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/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/procarray.h"
30 : #include "storage/procsignal.h"
31 : #include "storage/shmem.h"
32 : #include "tcop/tcopprot.h"
33 : #include "utils/ascii.h"
34 : #include "utils/memutils.h"
35 : #include "utils/ps_status.h"
36 : #include "utils/timeout.h"
37 :
38 : /*
39 : * The postmaster's list of registered background workers, in private memory.
40 : */
41 : dlist_head BackgroundWorkerList = DLIST_STATIC_INIT(BackgroundWorkerList);
42 :
43 : /*
44 : * BackgroundWorkerSlots exist in shared memory and can be accessed (via
45 : * the BackgroundWorkerArray) by both the postmaster and by regular backends.
46 : * However, the postmaster cannot take locks, even spinlocks, because this
47 : * might allow it to crash or become wedged if shared memory gets corrupted.
48 : * Such an outcome is intolerable. Therefore, we need a lockless protocol
49 : * for coordinating access to this data.
50 : *
51 : * The 'in_use' flag is used to hand off responsibility for the slot between
52 : * the postmaster and the rest of the system. When 'in_use' is false,
53 : * the postmaster will ignore the slot entirely, except for the 'in_use' flag
54 : * itself, which it may read. In this state, regular backends may modify the
55 : * slot. Once a backend sets 'in_use' to true, the slot becomes the
56 : * responsibility of the postmaster. Regular backends may no longer modify it,
57 : * but the postmaster may examine it. Thus, a backend initializing a slot
58 : * must fully initialize the slot - and insert a write memory barrier - before
59 : * marking it as in use.
60 : *
61 : * As an exception, however, even when the slot is in use, regular backends
62 : * may set the 'terminate' flag for a slot, telling the postmaster not
63 : * to restart it. Once the background worker is no longer running, the slot
64 : * will be released for reuse.
65 : *
66 : * In addition to coordinating with the postmaster, backends modifying this
67 : * data structure must coordinate with each other. Since they can take locks,
68 : * this is straightforward: any backend wishing to manipulate a slot must
69 : * take BackgroundWorkerLock in exclusive mode. Backends wishing to read
70 : * data that might get concurrently modified by other backends should take
71 : * this lock in shared mode. No matter what, backends reading this data
72 : * structure must be able to tolerate concurrent modifications by the
73 : * postmaster.
74 : */
75 : typedef struct BackgroundWorkerSlot
76 : {
77 : bool in_use;
78 : bool terminate;
79 : pid_t pid; /* InvalidPid = not started yet; 0 = dead */
80 : uint64 generation; /* incremented when slot is recycled */
81 : BackgroundWorker worker;
82 : } BackgroundWorkerSlot;
83 :
84 : /*
85 : * In order to limit the total number of parallel workers (according to
86 : * max_parallel_workers GUC), we maintain the number of active parallel
87 : * workers. Since the postmaster cannot take locks, two variables are used for
88 : * this purpose: the number of registered parallel workers (modified by the
89 : * backends, protected by BackgroundWorkerLock) and the number of terminated
90 : * parallel workers (modified only by the postmaster, lockless). The active
91 : * number of parallel workers is the number of registered workers minus the
92 : * terminated ones. These counters can of course overflow, but it's not
93 : * important here since the subtraction will still give the right number.
94 : */
95 : typedef struct BackgroundWorkerArray
96 : {
97 : int total_slots;
98 : uint32 parallel_register_count;
99 : uint32 parallel_terminate_count;
100 : BackgroundWorkerSlot slot[FLEXIBLE_ARRAY_MEMBER];
101 : } BackgroundWorkerArray;
102 :
103 : struct BackgroundWorkerHandle
104 : {
105 : int slot;
106 : uint64 generation;
107 : };
108 :
109 : static BackgroundWorkerArray *BackgroundWorkerData;
110 :
111 : /*
112 : * List of internal background worker entry points. We need this for
113 : * reasons explained in LookupBackgroundWorkerFunction(), below.
114 : */
115 : static const struct
116 : {
117 : const char *fn_name;
118 : bgworker_main_type fn_addr;
119 : } InternalBGWorkers[] =
120 :
121 : {
122 : {
123 : .fn_name = "ParallelWorkerMain",
124 : .fn_addr = ParallelWorkerMain
125 : },
126 : {
127 : .fn_name = "ApplyLauncherMain",
128 : .fn_addr = ApplyLauncherMain
129 : },
130 : {
131 : .fn_name = "ApplyWorkerMain",
132 : .fn_addr = ApplyWorkerMain
133 : },
134 : {
135 : .fn_name = "ParallelApplyWorkerMain",
136 : .fn_addr = ParallelApplyWorkerMain
137 : },
138 : {
139 : .fn_name = "TableSyncWorkerMain",
140 : .fn_addr = TableSyncWorkerMain
141 : },
142 : {
143 : .fn_name = "SequenceSyncWorkerMain",
144 : .fn_addr = SequenceSyncWorkerMain
145 : }
146 : };
147 :
148 : /* Private functions. */
149 : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
150 :
151 :
152 : /*
153 : * Calculate shared memory needed.
154 : */
155 : Size
156 3297 : BackgroundWorkerShmemSize(void)
157 : {
158 : Size size;
159 :
160 : /* Array of workers is variably sized. */
161 3297 : size = offsetof(BackgroundWorkerArray, slot);
162 3297 : size = add_size(size, mul_size(max_worker_processes,
163 : sizeof(BackgroundWorkerSlot)));
164 :
165 3297 : return size;
166 : }
167 :
168 : /*
169 : * Initialize shared memory.
170 : */
171 : void
172 1150 : BackgroundWorkerShmemInit(void)
173 : {
174 : bool found;
175 :
176 1150 : BackgroundWorkerData = ShmemInitStruct("Background Worker Data",
177 : BackgroundWorkerShmemSize(),
178 : &found);
179 1150 : if (!IsUnderPostmaster)
180 : {
181 : dlist_iter iter;
182 1150 : int slotno = 0;
183 :
184 1150 : BackgroundWorkerData->total_slots = max_worker_processes;
185 1150 : BackgroundWorkerData->parallel_register_count = 0;
186 1150 : BackgroundWorkerData->parallel_terminate_count = 0;
187 :
188 : /*
189 : * Copy contents of worker list into shared memory. Record the shared
190 : * memory slot assigned to each worker. This ensures a 1-to-1
191 : * correspondence between the postmaster's private list and the array
192 : * in shared memory.
