Line data Source code
1 : /*--------------------------------------------------------------------
2 : * bgworker.c
3 : * POSTGRES pluggable background workers implementation
4 : *
5 : * Portions Copyright (c) 1996-2023, 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/interrupt.h"
22 : #include "postmaster/postmaster.h"
23 : #include "replication/logicallauncher.h"
24 : #include "replication/logicalworker.h"
25 : #include "storage/dsm.h"
26 : #include "storage/ipc.h"
27 : #include "storage/latch.h"
28 : #include "storage/lwlock.h"
29 : #include "storage/pg_shmem.h"
30 : #include "storage/pmsignal.h"
31 : #include "storage/proc.h"
32 : #include "storage/procsignal.h"
33 : #include "storage/shmem.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 :
40 : /*
41 : * The postmaster's list of registered background workers, in private memory.
42 : */
43 : slist_head BackgroundWorkerList = SLIST_STATIC_INIT(BackgroundWorkerList);
44 :
45 : /*
46 : * BackgroundWorkerSlots exist in shared memory and can be accessed (via
47 : * the BackgroundWorkerArray) by both the postmaster and by regular backends.
48 : * However, the postmaster cannot take locks, even spinlocks, because this
49 : * might allow it to crash or become wedged if shared memory gets corrupted.
50 : * Such an outcome is intolerable. Therefore, we need a lockless protocol
51 : * for coordinating access to this data.
52 : *
53 : * The 'in_use' flag is used to hand off responsibility for the slot between
54 : * the postmaster and the rest of the system. When 'in_use' is false,
55 : * the postmaster will ignore the slot entirely, except for the 'in_use' flag
56 : * itself, which it may read. In this state, regular backends may modify the
57 : * slot. Once a backend sets 'in_use' to true, the slot becomes the
58 : * responsibility of the postmaster. Regular backends may no longer modify it,
59 : * but the postmaster may examine it. Thus, a backend initializing a slot
60 : * must fully initialize the slot - and insert a write memory barrier - before
61 : * marking it as in use.
62 : *
63 : * As an exception, however, even when the slot is in use, regular backends
64 : * may set the 'terminate' flag for a slot, telling the postmaster not
65 : * to restart it. Once the background worker is no longer running, the slot
66 : * will be released for reuse.
67 : *
68 : * In addition to coordinating with the postmaster, backends modifying this
69 : * data structure must coordinate with each other. Since they can take locks,
70 : * this is straightforward: any backend wishing to manipulate a slot must
71 : * take BackgroundWorkerLock in exclusive mode. Backends wishing to read
72 : * data that might get concurrently modified by other backends should take
73 : * this lock in shared mode. No matter what, backends reading this data
74 : * structure must be able to tolerate concurrent modifications by the
75 : * postmaster.
76 : */
77 : typedef struct BackgroundWorkerSlot
78 : {
79 : bool in_use;
80 : bool terminate;
81 : pid_t pid; /* InvalidPid = not started yet; 0 = dead */
82 : uint64 generation; /* incremented when slot is recycled */
83 : BackgroundWorker worker;
84 : } BackgroundWorkerSlot;
85 :
86 : /*
87 : * In order to limit the total number of parallel workers (according to
88 : * max_parallel_workers GUC), we maintain the number of active parallel
89 : * workers. Since the postmaster cannot take locks, two variables are used for
90 : * this purpose: the number of registered parallel workers (modified by the
91 : * backends, protected by BackgroundWorkerLock) and the number of terminated
92 : * parallel workers (modified only by the postmaster, lockless). The active
93 : * number of parallel workers is the number of registered workers minus the
94 : * terminated ones. These counters can of course overflow, but it's not
95 : * important here since the subtraction will still give the right number.
96 : */
97 : typedef struct BackgroundWorkerArray
98 : {
99 : int total_slots;
100 : uint32 parallel_register_count;
101 : uint32 parallel_terminate_count;
102 : BackgroundWorkerSlot slot[FLEXIBLE_ARRAY_MEMBER];
103 : } BackgroundWorkerArray;
104 :
105 : struct BackgroundWorkerHandle
106 : {
107 : int slot;
108 : uint64 generation;
109 : };
110 :
111 : static BackgroundWorkerArray *BackgroundWorkerData;
112 :
113 : /*
114 : * List of internal background worker entry points. We need this for
115 : * reasons explained in LookupBackgroundWorkerFunction(), below.
116 : */
117 : static const struct
118 : {
119 : const char *fn_name;
120 : bgworker_main_type fn_addr;
121 : } InternalBGWorkers[] =
122 :
123 : {
124 : {
125 : "ParallelWorkerMain", ParallelWorkerMain
126 : },
127 : {
128 : "ApplyLauncherMain", ApplyLauncherMain
129 : },
130 : {
131 : "ApplyWorkerMain", ApplyWorkerMain
132 : },
133 : {
134 : "ParallelApplyWorkerMain", ParallelApplyWorkerMain
135 : },
136 : {
137 : "TablesyncWorkerMain", TablesyncWorkerMain
138 : }
139 : };
140 :
141 : /* Private functions. */
142 : static bgworker_main_type LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname);
143 :
144 :
145 : /*
146 : * Calculate shared memory needed.
147 : */
148 : Size
149 4496 : BackgroundWorkerShmemSize(void)
150 : {
151 : Size size;
152 :
153 : /* Array of workers is variably sized. */
154 4496 : size = offsetof(BackgroundWorkerArray, slot);
155 4496 : size = add_size(size, mul_size(max_worker_processes,
156 : sizeof(BackgroundWorkerSlot)));
157 :
158 4496 : return size;
159 : }
160 :
161 : /*
162 : * Initialize shared memory.
163 : */
164 : void
165 1562 : BackgroundWorkerShmemInit(void)
166 : {
167 : bool found;
168 :
169 1562 : BackgroundWorkerData = ShmemInitStruct("Background Worker Data",
170 : BackgroundWorkerShmemSize(),
171 : &found);
172 1562 : if (!IsUnderPostmaster)
173 : {
174 : slist_iter siter;
175 1562 : int slotno = 0;
176 :
177 1562 : BackgroundWorkerData->total_slots = max_worker_processes;
178 1562 : BackgroundWorkerData->parallel_register_count = 0;
179 1562 : BackgroundWorkerData->parallel_terminate_count = 0;
180 :
181 : /*
182 : * Copy contents of worker list into shared memory. Record the shared
183 : * memory slot assigned to each worker. This ensures a 1-to-1
184 : * correspondence between the postmaster's private list and the array
185 : * in shared memory.
