Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * portalmem.c
4 : * backend portal memory management
5 : *
6 : * Portals are objects representing the execution state of a query.
7 : * This module provides memory management services for portals, but it
8 : * doesn't actually run the executor for them.
9 : *
10 : *
11 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
12 : * Portions Copyright (c) 1994, Regents of the University of California
13 : *
14 : * IDENTIFICATION
15 : * src/backend/utils/mmgr/portalmem.c
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 : #include "postgres.h"
20 :
21 : #include "access/xact.h"
22 : #include "commands/portalcmds.h"
23 : #include "funcapi.h"
24 : #include "miscadmin.h"
25 : #include "storage/ipc.h"
26 : #include "utils/builtins.h"
27 : #include "utils/memutils.h"
28 : #include "utils/snapmgr.h"
29 : #include "utils/timestamp.h"
30 :
31 : /*
32 : * Estimate of the maximum number of open portals a user would have,
33 : * used in initially sizing the PortalHashTable in EnablePortalManager().
34 : * Since the hash table can expand, there's no need to make this overly
35 : * generous, and keeping it small avoids unnecessary overhead in the
36 : * hash_seq_search() calls executed during transaction end.
37 : */
38 : #define PORTALS_PER_USER 16
39 :
40 :
41 : /* ----------------
42 : * Global state
43 : * ----------------
44 : */
45 :
46 : #define MAX_PORTALNAME_LEN NAMEDATALEN
47 :
48 : typedef struct portalhashent
49 : {
50 : char portalname[MAX_PORTALNAME_LEN];
51 : Portal portal;
52 : } PortalHashEnt;
53 :
54 : static HTAB *PortalHashTable = NULL;
55 :
56 : #define PortalHashTableLookup(NAME, PORTAL) \
57 : do { \
58 : PortalHashEnt *hentry; \
59 : \
60 : hentry = (PortalHashEnt *) hash_search(PortalHashTable, \
61 : (NAME), HASH_FIND, NULL); \
62 : if (hentry) \
63 : PORTAL = hentry->portal; \
64 : else \
65 : PORTAL = NULL; \
66 : } while(0)
67 :
68 : #define PortalHashTableInsert(PORTAL, NAME) \
69 : do { \
70 : PortalHashEnt *hentry; bool found; \
71 : \
72 : hentry = (PortalHashEnt *) hash_search(PortalHashTable, \
73 : (NAME), HASH_ENTER, &found); \
74 : if (found) \
75 : elog(ERROR, "duplicate portal name"); \
76 : hentry->portal = PORTAL; \
77 : /* To avoid duplicate storage, make PORTAL->name point to htab entry */ \
78 : PORTAL->name = hentry->portalname; \
79 : } while(0)
80 :
81 : #define PortalHashTableDelete(PORTAL) \
82 : do { \
83 : PortalHashEnt *hentry; \
84 : \
85 : hentry = (PortalHashEnt *) hash_search(PortalHashTable, \
86 : PORTAL->name, HASH_REMOVE, NULL); \
87 : if (hentry == NULL) \
88 : elog(WARNING, "trying to delete portal name that does not exist"); \
89 : } while(0)
90 :
91 : static MemoryContext TopPortalContext = NULL;
92 :
93 :
94 : /* ----------------------------------------------------------------
95 : * public portal interface functions
96 : * ----------------------------------------------------------------
97 : */
98 :
99 : /*
100 : * EnablePortalManager
101 : * Enables the portal management module at backend startup.
102 : */
103 : void
104 28608 : EnablePortalManager(void)
105 : {
106 : HASHCTL ctl;
107 :
108 : Assert(TopPortalContext == NULL);
109 :
110 28608 : TopPortalContext = AllocSetContextCreate(TopMemoryContext,
111 : "TopPortalContext",
112 : ALLOCSET_DEFAULT_SIZES);
113 :
114 28608 : ctl.keysize = MAX_PORTALNAME_LEN;
115 28608 : ctl.entrysize = sizeof(PortalHashEnt);
116 :
117 : /*
118 : * use PORTALS_PER_USER as a guess of how many hash table entries to
119 : * create, initially
120 : */
121 28608 : PortalHashTable = hash_create("Portal hash", PORTALS_PER_USER,
122 : &ctl, HASH_ELEM | HASH_STRINGS);
123 28608 : }
124 :
125 : /*
126 : * GetPortalByName
127 : * Returns a portal given a portal name, or NULL if name not found.
128 : */
129 : Portal
130 780098 : GetPortalByName(const char *name)
131 : {
132 : Portal portal;
133 :
134 780098 : if (PointerIsValid(name))
135 780098 : PortalHashTableLookup(name, portal);
136 : else
137 0 : portal = NULL;
138 :
139 780098 : return portal;
140 : }
141 :
142 : /*
143 : * PortalGetPrimaryStmt
144 : * Get the "primary" stmt within a portal, ie, the one marked canSetTag.
145 : *
146 : * Returns NULL if no such stmt. If multiple PlannedStmt structs within the
147 : * portal are marked canSetTag, returns the first one. Neither of these
148 : * cases should occur in present usages of this function.
149 : */
150 : PlannedStmt *
151 327944 : PortalGetPrimaryStmt(Portal portal)
152 : {
153 : ListCell *lc;
154 :
155 327944 : foreach(lc, portal->stmts)
156 : {
157 327944 : PlannedStmt *stmt = lfirst_node(PlannedStmt, lc);
158 :
159 327944 : if (stmt->canSetTag)
160 327944 : return stmt;
161 : }
162 0 : return NULL;
163 : }
164 :
165 : /*
166 : * CreatePortal
167 : * Returns a new portal given a name.
168 : *
169 : * allowDup: if true, automatically drop any pre-existing portal of the
170 : * same name (if false, an error is raised).
171 : *
172 : * dupSilent: if true, don't even emit a WARNING.