193 : */
194 2025 : dlist_foreach(iter, &BackgroundWorkerList)
195 : {
196 875 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
197 : RegisteredBgWorker *rw;
198 :
199 875 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
200 : Assert(slotno < max_worker_processes);
201 875 : slot->in_use = true;
202 875 : slot->terminate = false;
203 875 : slot->pid = InvalidPid;
204 875 : slot->generation = 0;
205 875 : rw->rw_shmem_slot = slotno;
206 875 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
207 875 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
208 875 : ++slotno;
209 : }
210 :
211 : /*
212 : * Mark any remaining slots as not in use.
213 : */
214 9466 : while (slotno < max_worker_processes)
215 : {
216 8316 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
217 :
218 8316 : slot->in_use = false;
219 8316 : ++slotno;
220 : }
221 : }
222 : else
223 : Assert(found);
224 1150 : }
225 :
226 : /*
227 : * Search the postmaster's backend-private list of RegisteredBgWorker objects
228 : * for the one that maps to the given slot number.
229 : */
230 : static RegisteredBgWorker *
231 4342 : FindRegisteredWorkerBySlotNumber(int slotno)
232 : {
233 : dlist_iter iter;
234 :
235 10795 : dlist_foreach(iter, &BackgroundWorkerList)
236 : {
237 : RegisteredBgWorker *rw;
238 :
239 8742 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
240 8742 : if (rw->rw_shmem_slot == slotno)
241 2289 : return rw;
242 : }
243 :
244 2053 : return NULL;
245 : }
246 :
247 : /*
248 : * Notice changes to shared memory made by other backends.
249 : * Accept new worker requests only if allow_new_workers is true.
250 : *
251 : * This code runs in the postmaster, so we must be very careful not to assume
252 : * that shared memory contents are sane. Otherwise, a rogue backend could
253 : * take out the postmaster.
254 : */
255 : void
256 1285 : BackgroundWorkerStateChange(bool allow_new_workers)
257 : {
258 : int slotno;
259 :
260 : /*
261 : * The total number of slots stored in shared memory should match our
262 : * notion of max_worker_processes. If it does not, something is very
263 : * wrong. Further down, we always refer to this value as
264 : * max_worker_processes, in case shared memory gets corrupted while we're
265 : * looping.
266 : */
267 1285 : if (max_worker_processes != BackgroundWorkerData->total_slots)
268 : {
269 0 : ereport(LOG,
270 : (errmsg("inconsistent background worker state (\"max_worker_processes\"=%d, total slots=%d)",
271 : max_worker_processes,
272 : BackgroundWorkerData->total_slots)));
273 0 : return;
274 : }
275 :
276 : /*
277 : * Iterate through slots, looking for newly-registered workers or workers
278 : * who must die.
279 : */
280 11709 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
281 : {
282 10424 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
283 : RegisteredBgWorker *rw;
284 :
285 10424 : if (!slot->in_use)
286 6082 : continue;
287 :
288 : /*
289 : * Make sure we don't see the in_use flag before the updated slot
290 : * contents.
291 : */
292 4342 : pg_read_barrier();
293 :
294 : /* See whether we already know about this worker. */
295 4342 : rw = FindRegisteredWorkerBySlotNumber(slotno);
296 4342 : if (rw != NULL)
297 : {
298 : /*
299 : * In general, the worker data can't change after it's initially
300 : * registered. However, someone can set the terminate flag.
301 : */
302 2289 : if (slot->terminate && !rw->rw_terminate)
303 : {
304 10 : rw->rw_terminate = true;
305 10 : if (rw->rw_pid != 0)
306 10 : kill(rw->rw_pid, SIGTERM);
307 : else
308 : {
309 : /* Report never-started, now-terminated worker as dead. */
310 0 : ReportBackgroundWorkerPID(rw);
311 : }
312 : }
313 2289 : continue;
314 : }
315 :
316 : /*
317 : * If we aren't allowing new workers, then immediately mark it for
318 : * termination; the next stanza will take care of cleaning it up.
319 : * Doing this ensures that any process waiting for the worker will get
320 : * awoken, even though the worker will never be allowed to run.
321 : */
322 2053 : if (!allow_new_workers)
323 1 : slot->terminate = true;
324 :
325 : /*
326 : * If the worker is marked for termination, we don't need to add it to
327 : * the registered workers list; we can just free the slot. However, if
328 : * bgw_notify_pid is set, the process that registered the worker may
329 : * need to know that we've processed the terminate request, so be sure
330 : * to signal it.
331 : */
332 2053 : if (slot->terminate)
333 1 : {
334 : int notify_pid;
335 :
336 : /*
337 : * We need a memory barrier here to make sure that the load of
338 : * bgw_notify_pid and the update of parallel_terminate_count
339 : * complete before the store to in_use.
340 : */
341 1 : notify_pid = slot->worker.bgw_notify_pid;
342 1 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
343 0 : BackgroundWorkerData->parallel_terminate_count++;
344 1 : slot->pid = 0;
345 :
346 1 : pg_memory_barrier();
347 1 : slot->in_use = false;
348 :
349 1 : if (notify_pid != 0)
350 1 : kill(notify_pid, SIGUSR1);
351 :
352 1 : continue;
353 : }
354 :
355 : /*
356 : * Copy the registration data into the registered workers list.
357 : */
358 2052 : rw = MemoryContextAllocExtended(PostmasterContext,
359 : sizeof(RegisteredBgWorker),
360 : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
361 2052 : if (rw == NULL)
362 : {
363 0 : ereport(LOG,
364 : (errcode(ERRCODE_OUT_OF_MEMORY),
365 : errmsg("out of memory")));
366 0 : return;
367 : }
368 :
369 : /*
370 : * Copy strings in a paranoid way. If shared memory is corrupted, the
371 : * source data might not even be NUL-terminated.
