
917 2656 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
918 0 : ereport(FATAL,
919 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
920 : errmsg("database connection requirement not indicated during registration")));
921 :
922 2656 : InitPostgres(NULL, dboid, /* database to connect to */
923 : NULL, useroid, /* role to connect as */
924 : init_flags,
925 : NULL); /* no out_dbname */
926 :
927 : /* it had better not gotten out of "init" mode yet */
928 2653 : if (!IsInitProcessingMode())
929 0 : ereport(ERROR,
930 : (errmsg("invalid processing mode in background worker")));
931 2653 : SetProcessingMode(NormalProcessing);
932 2653 : }
933 :
934 : /*
935 : * Block/unblock signals in a background worker
936 : */
937 : void
938 0 : BackgroundWorkerBlockSignals(void)
939 : {
940 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
941 0 : }
942 :
943 : void
944 3330 : BackgroundWorkerUnblockSignals(void)
945 : {
946 3330 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
947 3330 : }
948 :
949 : /*
950 : * Register a new static background worker.
951 : *
952 : * This can only be called directly from postmaster or in the _PG_init
953 : * function of a module library that's loaded by shared_preload_libraries;
954 : * otherwise it will have no effect.
955 : */
956 : void
957 931 : RegisterBackgroundWorker(BackgroundWorker *worker)
958 : {
959 : RegisteredBgWorker *rw;
960 : static int numworkers = 0;
961 :
962 : /*
963 : * Static background workers can only be registered in the postmaster
964 : * process.
965 : */
966 931 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
967 : {
968 : /*
969 : * In EXEC_BACKEND or single-user mode, we process
970 : * shared_preload_libraries in backend processes too. We cannot
971 : * register static background workers at that stage, but many
972 : * libraries' _PG_init() functions don't distinguish whether they're
973 : * being loaded in the postmaster or in a backend, they just check
974 : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
975 : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
976 : * the background workers should already have been registered when the
977 : * library was loaded in postmaster.
978 : */
979 0 : if (process_shared_preload_libraries_in_progress)
980 0 : return;
981 0 : ereport(LOG,
982 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
983 : errmsg("background worker \"%s\": must be registered in \"shared_preload_libraries\"",
984 : worker->bgw_name)));
985 0 : return;
986 : }
987 :
988 : /*
989 : * Cannot register static background workers after calling
990 : * BackgroundWorkerShmemInit().
991 : */
992 931 : if (BackgroundWorkerData != NULL)
993 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
994 : worker->bgw_name);
995 :
996 931 : ereport(DEBUG1,
997 : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
998 :
999 931 : if (!SanityCheckBackgroundWorker(worker, LOG))
1000 0 : return;
1001 :
1002 931 : if (worker->bgw_notify_pid != 0)
1003 : {
1004 0 : ereport(LOG,
1005 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1006 : errmsg("background worker \"%s\": only dynamic background workers can request notification",
1007 : worker->bgw_name)));
1008 0 : return;
1009 : }
1010 :
1011 : /*
1012 : * Enforce maximum number of workers. Note this is overly restrictive: we
1013 : * could allow more non-shmem-connected workers, because these don't count
1014 : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
1015 : * important to relax this restriction.
1016 : */
1017 931 : if (++numworkers > max_worker_processes)
1018 : {
1019 0 : ereport(LOG,
1020 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
1021 : errmsg("too many background workers"),
1022 : errdetail_plural("Up to %d background worker can be registered with the current settings.",
1023 : "Up to %d background workers can be registered with the current settings.",
1024 : max_worker_processes,
1025 : max_worker_processes),
1026 : errhint("Consider increasing the configuration parameter \"%s\".", "max_worker_processes")));
1027 0 : return;
1028 : }
1029 :
1030 : /*
1031 : * Copy the registration data into the registered workers list.
1032 : */
1033 931 : rw = MemoryContextAllocExtended(PostmasterContext,
1034 : sizeof(RegisteredBgWorker),
1035 : MCXT_ALLOC_NO_OOM);
1036 931 : if (rw == NULL)
1037 : {
1038 0 : ereport(LOG,
1039 : (errcode(ERRCODE_OUT_OF_MEMORY),
1040 : errmsg("out of memory")));
1041 0 : return;
1042 : }
1043 :
1044 931 : rw->rw_worker = *worker;
1045 931 : rw->rw_pid = 0;
1046 931 : rw->rw_crashed_at = 0;
1047 931 : rw->rw_terminate = false;
1048 :
1049 931 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
1050 : }
1051 :
1052 : /*
1053 : * Register a new background worker from a regular backend.
1054 : *
1055 : * Returns true on success and false on failure. Failure typically indicates
1056 : * that no background worker slots are currently available.