186 : */
187 2830 : slist_foreach(siter, &BackgroundWorkerList)
188 : {
189 1268 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
190 : RegisteredBgWorker *rw;
191 :
192 1268 : rw = slist_container(RegisteredBgWorker, rw_lnode, siter.cur);
193 : Assert(slotno < max_worker_processes);
194 1268 : slot->in_use = true;
195 1268 : slot->terminate = false;
196 1268 : slot->pid = InvalidPid;
197 1268 : slot->generation = 0;
198 1268 : rw->rw_shmem_slot = slotno;
199 1268 : rw->rw_worker.bgw_notify_pid = 0; /* might be reinit after crash */
200 1268 : memcpy(&slot->worker, &rw->rw_worker, sizeof(BackgroundWorker));
201 1268 : ++slotno;
202 : }
203 :
204 : /*
205 : * Mark any remaining slots as not in use.
206 : */
207 12838 : while (slotno < max_worker_processes)
208 : {
209 11276 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
210 :
211 11276 : slot->in_use = false;
212 11276 : ++slotno;
213 : }
214 : }
215 : else
216 : Assert(found);
217 1562 : }
218 :
219 : /*
220 : * Search the postmaster's backend-private list of RegisteredBgWorker objects
221 : * for the one that maps to the given slot number.
222 : */
223 : static RegisteredBgWorker *
224 6372 : FindRegisteredWorkerBySlotNumber(int slotno)
225 : {
226 : slist_iter siter;
227 :
228 15426 : slist_foreach(siter, &BackgroundWorkerList)
229 : {
230 : RegisteredBgWorker *rw;
231 :
232 12106 : rw = slist_container(RegisteredBgWorker, rw_lnode, siter.cur);
233 12106 : if (rw->rw_shmem_slot == slotno)
234 3052 : return rw;
235 : }
236 :
237 3320 : return NULL;
238 : }
239 :
240 : /*
241 : * Notice changes to shared memory made by other backends.
242 : * Accept new worker requests only if allow_new_workers is true.
243 : *
244 : * This code runs in the postmaster, so we must be very careful not to assume
245 : * that shared memory contents are sane. Otherwise, a rogue backend could
246 : * take out the postmaster.
247 : */
248 : void
249 1906 : BackgroundWorkerStateChange(bool allow_new_workers)
250 : {
251 : int slotno;
252 :
253 : /*
254 : * The total number of slots stored in shared memory should match our
255 : * notion of max_worker_processes. If it does not, something is very
256 : * wrong. Further down, we always refer to this value as
257 : * max_worker_processes, in case shared memory gets corrupted while we're
258 : * looping.
259 : */
260 1906 : if (max_worker_processes != BackgroundWorkerData->total_slots)
261 : {
262 0 : ereport(LOG,
263 : (errmsg("inconsistent background worker state (max_worker_processes=%d, total_slots=%d)",
264 : max_worker_processes,
265 : BackgroundWorkerData->total_slots)));
266 0 : return;
267 : }
268 :
269 : /*
270 : * Iterate through slots, looking for newly-registered workers or workers
271 : * who must die.
272 : */
273 17442 : for (slotno = 0; slotno < max_worker_processes; ++slotno)
274 : {
275 15536 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
276 : RegisteredBgWorker *rw;
277 :
278 15536 : if (!slot->in_use)
279 9164 : continue;
280 :
281 : /*
282 : * Make sure we don't see the in_use flag before the updated slot
283 : * contents.
284 : */
285 6372 : pg_read_barrier();
286 :
287 : /* See whether we already know about this worker. */
288 6372 : rw = FindRegisteredWorkerBySlotNumber(slotno);
289 6372 : if (rw != NULL)
290 : {
291 : /*
292 : * In general, the worker data can't change after it's initially
293 : * registered. However, someone can set the terminate flag.
294 : */
295 3052 : if (slot->terminate && !rw->rw_terminate)
296 : {
297 6 : rw->rw_terminate = true;
298 6 : if (rw->rw_pid != 0)
299 6 : kill(rw->rw_pid, SIGTERM);
300 : else
301 : {
302 : /* Report never-started, now-terminated worker as dead. */
303 0 : ReportBackgroundWorkerPID(rw);
304 : }
305 : }
306 3052 : continue;
307 : }
308 :
309 : /*
310 : * If we aren't allowing new workers, then immediately mark it for
311 : * termination; the next stanza will take care of cleaning it up.
312 : * Doing this ensures that any process waiting for the worker will get
313 : * awoken, even though the worker will never be allowed to run.
314 : */
315 3320 : if (!allow_new_workers)
316 0 : slot->terminate = true;
317 :
318 : /*
319 : * If the worker is marked for termination, we don't need to add it to
320 : * the registered workers list; we can just free the slot. However, if
321 : * bgw_notify_pid is set, the process that registered the worker may
322 : * need to know that we've processed the terminate request, so be sure
323 : * to signal it.
324 : */
325 3320 : if (slot->terminate)
326 : {
327 : int notify_pid;
328 :
329 : /*
330 : * We need a memory barrier here to make sure that the load of
331 : * bgw_notify_pid and the update of parallel_terminate_count
332 : * complete before the store to in_use.
333 : */
334 0 : notify_pid = slot->worker.bgw_notify_pid;
335 0 : if ((slot->worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
336 0 : BackgroundWorkerData->parallel_terminate_count++;
337 0 : slot->pid = 0;
338 :
339 0 : pg_memory_barrier();
340 0 : slot->in_use = false;
341 :
342 0 : if (notify_pid != 0)
343 0 : kill(notify_pid, SIGUSR1);
344 :
345 0 : continue;
346 : }
347 :
348 : /*
349 : * Copy the registration data into the registered workers list.
