Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * indexam.c
4 : * general index access method routines
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/index/indexam.c
12 : *
13 : * INTERFACE ROUTINES
14 : * index_open - open an index relation by relation OID
15 : * index_close - close an index relation
16 : * index_beginscan - start a scan of an index with amgettuple
17 : * index_beginscan_bitmap - start a scan of an index with amgetbitmap
18 : * index_rescan - restart a scan of an index
19 : * index_endscan - end a scan
20 : * index_insert - insert an index tuple into a relation
21 : * index_markpos - mark a scan position
22 : * index_restrpos - restore a scan position
23 : * index_parallelscan_estimate - estimate shared memory for parallel scan
24 : * index_parallelscan_initialize - initialize parallel scan
25 : * index_parallelrescan - (re)start a parallel scan of an index
26 : * index_beginscan_parallel - join parallel index scan
27 : * index_getnext_tid - get the next TID from a scan
28 : * index_fetch_heap - get the scan's next heap tuple
29 : * index_getnext_slot - get the next tuple from a scan
30 : * index_getbitmap - get all tuples from a scan
31 : * index_bulk_delete - bulk deletion of index tuples
32 : * index_vacuum_cleanup - post-deletion cleanup of an index
33 : * index_can_return - does index support index-only scans?
34 : * index_getprocid - get a support procedure OID
35 : * index_getprocinfo - get a support procedure's lookup info
36 : *
37 : * NOTES
38 : * This file contains the index_ routines which used
39 : * to be a scattered collection of stuff in access/genam.
40 : *
41 : *-------------------------------------------------------------------------
42 : */
43 :
44 : #include "postgres.h"
45 :
46 : #include "access/amapi.h"
47 : #include "access/relation.h"
48 : #include "access/reloptions.h"
49 : #include "access/relscan.h"
50 : #include "access/tableam.h"
51 : #include "catalog/index.h"
52 : #include "catalog/pg_type.h"
53 : #include "nodes/execnodes.h"
54 : #include "pgstat.h"
55 : #include "storage/lmgr.h"
56 : #include "storage/predicate.h"
57 : #include "utils/ruleutils.h"
58 : #include "utils/snapmgr.h"
59 : #include "utils/syscache.h"
60 :
61 :
62 : /* ----------------------------------------------------------------
63 : * macros used in index_ routines
64 : *
65 : * Note: the ReindexIsProcessingIndex() check in RELATION_CHECKS is there
66 : * to check that we don't try to scan or do retail insertions into an index
67 : * that is currently being rebuilt or pending rebuild. This helps to catch
68 : * things that don't work when reindexing system catalogs, as well as prevent
69 : * user errors like index expressions that access their own tables. The check
70 : * doesn't prevent the actual rebuild because we don't use RELATION_CHECKS
71 : * when calling the index AM's ambuild routine, and there is no reason for
72 : * ambuild to call its subsidiary routines through this file.
73 : * ----------------------------------------------------------------
74 : */
75 : #define RELATION_CHECKS \
76 : do { \
77 : Assert(RelationIsValid(indexRelation)); \
78 : Assert(PointerIsValid(indexRelation->rd_indam)); \
79 : if (unlikely(ReindexIsProcessingIndex(RelationGetRelid(indexRelation)))) \
80 : ereport(ERROR, \
81 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), \
82 : errmsg("cannot access index \"%s\" while it is being reindexed", \
83 : RelationGetRelationName(indexRelation)))); \
84 : } while(0)
85 :
86 : #define SCAN_CHECKS \
87 : ( \
88 : AssertMacro(IndexScanIsValid(scan)), \
89 : AssertMacro(RelationIsValid(scan->indexRelation)), \
90 : AssertMacro(PointerIsValid(scan->indexRelation->rd_indam)) \
91 : )
92 :
93 : #define CHECK_REL_PROCEDURE(pname) \
94 : do { \
95 : if (indexRelation->rd_indam->pname == NULL) \
96 : elog(ERROR, "function \"%s\" is not defined for index \"%s\"", \
97 : CppAsString(pname), RelationGetRelationName(indexRelation)); \
98 : } while(0)
99 :
100 : #define CHECK_SCAN_PROCEDURE(pname) \
101 : do { \
102 : if (scan->indexRelation->rd_indam->pname == NULL) \
103 : elog(ERROR, "function \"%s\" is not defined for index \"%s\"", \
104 : CppAsString(pname), RelationGetRelationName(scan->indexRelation)); \
105 : } while(0)
106 :
107 : static IndexScanDesc index_beginscan_internal(Relation indexRelation,
108 : int nkeys, int norderbys, Snapshot snapshot,
109 : ParallelIndexScanDesc pscan, bool temp_snap);
110 : static inline void validate_relation_kind(Relation r);
111 :
112 :
113 : /* ----------------------------------------------------------------
114 : * index_ interface functions
115 : * ----------------------------------------------------------------
116 : */
117 :
118 : /* ----------------
119 : * index_open - open an index relation by relation OID
120 : *
121 : * If lockmode is not "NoLock", the specified kind of lock is
122 : * obtained on the index. (Generally, NoLock should only be
123 : * used if the caller knows it has some appropriate lock on the
124 : * index already.)