173 : */
174 : Portal
175 680406 : CreatePortal(const char *name, bool allowDup, bool dupSilent)
176 : {
177 : Portal portal;
178 :
179 : Assert(PointerIsValid(name));
180 :
181 680406 : portal = GetPortalByName(name);
182 680406 : if (PortalIsValid(portal))
183 : {
184 11586 : if (!allowDup)
185 0 : ereport(ERROR,
186 : (errcode(ERRCODE_DUPLICATE_CURSOR),
187 : errmsg("cursor \"%s\" already exists", name)));
188 11586 : if (!dupSilent)
189 0 : ereport(WARNING,
190 : (errcode(ERRCODE_DUPLICATE_CURSOR),
191 : errmsg("closing existing cursor \"%s\"",
192 : name)));
193 11586 : PortalDrop(portal, false);
194 : }
195 :
196 : /* make new portal structure */
197 680406 : portal = (Portal) MemoryContextAllocZero(TopPortalContext, sizeof *portal);
198 :
199 : /* initialize portal context; typically it won't store much */
200 680406 : portal->portalContext = AllocSetContextCreate(TopPortalContext,
201 : "PortalContext",
202 : ALLOCSET_SMALL_SIZES);
203 :
204 : /* create a resource owner for the portal */
205 680406 : portal->resowner = ResourceOwnerCreate(CurTransactionResourceOwner,
206 : "Portal");
207 :
208 : /* initialize portal fields that don't start off zero */
209 680406 : portal->status = PORTAL_NEW;
210 680406 : portal->cleanup = PortalCleanup;
211 680406 : portal->createSubid = GetCurrentSubTransactionId();
212 680406 : portal->activeSubid = portal->createSubid;
213 680406 : portal->createLevel = GetCurrentTransactionNestLevel();
214 680406 : portal->strategy = PORTAL_MULTI_QUERY;
215 680406 : portal->cursorOptions = CURSOR_OPT_NO_SCROLL;
216 680406 : portal->atStart = true;
217 680406 : portal->atEnd = true; /* disallow fetches until query is set */
218 680406 : portal->visible = true;
219 680406 : portal->creation_time = GetCurrentStatementStartTimestamp();
220 :
221 : /* put portal in table (sets portal->name) */
222 680406 : PortalHashTableInsert(portal, name);
223 :
224 : /* for named portals reuse portal->name copy */
225 680406 : MemoryContextSetIdentifier(portal->portalContext, portal->name[0] ? portal->name : "<unnamed>");
226 :
227 680406 : return portal;
228 : }
229 :
230 : /*
231 : * CreateNewPortal
232 : * Create a new portal, assigning it a random nonconflicting name.
233 : */
234 : Portal
235 27096 : CreateNewPortal(void)
236 : {
237 : static unsigned int unnamed_portal_count = 0;
238 :
239 : char portalname[MAX_PORTALNAME_LEN];
240 :
241 : /* Select a nonconflicting name */
242 : for (;;)
243 : {
244 27096 : unnamed_portal_count++;
245 27096 : sprintf(portalname, "<unnamed portal %u>", unnamed_portal_count);
246 27096 : if (GetPortalByName(portalname) == NULL)
247 27096 : break;
248 : }
249 :
250 27096 : return CreatePortal(portalname, false, false);
251 : }
252 :
253 : /*
254 : * PortalDefineQuery
255 : * A simple subroutine to establish a portal's query.
256 : *
257 : * Notes: as of PG 8.4, caller MUST supply a sourceText string; it is not
258 : * allowed anymore to pass NULL. (If you really don't have source text,
259 : * you can pass a constant string, perhaps "(query not available)".)
260 : *
261 : * commandTag shall be NULL if and only if the original query string
262 : * (before rewriting) was an empty string. Also, the passed commandTag must
263 : * be a pointer to a constant string, since it is not copied.
264 : *
265 : * If cplan is provided, then it is a cached plan containing the stmts, and
266 : * the caller must have done GetCachedPlan(), causing a refcount increment.
267 : * The refcount will be released when the portal is destroyed.
268 : *
269 : * If cplan is NULL, then it is the caller's responsibility to ensure that
270 : * the passed plan trees have adequate lifetime. Typically this is done by
271 : * copying them into the portal's context.
272 : *
273 : * The caller is also responsible for ensuring that the passed prepStmtName
274 : * (if not NULL) and sourceText have adequate lifetime.
275 : *
276 : * NB: this function mustn't do much beyond storing the passed values; in
277 : * particular don't do anything that risks elog(ERROR). If that were to
278 : * happen here before storing the cplan reference, we'd leak the plancache
279 : * refcount that the caller is trying to hand off to us.
280 : */
281 : void
282 680372 : PortalDefineQuery(Portal portal,
283 : const char *prepStmtName,
284 : const char *sourceText,
285 : CommandTag commandTag,
286 : List *stmts,
287 : CachedPlan *cplan)
288 : {
289 : Assert(PortalIsValid(portal));
290 : Assert(portal->status == PORTAL_NEW);
291 :
292 : Assert(sourceText != NULL);
293 : Assert(commandTag != CMDTAG_UNKNOWN || stmts == NIL);
294 :
295 680372 : portal->prepStmtName = prepStmtName;
296 680372 : portal->sourceText = sourceText;
297 680372 : portal->qc.commandTag = commandTag;
298 680372 : portal->qc.nprocessed = 0;
299 680372 : portal->commandTag = commandTag;
300 680372 : portal->stmts = stmts;
301 680372 : portal->cplan = cplan;
302 680372 : portal->status = PORTAL_DEFINED;
303 680372 : }
304 :
305 : /*
306 : * PortalReleaseCachedPlan
307 : * Release a portal's reference to its cached plan, if any.
308 : */
309 : static void
310 708346 : PortalReleaseCachedPlan(Portal portal)
311 : {
312 708346 : if (portal->cplan)
313 : {
314 38090 : ReleaseCachedPlan(portal->cplan, NULL);
315 38090 : portal->cplan = NULL;
316 :
317 : /*
318 : * We must also clear portal->stmts which is now a dangling reference
319 : * to the cached plan's plan list. This protects any code that might
320 : * try to examine the Portal later.
321 : */
322 38090 : portal->stmts = NIL;
323 : }
324 708346 : }
325 :
326 : /*
327 : * PortalCreateHoldStore
328 : * Create the tuplestore for a portal.