350 : */
351 3320 : rw = MemoryContextAllocExtended(PostmasterContext,
352 : sizeof(RegisteredBgWorker),
353 : MCXT_ALLOC_NO_OOM);
354 3320 : if (rw == NULL)
355 : {
356 0 : ereport(LOG,
357 : (errcode(ERRCODE_OUT_OF_MEMORY),
358 : errmsg("out of memory")));
359 0 : return;
360 : }
361 :
362 : /*
363 : * Copy strings in a paranoid way. If shared memory is corrupted, the
364 : * source data might not even be NUL-terminated.
365 : */
366 3320 : ascii_safe_strlcpy(rw->rw_worker.bgw_name,
367 3320 : slot->worker.bgw_name, BGW_MAXLEN);
368 3320 : ascii_safe_strlcpy(rw->rw_worker.bgw_type,
369 3320 : slot->worker.bgw_type, BGW_MAXLEN);
370 3320 : ascii_safe_strlcpy(rw->rw_worker.bgw_library_name,
371 3320 : slot->worker.bgw_library_name, MAXPGPATH);
372 3320 : ascii_safe_strlcpy(rw->rw_worker.bgw_function_name,
373 3320 : slot->worker.bgw_function_name, BGW_MAXLEN);
374 :
375 : /*
376 : * Copy various fixed-size fields.
377 : *
378 : * flags, start_time, and restart_time are examined by the postmaster,
379 : * but nothing too bad will happen if they are corrupted. The
380 : * remaining fields will only be examined by the child process. It
381 : * might crash, but we won't.
382 : */
383 3320 : rw->rw_worker.bgw_flags = slot->worker.bgw_flags;
384 3320 : rw->rw_worker.bgw_start_time = slot->worker.bgw_start_time;
385 3320 : rw->rw_worker.bgw_restart_time = slot->worker.bgw_restart_time;
386 3320 : rw->rw_worker.bgw_main_arg = slot->worker.bgw_main_arg;
387 3320 : memcpy(rw->rw_worker.bgw_extra, slot->worker.bgw_extra, BGW_EXTRALEN);
388 :
389 : /*
390 : * Copy the PID to be notified about state changes, but only if the
391 : * postmaster knows about a backend with that PID. It isn't an error
392 : * if the postmaster doesn't know about the PID, because the backend
393 : * that requested the worker could have died (or been killed) just
394 : * after doing so. Nonetheless, at least until we get some experience
395 : * with how this plays out in the wild, log a message at a relative
396 : * high debug level.
397 : */
398 3320 : rw->rw_worker.bgw_notify_pid = slot->worker.bgw_notify_pid;
399 3320 : if (!PostmasterMarkPIDForWorkerNotify(rw->rw_worker.bgw_notify_pid))
400 : {
401 0 : elog(DEBUG1, "worker notification PID %d is not valid",
402 : (int) rw->rw_worker.bgw_notify_pid);
403 0 : rw->rw_worker.bgw_notify_pid = 0;
404 : }
405 :
406 : /* Initialize postmaster bookkeeping. */
407 3320 : rw->rw_backend = NULL;
408 3320 : rw->rw_pid = 0;
409 3320 : rw->rw_child_slot = 0;
410 3320 : rw->rw_crashed_at = 0;
411 3320 : rw->rw_shmem_slot = slotno;
412 3320 : rw->rw_terminate = false;
413 :
414 : /* Log it! */
415 3320 : ereport(DEBUG1,
416 : (errmsg_internal("registering background worker \"%s\"",
417 : rw->rw_worker.bgw_name)));
418 :
419 3320 : slist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
420 : }
421 : }
422 :
423 : /*
424 : * Forget about a background worker that's no longer needed.
425 : *
426 : * The worker must be identified by passing an slist_mutable_iter that
427 : * points to it. This convention allows deletion of workers during
428 : * searches of the worker list, and saves having to search the list again.
429 : *
430 : * Caller is responsible for notifying bgw_notify_pid, if appropriate.
431 : *
432 : * This function must be invoked only in the postmaster.
433 : */
434 : void
435 3320 : ForgetBackgroundWorker(slist_mutable_iter *cur)
436 : {
437 : RegisteredBgWorker *rw;
438 : BackgroundWorkerSlot *slot;
439 :
440 3320 : rw = slist_container(RegisteredBgWorker, rw_lnode, cur->cur);
441 :
442 : Assert(rw->rw_shmem_slot < max_worker_processes);
443 3320 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
444 : Assert(slot->in_use);
445 :
446 : /*
447 : * We need a memory barrier here to make sure that the update of
448 : * parallel_terminate_count completes before the store to in_use.
449 : */
450 3320 : if ((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
451 2594 : BackgroundWorkerData->parallel_terminate_count++;
452 :
453 3320 : pg_memory_barrier();
454 3320 : slot->in_use = false;
455 :
456 3320 : ereport(DEBUG1,
457 : (errmsg_internal("unregistering background worker \"%s\"",
458 : rw->rw_worker.bgw_name)));
459 :
460 3320 : slist_delete_current(cur);
461 3320 : pfree(rw);
462 3320 : }
463 :
464 : /*
465 : * Report the PID of a newly-launched background worker in shared memory.
466 : *
467 : * This function should only be called from the postmaster.
468 : */
469 : void
470 4430 : ReportBackgroundWorkerPID(RegisteredBgWorker *rw)
471 : {
472 : BackgroundWorkerSlot *slot;
473 :
474 : Assert(rw->rw_shmem_slot < max_worker_processes);
475 4430 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
476 4430 : slot->pid = rw->rw_pid;
477 :
478 4430 : if (rw->rw_worker.bgw_notify_pid != 0)
479 3320 : kill(rw->rw_worker.bgw_notify_pid, SIGUSR1);
480 4430 : }
481 :
482 : /*
483 : * Report that the PID of a background worker is now zero because a
484 : * previously-running background worker has exited.
485 : *
486 : * This function should only be called from the postmaster.
487 : */
488 : void
489 3940 : ReportBackgroundWorkerExit(slist_mutable_iter *cur)
490 : {
491 : RegisteredBgWorker *rw;
492 : BackgroundWorkerSlot *slot;
493 : int notify_pid;
494 :
495 3940 : rw = slist_container(RegisteredBgWorker, rw_lnode, cur->cur);
496 :
497 : Assert(rw->rw_shmem_slot < max_worker_processes);
498 3940 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
499 3940 : slot->pid = rw->rw_pid;
500 3940 : notify_pid = rw->rw_worker.bgw_notify_pid;
501 :
502 : /*
503 : * If this worker is slated for deregistration, do that before notifying
504 : * the process which started it. Otherwise, if that process tries to
505 : * reuse the slot immediately, it might not be available yet. In theory
506 : * that could happen anyway if the process checks slot->pid at just the
507 : * wrong moment, but this makes the window narrower.