125 : *
126 : * An error is raised if the index does not exist.
127 : *
128 : * This is a convenience routine adapted for indexscan use.
129 : * Some callers may prefer to use relation_open directly.
130 : * ----------------
131 : */
132 : Relation
133 20108752 : index_open(Oid relationId, LOCKMODE lockmode)
134 : {
135 : Relation r;
136 :
137 20108752 : r = relation_open(relationId, lockmode);
138 :
139 20108464 : validate_relation_kind(r);
140 :
141 20108442 : return r;
142 : }
143 :
144 : /* ----------------
145 : * try_index_open - open an index relation by relation OID
146 : *
147 : * Same as index_open, except return NULL instead of failing
148 : * if the relation does not exist.
149 : * ----------------
150 : */
151 : Relation
152 2130 : try_index_open(Oid relationId, LOCKMODE lockmode)
153 : {
154 : Relation r;
155 :
156 2130 : r = try_relation_open(relationId, lockmode);
157 :
158 : /* leave if index does not exist */
159 2130 : if (!r)
160 0 : return NULL;
161 :
162 2130 : validate_relation_kind(r);
163 :
164 2130 : return r;
165 : }
166 :
167 : /* ----------------
168 : * index_close - close an index relation
169 : *
170 : * If lockmode is not "NoLock", we then release the specified lock.
171 : *
172 : * Note that it is often sensible to hold a lock beyond index_close;
173 : * in that case, the lock is released automatically at xact end.
174 : * ----------------
175 : */
176 : void
177 20142912 : index_close(Relation relation, LOCKMODE lockmode)
178 : {
179 20142912 : LockRelId relid = relation->rd_lockInfo.lockRelId;
180 :
181 : Assert(lockmode >= NoLock && lockmode < MAX_LOCKMODES);
182 :
183 : /* The relcache does the real work... */
184 20142912 : RelationClose(relation);
185 :
186 20142912 : if (lockmode != NoLock)
187 18454098 : UnlockRelationId(&relid, lockmode);
188 20142912 : }
189 :
190 : /* ----------------
191 : * validate_relation_kind - check the relation's kind
192 : *
193 : * Make sure relkind is an index or a partitioned index.
194 : * ----------------
195 : */
196 : static inline void
197 20110594 : validate_relation_kind(Relation r)
198 : {
199 20110594 : if (r->rd_rel->relkind != RELKIND_INDEX &&
200 11312 : r->rd_rel->relkind != RELKIND_PARTITIONED_INDEX)
201 22 : ereport(ERROR,
202 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
203 : errmsg("\"%s\" is not an index",
204 : RelationGetRelationName(r))));
205 20110572 : }
206 :
207 :
208 : /* ----------------
209 : * index_insert - insert an index tuple into a relation
210 : * ----------------
211 : */
212 : bool
213 8635168 : index_insert(Relation indexRelation,
214 : Datum *values,
215 : bool *isnull,
216 : ItemPointer heap_t_ctid,
217 : Relation heapRelation,
218 : IndexUniqueCheck checkUnique,
219 : bool indexUnchanged,
220 : IndexInfo *indexInfo)
221 : {
222 8635168 : RELATION_CHECKS;
223 8635168 : CHECK_REL_PROCEDURE(aminsert);
224 :
225 8635168 : if (!(indexRelation->rd_indam->ampredlocks))
226 576810 : CheckForSerializableConflictIn(indexRelation,
227 : (ItemPointer) NULL,
228 : InvalidBlockNumber);
229 :
230 8635168 : return indexRelation->rd_indam->aminsert(indexRelation, values, isnull,
231 : heap_t_ctid, heapRelation,
232 : checkUnique, indexUnchanged,
233 : indexInfo);
234 : }
235 :
236 : /* -------------------------
237 : * index_insert_cleanup - clean up after all index inserts are done
238 : * -------------------------
239 : */
240 : void
241 3443290 : index_insert_cleanup(Relation indexRelation,
242 : IndexInfo *indexInfo)
243 : {
244 3443290 : RELATION_CHECKS;
245 :
246 3443290 : if (indexRelation->rd_indam->aminsertcleanup)
247 1154 : indexRelation->rd_indam->aminsertcleanup(indexRelation, indexInfo);
248 3443290 : }
249 :
250 : /*
251 : * index_beginscan - start a scan of an index with amgettuple
252 : *
253 : * Caller must be holding suitable locks on the heap and the index.