372 : */
373 2052 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
374 2052 : slot->worker.bgw_name, BGW_MAXLEN);
375 2052 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
376 2052 : slot->worker.bgw_type, BGW_MAXLEN);
377 2052 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
378 2052 : slot->worker.bgw_library_name, MAXPGPATH);
379 2052 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
380 2052 : slot->worker.bgw_function_name, BGW_MAXLEN);
381 :
382 : /*
383 : * Copy various fixed-size fields.
384 : *
385 : * flags, start_time, and restart_time are examined by the postmaster,
386 : * but nothing too bad will happen if they are corrupted. The
387 : * remaining fields will only be examined by the child process. It
388 : * might crash, but we won't.
389 : */
390 2052 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
391 2052 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
392 2052 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
393 2052 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
394 2052 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
395 :
396 : /*
397 : * Copy the PID to be notified about state changes, but only if the
398 : * postmaster knows about a backend with that PID. It isn't an error
399 : * if the postmaster doesn't know about the PID, because the backend
400 : * that requested the worker could have died (or been killed) just
401 : * after doing so. Nonetheless, at least until we get some experience
402 : * with how this plays out in the wild, log a message at a relative
403 : * high debug level.
404 : */
405 2052 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
406 2052 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
407 : {
408 0 : elog(DEBUG1, "worker notification PID %d is not valid",
409 : (int) rw->rw_worker.bgw_notify_pid);
410 0 : rw->rw_worker.bgw_notify_pid = 0;
411 : }
412 :
413 : /* Initialize postmaster bookkeeping. */
414 2052 : rw->rw_pid = 0;
415 2052 : rw->rw_crashed_at = 0;
416 2052 : rw->rw_shmem_slot = slotno;
417 2052 : rw->rw_terminate = false;
418 :
419 : /* Log it! */
420 2052 : ereport(DEBUG1,
421 : (errmsg_internal("registering background worker \"%s\"",
422 : rw->rw_worker.bgw_name)));
423 :
424 2052 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
425 : }
426 : }
427 :
428 : /*
429 : * Forget about a background worker that's no longer needed.
430 : *
431 : * NOTE: The entry is unlinked from BackgroundWorkerList. If the caller is
432 : * iterating through it, better use a mutable iterator!
433 : *
434 : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
435 : *
436 : * This function must be invoked only in the postmaster.
437 : */
438 : void
439 2026 : ForgetBackgroundWorker(RegisteredBgWorker *rw)
440 : {
441 : BackgroundWorkerSlot *slot;
442 :
443 : Assert(rw->rw_shmem_slot < max_worker_processes);
444 2026 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
445 : Assert(slot->in_use);
446 :
447 : /*
448 : * We need a memory barrier here to make sure that the update of
449 : * parallel_terminate_count completes before the store to in_use.
450 : */
451 2026 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
452 1482 : BackgroundWorkerData->parallel_terminate_count++;
453 :
454 2026 : pg_memory_barrier();
455 2026 : slot->in_use = false;
456 :
457 2026 : ereport(DEBUG1,
458 : (errmsg_internal("unregistering background worker \"%s\"",
459 : rw->rw_worker.bgw_name)));
460 :
461 2026 : dlist_delete(&rw->rw_lnode);
462 2026 : pfree(rw);
463 2026 : }
464 :
465 : /*
466 : * Report the PID of a newly-launched background worker in shared memory.
467 : *
468 : * This function should only be called from the postmaster.
469 : */
470 : void
471 2816 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
472 : {
473 : BackgroundWorkerSlot *slot;
474 :
475 : Assert(rw->rw_shmem_slot < max_worker_processes);
476 2816 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
477 2816 : slot->pid = rw->rw_pid;
478 :
479 2816 : if (rw->rw_worker.bgw_notify_pid != 0)
480 2052 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
481 2816 : }
482 :
483 : /*
484 : * Report that the PID of a background worker is now zero because a
485 : * previously-running background worker has exited.
486 : *
487 : * NOTE: The entry may be unlinked from BackgroundWorkerList. If the caller
488 : * is iterating through it, better use a mutable iterator!
489 : *
490 : * This function should only be called from the postmaster.
491 : */
492 : void
493 2485 : ReportBackgroundWorkerExit(RegisteredBgWorker *rw)
494 : {
495 : BackgroundWorkerSlot *slot;
496 : int notify_pid;
497 :
498 : Assert(rw->rw_shmem_slot < max_worker_processes);
499 2485 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
500 2485 : slot->pid = rw->rw_pid;
501 2485 : notify_pid = rw->rw_worker.bgw_notify_pid;
502 :
503 : /*
504 : * If this worker is slated for deregistration, do that before notifying
505 : * the process which started it. Otherwise, if that process tries to
506 : * reuse the slot immediately, it might not be available yet. In theory
507 : * that could happen anyway if the process checks slot->pid at just the
508 : * wrong moment, but this makes the window narrower.
509 : */
510 2485 : if (rw->rw_terminate ||
511 695 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
512 2026 : ForgetBackgroundWorker(rw);
513 :
514 2485 : if (notify_pid != 0)
515 1994 : kill(notify_pid, SIGUSR1);
516 2485 : }
517 :
518 : /*
519 : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
520 : *
521 : * This function should only be called from the postmaster.
522 : */
523 : void
524 268 : BackgroundWorkerStopNotifications(pid_t pid)
525 : {
526 : dlist_iter iter;
527 :
528 854 : dlist_foreach(iter, &BackgroundWorkerList)
529 : {
530 : RegisteredBgWorker *rw;
531 :
532 586 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
533 586 : if (rw->rw_worker.bgw_notify_pid == pid)
534 33 : rw->rw_worker.bgw_notify_pid = 0;
535 : }
536 268 : }
537 :
538 : /*
539 : * Cancel any not-yet-started worker requests that have waiting processes.
540 : *
541 : * This is called during a normal ("smart" or "fast") database shutdown.