508 : */
509 3940 : if (rw->rw_terminate ||
510 924 : rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
511 3298 : ForgetBackgroundWorker(cur);
512 :
513 3940 : if (notify_pid != 0)
514 3252 : kill(notify_pid, SIGUSR1);
515 3940 : }
516 :
517 : /*
518 : * Cancel SIGUSR1 notifications for a PID belonging to an exiting backend.
519 : *
520 : * This function should only be called from the postmaster.
521 : */
522 : void
523 396 : BackgroundWorkerStopNotifications(pid_t pid)
524 : {
525 : slist_iter siter;
526 :
527 1238 : slist_foreach(siter, &BackgroundWorkerList)
528 : {
529 : RegisteredBgWorker *rw;
530 :
531 842 : rw = slist_container(RegisteredBgWorker, rw_lnode, siter.cur);
532 842 : if (rw->rw_worker.bgw_notify_pid == pid)
533 50 : rw->rw_worker.bgw_notify_pid = 0;
534 : }
535 396 : }
536 :
537 : /*
538 : * Cancel any not-yet-started worker requests that have waiting processes.
539 : *
540 : * This is called during a normal ("smart" or "fast") database shutdown.
541 : * After this point, no new background workers will be started, so anything
542 : * that might be waiting for them needs to be kicked off its wait. We do
543 : * that by canceling the bgworker registration entirely, which is perhaps
544 : * overkill, but since we're shutting down it does not matter whether the
545 : * registration record sticks around.
546 : *
547 : * This function should only be called from the postmaster.
548 : */
549 : void
550 734 : ForgetUnstartedBackgroundWorkers(void)
551 : {
552 : slist_mutable_iter iter;
553 :
554 1496 : slist_foreach_modify(iter, &BackgroundWorkerList)
555 : {
556 : RegisteredBgWorker *rw;
557 : BackgroundWorkerSlot *slot;
558 :
559 762 : rw = slist_container(RegisteredBgWorker, rw_lnode, iter.cur);
560 : Assert(rw->rw_shmem_slot < max_worker_processes);
561 762 : slot = &BackgroundWorkerData->slot[rw->rw_shmem_slot];
562 :
563 : /* If it's not yet started, and there's someone waiting ... */
564 762 : if (slot->pid == InvalidPid &&
565 62 : rw->rw_worker.bgw_notify_pid != 0)
566 : {
567 : /* ... then zap it, and notify the waiter */
568 0 : int notify_pid = rw->rw_worker.bgw_notify_pid;
569 :
570 0 : ForgetBackgroundWorker(&iter);
571 0 : if (notify_pid != 0)
572 0 : kill(notify_pid, SIGUSR1);
573 : }
574 : }
575 734 : }
576 :
577 : /*
578 : * Reset background worker crash state.
579 : *
580 : * We assume that, after a crash-and-restart cycle, background workers without
581 : * the never-restart flag should be restarted immediately, instead of waiting
582 : * for bgw_restart_time to elapse. On the other hand, workers with that flag
583 : * should be forgotten immediately, since we won't ever restart them.
584 : *
585 : * This function should only be called from the postmaster.
586 : */
587 : void
588 8 : ResetBackgroundWorkerCrashTimes(void)
589 : {
590 : slist_mutable_iter iter;
591 :
592 16 : slist_foreach_modify(iter, &BackgroundWorkerList)
593 : {
594 : RegisteredBgWorker *rw;
595 :
596 8 : rw = slist_container(RegisteredBgWorker, rw_lnode, iter.cur);
597 :
598 8 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
599 : {
600 : /*
601 : * Workers marked BGW_NEVER_RESTART shouldn't get relaunched after
602 : * the crash, so forget about them. (If we wait until after the
603 : * crash to forget about them, and they are parallel workers,
604 : * parallel_terminate_count will get incremented after we've
605 : * already zeroed parallel_register_count, which would be bad.)
606 : */
607 0 : ForgetBackgroundWorker(&iter);
608 : }
609 : else
610 : {
611 : /*
612 : * The accounting which we do via parallel_register_count and
613 : * parallel_terminate_count would get messed up if a worker marked
614 : * parallel could survive a crash and restart cycle. All such
615 : * workers should be marked BGW_NEVER_RESTART, and thus control
616 : * should never reach this branch.
617 : */
618 : Assert((rw->rw_worker.bgw_flags & BGWORKER_CLASS_PARALLEL) == 0);
619 :
620 : /*
621 : * Allow this worker to be restarted immediately after we finish
622 : * resetting.
623 : */
624 8 : rw->rw_crashed_at = 0;
625 :
626 : /*
627 : * If there was anyone waiting for it, they're history.
628 : */
629 8 : rw->rw_worker.bgw_notify_pid = 0;
630 : }
631 : }
632 8 : }
633 :
634 : #ifdef EXEC_BACKEND
635 : /*
636 : * In EXEC_BACKEND mode, workers use this to retrieve their details from
637 : * shared memory.
638 : */
639 : BackgroundWorker *
640 : BackgroundWorkerEntry(int slotno)
641 : {
642 : static BackgroundWorker myEntry;
643 : BackgroundWorkerSlot *slot;
644 :
645 : Assert(slotno < BackgroundWorkerData->total_slots);
646 : slot = &BackgroundWorkerData->slot[slotno];
647 : Assert(slot->in_use);
648 :
649 : /* must copy this in case we don't intend to retain shmem access */
650 : memcpy(&myEntry, &slot->worker, sizeof myEntry);
651 : return &myEntry;
652 : }
653 : #endif
654 :
655 : /*
656 : * Complain about the BackgroundWorker definition using error level elevel.
657 : * Return true if it looks ok, false if not (unless elevel >= ERROR, in
658 : * which case we won't return at all in the not-OK case).
659 : */
660 : static bool
661 4494 : SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
662 : {
663 : /* sanity check for flags */
664 :
665 : /*
666 : * We used to support workers not connected to shared memory, but don't
667 : * anymore. Thus this is a required flag now. We're not removing the flag
668 : * for compatibility reasons and because the flag still provides some
669 : * signal when reading code.