254 : */
255 : IndexScanDesc
256 15115708 : index_beginscan(Relation heapRelation,
257 : Relation indexRelation,
258 : Snapshot snapshot,
259 : IndexScanInstrumentation *instrument,
260 : int nkeys, int norderbys)
261 : {
262 : IndexScanDesc scan;
263 :
264 : Assert(snapshot != InvalidSnapshot);
265 :
266 : /* Check that a historic snapshot is not used for non-catalog tables */
267 15115708 : if (IsHistoricMVCCSnapshot(snapshot) &&
268 30716 : !RelationIsAccessibleInLogicalDecoding(heapRelation))
269 : {
270 0 : ereport(ERROR,
271 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
272 : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
273 : RelationGetRelationName(heapRelation))));
274 : }
275 :
276 15115708 : scan = index_beginscan_internal(indexRelation, nkeys, norderbys, snapshot, NULL, false);
277 :
278 : /*
279 : * Save additional parameters into the scandesc. Everything else was set
280 : * up by RelationGetIndexScan.
281 : */
282 15115708 : scan->heapRelation = heapRelation;
283 15115708 : scan->xs_snapshot = snapshot;
284 15115708 : scan->instrument = instrument;
285 :
286 : /* prepare to fetch index matches from table */
287 15115708 : scan->xs_heapfetch = table_index_fetch_begin(heapRelation);
288 :
289 15115708 : return scan;
290 : }
291 :
292 : /*
293 : * index_beginscan_bitmap - start a scan of an index with amgetbitmap
294 : *
295 : * As above, caller had better be holding some lock on the parent heap
296 : * relation, even though it's not explicitly mentioned here.
297 : */
298 : IndexScanDesc
299 18094 : index_beginscan_bitmap(Relation indexRelation,
300 : Snapshot snapshot,
301 : IndexScanInstrumentation *instrument,
302 : int nkeys)
303 : {
304 : IndexScanDesc scan;
305 :
306 : Assert(snapshot != InvalidSnapshot);
307 :
308 18094 : scan = index_beginscan_internal(indexRelation, nkeys, 0, snapshot, NULL, false);
309 :
310 : /*
311 : * Save additional parameters into the scandesc. Everything else was set
312 : * up by RelationGetIndexScan.
313 : */
314 18094 : scan->xs_snapshot = snapshot;
315 18094 : scan->instrument = instrument;
316 :
317 18094 : return scan;
318 : }
319 :
320 : /*
321 : * index_beginscan_internal --- common code for index_beginscan variants
322 : */
323 : static IndexScanDesc
324 15134192 : index_beginscan_internal(Relation indexRelation,
325 : int nkeys, int norderbys, Snapshot snapshot,
326 : ParallelIndexScanDesc pscan, bool temp_snap)
327 : {
328 : IndexScanDesc scan;
329 :
330 15134192 : RELATION_CHECKS;
331 15134192 : CHECK_REL_PROCEDURE(ambeginscan);
332 :
333 15134192 : if (!(indexRelation->rd_indam->ampredlocks))
334 4624 : PredicateLockRelation(indexRelation, snapshot);
335 :
336 : /*
337 : * We hold a reference count to the relcache entry throughout the scan.
338 : */
339 15134192 : RelationIncrementReferenceCount(indexRelation);
340 :
341 : /*
342 : * Tell the AM to open a scan.
343 : */
344 15134192 : scan = indexRelation->rd_indam->ambeginscan(indexRelation, nkeys,
345 : norderbys);
346 : /* Initialize information for parallel scan. */
347 15134192 : scan->parallel_scan = pscan;
348 15134192 : scan->xs_temp_snap = temp_snap;
349 :
350 15134192 : return scan;
351 : }
352 :
353 : /* ----------------
354 : * index_rescan - (re)start a scan of an index
355 : *
356 : * During a restart, the caller may specify a new set of scankeys and/or
357 : * orderbykeys; but the number of keys cannot differ from what index_beginscan
358 : * was told. (Later we might relax that to "must not exceed", but currently
359 : * the index AMs tend to assume that scan->numberOfKeys is what to believe.)
360 : * To restart the scan without changing keys, pass NULL for the key arrays.
361 : * (Of course, keys *must* be passed on the first call, unless
362 : * scan->numberOfKeys is zero.)