329 : */
330 : void
331 46430 : PortalCreateHoldStore(Portal portal)
332 : {
333 : MemoryContext oldcxt;
334 :
335 : Assert(portal->holdContext == NULL);
336 : Assert(portal->holdStore == NULL);
337 : Assert(portal->holdSnapshot == NULL);
338 :
339 : /*
340 : * Create the memory context that is used for storage of the tuple set.
341 : * Note this is NOT a child of the portal's portalContext.
342 : */
343 46430 : portal->holdContext =
344 46430 : AllocSetContextCreate(TopPortalContext,
345 : "PortalHoldContext",
346 : ALLOCSET_DEFAULT_SIZES);
347 :
348 : /*
349 : * Create the tuple store, selecting cross-transaction temp files, and
350 : * enabling random access only if cursor requires scrolling.
351 : *
352 : * XXX: Should maintenance_work_mem be used for the portal size?
353 : */
354 46430 : oldcxt = MemoryContextSwitchTo(portal->holdContext);
355 :
356 46430 : portal->holdStore =
357 46430 : tuplestore_begin_heap(portal->cursorOptions & CURSOR_OPT_SCROLL,
358 : true, work_mem);
359 :
360 46430 : MemoryContextSwitchTo(oldcxt);
361 46430 : }
362 :
363 : /*
364 : * PinPortal
365 : * Protect a portal from dropping.
366 : *
367 : * A pinned portal is still unpinned and dropped at transaction or
368 : * subtransaction abort.
369 : */
370 : void
371 11876 : PinPortal(Portal portal)
372 : {
373 11876 : if (portal->portalPinned)
374 0 : elog(ERROR, "portal already pinned");
375 :
376 11876 : portal->portalPinned = true;
377 11876 : }
378 :
379 : void
380 11828 : UnpinPortal(Portal portal)
381 : {
382 11828 : if (!portal->portalPinned)
383 0 : elog(ERROR, "portal not pinned");
384 :
385 11828 : portal->portalPinned = false;
386 11828 : }
387 :
388 : /*
389 : * MarkPortalActive
390 : * Transition a portal from READY to ACTIVE state.
391 : *
392 : * NOTE: never set portal->status = PORTAL_ACTIVE directly; call this instead.
393 : */
394 : void
395 714690 : MarkPortalActive(Portal portal)
396 : {
397 : /* For safety, this is a runtime test not just an Assert */
398 714690 : if (portal->status != PORTAL_READY)
399 18 : ereport(ERROR,
400 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
401 : errmsg("portal \"%s\" cannot be run", portal->name)));
402 : /* Perform the state transition */
403 714672 : portal->status = PORTAL_ACTIVE;
404 714672 : portal->activeSubid = GetCurrentSubTransactionId();
405 714672 : }
406 :
407 : /*
408 : * MarkPortalDone
409 : * Transition a portal from ACTIVE to DONE state.
410 : *
411 : * NOTE: never set portal->status = PORTAL_DONE directly; call this instead.
412 : */
413 : void
414 354106 : MarkPortalDone(Portal portal)
415 : {
416 : /* Perform the state transition */
417 : Assert(portal->status == PORTAL_ACTIVE);
418 354106 : portal->status = PORTAL_DONE;
419 :
420 : /*
421 : * Allow portalcmds.c to clean up the state it knows about. We might as
422 : * well do that now, since the portal can't be executed any more.
423 : *
424 : * In some cases involving execution of a ROLLBACK command in an already
425 : * aborted transaction, this is necessary, or we'd reach AtCleanup_Portals
426 : * with the cleanup hook still unexecuted.
427 : */
428 354106 : if (PointerIsValid(portal->cleanup))
429 : {
430 354056 : portal->cleanup(portal);
431 354056 : portal->cleanup = NULL;
432 : }
433 354106 : }
434 :
435 : /*
436 : * MarkPortalFailed
437 : * Transition a portal into FAILED state.
438 : *
439 : * NOTE: never set portal->status = PORTAL_FAILED directly; call this instead.
440 : */
441 : void
442 27658 : MarkPortalFailed(Portal portal)
443 : {
444 : /* Perform the state transition */
445 : Assert(portal->status != PORTAL_DONE);
446 27658 : portal->status = PORTAL_FAILED;
447 :
448 : /*
449 : * Allow portalcmds.c to clean up the state it knows about. We might as
450 : * well do that now, since the portal can't be executed any more.
451 : *
452 : * In some cases involving cleanup of an already aborted transaction, this
453 : * is necessary, or we'd reach AtCleanup_Portals with the cleanup hook
454 : * still unexecuted.
455 : */
456 27658 : if (PointerIsValid(portal->cleanup))
457 : {
458 27644 : portal->cleanup(portal);
459 27644 : portal->cleanup = NULL;
460 : }
461 27658 : }
462 :
463 : /*
464 : * PortalDrop
465 : * Destroy the portal.
466 : */
467 : void
468 680388 : PortalDrop(Portal portal, bool isTopCommit)
469 : {
470 : Assert(PortalIsValid(portal));
471 :
472 : /*
473 : * Don't allow dropping a pinned portal, it's still needed by whoever
474 : * pinned it.
475 : */
476 680388 : if (portal->portalPinned)
477 0 : ereport(ERROR,
478 : (errcode(ERRCODE_INVALID_CURSOR_STATE),
479 : errmsg("cannot drop pinned portal \"%s\"", portal->name)));
480 :
481 : /*
482 : * Not sure if the PORTAL_ACTIVE case can validly happen or not...
483 : */
484 680388 : if (portal->status == PORTAL_ACTIVE)
485 0 : ereport(ERROR,
486 : (errcode(ERRCODE_INVALID_CURSOR_STATE),
487 : errmsg("cannot drop active portal \"%s\"", portal->name)));
488 :
489 : /*
490 : * Allow portalcmds.c to clean up the state it knows about, in particular
491 : * shutting down the executor if still active. This step potentially runs
492 : * user-defined code so failure has to be expected. It's the cleanup
493 : * hook's responsibility to not try to do that more than once, in the case
494 : * that failure occurs and then we come back to drop the portal again
495 : * during transaction abort.