670 : */
671 4494 : if (!(worker->bgw_flags & BGWORKER_SHMEM_ACCESS))
672 : {
673 0 : ereport(elevel,
674 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
675 : errmsg("background worker \"%s\": background workers without shared memory access are not supported",
676 : worker->bgw_name)));
677 0 : return false;
678 : }
679 :
680 4494 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
681 : {
682 4476 : if (worker->bgw_start_time == BgWorkerStart_PostmasterStart)
683 : {
684 0 : ereport(elevel,
685 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
686 : errmsg("background worker \"%s\": cannot request database access if starting at postmaster start",
687 : worker->bgw_name)));
688 0 : return false;
689 : }
690 :
691 : /* XXX other checks? */
692 : }
693 :
694 4494 : if ((worker->bgw_restart_time < 0 &&
695 3230 : worker->bgw_restart_time != BGW_NEVER_RESTART) ||
696 4494 : (worker->bgw_restart_time > USECS_PER_DAY / 1000))
697 : {
698 0 : ereport(elevel,
699 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
700 : errmsg("background worker \"%s\": invalid restart interval",
701 : worker->bgw_name)));
702 0 : return false;
703 : }
704 :
705 : /*
706 : * Parallel workers may not be configured for restart, because the
707 : * parallel_register_count/parallel_terminate_count accounting can't
708 : * handle parallel workers lasting through a crash-and-restart cycle.
709 : */
710 4494 : if (worker->bgw_restart_time != BGW_NEVER_RESTART &&
711 1264 : (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0)
712 : {
713 0 : ereport(elevel,
714 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
715 : errmsg("background worker \"%s\": parallel workers may not be configured for restart",
716 : worker->bgw_name)));
717 0 : return false;
718 : }
719 :
720 : /*
721 : * If bgw_type is not filled in, use bgw_name.
722 : */
723 4494 : if (strcmp(worker->bgw_type, "") == 0)
724 0 : strcpy(worker->bgw_type, worker->bgw_name);
725 :
726 4494 : return true;
727 : }
728 :
729 : /*
730 : * Standard SIGTERM handler for background workers
731 : */
732 : static void
733 0 : bgworker_die(SIGNAL_ARGS)
734 : {
735 0 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
736 :
737 0 : ereport(FATAL,
738 : (errcode(ERRCODE_ADMIN_SHUTDOWN),
739 : errmsg("terminating background worker \"%s\" due to administrator command",
740 : MyBgworkerEntry->bgw_type)));
741 : }
742 :
743 : /*
744 : * Main entry point for background worker processes.
745 : */
746 : void
747 3946 : BackgroundWorkerMain(void)
748 : {
749 : sigjmp_buf local_sigjmp_buf;
750 3946 : BackgroundWorker *worker = MyBgworkerEntry;
751 : bgworker_main_type entrypt;
752 :
753 3946 : if (worker == NULL)
754 0 : elog(FATAL, "unable to find bgworker entry");
755 :
756 3946 : IsBackgroundWorker = true;
757 :
758 3946 : MyBackendType = B_BG_WORKER;
759 3946 : init_ps_display(worker->bgw_name);
760 :
761 3946 : SetProcessingMode(InitProcessing);
762 :
763 : /* Apply PostAuthDelay */
764 3946 : if (PostAuthDelay > 0)
765 0 : pg_usleep(PostAuthDelay * 1000000L);
766 :
767 : /*
768 : * Set up signal handlers.
769 : */
770 3946 : if (worker->bgw_flags & BGWORKER_BACKEND_DATABASE_CONNECTION)
771 : {
772 : /*
773 : * SIGINT is used to signal canceling the current action
774 : */
775 3928 : pqsignal(SIGINT, StatementCancelHandler);
776 3928 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
777 3928 : pqsignal(SIGFPE, FloatExceptionHandler);
778 :
779 : /* XXX Any other handlers needed here? */
780 : }
781 : else
782 : {
783 18 : pqsignal(SIGINT, SIG_IGN);
784 18 : pqsignal(SIGUSR1, SIG_IGN);
785 18 : pqsignal(SIGFPE, SIG_IGN);
786 : }
787 3946 : pqsignal(SIGTERM, bgworker_die);
788 : /* SIGQUIT handler was already set up by InitPostmasterChild */
789 3946 : pqsignal(SIGHUP, SIG_IGN);
790 :
791 3946 : InitializeTimeouts(); /* establishes SIGALRM handler */
792 :
793 3946 : pqsignal(SIGPIPE, SIG_IGN);
794 3946 : pqsignal(SIGUSR2, SIG_IGN);
795 3946 : 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 3946 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
803 : {
804 : /* Since not using PG_TRY, must reset error stack by hand */
805 136 : error_context_stack = NULL;
806 :
807 : /* Prevent interrupts while cleaning up */
808 136 : 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 : */
815 136 : BackgroundWorkerUnblockSignals();
816 :
817 : /* Report the error to the parallel leader and the server log */
818 136 : 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 */
827 136 : proc_exit(1);
828 : }
829 :
830 : /* We can now handle ereport(ERROR) */
831 3946 : PG_exception_stack = &local_sigjmp_buf;
832 :
833 : /*
834 : * Create a per-backend PGPROC struct in shared memory, except in the
835 : * EXEC_BACKEND case where this was done in SubPostmasterMain. We must do
836 : * this before we can use LWLocks (and in the EXEC_BACKEND case we already
837 : * had to do some stuff with LWLocks).
838 : */
839 : #ifndef EXEC_BACKEND
840 3946 : InitProcess();
841 : #endif
842 :
843 : /*
844 : * Early initialization.
845 : */
846 3946 : BaseInit();
847 :
848 : /*
849 : * Look up the entry point function, loading its library if necessary.
850 : */
851 7892 : entrypt = LookupBackgroundWorkerFunction(worker->bgw_library_name,
852 3946 : worker->bgw_function_name);
853 :
854 : /*
855 : * Note that in normal processes, we would call InitPostgres here. For a
856 : * worker, however, we don't know what database to connect to, yet; so we
857 : * need to wait until the user code does it via
858 : * BackgroundWorkerInitializeConnection().