363 : * ----------------
364 : */
365 : void
366 15765766 : index_rescan(IndexScanDesc scan,
367 : ScanKey keys, int nkeys,
368 : ScanKey orderbys, int norderbys)
369 : {
370 : SCAN_CHECKS;
371 15765766 : CHECK_SCAN_PROCEDURE(amrescan);
372 :
373 : Assert(nkeys == scan->numberOfKeys);
374 : Assert(norderbys == scan->numberOfOrderBys);
375 :
376 : /* Release resources (like buffer pins) from table accesses */
377 15765766 : if (scan->xs_heapfetch)
378 15743794 : table_index_fetch_reset(scan->xs_heapfetch);
379 :
380 15765766 : scan->kill_prior_tuple = false; /* for safety */
381 15765766 : scan->xs_heap_continue = false;
382 :
383 15765766 : scan->indexRelation->rd_indam->amrescan(scan, keys, nkeys,
384 : orderbys, norderbys);
385 15765766 : }
386 :
387 : /* ----------------
388 : * index_endscan - end a scan
389 : * ----------------
390 : */
391 : void
392 15132320 : index_endscan(IndexScanDesc scan)
393 : {
394 : SCAN_CHECKS;
395 15132320 : CHECK_SCAN_PROCEDURE(amendscan);
396 :
397 : /* Release resources (like buffer pins) from table accesses */
398 15132320 : if (scan->xs_heapfetch)
399 : {
400 15114328 : table_index_fetch_end(scan->xs_heapfetch);
401 15114328 : scan->xs_heapfetch = NULL;
402 : }
403 :
404 : /* End the AM's scan */
405 15132320 : scan->indexRelation->rd_indam->amendscan(scan);
406 :
407 : /* Release index refcount acquired by index_beginscan */
408 15132320 : RelationDecrementReferenceCount(scan->indexRelation);
409 :
410 15132320 : if (scan->xs_temp_snap)
411 390 : UnregisterSnapshot(scan->xs_snapshot);
412 :
413 : /* Release the scan data structure itself */
414 15132320 : IndexScanEnd(scan);
415 15132320 : }
416 :
417 : /* ----------------
418 : * index_markpos - mark a scan position
419 : * ----------------
420 : */
421 : void
422 130108 : index_markpos(IndexScanDesc scan)
423 : {
424 : SCAN_CHECKS;
425 130108 : CHECK_SCAN_PROCEDURE(ammarkpos);
426 :
427 130108 : scan->indexRelation->rd_indam->ammarkpos(scan);
428 130108 : }
429 :
430 : /* ----------------
431 : * index_restrpos - restore a scan position
432 : *
433 : * NOTE: this only restores the internal scan state of the index AM. See
434 : * comments for ExecRestrPos().
435 : *
436 : * NOTE: For heap, in the presence of HOT chains, mark/restore only works
437 : * correctly if the scan's snapshot is MVCC-safe; that ensures that there's at
438 : * most one returnable tuple in each HOT chain, and so restoring the prior
439 : * state at the granularity of the index AM is sufficient. Since the only
440 : * current user of mark/restore functionality is nodeMergejoin.c, this
441 : * effectively means that merge-join plans only work for MVCC snapshots. This
442 : * could be fixed if necessary, but for now it seems unimportant.
443 : * ----------------
444 : */
445 : void
446 54042 : index_restrpos(IndexScanDesc scan)
447 : {
448 : Assert(IsMVCCSnapshot(scan->xs_snapshot));
449 :
450 : SCAN_CHECKS;
451 54042 : CHECK_SCAN_PROCEDURE(amrestrpos);
452 :
453 : /* release resources (like buffer pins) from table accesses */
454 54042 : if (scan->xs_heapfetch)
455 54042 : table_index_fetch_reset(scan->xs_heapfetch);
456 :
457 54042 : scan->kill_prior_tuple = false; /* for safety */
458 54042 : scan->xs_heap_continue = false;
459 :
460 54042 : scan->indexRelation->rd_indam->amrestrpos(scan);
461 54042 : }
462 :
463 : /*
464 : * index_parallelscan_estimate - estimate shared memory for parallel scan
465 : *
466 : * When instrument=true, estimate includes SharedIndexScanInstrumentation
467 : * space. When parallel_aware=true, estimate includes whatever space the
468 : * index AM's amestimateparallelscan routine requested when called.
469 : */
470 : Size
471 334 : index_parallelscan_estimate(Relation indexRelation, int nkeys, int norderbys,
472 : Snapshot snapshot, bool instrument,
473 : bool parallel_aware, int nworkers)
474 : {
475 : Size nbytes;
476 :
477 : Assert(instrument || parallel_aware);
478 :
479 334 : RELATION_CHECKS;
480 :
481 334 : nbytes = offsetof(ParallelIndexScanDescData, ps_snapshot_data);
482 334 : nbytes = add_size(nbytes, EstimateSnapshotSpace(snapshot));
483 334 : nbytes = MAXALIGN(nbytes);
484 :
485 334 : if (instrument)
486 : {
487 : Size sharedinfosz;
488 :
489 270 : sharedinfosz = offsetof(SharedIndexScanInstrumentation, winstrument) +
490 : nworkers * sizeof(IndexScanInstrumentation);
491 270 : nbytes = add_size(nbytes, sharedinfosz);
492 270 : nbytes = MAXALIGN(nbytes);
493 : }
494 :
495 : /*
496 : * If parallel scan index AM interface can't be used (or index AM provides
497 : * no such interface), assume there is no AM-specific data needed
498 : */
499 334 : if (parallel_aware &&
500 64 : indexRelation->rd_indam->amestimateparallelscan != NULL)
501 64 : nbytes = add_size(nbytes,
502 64 : indexRelation->rd_indam->amestimateparallelscan(indexRelation,
503 : nkeys,
504 : norderbys));
505 :
506 334 : return nbytes;
507 : }
508 :
509 : /*
510 : * index_parallelscan_initialize - initialize parallel scan
511 : *
512 : * We initialize both the ParallelIndexScanDesc proper and the AM-specific
513 : * information which follows it.