496 : *
497 : * Note: in most paths of control, this will have been done already in
498 : * MarkPortalDone or MarkPortalFailed. We're just making sure.
499 : */
500 680388 : if (PointerIsValid(portal->cleanup))
501 : {
502 298588 : portal->cleanup(portal);
503 298588 : portal->cleanup = NULL;
504 : }
505 :
506 : /* There shouldn't be an active snapshot anymore, except after error */
507 : Assert(portal->portalSnapshot == NULL || !isTopCommit);
508 :
509 : /*
510 : * Remove portal from hash table. Because we do this here, we will not
511 : * come back to try to remove the portal again if there's any error in the
512 : * subsequent steps. Better to leak a little memory than to get into an
513 : * infinite error-recovery loop.
514 : */
515 680388 : PortalHashTableDelete(portal);
516 :
517 : /* drop cached plan reference, if any */
518 680388 : PortalReleaseCachedPlan(portal);
519 :
520 : /*
521 : * If portal has a snapshot protecting its data, release that. This needs
522 : * a little care since the registration will be attached to the portal's
523 : * resowner; if the portal failed, we will already have released the
524 : * resowner (and the snapshot) during transaction abort.
525 : */
526 680388 : if (portal->holdSnapshot)
527 : {
528 39908 : if (portal->resowner)
529 39600 : UnregisterSnapshotFromOwner(portal->holdSnapshot,
530 : portal->resowner);
531 39908 : portal->holdSnapshot = NULL;
532 : }
533 :
534 : /*
535 : * Release any resources still attached to the portal. There are several
536 : * cases being covered here:
537 : *
538 : * Top transaction commit (indicated by isTopCommit): normally we should
539 : * do nothing here and let the regular end-of-transaction resource
540 : * releasing mechanism handle these resources too. However, if we have a
541 : * FAILED portal (eg, a cursor that got an error), we'd better clean up
542 : * its resources to avoid resource-leakage warning messages.
543 : *
544 : * Sub transaction commit: never comes here at all, since we don't kill
545 : * any portals in AtSubCommit_Portals().
546 : *
547 : * Main or sub transaction abort: we will do nothing here because
548 : * portal->resowner was already set NULL; the resources were already
549 : * cleaned up in transaction abort.
550 : *
551 : * Ordinary portal drop: must release resources. However, if the portal
552 : * is not FAILED then we do not release its locks. The locks become the
553 : * responsibility of the transaction's ResourceOwner (since it is the
554 : * parent of the portal's owner) and will be released when the transaction
555 : * eventually ends.
556 : */
557 680388 : if (portal->resowner &&
558 644930 : (!isTopCommit || portal->status == PORTAL_FAILED))
559 : {
560 634392 : bool isCommit = (portal->status != PORTAL_FAILED);
561 :
562 634392 : ResourceOwnerRelease(portal->resowner,
563 : RESOURCE_RELEASE_BEFORE_LOCKS,
564 : isCommit, false);
565 634392 : ResourceOwnerRelease(portal->resowner,
566 : RESOURCE_RELEASE_LOCKS,
567 : isCommit, false);
568 634392 : ResourceOwnerRelease(portal->resowner,
569 : RESOURCE_RELEASE_AFTER_LOCKS,
570 : isCommit, false);
571 634392 : ResourceOwnerDelete(portal->resowner);
572 : }
573 680388 : portal->resowner = NULL;
574 :
575 : /*
576 : * Delete tuplestore if present. We should do this even under error
577 : * conditions; since the tuplestore would have been using cross-
578 : * transaction storage, its temp files need to be explicitly deleted.
579 : */
580 680388 : if (portal->holdStore)
581 : {
582 : MemoryContext oldcontext;
583 :
584 46412 : oldcontext = MemoryContextSwitchTo(portal->holdContext);
585 46412 : tuplestore_end(portal->holdStore);
586 46412 : MemoryContextSwitchTo(oldcontext);
587 46412 : portal->holdStore = NULL;
588 : }
589 :
590 : /* delete tuplestore storage, if any */
591 680388 : if (portal->holdContext)
592 46412 : MemoryContextDelete(portal->holdContext);
593 :
594 : /* release subsidiary storage */
595 680388 : MemoryContextDelete(portal->portalContext);
596 :
597 : /* release portal struct (it's in TopPortalContext) */
598 680388 : pfree(portal);
599 680388 : }
600 :
601 : /*
602 : * Delete all declared cursors.
603 : *
604 : * Used by commands: CLOSE ALL, DISCARD ALL
605 : */
606 : void
607 18 : PortalHashTableDeleteAll(void)
608 : {
609 : HASH_SEQ_STATUS status;
610 : PortalHashEnt *hentry;
611 :
612 18 : if (PortalHashTable == NULL)
613 0 : return;
614 :
615 18 : hash_seq_init(&status, PortalHashTable);
616 72 : while ((hentry = hash_seq_search(&status)) != NULL)
617 : {
618 54 : Portal portal = hentry->portal;
619 :
620 : /* Can't close the active portal (the one running the command) */
621 54 : if (portal->status == PORTAL_ACTIVE)
622 30 : continue;
623 :
624 24 : PortalDrop(portal, false);
625 :
626 : /* Restart the iteration in case that led to other drops */
627 24 : hash_seq_term(&status);
628 24 : hash_seq_init(&status, PortalHashTable);
629 : }
630 : }
631 :
632 : /*
633 : * "Hold" a portal. Prepare it for access by later transactions.
634 : */
635 : static void
636 82 : HoldPortal(Portal portal)
637 : {
638 : /*
639 : * Note that PersistHoldablePortal() must release all resources used by
640 : * the portal that are local to the creating transaction.
641 : */
642 82 : PortalCreateHoldStore(portal);
643 82 : PersistHoldablePortal(portal);
644 :
645 : /* drop cached plan reference, if any */
646 78 : PortalReleaseCachedPlan(portal);
647 :
648 : /*
649 : * Any resources belonging to the portal will be released in the upcoming
650 : * transaction-wide cleanup; the portal will no longer have its own
651 : * resources.