542 : * After this point, no new background workers will be started, so anything
543 : * that might be waiting for them needs to be kicked off its wait. We do
544 : * that by canceling the bgworker registration entirely, which is perhaps
545 : * overkill, but since we're shutting down it does not matter whether the
546 : * registration record sticks around.
547 : *
548 : * This function should only be called from the postmaster.
549 : */
550 : void
551 574 : ForgetUnstartedBackgroundWorkers(void)
552 : {
553 : dlist_mutable_iter iter;
554 :
555 1140 : dlist_foreach_modify(iter, &BackgroundWorkerList)
556 : {
557 : RegisteredBgWorker *rw;
558 : BackgroundWorkerSlot *slot;
559 :
560 566 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
561 : Assert(rw->rw_shmem_slot < max_worker_processes);
562 566 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
563 :
564 : /* If it's not yet started, and there's someone waiting ... */
565 566 : if (slot->pid == InvalidPid &&
566 58 : rw->rw_worker.bgw_notify_pid != 0)
567 : {
568 : /* ... then zap it, and notify the waiter */
569 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
570 :
571 0 : ForgetBackgroundWorker(rw);
572 0 : if (notify_pid != 0)
573 0 : kill(notify_pid, SIGUSR1);
574 : }
575 : }
576 574 : }
577 :
578 : /*
579 : * Reset background worker crash state.
580 : *
581 : * We assume that, after a crash-and-restart cycle, background workers without
582 : * the never-restart flag should be restarted immediately, instead of waiting
583 : * for bgw_restart_time to elapse. On the other hand, workers with that flag
584 : * should be forgotten immediately, since we won't ever restart them.
585 : *
586 : * This function should only be called from the postmaster.
587 : */
588 : void
589 5 : ResetBackgroundWorkerCrashTimes(void)
590 : {
591 : dlist_mutable_iter iter;
592 :
593 10 : dlist_foreach_modify(iter, &BackgroundWorkerList)
594 : {
595 : RegisteredBgWorker *rw;
596 :
597 5 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
598 :
599 5 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
600 : {
601 : /*
602 : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
603 : * the crash, so forget about them. (If we wait until after the
604 : * crash to forget about them, and they are parallel workers,
605 : * parallel_terminate_count will get incremented after we've
606 : * already zeroed parallel_register_count, which would be bad.)
607 : */
608 0 : ForgetBackgroundWorker(rw);
609 : }
610 : else
611 : {
612 : /*
613 : * The accounting which we do via parallel_register_count and
614 : * parallel_terminate_count would get messed up if a worker marked
615 : * parallel could survive a crash and restart cycle. All such
616 : * workers should be marked BGW_NEVER_RESTART, and thus control
617 : * should never reach this branch.
618 : */
619 : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
620 :
621 : /*
622 : * Allow this worker to be restarted immediately after we finish
623 : * resetting.
624 : */
625 5 : rw->rw_crashed_at = 0;
626 5 : rw->rw_pid = 0;
627 :
628 : /*
629 : * If there was anyone waiting for it, they're history.
630 : */
631 5 : rw->rw_worker.bgw_notify_pid = 0;
632 : }
633 : }
634 5 : }
635 :
636 : /*
637 : * Complain about the BackgroundWorker definition using error level elevel.
638 : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
639 : * which case we won't return at all in the not-OK case).
640 : */
641 : static bool
642 2819 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
643 : {
644 : /* sanity check for flags */
645 :
646 : /*
647 : * We used to support workers not connected to shared memory, but don't
648 : * anymore. Thus this is a required flag now. We're not removing the flag
649 : * for compatibility reasons and because the flag still provides some
650 : * signal when reading code.
651 : */
652 2819 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
653 : {
654 0 : ereport(elevel,
655 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
656 : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
657 : worker->bgw_name)));
658 0 : return false;
659 : }
660 :
661 2819 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
662 : {
663 2810 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
664 : {
665 0 : ereport(elevel,
666 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
667 : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
668 : worker->bgw_name)));
669 0 : return false;
670 : }
671 :
672 : /* XXX other checks? */
673 : }
674 :
675 : /* Interruptible workers require a database connection */
676 2819 : if ((worker->bgw_flags & BGWORKER_INTERRUPTIBLE) &&
677 4 : !(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
678 : {
679 0 : ereport(elevel,
680 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
681 : errmsg("background worker \"%s\": cannot make background workers interruptible without database access",
682 : worker->bgw_name)));
683 0 : return false;
684 : }
685 :
686 2819 : if ((worker->bgw_restart_time < 0 &&
687 1945 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
688 2819 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
689 : {
690 0 : ereport(elevel,
691 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
692 : errmsg("background worker \"%s\": invalid restart interval",
693 : worker->bgw_name)));
694 0 : return false;
695 : }
696 :
697 : /*
698 : * Parallel workers may not be configured for restart, because the
699 : * parallel_register_count/parallel_terminate_count accounting can't
700 : * handle parallel workers lasting through a crash-and-restart cycle.
701 : */
702 2819 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
703 874 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
704 : {
705 0 : ereport(elevel,
706 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
707 : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
708 : worker->bgw_name)));
709 0 : return false;
710 : }
711 :
712 : /*
713 : * If bgw_type is not filled in, use bgw_name.
714 : */
715 2819 : if (strcmp(worker->bgw_type, "") == 0)
716 0 : strcpy(worker->bgw_type, worker->bgw_name);
717 :
718 2819 : return true;
719 : }
720 :
721 : /*
722 : * Main entry point for background worker processes.
723 : */
724 : void
725 2503 : BackgroundWorkerMain(const void *startup_data, size_t startup_data_len)
726 : {
727 : sigjmp_buf local_sigjmp_buf;
728 : BackgroundWorker *worker;
729 : bgworker_main_type entrypt;
730 :
731 2503 : if (startup_data == NULL)
732 0 : elog(FATAL, "unable to find bgworker entry");
733 : Assert(startup_data_len == sizeof(BackgroundWorker));
734 2503 : worker = MemoryContextAlloc(TopMemoryContext, sizeof(BackgroundWorker));
735 2503 : memcpy(worker, startup_data, sizeof(BackgroundWorker));
736 :
737 : /*
738 : * Now that we're done reading the startup data, release postmaster's
739 : * working memory context.