1057 : *
1058 : * If handle != NULL, we'll set *handle to a pointer that can subsequently
1059 : * be used as an argument to GetBackgroundWorkerPid(). The caller can
1060 : * free this pointer using pfree(), if desired.
1061 : */
1062 : bool
1063 2553 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
1064 : BackgroundWorkerHandle **handle)
1065 : {
1066 : int slotno;
1067 2553 : bool success = false;
1068 : bool parallel;
1069 2553 : uint64 generation = 0;
1070 :
1071 : /*
1072 : * We can't register dynamic background workers from the postmaster. If
1073 : * this is a standalone backend, we're the only process and can't start
1074 : * any more. In a multi-process environment, it might be theoretically
1075 : * possible, but we don't currently support it due to locking
1076 : * considerations; see comments on the BackgroundWorkerSlot data
1077 : * structure.
1078 : */
1079 2553 : if (!IsUnderPostmaster)
1080 0 : return false;
1081 :
1082 2553 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1083 0 : return false;
1084 :
1085 2553 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1086 :
1087 2553 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1088 :
1089 : /*
1090 : * If this is a parallel worker, check whether there are already too many
1091 : * parallel workers; if so, don't register another one. Our view of
1092 : * parallel_terminate_count may be slightly stale, but that doesn't really
1093 : * matter: we would have gotten the same result if we'd arrived here
1094 : * slightly earlier anyway. There's no help for it, either, since the
1095 : * postmaster must not take locks; a memory barrier wouldn't guarantee
1096 : * anything useful.
1097 : */
1098 2553 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1099 2034 : BackgroundWorkerData->parallel_terminate_count) >=
1100 : max_parallel_workers)
1101 : {
1102 : Assert(BackgroundWorkerData->parallel_register_count -
1103 : BackgroundWorkerData->parallel_terminate_count <=
1104 : MAX_PARALLEL_WORKER_LIMIT);
1105 13 : LWLockRelease(BackgroundWorkerLock);
1106 13 : return false;
1107 : }
1108 :
1109 : /*
1110 : * Look for an unused slot. If we find one, grab it.
1111 : */
1112 8059 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1113 : {
1114 8046 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1115 :
1116 8046 : if (!slot->in_use)
1117 : {
1118 2527 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1119 2527 : slot->pid = InvalidPid; /* indicates not started yet */
1120 2527 : slot->generation++;
1121 2527 : slot->terminate = false;
1122 2527 : generation = slot->generation;
1123 2527 : if (parallel)
1124 2013 : BackgroundWorkerData->parallel_register_count++;
1125 :
1126 : /*
1127 : * Make sure postmaster doesn't see the slot as in use before it
1128 : * sees the new contents.
1129 : */
1130 2527 : pg_write_barrier();
1131 :
1132 2527 : slot->in_use = true;
1133 2527 : success = true;
1134 2527 : break;
1135 : }
1136 : }
1137 :
1138 2540 : LWLockRelease(BackgroundWorkerLock);
1139 :
1140 : /* If we found a slot, tell the postmaster to notice the change. */
1141 2540 : if (success)
1142 2527 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1143 :
1144 : /*
1145 : * If we found a slot and the user has provided a handle, initialize it.
1146 : */
1147 2540 : if (success && handle)
1148 : {
1149 2527 : *handle = palloc_object(BackgroundWorkerHandle);
1150 2527 : (*handle)->slot = slotno;
1151 2527 : (*handle)->generation = generation;
1152 : }
1153 :
1154 2540 : return success;
1155 : }
1156 :
1157 : /*
1158 : * Get the PID of a dynamically-registered background worker.
1159 : *
1160 : * If the worker is determined to be running, the return value will be
1161 : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1162 : * postmaster has not yet attempted to start the worker, the return value will
1163 : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1164 : *
1165 : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1166 : * (because it is configured for automatic restart and exited non-zero),
1167 : * or that the worker is permanently stopped (because it exited with exit
1168 : * code 0, or was not configured for automatic restart), or even that the
1169 : * worker was unregistered without ever starting (either because startup
1170 : * failed and the worker is not configured for automatic restart, or because
1171 : * TerminateBackgroundWorker was used before the worker was successfully
1172 : * started).
1173 : */
1174 : BgwHandleStatus
1175 10056194 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1176 : {
1177 : BackgroundWorkerSlot *slot;
1178 : pid_t pid;
1179 :
1180 : Assert(handle->slot < max_worker_processes);
1181 10056194 : slot = &BackgroundWorkerData->slot[handle->slot];
1182 :
1183 : /*
1184 : * We could probably arrange to synchronize access to data using memory
1185 : * barriers only, but for now, let's just keep it simple and grab the
1186 : * lock. It seems unlikely that there will be enough traffic here to
1187 : * result in meaningful contention.