652 : */
653 78 : portal->resowner = NULL;
654 :
655 : /*
656 : * Having successfully exported the holdable cursor, mark it as not
657 : * belonging to this transaction.
658 : */
659 78 : portal->createSubid = InvalidSubTransactionId;
660 78 : portal->activeSubid = InvalidSubTransactionId;
661 78 : portal->createLevel = 0;
662 78 : }
663 :
664 : /*
665 : * Pre-commit processing for portals.
666 : *
667 : * Holdable cursors created in this transaction need to be converted to
668 : * materialized form, since we are going to close down the executor and
669 : * release locks. Non-holdable portals created in this transaction are
670 : * simply removed. Portals remaining from prior transactions should be
671 : * left untouched.
672 : *
673 : * Returns true if any portals changed state (possibly causing user-defined
674 : * code to be run), false if not.
675 : */
676 : bool
677 707004 : PreCommit_Portals(bool isPrepare)
678 : {
679 707004 : bool result = false;
680 : HASH_SEQ_STATUS status;
681 : PortalHashEnt *hentry;
682 :
683 707004 : hash_seq_init(&status, PortalHashTable);
684 :
685 766746 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
686 : {
687 59742 : Portal portal = hentry->portal;
688 :
689 : /*
690 : * There should be no pinned portals anymore. Complain if someone
691 : * leaked one. Auto-held portals are allowed; we assume that whoever
692 : * pinned them is managing them.
693 : */
694 59742 : if (portal->portalPinned && !portal->autoHeld)
695 0 : elog(ERROR, "cannot commit while a portal is pinned");
696 :
697 : /*
698 : * Do not touch active portals --- this can only happen in the case of
699 : * a multi-transaction utility command, such as VACUUM, or a commit in
700 : * a procedure.
701 : *
702 : * Note however that any resource owner attached to such a portal is
703 : * still going to go away, so don't leave a dangling pointer. Also
704 : * unregister any snapshots held by the portal, mainly to avoid
705 : * snapshot leak warnings from ResourceOwnerRelease().
706 : */
707 59742 : if (portal->status == PORTAL_ACTIVE)
708 : {
709 48588 : if (portal->holdSnapshot)
710 : {
711 2 : if (portal->resowner)
712 2 : UnregisterSnapshotFromOwner(portal->holdSnapshot,
713 : portal->resowner);
714 2 : portal->holdSnapshot = NULL;
715 : }
716 48588 : portal->resowner = NULL;
717 : /* Clear portalSnapshot too, for cleanliness */
718 48588 : portal->portalSnapshot = NULL;
719 48588 : continue;
720 : }
721 :
722 : /* Is it a holdable portal created in the current xact? */
723 11154 : if ((portal->cursorOptions & CURSOR_OPT_HOLD) &&
724 492 : portal->createSubid != InvalidSubTransactionId &&
725 46 : portal->status == PORTAL_READY)
726 : {
727 : /*
728 : * We are exiting the transaction that created a holdable cursor.
729 : * Instead of dropping the portal, prepare it for access by later
730 : * transactions.
731 : *
732 : * However, if this is PREPARE TRANSACTION rather than COMMIT,
733 : * refuse PREPARE, because the semantics seem pretty unclear.
734 : */
735 46 : if (isPrepare)
736 0 : ereport(ERROR,
737 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
738 : errmsg("cannot PREPARE a transaction that has created a cursor WITH HOLD")));
739 :
740 46 : HoldPortal(portal);
741 :
742 : /* Report we changed state */
743 46 : result = true;
744 : }
745 11108 : else if (portal->createSubid == InvalidSubTransactionId)
746 : {
747 : /*
748 : * Do nothing to cursors held over from a previous transaction
749 : * (including ones we just froze in a previous cycle of this loop)
750 : */
751 530 : continue;
752 : }
753 : else
754 : {
755 : /* Zap all non-holdable portals */
756 10578 : PortalDrop(portal, true);
757 :
758 : /* Report we changed state */
759 10578 : result = true;
760 : }
761 :
762 : /*
763 : * After either freezing or dropping a portal, we have to restart the
764 : * iteration, because we could have invoked user-defined code that
765 : * caused a drop of the next portal in the hash chain.
766 : */
767 10624 : hash_seq_term(&status);
768 10624 : hash_seq_init(&status, PortalHashTable);
769 : }
770 :
771 707004 : return result;
772 : }
773 :
774 : /*
775 : * Abort processing for portals.
776 : *
777 : * At this point we run the cleanup hook if present, but we can't release the
778 : * portal's memory until the cleanup call.
779 : */
780 : void
781 46610 : AtAbort_Portals(void)
782 : {
783 : HASH_SEQ_STATUS status;
784 : PortalHashEnt *hentry;
785 :
786 46610 : hash_seq_init(&status, PortalHashTable);
787 :
788 74386 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
789 : {
790 27776 : Portal portal = hentry->portal;
791 :
792 : /*
793 : * When elog(FATAL) is progress, we need to set the active portal to
794 : * failed, so that PortalCleanup() doesn't run the executor shutdown.
795 : */
796 27776 : if (portal->status == PORTAL_ACTIVE && shmem_exit_inprogress)
797 8 : MarkPortalFailed(portal);
798 :
799 : /*
800 : * Do nothing else to cursors held over from a previous transaction.
801 : */
802 27776 : if (portal->createSubid == InvalidSubTransactionId)
803 136 : continue;
804 :
805 : /*
806 : * Do nothing to auto-held cursors. This is similar to the case of a
807 : * cursor from a previous transaction, but it could also be that the
808 : * cursor was auto-held in this transaction, so it wants to live on.
809 : */
810 27640 : if (portal->autoHeld)
811 0 : continue;
812 :
813 : /*
814 : * If it was created in the current transaction, we can't do normal
815 : * shutdown on a READY portal either; it might refer to objects
816 : * created in the failed transaction. See comments in
817 : * AtSubAbort_Portals.