740 : */
741 2503 : if (PostmasterContext)
742 : {
743 2503 : MemoryContextDelete(PostmasterContext);
744 2503 : PostmasterContext = NULL;
745 : }
746 :
747 2503 : MyBgworkerEntry = worker;
748 2503 : init_ps_display(worker->bgw_name);
749 :
750 : Assert(GetProcessingMode() == InitProcessing);
751 :
752 : /* Apply PostAuthDelay */
753 2503 : if (PostAuthDelay > 0)
754 0 : pg_usleep(PostAuthDelay * 1000000L);
755 :
756 : /*
757 : * Set up signal handlers.
758 : */
759 2503 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
760 : {
761 : /*
762 : * SIGINT is used to signal canceling the current action
763 : */
764 2494 : pqsignal(SIGINT, StatementCancelHandler);
765 2494 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
766 2494 : pqsignal(SIGFPE, FloatExceptionHandler);
767 :
768 : /* XXX Any other handlers needed here? */
769 : }
770 : else
771 : {
772 9 : pqsignal(SIGINT, SIG_IGN);
773 9 : pqsignal(SIGUSR1, SIG_IGN);
774 9 : pqsignal(SIGFPE, SIG_IGN);
775 : }
776 2503 : pqsignal(SIGTERM, die);
777 : /* SIGQUIT handler was already set up by InitPostmasterChild */
778 2503 : pqsignal(SIGHUP, SIG_IGN);
779 :
780 2503 : InitializeTimeouts(); /* establishes SIGALRM handler */
781 :
782 2503 : pqsignal(SIGPIPE, SIG_IGN);
783 2503 : pqsignal(SIGUSR2, SIG_IGN);
784 2503 : pqsignal(SIGCHLD, SIG_DFL);
785 :
786 : /*
787 : * If an exception is encountered, processing resumes here.
788 : *
789 : * We just need to clean up, report the error, and go away.
790 : */
791 2503 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
792 : {
793 : /* Since not using PG_TRY, must reset error stack by hand */
794 137 : error_context_stack = NULL;
795 :
796 : /* Prevent interrupts while cleaning up */
797 137 : HOLD_INTERRUPTS();
798 :
799 : /*
800 : * sigsetjmp will have blocked all signals, but we may need to accept
801 : * signals while communicating with our parallel leader. Once we've
802 : * done HOLD_INTERRUPTS() it should be safe to unblock signals.
803 : */
804 137 : BackgroundWorkerUnblockSignals();
805 :
806 : /* Report the error to the parallel leader and the server log */
807 137 : EmitErrorReport();
808 :
809 : /*
810 : * Do we need more cleanup here? For shmem-connected bgworkers, we
811 : * will call InitProcess below, which will install ProcKill as exit
812 : * callback. That will take care of releasing locks, etc.
813 : */
814 :
815 : /* and go away */
816 137 : proc_exit(1);
817 : }
818 :
819 : /* We can now handle ereport(ERROR) */
820 2503 : PG_exception_stack = &local_sigjmp_buf;
821 :
822 : /*
823 : * Create a per-backend PGPROC struct in shared memory. We must do this
824 : * before we can use LWLocks or access any shared memory.
825 : */
826 2503 : InitProcess();
827 :
828 : /*
829 : * Early initialization.
830 : */
831 2503 : BaseInit();
832 :
833 : /*
834 : * Look up the entry point function, loading its library if necessary.
835 : */
836 5006 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
837 2503 : worker->bgw_function_name);
838 :
839 : /*
840 : * Note that in normal processes, we would call InitPostgres here. For a
841 : * worker, however, we don't know what database to connect to, yet; so we
842 : * need to wait until the user code does it via
843 : * BackgroundWorkerInitializeConnection().
844 : */
845 :
846 : /*
847 : * Now invoke the user-defined worker code
848 : */
849 2503 : entrypt(worker->bgw_main_arg);
850 :
851 : /* ... and if it returns, we're done */
852 1479 : proc_exit(0);
853 : }
854 :
855 : /*
856 : * Connect background worker to a database.
857 : */
858 : void
859 462 : BackgroundWorkerInitializeConnection(const char *dbname, const char *username, uint32 flags)
860 : {
861 462 : BackgroundWorker *worker = MyBgworkerEntry;
862 462 : bits32 init_flags = 0; /* never honor session_preload_libraries */
863 :
864 : /* ignore datallowconn and ACL_CONNECT? */
865 462 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
866 0 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
867 : /* ignore rolcanlogin? */
868 462 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
869 0 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
870 :
871 : /* XXX is this the right errcode? */
872 462 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
873 0 : ereport(FATAL,
874 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
875 : errmsg("database connection requirement not indicated during registration")));
876 :
877 462 : InitPostgres(dbname, InvalidOid, /* database to connect to */
878 : username, InvalidOid, /* role to connect as */
879 : init_flags,
880 : NULL); /* no out_dbname */
881 :
882 : /* it had better not gotten out of "init" mode yet */
883 462 : if (!IsInitProcessingMode())
884 0 : ereport(ERROR,
885 : (errmsg("invalid processing mode in background worker")));
886 462 : SetProcessingMode(NormalProcessing);
887 462 : }
888 :
889 : /*
890 : * Connect background worker to a database using OIDs.