514 : *
515 : * This function calls access method specific initialization routine to
516 : * initialize am specific information. Call this just once in the leader
517 : * process; then, individual workers attach via index_beginscan_parallel.
518 : */
519 : void
520 334 : index_parallelscan_initialize(Relation heapRelation, Relation indexRelation,
521 : Snapshot snapshot, bool instrument,
522 : bool parallel_aware, int nworkers,
523 : SharedIndexScanInstrumentation **sharedinfo,
524 : ParallelIndexScanDesc target)
525 : {
526 : Size offset;
527 :
528 : Assert(instrument || parallel_aware);
529 :
530 334 : RELATION_CHECKS;
531 :
532 334 : offset = add_size(offsetof(ParallelIndexScanDescData, ps_snapshot_data),
533 : EstimateSnapshotSpace(snapshot));
534 334 : offset = MAXALIGN(offset);
535 :
536 334 : target->ps_locator = heapRelation->rd_locator;
537 334 : target->ps_indexlocator = indexRelation->rd_locator;
538 334 : target->ps_offset_ins = 0;
539 334 : target->ps_offset_am = 0;
540 334 : SerializeSnapshot(snapshot, target->ps_snapshot_data);
541 :
542 334 : if (instrument)
543 : {
544 : Size sharedinfosz;
545 :
546 270 : target->ps_offset_ins = offset;
547 270 : sharedinfosz = offsetof(SharedIndexScanInstrumentation, winstrument) +
548 : nworkers * sizeof(IndexScanInstrumentation);
549 270 : offset = add_size(offset, sharedinfosz);
550 270 : offset = MAXALIGN(offset);
551 :
552 : /* Set leader's *sharedinfo pointer, and initialize stats */
553 270 : *sharedinfo = (SharedIndexScanInstrumentation *)
554 270 : OffsetToPointer(target, target->ps_offset_ins);
555 270 : memset(*sharedinfo, 0, sharedinfosz);
556 270 : (*sharedinfo)->num_workers = nworkers;
557 : }
558 :
559 : /* aminitparallelscan is optional; assume no-op if not provided by AM */
560 334 : if (parallel_aware && indexRelation->rd_indam->aminitparallelscan != NULL)
561 : {
562 : void *amtarget;
563 :
564 64 : target->ps_offset_am = offset;
565 64 : amtarget = OffsetToPointer(target, target->ps_offset_am);
566 64 : indexRelation->rd_indam->aminitparallelscan(amtarget);
567 : }
568 334 : }
569 :
570 : /* ----------------
571 : * index_parallelrescan - (re)start a parallel scan of an index
572 : * ----------------
573 : */
574 : void
575 24 : index_parallelrescan(IndexScanDesc scan)
576 : {
577 : SCAN_CHECKS;
578 :
579 24 : if (scan->xs_heapfetch)
580 24 : table_index_fetch_reset(scan->xs_heapfetch);
581 :
582 : /* amparallelrescan is optional; assume no-op if not provided by AM */
583 24 : if (scan->indexRelation->rd_indam->amparallelrescan != NULL)
584 24 : scan->indexRelation->rd_indam->amparallelrescan(scan);
585 24 : }
586 :
587 : /*
588 : * index_beginscan_parallel - join parallel index scan
589 : *
590 : * Caller must be holding suitable locks on the heap and the index.
591 : */
592 : IndexScanDesc
593 390 : index_beginscan_parallel(Relation heaprel, Relation indexrel,
594 : IndexScanInstrumentation *instrument,
595 : int nkeys, int norderbys,
596 : ParallelIndexScanDesc pscan)
597 : {
598 : Snapshot snapshot;
599 : IndexScanDesc scan;
600 :
601 : Assert(RelFileLocatorEquals(heaprel->rd_locator, pscan->ps_locator));
602 : Assert(RelFileLocatorEquals(indexrel->rd_locator, pscan->ps_indexlocator));
603 :
604 390 : snapshot = RestoreSnapshot(pscan->ps_snapshot_data);
605 390 : RegisterSnapshot(snapshot);
606 390 : scan = index_beginscan_internal(indexrel, nkeys, norderbys, snapshot,
607 : pscan, true);
608 :
609 : /*
610 : * Save additional parameters into the scandesc. Everything else was set
611 : * up by index_beginscan_internal.