818 : */
819 27640 : if (portal->status == PORTAL_READY)
820 286 : MarkPortalFailed(portal);
821 :
822 : /*
823 : * Allow portalcmds.c to clean up the state it knows about, if we
824 : * haven't already.
825 : */
826 27640 : if (PointerIsValid(portal->cleanup))
827 : {
828 100 : portal->cleanup(portal);
829 100 : portal->cleanup = NULL;
830 : }
831 :
832 : /* drop cached plan reference, if any */
833 27640 : PortalReleaseCachedPlan(portal);
834 :
835 : /*
836 : * Any resources belonging to the portal will be released in the
837 : * upcoming transaction-wide cleanup; they will be gone before we run
838 : * PortalDrop.
839 : */
840 27640 : portal->resowner = NULL;
841 :
842 : /*
843 : * Although we can't delete the portal data structure proper, we can
844 : * release any memory in subsidiary contexts, such as executor state.
845 : * The cleanup hook was the last thing that might have needed data
846 : * there. But leave active portals alone.
847 : */
848 27640 : if (portal->status != PORTAL_ACTIVE)
849 27460 : MemoryContextDeleteChildren(portal->portalContext);
850 : }
851 46610 : }
852 :
853 : /*
854 : * Post-abort cleanup for portals.
855 : *
856 : * Delete all portals not held over from prior transactions. */
857 : void
858 46588 : AtCleanup_Portals(void)
859 : {
860 : HASH_SEQ_STATUS status;
861 : PortalHashEnt *hentry;
862 :
863 46588 : hash_seq_init(&status, PortalHashTable);
864 :
865 73092 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
866 : {
867 26504 : Portal portal = hentry->portal;
868 :
869 : /*
870 : * Do not touch active portals --- this can only happen in the case of
871 : * a multi-transaction command.
872 : */
873 26504 : if (portal->status == PORTAL_ACTIVE)
874 180 : continue;
875 :
876 : /*
877 : * Do nothing to cursors held over from a previous transaction or
878 : * auto-held ones.
879 : */
880 26324 : if (portal->createSubid == InvalidSubTransactionId || portal->autoHeld)
881 : {
882 : Assert(portal->status != PORTAL_ACTIVE);
883 : Assert(portal->resowner == NULL);
884 136 : continue;
885 : }
886 :
887 : /*
888 : * If a portal is still pinned, forcibly unpin it. PortalDrop will not
889 : * let us drop the portal otherwise. Whoever pinned the portal was
890 : * interrupted by the abort too and won't try to use it anymore.
891 : */
892 26188 : if (portal->portalPinned)
893 38 : portal->portalPinned = false;
894 :
895 : /*
896 : * We had better not call any user-defined code during cleanup, so if
897 : * the cleanup hook hasn't been run yet, too bad; we'll just skip it.
898 : */
899 26188 : if (PointerIsValid(portal->cleanup))
900 : {
901 0 : elog(WARNING, "skipping cleanup for portal \"%s\"", portal->name);
902 0 : portal->cleanup = NULL;
903 : }
904 :
905 : /* Zap it. */
906 26188 : PortalDrop(portal, false);
907 : }
908 46588 : }
909 :
910 : /*
911 : * Portal-related cleanup when we return to the main loop on error.
912 : *
913 : * This is different from the cleanup at transaction abort. Auto-held portals
914 : * are cleaned up on error but not on transaction abort.
915 : */
916 : void
917 41156 : PortalErrorCleanup(void)
918 : {
919 : HASH_SEQ_STATUS status;
920 : PortalHashEnt *hentry;
921 :
922 41156 : hash_seq_init(&status, PortalHashTable);
923 :
924 42934 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
925 : {
926 1778 : Portal portal = hentry->portal;
927 :
928 1778 : if (portal->autoHeld)
929 : {
930 4 : portal->portalPinned = false;
931 4 : PortalDrop(portal, false);
932 : }
933 : }
934 41156 : }
935 :
936 : /*
937 : * Pre-subcommit processing for portals.
938 : *
939 : * Reassign portals created or used in the current subtransaction to the
940 : * parent subtransaction.
941 : */
942 : void
943 10718 : AtSubCommit_Portals(SubTransactionId mySubid,
944 : SubTransactionId parentSubid,
945 : int parentLevel,
946 : ResourceOwner parentXactOwner)
947 : {
948 : HASH_SEQ_STATUS status;
949 : PortalHashEnt *hentry;
950 :
951 10718 : hash_seq_init(&status, PortalHashTable);
952 :
953 20162 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
954 : {
955 9444 : Portal portal = hentry->portal;
956 :
957 9444 : if (portal->createSubid == mySubid)
958 : {
959 60 : portal->createSubid = parentSubid;
960 60 : portal->createLevel = parentLevel;
961 60 : if (portal->resowner)
962 60 : ResourceOwnerNewParent(portal->resowner, parentXactOwner);
963 : }
964 9444 : if (portal->activeSubid == mySubid)
965 220 : portal->activeSubid = parentSubid;
966 : }
967 10718 : }
968 :
969 : /*
970 : * Subtransaction abort handling for portals.
971 : *
972 : * Deactivate portals created or used during the failed subtransaction.
973 : * Note that per AtSubCommit_Portals, this will catch portals created/used
974 : * in descendants of the subtransaction too.
975 : *
976 : * We don't destroy any portals here; that's done in AtSubCleanup_Portals.
977 : */
978 : void
979 9280 : AtSubAbort_Portals(SubTransactionId mySubid,
980 : SubTransactionId parentSubid,
981 : ResourceOwner myXactOwner,
982 : ResourceOwner parentXactOwner)
983 : {
984 : HASH_SEQ_STATUS status;
985 : PortalHashEnt *hentry;
986 :
987 9280 : hash_seq_init(&status, PortalHashTable);
988 :
989 21426 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
990 : {
991 12146 : Portal portal = hentry->portal;
992 :
993 : /* Was it created in this subtransaction? */
994 12146 : if (portal->createSubid != mySubid)
995 : {
996 : /* No, but maybe it was used in this subtransaction? */
997 11906 : if (portal->activeSubid == mySubid)
998 : {
999 : /* Maintain activeSubid until the portal is removed */
1000 44 : portal->activeSubid = parentSubid;
1001 :
1002 : /*
1003 : * A MarkPortalActive() caller ran an upper-level portal in
1004 : * this subtransaction and left the portal ACTIVE. This can't
1005 : * happen, but force the portal into FAILED state for the same
1006 : * reasons discussed below.