891 : */
892 : void
893 2032 : BackgroundWorkerInitializeConnectionByOid(Oid dboid, Oid useroid, uint32 flags)
894 : {
895 2032 : BackgroundWorker *worker = MyBgworkerEntry;
896 2032 : bits32 init_flags = 0; /* never honor session_preload_libraries */
897 :
898 : /* ignore datallowconn and ACL_CONNECT? */
899 2032 : if (flags & BGWORKER_BYPASS_ALLOWCONN)
900 1482 : init_flags |= INIT_PG_OVERRIDE_ALLOW_CONNS;
901 : /* ignore rolcanlogin? */
902 2032 : if (flags & BGWORKER_BYPASS_ROLELOGINCHECK)
903 1482 : init_flags |= INIT_PG_OVERRIDE_ROLE_LOGIN;
904 :
905 : /* XXX is this the right errcode? */
906 2032 : if (!(worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION))
907 0 : ereport(FATAL,
908 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
909 : errmsg("database connection requirement not indicated during registration")));
910 :
911 2032 : InitPostgres(NULL, dboid, /* database to connect to */
912 : NULL, useroid, /* role to connect as */
913 : init_flags,
914 : NULL); /* no out_dbname */
915 :
916 : /* it had better not gotten out of "init" mode yet */
917 2027 : if (!IsInitProcessingMode())
918 0 : ereport(ERROR,
919 : (errmsg("invalid processing mode in background worker")));
920 2027 : SetProcessingMode(NormalProcessing);
921 2027 : }
922 :
923 : /*
924 : * Block/unblock signals in a background worker
925 : */
926 : void
927 0 : BackgroundWorkerBlockSignals(void)
928 : {
929 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
930 0 : }
931 :
932 : void
933 2640 : BackgroundWorkerUnblockSignals(void)
934 : {
935 2640 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
936 2640 : }
937 :
938 : /*
939 : * Register a new static background worker.
940 : *
941 : * This can only be called directly from postmaster or in the _PG_init
942 : * function of a module library that's loaded by shared_preload_libraries;
943 : * otherwise it will have no effect.
944 : */
945 : void
946 877 : RegisterBackgroundWorker(BackgroundWorker *worker)
947 : {
948 : RegisteredBgWorker *rw;
949 : static int numworkers = 0;
950 :
951 : /*
952 : * Static background workers can only be registered in the postmaster
953 : * process.
954 : */
955 877 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
956 : {
957 : /*
958 : * In EXEC_BACKEND or single-user mode, we process
959 : * shared_preload_libraries in backend processes too. We cannot
960 : * register static background workers at that stage, but many
961 : * libraries' _PG_init() functions don't distinguish whether they're
962 : * being loaded in the postmaster or in a backend, they just check
963 : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
964 : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
965 : * the background workers should already have been registered when the
966 : * library was loaded in postmaster.
967 : */
968 0 : if (process_shared_preload_libraries_in_progress)
969 0 : return;
970 0 : ereport(LOG,
971 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
972 : errmsg("background worker \"%s\": must be registered in \"shared_preload_libraries\"",
973 : worker->bgw_name)));
974 0 : return;
975 : }
976 :
977 : /*
978 : * Cannot register static background workers after calling
979 : * BackgroundWorkerShmemInit().
980 : */
981 877 : if (BackgroundWorkerData != NULL)
982 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
983 : worker->bgw_name);
984 :
985 877 : ereport(DEBUG1,
986 : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
987 :
988 877 : if (!SanityCheckBackgroundWorker(worker, LOG))
989 0 : return;
990 :
991 877 : if (worker->bgw_notify_pid != 0)
992 : {
993 0 : ereport(LOG,
994 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
995 : errmsg("background worker \"%s\": only dynamic background workers can request notification",
996 : worker->bgw_name)));
997 0 : return;
998 : }
999 :
1000 : /*
1001 : * Enforce maximum number of workers. Note this is overly restrictive: we
1002 : * could allow more non-shmem-connected workers, because these don't count
1003 : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
1004 : * important to relax this restriction.
1005 : */
1006 877 : if (++numworkers > max_worker_processes)
1007 : {
1008 0 : ereport(LOG,
1009 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
1010 : errmsg("too many background workers"),
1011 : errdetail_plural("Up to %d background worker can be registered with the current settings.",
1012 : "Up to %d background workers can be registered with the current settings.",
1013 : max_worker_processes,
1014 : max_worker_processes),
1015 : errhint("Consider increasing the configuration parameter \"%s\".", "max_worker_processes")));
1016 0 : return;
1017 : }
1018 :
1019 : /*
1020 : * Copy the registration data into the registered workers list.
1021 : */
1022 877 : rw = MemoryContextAllocExtended(PostmasterContext,
1023 : sizeof(RegisteredBgWorker),
1024 : MCXT_ALLOC_NO_OOM);
1025 877 : if (rw == NULL)
1026 : {
1027 0 : ereport(LOG,
1028 : (errcode(ERRCODE_OUT_OF_MEMORY),
1029 : errmsg("out of memory")));
1030 0 : return;
1031 : }
1032 :
1033 877 : rw->rw_worker = *worker;
1034 877 : rw->rw_pid = 0;
1035 877 : rw->rw_crashed_at = 0;
1036 877 : rw->rw_terminate = false;
1037 :
1038 877 : dlist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
1039 : }
1040 :
1041 : /*
1042 : * Register a new background worker from a regular backend.
1043 : *
1044 : * Returns true on success and false on failure. Failure typically indicates
1045 : * that no background worker slots are currently available.
1046 : *
1047 : * If handle != NULL, we'll set *handle to a pointer that can subsequently
1048 : * be used as an argument to GetBackgroundWorkerPid(). The caller can
1049 : * free this pointer using pfree(), if desired.
1050 : */
1051 : bool
1052 1942 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
1053 : BackgroundWorkerHandle **handle)
1054 : {
1055 : int slotno;
1056 1942 : bool success = false;
1057 : bool parallel;
1058 1942 : uint64 generation = 0;
1059 :
1060 : /*
1061 : * We can't register dynamic background workers from the postmaster. If
1062 : * this is a standalone backend, we're the only process and can't start
1063 : * any more. In a multi-process environment, it might be theoretically
1064 : * possible, but we don't currently support it due to locking
1065 : * considerations; see comments on the BackgroundWorkerSlot data
1066 : * structure.
1067 : */
1068 1942 : if (!IsUnderPostmaster)
1069 0 : return false;
1070 :
1071 1942 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1072 0 : return false;
1073 :
1074 1942 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1075 :
1076 1942 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1077 :
1078 : /*
1079 : * If this is a parallel worker, check whether there are already too many
1080 : * parallel workers; if so, don't register another one. Our view of
1081 : * parallel_terminate_count may be slightly stale, but that doesn't really
1082 : * matter: we would have gotten the same result if we'd arrived here
1083 : * slightly earlier anyway. There's no help for it, either, since the
1084 : * postmaster must not take locks; a memory barrier wouldn't guarantee
1085 : * anything useful.