859 : */
860 :
861 : /*
862 : * Now invoke the user-defined worker code
863 : */
864 3946 : entrypt(worker->bgw_main_arg);
865 :
866 : /* ... and if it returns, we're done */
867 2594 : proc_exit(0);
868 : }
869 :
870 : /*
871 : * Register a new static background worker.
872 : *
873 : * This can only be called directly from postmaster or in the _PG_init
874 : * function of a module library that's loaded by shared_preload_libraries;
875 : * otherwise it will have no effect.
876 : */
877 : void
878 1270 : RegisterBackgroundWorker(BackgroundWorker *worker)
879 : {
880 : RegisteredBgWorker *rw;
881 : static int numworkers = 0;
882 :
883 : /*
884 : * Static background workers can only be registered in the postmaster
885 : * process.
886 : */
887 1270 : if (IsUnderPostmaster || !IsPostmasterEnvironment)
888 : {
889 : /*
890 : * In EXEC_BACKEND or single-user mode, we process
891 : * shared_preload_libraries in backend processes too. We cannot
892 : * register static background workers at that stage, but many
893 : * libraries' _PG_init() functions don't distinguish whether they're
894 : * being loaded in the postmaster or in a backend, they just check
895 : * process_shared_preload_libraries_in_progress. It's a bit sloppy,
896 : * but for historical reasons we tolerate it. In EXEC_BACKEND mode,
897 : * the background workers should already have been registered when the
898 : * library was loaded in postmaster.
899 : */
900 0 : if (process_shared_preload_libraries_in_progress)
901 0 : return;
902 0 : ereport(LOG,
903 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
904 : errmsg("background worker \"%s\": must be registered in shared_preload_libraries",
905 : worker->bgw_name)));
906 0 : return;
907 : }
908 :
909 : /*
910 : * Cannot register static background workers after calling
911 : * BackgroundWorkerShmemInit().
912 : */
913 1270 : if (BackgroundWorkerData != NULL)
914 0 : elog(ERROR, "cannot register background worker \"%s\" after shmem init",
915 : worker->bgw_name);
916 :
917 1270 : ereport(DEBUG1,
918 : (errmsg_internal("registering background worker \"%s\"", worker->bgw_name)));
919 :
920 1270 : if (!SanityCheckBackgroundWorker(worker, LOG))
921 0 : return;
922 :
923 1270 : if (worker->bgw_notify_pid != 0)
924 : {
925 0 : ereport(LOG,
926 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
927 : errmsg("background worker \"%s\": only dynamic background workers can request notification",
928 : worker->bgw_name)));
929 0 : return;
930 : }
931 :
932 : /*
933 : * Enforce maximum number of workers. Note this is overly restrictive: we
934 : * could allow more non-shmem-connected workers, because these don't count
935 : * towards the MAX_BACKENDS limit elsewhere. For now, it doesn't seem
936 : * important to relax this restriction.
937 : */
938 1270 : if (++numworkers > max_worker_processes)
939 : {
940 0 : ereport(LOG,
941 : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
942 : errmsg("too many background workers"),
943 : errdetail_plural("Up to %d background worker can be registered with the current settings.",
944 : "Up to %d background workers can be registered with the current settings.",
945 : max_worker_processes,
946 : max_worker_processes),
947 : errhint("Consider increasing the configuration parameter max_worker_processes.")));
948 0 : return;
949 : }
950 :
951 : /*
952 : * Copy the registration data into the registered workers list.
953 : */
954 1270 : rw = MemoryContextAllocExtended(PostmasterContext,
955 : sizeof(RegisteredBgWorker),
956 : MCXT_ALLOC_NO_OOM);
957 1270 : if (rw == NULL)
958 : {
959 0 : ereport(LOG,
960 : (errcode(ERRCODE_OUT_OF_MEMORY),
961 : errmsg("out of memory")));
962 0 : return;
963 : }
964 :
965 1270 : rw->rw_worker = *worker;
966 1270 : rw->rw_backend = NULL;
967 1270 : rw->rw_pid = 0;
968 1270 : rw->rw_child_slot = 0;
969 1270 : rw->rw_crashed_at = 0;
970 1270 : rw->rw_terminate = false;
971 :
972 1270 : slist_push_head(&BackgroundWorkerList, &rw->rw_lnode);
973 : }
974 :
975 : /*
976 : * Register a new background worker from a regular backend.
977 : *
978 : * Returns true on success and false on failure. Failure typically indicates
979 : * that no background worker slots are currently available.
980 : *
981 : * If handle != NULL, we'll set *handle to a pointer that can subsequently
982 : * be used as an argument to GetBackgroundWorkerPid(). The caller can
983 : * free this pointer using pfree(), if desired.
984 : */
985 : bool
986 3224 : RegisterDynamicBackgroundWorker(BackgroundWorker *worker,
987 : BackgroundWorkerHandle **handle)
988 : {
989 : int slotno;
990 3224 : bool success = false;
991 : bool parallel;
992 3224 : uint64 generation = 0;
993 :
994 : /*
995 : * We can't register dynamic background workers from the postmaster. If
996 : * this is a standalone backend, we're the only process and can't start
997 : * any more. In a multi-process environment, it might be theoretically
998 : * possible, but we don't currently support it due to locking
999 : * considerations; see comments on the BackgroundWorkerSlot data
1000 : * structure.
1001 : */
1002 3224 : if (!IsUnderPostmaster)
1003 0 : return false;
1004 :
1005 3224 : if (!SanityCheckBackgroundWorker(worker, ERROR))
1006 0 : return false;
1007 :
1008 3224 : parallel = (worker->bgw_flags & BGWORKER_CLASS_PARALLEL) != 0;
1009 :
1010 3224 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1011 :
1012 : /*
1013 : * If this is a parallel worker, check whether there are already too many
1014 : * parallel workers; if so, don't register another one. Our view of
1015 : * parallel_terminate_count may be slightly stale, but that doesn't really
1016 : * matter: we would have gotten the same result if we'd arrived here
1017 : * slightly earlier anyway. There's no help for it, either, since the
1018 : * postmaster must not take locks; a memory barrier wouldn't guarantee
1019 : * anything useful.