1007 : *
1008 : * We assume we can get away without forcing upper-level READY
1009 : * portals to fail, even if they were run and then suspended.
1010 : * In theory a suspended upper-level portal could have
1011 : * acquired some references to objects that are about to be
1012 : * destroyed, but there should be sufficient defenses against
1013 : * such cases: the portal's original query cannot contain such
1014 : * references, and any references within, say, cached plans of
1015 : * PL/pgSQL functions are not from active queries and should
1016 : * be protected by revalidation logic.
1017 : */
1018 44 : if (portal->status == PORTAL_ACTIVE)
1019 0 : MarkPortalFailed(portal);
1020 :
1021 : /*
1022 : * Also, if we failed it during the current subtransaction
1023 : * (either just above, or earlier), reattach its resource
1024 : * owner to the current subtransaction's resource owner, so
1025 : * that any resources it still holds will be released while
1026 : * cleaning up this subtransaction. This prevents some corner
1027 : * cases wherein we might get Asserts or worse while cleaning
1028 : * up objects created during the current subtransaction
1029 : * (because they're still referenced within this portal).
1030 : */
1031 44 : if (portal->status == PORTAL_FAILED && portal->resowner)
1032 : {
1033 14 : ResourceOwnerNewParent(portal->resowner, myXactOwner);
1034 14 : portal->resowner = NULL;
1035 : }
1036 : }
1037 : /* Done if it wasn't created in this subtransaction */
1038 11906 : continue;
1039 : }
1040 :
1041 : /*
1042 : * Force any live portals of my own subtransaction into FAILED state.
1043 : * We have to do this because they might refer to objects created or
1044 : * changed in the failed subtransaction, leading to crashes within
1045 : * ExecutorEnd when portalcmds.c tries to close down the portal.
1046 : * Currently, every MarkPortalActive() caller ensures it updates the
1047 : * portal status again before relinquishing control, so ACTIVE can't
1048 : * happen here. If it does happen, dispose the portal like existing
1049 : * MarkPortalActive() callers would.
1050 : */
1051 240 : if (portal->status == PORTAL_READY ||
1052 228 : portal->status == PORTAL_ACTIVE)
1053 12 : MarkPortalFailed(portal);
1054 :
1055 : /*
1056 : * Allow portalcmds.c to clean up the state it knows about, if we
1057 : * haven't already.
1058 : */
1059 240 : if (PointerIsValid(portal->cleanup))
1060 : {
1061 0 : portal->cleanup(portal);
1062 0 : portal->cleanup = NULL;
1063 : }
1064 :
1065 : /* drop cached plan reference, if any */
1066 240 : PortalReleaseCachedPlan(portal);
1067 :
1068 : /*
1069 : * Any resources belonging to the portal will be released in the
1070 : * upcoming transaction-wide cleanup; they will be gone before we run
1071 : * PortalDrop.
1072 : */
1073 240 : portal->resowner = NULL;
1074 :
1075 : /*
1076 : * Although we can't delete the portal data structure proper, we can
1077 : * release any memory in subsidiary contexts, such as executor state.
1078 : * The cleanup hook was the last thing that might have needed data
1079 : * there.
1080 : */
1081 240 : MemoryContextDeleteChildren(portal->portalContext);
1082 : }
1083 9280 : }
1084 :
1085 : /*
1086 : * Post-subabort cleanup for portals.
1087 : *
1088 : * Drop all portals created in the failed subtransaction (but note that
1089 : * we will not drop any that were reassigned to the parent above).
1090 : */
1091 : void
1092 9280 : AtSubCleanup_Portals(SubTransactionId mySubid)
1093 : {
1094 : HASH_SEQ_STATUS status;
1095 : PortalHashEnt *hentry;
1096 :
1097 9280 : hash_seq_init(&status, PortalHashTable);
1098 :
1099 21200 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
1100 : {
1101 11920 : Portal portal = hentry->portal;
1102 :
1103 11920 : if (portal->createSubid != mySubid)
1104 11906 : continue;
1105 :
1106 : /*
1107 : * If a portal is still pinned, forcibly unpin it. PortalDrop will not
1108 : * let us drop the portal otherwise. Whoever pinned the portal was
1109 : * interrupted by the abort too and won't try to use it anymore.
1110 : */
1111 14 : if (portal->portalPinned)
1112 6 : portal->portalPinned = false;
1113 :
1114 : /*
1115 : * We had better not call any user-defined code during cleanup, so if
1116 : * the cleanup hook hasn't been run yet, too bad; we'll just skip it.
1117 : */
1118 14 : if (PointerIsValid(portal->cleanup))
1119 : {
1120 0 : elog(WARNING, "skipping cleanup for portal \"%s\"", portal->name);
1121 0 : portal->cleanup = NULL;
1122 : }
1123 :
1124 : /* Zap it. */
1125 14 : PortalDrop(portal, false);
1126 : }
1127 9280 : }
1128 :
1129 : /* Find all available cursors */
1130 : Datum
1131 120 : pg_cursor(PG_FUNCTION_ARGS)
1132 : {
1133 120 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1134 : HASH_SEQ_STATUS hash_seq;
1135 : PortalHashEnt *hentry;
1136 :
1137 : /*
1138 : * We put all the tuples into a tuplestore in one scan of the hashtable.
1139 : * This avoids any issue of the hashtable possibly changing between calls.