1086 : */
1087 1942 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1088 1497 : BackgroundWorkerData->parallel_terminate_count) >=
1089 : max_parallel_workers)
1090 : {
1091 : Assert(BackgroundWorkerData->parallel_register_count -
1092 : BackgroundWorkerData->parallel_terminate_count <=
1093 : MAX_PARALLEL_WORKER_LIMIT);
1094 10 : LWLockRelease(BackgroundWorkerLock);
1095 10 : return false;
1096 : }
1097 :
1098 : /*
1099 : * Look for an unused slot. If we find one, grab it.
1100 : */
1101 6134 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1102 : {
1103 6129 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1104 :
1105 6129 : if (!slot->in_use)
1106 : {
1107 1927 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1108 1927 : slot->pid = InvalidPid; /* indicates not started yet */
1109 1927 : slot->generation++;
1110 1927 : slot->terminate = false;
1111 1927 : generation = slot->generation;
1112 1927 : if (parallel)
1113 1482 : BackgroundWorkerData->parallel_register_count++;
1114 :
1115 : /*
1116 : * Make sure postmaster doesn't see the slot as in use before it
1117 : * sees the new contents.
1118 : */
1119 1927 : pg_write_barrier();
1120 :
1121 1927 : slot->in_use = true;
1122 1927 : success = true;
1123 1927 : break;
1124 : }
1125 : }
1126 :
1127 1932 : LWLockRelease(BackgroundWorkerLock);
1128 :
1129 : /* If we found a slot, tell the postmaster to notice the change. */
1130 1932 : if (success)
1131 1927 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1132 :
1133 : /*
1134 : * If we found a slot and the user has provided a handle, initialize it.
1135 : */
1136 1932 : if (success && handle)
1137 : {
1138 1927 : *handle = palloc_object(BackgroundWorkerHandle);
1139 1927 : (*handle)->slot = slotno;
1140 1927 : (*handle)->generation = generation;
1141 : }
1142 :
1143 1932 : return success;
1144 : }
1145 :
1146 : /*
1147 : * Get the PID of a dynamically-registered background worker.
1148 : *
1149 : * If the worker is determined to be running, the return value will be
1150 : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1151 : * postmaster has not yet attempted to start the worker, the return value will
1152 : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1153 : *
1154 : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1155 : * (because it is configured for automatic restart and exited non-zero),
1156 : * or that the worker is permanently stopped (because it exited with exit
1157 : * code 0, or was not configured for automatic restart), or even that the
1158 : * worker was unregistered without ever starting (either because startup
1159 : * failed and the worker is not configured for automatic restart, or because
1160 : * TerminateBackgroundWorker was used before the worker was successfully
1161 : * started).
1162 : */
1163 : BgwHandleStatus
1164 5827345 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1165 : {
1166 : BackgroundWorkerSlot *slot;
1167 : pid_t pid;
1168 :
1169 : Assert(handle->slot < max_worker_processes);
1170 5827345 : slot = &BackgroundWorkerData->slot[handle->slot];
1171 :
1172 : /*
1173 : * We could probably arrange to synchronize access to data using memory
1174 : * barriers only, but for now, let's just keep it simple and grab the
1175 : * lock. It seems unlikely that there will be enough traffic here to
1176 : * result in meaningful contention.
1177 : */
1178 5827345 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1179 :
1180 : /*
1181 : * The generation number can't be concurrently changed while we hold the
1182 : * lock. The pid, which is updated by the postmaster, can change at any
1183 : * time, but we assume such changes are atomic. So the value we read
1184 : * won't be garbage, but it might be out of date by the time the caller
1185 : * examines it (but that's unavoidable anyway).
1186 : *
1187 : * The in_use flag could be in the process of changing from true to false,
1188 : * but if it is already false then it can't change further.
1189 : */
1190 5827345 : if (handle->generation != slot->generation || !slot->in_use)
1191 1483 : pid = 0;
1192 : else
1193 5825862 : pid = slot->pid;
1194 :
1195 : /* All done. */
1196 5827345 : LWLockRelease(BackgroundWorkerLock);
1197 :
1198 5827345 : if (pid == 0)
1199 1483 : return BGWH_STOPPED;
1200 5825862 : else if (pid == InvalidPid)
1201 482835 : return BGWH_NOT_YET_STARTED;
1202 5343027 : *pidp = pid;
1203 5343027 : return BGWH_STARTED;
1204 : }
1205 :
1206 : /*
1207 : * Wait for a background worker to start up.
1208 : *
1209 : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1210 : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1211 : * returned. However, if the postmaster has died, we give up and return
1212 : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1213 : * take place.
1214 : *
1215 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1216 : * else we will not be awoken promptly when the worker's state changes.
1217 : */
1218 : BgwHandleStatus
1219 11 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1220 : {
1221 : BgwHandleStatus status;
1222 : int rc;
1223 :
1224 : for (;;)
1225 7 : {
1226 : pid_t pid;
1227 :
1228 18 : CHECK_FOR_INTERRUPTS();
1229 :
1230 18 : status = GetBackgroundWorkerPid(handle, &pid);
1231 18 : if (status == BGWH_STARTED)
1232 11 : *pidp = pid;
1233 18 : if (status != BGWH_NOT_YET_STARTED)
1234 11 : break;
1235 :
1236 7 : rc = WaitLatch(MyLatch,
1237 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1238 : WAIT_EVENT_BGWORKER_STARTUP);
1239 :
1240 7 : if (rc & WL_POSTMASTER_DEATH)
1241 : {
1242 0 : status = BGWH_POSTMASTER_DIED;
1243 0 : break;
1244 : }
1245 :
1246 7 : ResetLatch(MyLatch);
1247 : }
1248 :
1249 11 : return status;
1250 : }
1251 :
1252 : /*
1253 : * Wait for a background worker to stop.
1254 : *
1255 : * If the worker hasn't yet started, or is running, we wait for it to stop
1256 : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1257 : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1258 : * notifies us when a worker's state changes.