1188 : */
1189 10056194 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1190 :
1191 : /*
1192 : * The generation number can't be concurrently changed while we hold the
1193 : * lock. The pid, which is updated by the postmaster, can change at any
1194 : * time, but we assume such changes are atomic. So the value we read
1195 : * won't be garbage, but it might be out of date by the time the caller
1196 : * examines it (but that's unavoidable anyway).
1197 : *
1198 : * The in_use flag could be in the process of changing from true to false,
1199 : * but if it is already false then it can't change further.
1200 : */
1201 10056194 : if (handle->generation != slot->generation || !slot->in_use)
1202 2033 : pid = 0;
1203 : else
1204 10054161 : pid = slot->pid;
1205 :
1206 : /* All done. */
1207 10056194 : LWLockRelease(BackgroundWorkerLock);
1208 :
1209 10056194 : if (pid == 0)
1210 2033 : return BGWH_STOPPED;
1211 10054161 : else if (pid == InvalidPid)
1212 1203887 : return BGWH_NOT_YET_STARTED;
1213 8850274 : *pidp = pid;
1214 8850274 : return BGWH_STARTED;
1215 : }
1216 :
1217 : /*
1218 : * Wait for a background worker to start up.
1219 : *
1220 : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1221 : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1222 : * returned. However, if the postmaster has died, we give up and return
1223 : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1224 : * take place.
1225 : *
1226 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1227 : * else we will not be awoken promptly when the worker's state changes.
1228 : */
1229 : BgwHandleStatus
1230 32 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1231 : {
1232 : BgwHandleStatus status;
1233 : int rc;
1234 :
1235 : for (;;)
1236 23 : {
1237 : pid_t pid;
1238 :
1239 55 : CHECK_FOR_INTERRUPTS();
1240 :
1241 55 : status = GetBackgroundWorkerPid(handle, &pid);
1242 55 : if (status == BGWH_STARTED)
1243 32 : *pidp = pid;
1244 55 : if (status != BGWH_NOT_YET_STARTED)
1245 32 : break;
1246 :
1247 23 : rc = WaitLatch(MyLatch,
1248 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1249 : WAIT_EVENT_BGWORKER_STARTUP);
1250 :
1251 23 : if (rc & WL_POSTMASTER_DEATH)
1252 : {
1253 0 : status = BGWH_POSTMASTER_DIED;
1254 0 : break;
1255 : }
1256 :
1257 23 : ResetLatch(MyLatch);
1258 : }
1259 :
1260 32 : return status;
1261 : }
1262 :
1263 : /*
1264 : * Wait for a background worker to stop.
1265 : *
1266 : * If the worker hasn't yet started, or is running, we wait for it to stop
1267 : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1268 : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1269 : * notifies us when a worker's state changes.
1270 : *
1271 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1272 : * else we will not be awoken promptly when the worker's state changes.
1273 : */
1274 : BgwHandleStatus
1275 2034 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1276 : {
1277 : BgwHandleStatus status;
1278 : int rc;
1279 :
1280 : for (;;)
1281 2427 : {
1282 : pid_t pid;
1283 :
1284 4461 : CHECK_FOR_INTERRUPTS();
1285 :
1286 4460 : status = GetBackgroundWorkerPid(handle, &pid);
1287 4460 : if (status == BGWH_STOPPED)
1288 2033 : break;
1289 :
1290 2427 : rc = WaitLatch(MyLatch,
1291 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1292 : WAIT_EVENT_BGWORKER_SHUTDOWN);
1293 :
1294 2427 : if (rc & WL_POSTMASTER_DEATH)
1295 : {
1296 0 : status = BGWH_POSTMASTER_DIED;
1297 0 : break;
1298 : }
1299 :
1300 2427 : ResetLatch(MyLatch);
1301 : }
1302 :
1303 2033 : return status;
1304 : }
1305 :
1306 : /*
1307 : * Instruct the postmaster to terminate a background worker.
1308 : *
1309 : * Note that it's safe to do this without regard to whether the worker is
1310 : * still running, or even if the worker may already have exited and been
1311 : * unregistered.