1140 : */
1141 120 : InitMaterializedSRF(fcinfo, 0);
1142 :
1143 120 : hash_seq_init(&hash_seq, PortalHashTable);
1144 372 : while ((hentry = hash_seq_search(&hash_seq)) != NULL)
1145 : {
1146 252 : Portal portal = hentry->portal;
1147 : Datum values[6];
1148 252 : bool nulls[6] = {0};
1149 :
1150 : /* report only "visible" entries */
1151 252 : if (!portal->visible)
1152 126 : continue;
1153 : /* also ignore it if PortalDefineQuery hasn't been called yet */
1154 126 : if (!portal->sourceText)
1155 0 : continue;
1156 :
1157 126 : values[0] = CStringGetTextDatum(portal->name);
1158 126 : values[1] = CStringGetTextDatum(portal->sourceText);
1159 126 : values[2] = BoolGetDatum(portal->cursorOptions & CURSOR_OPT_HOLD);
1160 126 : values[3] = BoolGetDatum(portal->cursorOptions & CURSOR_OPT_BINARY);
1161 126 : values[4] = BoolGetDatum(portal->cursorOptions & CURSOR_OPT_SCROLL);
1162 126 : values[5] = TimestampTzGetDatum(portal->creation_time);
1163 :
1164 126 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
1165 : }
1166 :
1167 120 : return (Datum) 0;
1168 : }
1169 :
1170 : bool
1171 60 : ThereAreNoReadyPortals(void)
1172 : {
1173 : HASH_SEQ_STATUS status;
1174 : PortalHashEnt *hentry;
1175 :
1176 60 : hash_seq_init(&status, PortalHashTable);
1177 :
1178 120 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
1179 : {
1180 60 : Portal portal = hentry->portal;
1181 :
1182 60 : if (portal->status == PORTAL_READY)
1183 0 : return false;
1184 : }
1185 :
1186 60 : return true;
1187 : }
1188 :
1189 : /*
1190 : * Hold all pinned portals.
1191 : *
1192 : * When initiating a COMMIT or ROLLBACK inside a procedure, this must be
1193 : * called to protect internally-generated cursors from being dropped during
1194 : * the transaction shutdown. Currently, SPI calls this automatically; PLs
1195 : * that initiate COMMIT or ROLLBACK some other way are on the hook to do it
1196 : * themselves. (Note that we couldn't do this in, say, AtAbort_Portals
1197 : * because we need to run user-defined code while persisting a portal.
1198 : * It's too late to do that once transaction abort has started.)
1199 : *
1200 : * We protect such portals by converting them to held cursors. We mark them
1201 : * as "auto-held" so that exception exit knows to clean them up. (In normal,
1202 : * non-exception code paths, the PL needs to clean such portals itself, since
1203 : * transaction end won't do it anymore; but that should be normal practice
1204 : * anyway.)
1205 : */
1206 : void
1207 4414 : HoldPinnedPortals(void)
1208 : {
1209 : HASH_SEQ_STATUS status;
1210 : PortalHashEnt *hentry;
1211 :
1212 4414 : hash_seq_init(&status, PortalHashTable);
1213 :
1214 8948 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
1215 : {
1216 4540 : Portal portal = hentry->portal;
1217 :
1218 4540 : if (portal->portalPinned && !portal->autoHeld)
1219 : {
1220 : /*
1221 : * Doing transaction control, especially abort, inside a cursor
1222 : * loop that is not read-only, for example using UPDATE ...
1223 : * RETURNING, has weird semantics issues. Also, this
1224 : * implementation wouldn't work, because such portals cannot be
1225 : * held. (The core grammar enforces that only SELECT statements
1226 : * can drive a cursor, but for example PL/pgSQL does not restrict
1227 : * it.)
1228 : */
1229 38 : if (portal->strategy != PORTAL_ONE_SELECT)
1230 2 : ereport(ERROR,
1231 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1232 : errmsg("cannot perform transaction commands inside a cursor loop that is not read-only")));
1233 :
1234 : /* Verify it's in a suitable state to be held */
1235 36 : if (portal->status != PORTAL_READY)
1236 0 : elog(ERROR, "pinned portal is not ready to be auto-held");
1237 :
1238 36 : HoldPortal(portal);
1239 32 : portal->autoHeld = true;
1240 : }
1241 : }
1242 4408 : }
1243 :
1244 : /*
1245 : * Drop the outer active snapshots for all portals, so that no snapshots
1246 : * remain active.
1247 : *
1248 : * Like HoldPinnedPortals, this must be called when initiating a COMMIT or
1249 : * ROLLBACK inside a procedure. This has to be separate from that since it
1250 : * should not be run until we're done with steps that are likely to fail.
1251 : *
1252 : * It's tempting to fold this into PreCommit_Portals, but to do so, we'd
1253 : * need to clean up snapshot management in VACUUM and perhaps other places.
1254 : */
1255 : void
1256 4408 : ForgetPortalSnapshots(void)
1257 : {
1258 : HASH_SEQ_STATUS status;
1259 : PortalHashEnt *hentry;
1260 4408 : int numPortalSnaps = 0;
1261 4408 : int numActiveSnaps = 0;
1262 :
1263 : /* First, scan PortalHashTable and clear portalSnapshot fields */
1264 4408 : hash_seq_init(&status, PortalHashTable);
1265 :
1266 8942 : while ((hentry = (PortalHashEnt *) hash_seq_search(&status)) != NULL)
1267 : {
1268 4534 : Portal portal = hentry->portal;
1269 :
1270 4534 : if (portal->portalSnapshot != NULL)
1271 : {
1272 4408 : portal->portalSnapshot = NULL;
1273 4408 : numPortalSnaps++;
1274 : }
1275 : /* portal->holdSnapshot will be cleaned up in PreCommit_Portals */
1276 : }
1277 :
1278 : /*
1279 : * Now, pop all the active snapshots, which should be just those that were
1280 : * portal snapshots. Ideally we'd drive this directly off the portal
1281 : * scan, but there's no good way to visit the portals in the correct
1282 : * order. So just cross-check after the fact.
1283 : */
1284 8816 : while (ActiveSnapshotSet())
1285 : {
1286 4408 : PopActiveSnapshot();
1287 4408 : numActiveSnaps++;
1288 : }
1289 :
1290 4408 : if (numPortalSnaps != numActiveSnaps)
1291 0 : elog(ERROR, "portal snapshots (%d) did not account for all active snapshots (%d)",
1292 : numPortalSnaps, numActiveSnaps);
1293 4408 : }
|