1259 : *
1260 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1261 : * else we will not be awoken promptly when the worker's state changes.
1262 : */
1263 : BgwHandleStatus
1264 1483 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1265 : {
1266 : BgwHandleStatus status;
1267 : int rc;
1268 :
1269 : for (;;)
1270 1718 : {
1271 : pid_t pid;
1272 :
1273 3201 : CHECK_FOR_INTERRUPTS();
1274 :
1275 3201 : status = GetBackgroundWorkerPid(handle, &pid);
1276 3201 : if (status == BGWH_STOPPED)
1277 1483 : break;
1278 :
1279 1718 : rc = WaitLatch(MyLatch,
1280 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1281 : WAIT_EVENT_BGWORKER_SHUTDOWN);
1282 :
1283 1718 : if (rc & WL_POSTMASTER_DEATH)
1284 : {
1285 0 : status = BGWH_POSTMASTER_DIED;
1286 0 : break;
1287 : }
1288 :
1289 1718 : ResetLatch(MyLatch);
1290 : }
1291 :
1292 1483 : return status;
1293 : }
1294 :
1295 : /*
1296 : * Instruct the postmaster to terminate a background worker.
1297 : *
1298 : * Note that it's safe to do this without regard to whether the worker is
1299 : * still running, or even if the worker may already have exited and been
1300 : * unregistered.
1301 : */
1302 : void
1303 6 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1304 : {
1305 : BackgroundWorkerSlot *slot;
1306 6 : bool signal_postmaster = false;
1307 :
1308 : Assert(handle->slot < max_worker_processes);
1309 6 : slot = &BackgroundWorkerData->slot[handle->slot];
1310 :
1311 : /* Set terminate flag in shared memory, unless slot has been reused. */
1312 6 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1313 6 : if (handle->generation == slot->generation)
1314 : {
1315 6 : slot->terminate = true;
1316 6 : signal_postmaster = true;
1317 : }
1318 6 : LWLockRelease(BackgroundWorkerLock);
1319 :
1320 : /* Make sure the postmaster notices the change to shared memory. */
1321 6 : if (signal_postmaster)
1322 6 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1323 6 : }
1324 :
1325 : /*
1326 : * Look up (and possibly load) a bgworker entry point function.
1327 : *
1328 : * For functions contained in the core code, we use library name "postgres"
1329 : * and consult the InternalBGWorkers array. External functions are
1330 : * looked up, and loaded if necessary, using load_external_function().
1331 : *
1332 : * The point of this is to pass function names as strings across process
1333 : * boundaries. We can't pass actual function addresses because of the
1334 : * possibility that the function has been loaded at a different address
1335 : * in a different process. This is obviously a hazard for functions in
1336 : * loadable libraries, but it can happen even for functions in the core code
1337 : * on platforms using EXEC_BACKEND (e.g., Windows).
1338 : *
1339 : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1340 : * in favor of applying load_external_function() for core functions too;
1341 : * but that raises portability issues that are not worth addressing now.
1342 : */
1343 : static bgworker_main_type
1344 2503 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1345 : {
1346 : /*
1347 : * If the function is to be loaded from postgres itself, search the
1348 : * InternalBGWorkers array.
1349 : */
1350 2503 : if (strcmp(libraryname, "postgres") == 0)
1351 : {
1352 : int i;
1353 :
1354 4470 : for (i = 0; i < lengthof(InternalBGWorkers); i++)
1355 : {
1356 4470 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1357 2485 : return InternalBGWorkers[i].fn_addr;
1358 : }
1359 :
1360 : /* We can only reach this by programming error. */
1361 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1362 : }
1363 :
1364 : /* Otherwise load from external library. */
1365 18 : return (bgworker_main_type)
1366 18 : load_external_function(libraryname, funcname, true, NULL);
1367 : }
1368 :
1369 : /*
1370 : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1371 : * not a valid background worker.
1372 : *
1373 : * The return value is in static memory belonging to this function, so it has
1374 : * to be used before calling this function again. This is so that the caller
1375 : * doesn't have to worry about the background worker locking protocol.
1376 : */
1377 : const char *
1378 1184 : GetBackgroundWorkerTypeByPid(pid_t pid)
1379 : {
1380 : int slotno;
1381 1184 : bool found = false;
1382 : static char result[BGW_MAXLEN];
1383 :
1384 1184 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1385 :
1386 1312 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1387 : {
1388 1312 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1389 :
1390 1312 : if (slot->pid > 0 && slot->pid == pid)
1391 : {
1392 1184 : strcpy(result, slot->worker.bgw_type);
1393 1184 : found = true;
1394 1184 : break;
1395 : }
1396 : }
1397 :
1398 1184 : LWLockRelease(BackgroundWorkerLock);
1399 :
1400 1184 : if (!found)
1401 0 : return NULL;
1402 :
1403 1184 : return result;
1404 : }
1405 :
1406 : /*
1407 : * Terminate all background workers connected to the given database, if the
1408 : * workers can be interrupted.
1409 : */
1410 : void
1411 6 : TerminateBackgroundWorkersForDatabase(Oid databaseId)
1412 : {
1413 6 : bool signal_postmaster = false;
1414 :
1415 6 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1416 :
1417 : /*
1418 : * Iterate through slots, looking for workers connected to the given
1419 : * database.
1420 : */
1421 54 : for (int slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1422 : {
1423 48 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1424 :
1425 48 : if (slot->in_use &&
1426 12 : (slot->worker.bgw_flags & BGWORKER_INTERRUPTIBLE))
1427 : {
1428 3 : PGPROC *proc = BackendPidGetProc(slot->pid);
1429 :
1430 3 : if (proc && proc->databaseId == databaseId)
1431 : {
1432 3 : slot->terminate = true;
1433 3 : signal_postmaster = true;
1434 : }
1435 : }
1436 : }
1437 :
1438 6 : LWLockRelease(BackgroundWorkerLock);
1439 :
1440 : /* Make sure the postmaster notices the change to shared memory. */
1441 6 : if (signal_postmaster)
1442 3 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1443 6 : }
|