1020 : */
1021 3224 : if (parallel && (BackgroundWorkerData->parallel_register_count -
1022 2614 : BackgroundWorkerData->parallel_terminate_count) >=
1023 : max_parallel_workers)
1024 : {
1025 : Assert(BackgroundWorkerData->parallel_register_count -
1026 : BackgroundWorkerData->parallel_terminate_count <=
1027 : MAX_PARALLEL_WORKER_LIMIT);
1028 18 : LWLockRelease(BackgroundWorkerLock);
1029 18 : return false;
1030 : }
1031 :
1032 : /*
1033 : * Look for an unused slot. If we find one, grab it.
1034 : */
1035 9940 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
1036 : {
1037 9938 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1038 :
1039 9938 : if (!slot->in_use)
1040 : {
1041 3204 : memcpy(&slot->worker, worker, sizeof(BackgroundWorker));
1042 3204 : slot->pid = InvalidPid; /* indicates not started yet */
1043 3204 : slot->generation++;
1044 3204 : slot->terminate = false;
1045 3204 : generation = slot->generation;
1046 3204 : if (parallel)
1047 2594 : BackgroundWorkerData->parallel_register_count++;
1048 :
1049 : /*
1050 : * Make sure postmaster doesn't see the slot as in use before it
1051 : * sees the new contents.
1052 : */
1053 3204 : pg_write_barrier();
1054 :
1055 3204 : slot->in_use = true;
1056 3204 : success = true;
1057 3204 : break;
1058 : }
1059 : }
1060 :
1061 3206 : LWLockRelease(BackgroundWorkerLock);
1062 :
1063 : /* If we found a slot, tell the postmaster to notice the change. */
1064 3206 : if (success)
1065 3204 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1066 :
1067 : /*
1068 : * If we found a slot and the user has provided a handle, initialize it.
1069 : */
1070 3206 : if (success && handle)
1071 : {
1072 3204 : *handle = palloc(sizeof(BackgroundWorkerHandle));
1073 3204 : (*handle)->slot = slotno;
1074 3204 : (*handle)->generation = generation;
1075 : }
1076 :
1077 3206 : return success;
1078 : }
1079 :
1080 : /*
1081 : * Get the PID of a dynamically-registered background worker.
1082 : *
1083 : * If the worker is determined to be running, the return value will be
1084 : * BGWH_STARTED and *pidp will get the PID of the worker process. If the
1085 : * postmaster has not yet attempted to start the worker, the return value will
1086 : * be BGWH_NOT_YET_STARTED. Otherwise, the return value is BGWH_STOPPED.
1087 : *
1088 : * BGWH_STOPPED can indicate either that the worker is temporarily stopped
1089 : * (because it is configured for automatic restart and exited non-zero),
1090 : * or that the worker is permanently stopped (because it exited with exit
1091 : * code 0, or was not configured for automatic restart), or even that the
1092 : * worker was unregistered without ever starting (either because startup
1093 : * failed and the worker is not configured for automatic restart, or because
1094 : * TerminateBackgroundWorker was used before the worker was successfully
1095 : * started).
1096 : */
1097 : BgwHandleStatus
1098 16819046 : GetBackgroundWorkerPid(BackgroundWorkerHandle *handle, pid_t *pidp)
1099 : {
1100 : BackgroundWorkerSlot *slot;
1101 : pid_t pid;
1102 :
1103 : Assert(handle->slot < max_worker_processes);
1104 16819046 : slot = &BackgroundWorkerData->slot[handle->slot];
1105 :
1106 : /*
1107 : * We could probably arrange to synchronize access to data using memory
1108 : * barriers only, but for now, let's just keep it simple and grab the
1109 : * lock. It seems unlikely that there will be enough traffic here to
1110 : * result in meaningful contention.
1111 : */
1112 16819046 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1113 :
1114 : /*
1115 : * The generation number can't be concurrently changed while we hold the
1116 : * lock. The pid, which is updated by the postmaster, can change at any
1117 : * time, but we assume such changes are atomic. So the value we read
1118 : * won't be garbage, but it might be out of date by the time the caller
1119 : * examines it (but that's unavoidable anyway).
1120 : *
1121 : * The in_use flag could be in the process of changing from true to false,
1122 : * but if it is already false then it can't change further.
1123 : */
1124 16819046 : if (handle->generation != slot->generation || !slot->in_use)
1125 2596 : pid = 0;
1126 : else
1127 16816450 : pid = slot->pid;
1128 :
1129 : /* All done. */
1130 16819046 : LWLockRelease(BackgroundWorkerLock);
1131 :
1132 16819046 : if (pid == 0)
1133 2596 : return BGWH_STOPPED;
1134 16816450 : else if (pid == InvalidPid)
1135 5108798 : return BGWH_NOT_YET_STARTED;
1136 11707652 : *pidp = pid;
1137 11707652 : return BGWH_STARTED;
1138 : }
1139 :
1140 : /*
1141 : * Wait for a background worker to start up.
1142 : *
1143 : * This is like GetBackgroundWorkerPid(), except that if the worker has not
1144 : * yet started, we wait for it to do so; thus, BGWH_NOT_YET_STARTED is never
1145 : * returned. However, if the postmaster has died, we give up and return
1146 : * BGWH_POSTMASTER_DIED, since it that case we know that startup will not
1147 : * take place.
1148 : *
1149 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1150 : * else we will not be awoken promptly when the worker's state changes.
1151 : */
1152 : BgwHandleStatus
1153 38 : WaitForBackgroundWorkerStartup(BackgroundWorkerHandle *handle, pid_t *pidp)
1154 : {
1155 : BgwHandleStatus status;
1156 : int rc;
1157 :
1158 : for (;;)
1159 24 : {
1160 : pid_t pid;
1161 :
1162 38 : CHECK_FOR_INTERRUPTS();
1163 :
1164 38 : status = GetBackgroundWorkerPid(handle, &pid);
1165 38 : if (status == BGWH_STARTED)
1166 14 : *pidp = pid;
1167 38 : if (status != BGWH_NOT_YET_STARTED)
1168 14 : break;
1169 :
1170 24 : rc = WaitLatch(MyLatch,
1171 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1172 : WAIT_EVENT_BGWORKER_STARTUP);
1173 :
1174 24 : if (rc & WL_POSTMASTER_DEATH)
1175 : {
1176 0 : status = BGWH_POSTMASTER_DIED;
1177 0 : break;
1178 : }
1179 :
1180 24 : ResetLatch(MyLatch);
1181 : }
1182 :
1183 14 : return status;
1184 : }
1185 :
1186 : /*
1187 : * Wait for a background worker to stop.