612 : */
613 390 : scan->heapRelation = heaprel;
614 390 : scan->xs_snapshot = snapshot;
615 390 : scan->instrument = instrument;
616 :
617 : /* prepare to fetch index matches from table */
618 390 : scan->xs_heapfetch = table_index_fetch_begin(heaprel);
619 :
620 390 : return scan;
621 : }
622 :
623 : /* ----------------
624 : * index_getnext_tid - get the next TID from a scan
625 : *
626 : * The result is the next TID satisfying the scan keys,
627 : * or NULL if no more matching tuples exist.
628 : * ----------------
629 : */
630 : ItemPointer
631 36780504 : index_getnext_tid(IndexScanDesc scan, ScanDirection direction)
632 : {
633 : bool found;
634 :
635 : SCAN_CHECKS;
636 36780504 : CHECK_SCAN_PROCEDURE(amgettuple);
637 :
638 : /* XXX: we should assert that a snapshot is pushed or registered */
639 : Assert(TransactionIdIsValid(RecentXmin));
640 :
641 : /*
642 : * The AM's amgettuple proc finds the next index entry matching the scan
643 : * keys, and puts the TID into scan->xs_heaptid. It should also set
644 : * scan->xs_recheck and possibly scan->xs_itup/scan->xs_hitup, though we
645 : * pay no attention to those fields here.
646 : */
647 36780504 : found = scan->indexRelation->rd_indam->amgettuple(scan, direction);
648 :
649 : /* Reset kill flag immediately for safety */
650 36780500 : scan->kill_prior_tuple = false;
651 36780500 : scan->xs_heap_continue = false;
652 :
653 : /* If we're out of index entries, we're done */
654 36780500 : if (!found)
655 : {
656 : /* release resources (like buffer pins) from table accesses */
657 6957816 : if (scan->xs_heapfetch)
658 6957816 : table_index_fetch_reset(scan->xs_heapfetch);
659 :
660 6957816 : return NULL;
661 : }
662 : Assert(ItemPointerIsValid(&scan->xs_heaptid));
663 :
664 29822684 : pgstat_count_index_tuples(scan->indexRelation, 1);
665 :
666 : /* Return the TID of the tuple we found. */
667 29822684 : return &scan->xs_heaptid;
668 : }
669 :
670 : /* ----------------
671 : * index_fetch_heap - get the scan's next heap tuple
672 : *
673 : * The result is a visible heap tuple associated with the index TID most
674 : * recently fetched by index_getnext_tid, or NULL if no more matching tuples
675 : * exist. (There can be more than one matching tuple because of HOT chains,
676 : * although when using an MVCC snapshot it should be impossible for more than
677 : * one such tuple to exist.)
678 : *
679 : * On success, the buffer containing the heap tup is pinned (the pin will be
680 : * dropped in a future index_getnext_tid, index_fetch_heap or index_endscan
681 : * call).
682 : *
683 : * Note: caller must check scan->xs_recheck, and perform rechecking of the
684 : * scan keys if required. We do not do that here because we don't have
685 : * enough information to do it efficiently in the general case.
686 : * ----------------
687 : */
688 : bool
689 26438958 : index_fetch_heap(IndexScanDesc scan, TupleTableSlot *slot)
690 : {
691 26438958 : bool all_dead = false;
692 : bool found;
693 :
694 26438958 : found = table_index_fetch_tuple(scan->xs_heapfetch, &scan->xs_heaptid,
695 : scan->xs_snapshot, slot,
696 : &scan->xs_heap_continue, &all_dead);
697 :
698 26438948 : if (found)
699 25289398 : pgstat_count_heap_fetch(scan->indexRelation);
700 :
701 : /*
702 : * If we scanned a whole HOT chain and found only dead tuples, tell index
703 : * AM to kill its entry for that TID (this will take effect in the next
704 : * amgettuple call, in index_getnext_tid). We do not do this when in
705 : * recovery because it may violate MVCC to do so. See comments in
706 : * RelationGetIndexScan().
707 : */
708 26438948 : if (!scan->xactStartedInRecovery)
709 25992480 : scan->kill_prior_tuple = all_dead;
710 :
711 26438948 : return found;
712 : }
713 :
714 : /* ----------------
715 : * index_getnext_slot - get the next tuple from a scan
716 : *
717 : * The result is true if a tuple satisfying the scan keys and the snapshot was
718 : * found, false otherwise. The tuple is stored in the specified slot.
719 : *
720 : * On success, resources (like buffer pins) are likely to be held, and will be
721 : * dropped by a future index_getnext_tid, index_fetch_heap or index_endscan
722 : * call).