1312 : */
1313 : void
1314 8 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1315 : {
1316 : BackgroundWorkerSlot *slot;
1317 8 : bool signal_postmaster = false;
1318 :
1319 : Assert(handle->slot < max_worker_processes);
1320 8 : slot = &BackgroundWorkerData->slot[handle->slot];
1321 :
1322 : /* Set terminate flag in shared memory, unless slot has been reused. */
1323 8 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1324 8 : if (handle->generation == slot->generation)
1325 : {
1326 8 : slot->terminate = true;
1327 8 : signal_postmaster = true;
1328 : }
1329 8 : LWLockRelease(BackgroundWorkerLock);
1330 :
1331 : /* Make sure the postmaster notices the change to shared memory. */
1332 8 : if (signal_postmaster)
1333 8 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1334 8 : }
1335 :
1336 : /*
1337 : * Look up (and possibly load) a bgworker entry point function.
1338 : *
1339 : * For functions contained in the core code, we use library name "postgres"
1340 : * and consult the InternalBGWorkers array. External functions are
1341 : * looked up, and loaded if necessary, using load_external_function().
1342 : *
1343 : * The point of this is to pass function names as strings across process
1344 : * boundaries. We can't pass actual function addresses because of the
1345 : * possibility that the function has been loaded at a different address
1346 : * in a different process. This is obviously a hazard for functions in
1347 : * loadable libraries, but it can happen even for functions in the core code
1348 : * on platforms using EXEC_BACKEND (e.g., Windows).
1349 : *
1350 : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1351 : * in favor of applying load_external_function() for core functions too;
1352 : * but that raises portability issues that are not worth addressing now.
1353 : */
1354 : static bgworker_main_type
1355 3170 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1356 : {
1357 : /*
1358 : * If the function is to be loaded from postgres itself, search the
1359 : * InternalBGWorkers array.
1360 : */
1361 3170 : if (strcmp(libraryname, "postgres") == 0)
1362 : {
1363 : int i;
1364 :
1365 10912 : for (i = 0; i < lengthof(InternalBGWorkers); i++)
1366 : {
1367 10912 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1368 3152 : return InternalBGWorkers[i].fn_addr;
1369 : }
1370 :
1371 : /* We can only reach this by programming error. */
1372 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1373 : }
1374 :
1375 : /* Otherwise load from external library. */
1376 18 : return (bgworker_main_type)
1377 18 : load_external_function(libraryname, funcname, true, NULL);
1378 : }
1379 :
1380 : /*
1381 : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1382 : * not a valid background worker.
1383 : *
1384 : * The return value is in static memory belonging to this function, so it has
1385 : * to be used before calling this function again. This is so that the caller
1386 : * doesn't have to worry about the background worker locking protocol.
1387 : */
1388 : const char *
1389 926 : GetBackgroundWorkerTypeByPid(pid_t pid)
1390 : {
1391 : int slotno;
1392 926 : bool found = false;
1393 : static char result[BGW_MAXLEN];
1394 :
1395 926 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1396 :
1397 1062 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1398 : {
1399 1062 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1400 :
1401 1062 : if (slot->pid > 0 && slot->pid == pid)
1402 : {
1403 926 : strcpy(result, slot->worker.bgw_type);
1404 926 : found = true;
1405 926 : break;
1406 : }
1407 : }
1408 :
1409 926 : LWLockRelease(BackgroundWorkerLock);
1410 :
1411 926 : if (!found)
1412 0 : return NULL;
1413 :
1414 926 : return result;
1415 : }
1416 :
1417 : /*
1418 : * Terminate all background workers connected to the given database, if the
1419 : * workers can be interrupted.
1420 : */
1421 : void
1422 7 : TerminateBackgroundWorkersForDatabase(Oid databaseId)
1423 : {
1424 7 : bool signal_postmaster = false;
1425 :
1426 7 : elog(DEBUG1, "attempting worker termination for database %u",
1427 : databaseId);
1428 :
1429 7 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1430 :
1431 : /*
1432 : * Iterate through slots, looking for workers connected to the given
1433 : * database.
1434 : */
1435 63 : for (int slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1436 : {
1437 56 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1438 :
1439 56 : if (slot->in_use &&
1440 14 : (slot->worker.bgw_flags & BGWORKER_INTERRUPTIBLE))
1441 : {
1442 4 : PGPROC *proc = BackendPidGetProc(slot->pid);
1443 :
1444 4 : if (proc && proc->databaseId == databaseId)
1445 : {
1446 4 : slot->terminate = true;
1447 4 : signal_postmaster = true;
1448 :
1449 4 : elog(DEBUG1, "termination requested for worker (PID %d) on database %u",
1450 : (int) slot->pid, databaseId);
1451 : }
1452 : }
1453 : }
1454 :
1455 7 : LWLockRelease(BackgroundWorkerLock);
1456 :
1457 : /* Make sure the postmaster notices the change to shared memory. */
1458 7 : if (signal_postmaster)
1459 4 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1460 7 : }
|