1188 : *
1189 : * If the worker hasn't yet started, or is running, we wait for it to stop
1190 : * and then return BGWH_STOPPED. However, if the postmaster has died, we give
1191 : * up and return BGWH_POSTMASTER_DIED, because it's the postmaster that
1192 : * notifies us when a worker's state changes.
1193 : *
1194 : * The caller *must* have set our PID as the worker's bgw_notify_pid,
1195 : * else we will not be awoken promptly when the worker's state changes.
1196 : */
1197 : BgwHandleStatus
1198 5124 : WaitForBackgroundWorkerShutdown(BackgroundWorkerHandle *handle)
1199 : {
1200 : BgwHandleStatus status;
1201 : int rc;
1202 :
1203 : for (;;)
1204 2528 : {
1205 : pid_t pid;
1206 :
1207 5124 : CHECK_FOR_INTERRUPTS();
1208 :
1209 5124 : status = GetBackgroundWorkerPid(handle, &pid);
1210 5124 : if (status == BGWH_STOPPED)
1211 2596 : break;
1212 :
1213 2528 : rc = WaitLatch(MyLatch,
1214 : WL_LATCH_SET | WL_POSTMASTER_DEATH, 0,
1215 : WAIT_EVENT_BGWORKER_SHUTDOWN);
1216 :
1217 2528 : if (rc & WL_POSTMASTER_DEATH)
1218 : {
1219 0 : status = BGWH_POSTMASTER_DIED;
1220 0 : break;
1221 : }
1222 :
1223 2528 : ResetLatch(MyLatch);
1224 : }
1225 :
1226 2596 : return status;
1227 : }
1228 :
1229 : /*
1230 : * Instruct the postmaster to terminate a background worker.
1231 : *
1232 : * Note that it's safe to do this without regard to whether the worker is
1233 : * still running, or even if the worker may already have exited and been
1234 : * unregistered.
1235 : */
1236 : void
1237 6 : TerminateBackgroundWorker(BackgroundWorkerHandle *handle)
1238 : {
1239 : BackgroundWorkerSlot *slot;
1240 6 : bool signal_postmaster = false;
1241 :
1242 : Assert(handle->slot < max_worker_processes);
1243 6 : slot = &BackgroundWorkerData->slot[handle->slot];
1244 :
1245 : /* Set terminate flag in shared memory, unless slot has been reused. */
1246 6 : LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
1247 6 : if (handle->generation == slot->generation)
1248 : {
1249 6 : slot->terminate = true;
1250 6 : signal_postmaster = true;
1251 : }
1252 6 : LWLockRelease(BackgroundWorkerLock);
1253 :
1254 : /* Make sure the postmaster notices the change to shared memory. */
1255 6 : if (signal_postmaster)
1256 6 : SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
1257 6 : }
1258 :
1259 : /*
1260 : * Look up (and possibly load) a bgworker entry point function.
1261 : *
1262 : * For functions contained in the core code, we use library name "postgres"
1263 : * and consult the InternalBGWorkers array. External functions are
1264 : * looked up, and loaded if necessary, using load_external_function().
1265 : *
1266 : * The point of this is to pass function names as strings across process
1267 : * boundaries. We can't pass actual function addresses because of the
1268 : * possibility that the function has been loaded at a different address
1269 : * in a different process. This is obviously a hazard for functions in
1270 : * loadable libraries, but it can happen even for functions in the core code
1271 : * on platforms using EXEC_BACKEND (e.g., Windows).
1272 : *
1273 : * At some point it might be worthwhile to get rid of InternalBGWorkers[]
1274 : * in favor of applying load_external_function() for core functions too;
1275 : * but that raises portability issues that are not worth addressing now.
1276 : */
1277 : static bgworker_main_type
1278 3946 : LookupBackgroundWorkerFunction(const char *libraryname, const char *funcname)
1279 : {
1280 : /*
1281 : * If the function is to be loaded from postgres itself, search the
1282 : * InternalBGWorkers array.
1283 : */
1284 3946 : if (strcmp(libraryname, "postgres") == 0)
1285 : {
1286 : int i;
1287 :
1288 6604 : for (i = 0; i < lengthof(InternalBGWorkers); i++)
1289 : {
1290 6604 : if (strcmp(InternalBGWorkers[i].fn_name, funcname) == 0)
1291 3920 : return InternalBGWorkers[i].fn_addr;
1292 : }
1293 :
1294 : /* We can only reach this by programming error. */
1295 0 : elog(ERROR, "internal function \"%s\" not found", funcname);
1296 : }
1297 :
1298 : /* Otherwise load from external library. */
1299 26 : return (bgworker_main_type)
1300 26 : load_external_function(libraryname, funcname, true, NULL);
1301 : }
1302 :
1303 : /*
1304 : * Given a PID, get the bgw_type of the background worker. Returns NULL if
1305 : * not a valid background worker.
1306 : *
1307 : * The return value is in static memory belonging to this function, so it has
1308 : * to be used before calling this function again. This is so that the caller
1309 : * doesn't have to worry about the background worker locking protocol.
1310 : */
1311 : const char *
1312 1220 : GetBackgroundWorkerTypeByPid(pid_t pid)
1313 : {
1314 : int slotno;
1315 1220 : bool found = false;
1316 : static char result[BGW_MAXLEN];
1317 :
1318 1220 : LWLockAcquire(BackgroundWorkerLock, LW_SHARED);
1319 :
1320 1390 : for (slotno = 0; slotno < BackgroundWorkerData->total_slots; slotno++)
1321 : {
1322 1390 : BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
1323 :
1324 1390 : if (slot->pid > 0 && slot->pid == pid)
1325 : {
1326 1220 : strcpy(result, slot->worker.bgw_type);
1327 1220 : found = true;
1328 1220 : break;
1329 : }
1330 : }
1331 :
1332 1220 : LWLockRelease(BackgroundWorkerLock);
1333 :
1334 1220 : if (!found)
1335 0 : return NULL;
1336 :
1337 1220 : return result;
1338 : }
|