723 : *
724 : * Note: caller must check scan->xs_recheck, and perform rechecking of the
725 : * scan keys if required. We do not do that here because we don't have
726 : * enough information to do it efficiently in the general case.
727 : * ----------------
728 : */
729 : bool
730 30160576 : index_getnext_slot(IndexScanDesc scan, ScanDirection direction, TupleTableSlot *slot)
731 : {
732 : for (;;)
733 : {
734 31110888 : if (!scan->xs_heap_continue)
735 : {
736 : ItemPointer tid;
737 :
738 : /* Time to fetch the next TID from the index */
739 30948356 : tid = index_getnext_tid(scan, direction);
740 :
741 : /* If we're out of index entries, we're done */
742 30948352 : if (tid == NULL)
743 6754674 : break;
744 :
745 : Assert(ItemPointerEquals(tid, &scan->xs_heaptid));
746 : }
747 :
748 : /*
749 : * Fetch the next (or only) visible heap tuple for this index entry.
750 : * If we don't find anything, loop around and grab the next TID from
751 : * the index.
752 : */
753 : Assert(ItemPointerIsValid(&scan->xs_heaptid));
754 24356210 : if (index_fetch_heap(scan, slot))
755 23405888 : return true;
756 : }
757 :
758 6754674 : return false;
759 : }
760 :
761 : /* ----------------
762 : * index_getbitmap - get all tuples at once from an index scan
763 : *
764 : * Adds the TIDs of all heap tuples satisfying the scan keys to a bitmap.
765 : * Since there's no interlock between the index scan and the eventual heap
766 : * access, this is only safe to use with MVCC-based snapshots: the heap
767 : * item slot could have been replaced by a newer tuple by the time we get
768 : * to it.
769 : *
770 : * Returns the number of matching tuples found. (Note: this might be only
771 : * approximate, so it should only be used for statistical purposes.)
772 : * ----------------
773 : */
774 : int64
775 20654 : index_getbitmap(IndexScanDesc scan, TIDBitmap *bitmap)
776 : {
777 : int64 ntids;
778 :
779 : SCAN_CHECKS;
780 20654 : CHECK_SCAN_PROCEDURE(amgetbitmap);
781 :
782 : /* just make sure this is false... */
783 20654 : scan->kill_prior_tuple = false;
784 :
785 : /*
786 : * have the am's getbitmap proc do all the work.
787 : */
788 20654 : ntids = scan->indexRelation->rd_indam->amgetbitmap(scan, bitmap);
789 :
790 20654 : pgstat_count_index_tuples(scan->indexRelation, ntids);
791 :
792 20654 : return ntids;
793 : }
794 :
795 : /* ----------------
796 : * index_bulk_delete - do mass deletion of index entries
797 : *
798 : * callback routine tells whether a given main-heap tuple is
799 : * to be deleted
800 : *
801 : * return value is an optional palloc'd struct of statistics
802 : * ----------------
803 : */
804 : IndexBulkDeleteResult *
805 3086 : index_bulk_delete(IndexVacuumInfo *info,
806 : IndexBulkDeleteResult *istat,
807 : IndexBulkDeleteCallback callback,
808 : void *callback_state)
809 : {
810 3086 : Relation indexRelation = info->index;
811 :
812 3086 : RELATION_CHECKS;
813 3086 : CHECK_REL_PROCEDURE(ambulkdelete);
814 :
815 3086 : return indexRelation->rd_indam->ambulkdelete(info, istat,
816 : callback, callback_state);
817 : }
818 :
819 : /* ----------------
820 : * index_vacuum_cleanup - do post-deletion cleanup of an index
821 : *
822 : * return value is an optional palloc'd struct of statistics
823 : * ----------------
824 : */
825 : IndexBulkDeleteResult *
826 220342 : index_vacuum_cleanup(IndexVacuumInfo *info,
827 : IndexBulkDeleteResult *istat)
828 : {
829 220342 : Relation indexRelation = info->index;
830 :
831 220342 : RELATION_CHECKS;
832 220342 : CHECK_REL_PROCEDURE(amvacuumcleanup);
833 :
834 220342 : return indexRelation->rd_indam->amvacuumcleanup(info, istat);
835 : }
836 :
837 : /* ----------------
838 : * index_can_return
839 : *
840 : * Does the index access method support index-only scans for the given
841 : * column?
842 : * ----------------
843 : */
844 : bool
845 1362220 : index_can_return(Relation indexRelation, int attno)
846 : {
847 1362220 : RELATION_CHECKS;
848 :
849 : /* amcanreturn is optional; assume false if not provided by AM */
850 1362220 : if (indexRelation->rd_indam->amcanreturn == NULL)
851 273356 : return false;
852 :
853 1088864 : return indexRelation->rd_indam->amcanreturn(indexRelation, attno);
854 : }
855 :
856 : /* ----------------
857 : * index_getprocid
858 : *
859 : * Index access methods typically require support routines that are
860 : * not directly the implementation of any WHERE-clause query operator
861 : * and so cannot be kept in pg_amop. Instead, such routines are kept
862 : * in pg_amproc. These registered procedure OIDs are assigned numbers
863 : * according to a convention established by the access method.
864 : * The general index code doesn't know anything about the routines
865 : * involved; it just builds an ordered list of them for
866 : * each attribute on which an index is defined.
867 : *
868 : * As of Postgres 8.3, support routines within an operator family
869 : * are further subdivided by the "left type" and "right type" of the
870 : * query operator(s) that they support. The "default" functions for a
871 : * particular indexed attribute are those with both types equal to
872 : * the index opclass' opcintype (note that this is subtly different
873 : * from the indexed attribute's own type: it may be a binary-compatible
874 : * type instead). Only the default functions are stored in relcache
875 : * entries --- access methods can use the syscache to look up non-default
876 : * functions.
877 : *
878 : * This routine returns the requested default procedure OID for a
879 : * particular indexed attribute.
880 : * ----------------
881 : */
882 : RegProcedure
883 2050982 : index_getprocid(Relation irel,
884 : AttrNumber attnum,
885 : uint16 procnum)
886 : {
887 : RegProcedure *loc;
888 : int nproc;
889 : int procindex;
890 :
891 2050982 : nproc = irel->rd_indam->amsupport;
892 :
893 : Assert(procnum > 0 && procnum <= (uint16) nproc);
894 :
895 2050982 : procindex = (nproc * (attnum - 1)) + (procnum - 1);
896 :
897 2050982 : loc = irel->rd_support;
898 :
899 : Assert(loc != NULL);
900 :
901 2050982 : return loc[procindex];
902 : }
903 :
904 : /* ----------------
905 : * index_getprocinfo
906 : *
907 : * This routine allows index AMs to keep fmgr lookup info for
908 : * support procs in the relcache. As above, only the "default"
909 : * functions for any particular indexed attribute are cached.
910 : *
911 : * Note: the return value points into cached data that will be lost during
912 : * any relcache rebuild! Therefore, either use the callinfo right away,
913 : * or save it only after having acquired some type of lock on the index rel.
914 : * ----------------
915 : */
916 : FmgrInfo *
917 48230218 : index_getprocinfo(Relation irel,
918 : AttrNumber attnum,
919 : uint16 procnum)
920 : {
921 : FmgrInfo *locinfo;
922 : int nproc;
923 : int optsproc;
924 : int procindex;
925 :
926 48230218 : nproc = irel->rd_indam->amsupport;
927 48230218 : optsproc = irel->rd_indam->amoptsprocnum;
928 :
929 : Assert(procnum > 0 && procnum <= (uint16) nproc);
930 :
931 48230218 : procindex = (nproc * (attnum - 1)) + (procnum - 1);
932 :
933 48230218 : locinfo = irel->rd_supportinfo;
934 :
935 : Assert(locinfo != NULL);
936 :
937 48230218 : locinfo += procindex;
938 :
939 : /* Initialize the lookup info if first time through */
940 48230218 : if (locinfo->fn_oid == InvalidOid)
941 : {
942 1173982 : RegProcedure *loc = irel->rd_support;
943 : RegProcedure procId;
944 :
945 : Assert(loc != NULL);
946 :
947 1173982 : procId = loc[procindex];
948 :
949 : /*
950 : * Complain if function was not found during IndexSupportInitialize.
951 : * This should not happen unless the system tables contain bogus
952 : * entries for the index opclass. (If an AM wants to allow a support
953 : * function to be optional, it can use index_getprocid.)
954 : */
955 1173982 : if (!RegProcedureIsValid(procId))
956 0 : elog(ERROR, "missing support function %d for attribute %d of index \"%s\"",
957 : procnum, attnum, RelationGetRelationName(irel));
958 :
959 1173982 : fmgr_info_cxt(procId, locinfo, irel->rd_indexcxt);
960 :
961 1173982 : if (procnum != optsproc)
962 : {
963 : /* Initialize locinfo->fn_expr with opclass options Const */
964 1172092 : bytea **attoptions = RelationGetIndexAttOptions(irel, false);
965 1172092 : MemoryContext oldcxt = MemoryContextSwitchTo(irel->rd_indexcxt);
966 :
967 1172092 : set_fn_opclass_options(locinfo, attoptions[attnum - 1]);
968 :
969 1172092 : MemoryContextSwitchTo(oldcxt);
970 : }
971 : }
972 :
973 48230218 : return locinfo;
974 : }
975 :
976 : /* ----------------
977 : * index_store_float8_orderby_distances
978 : *
979 : * Convert AM distance function's results (that can be inexact)
980 : * to ORDER BY types and save them into xs_orderbyvals/xs_orderbynulls
981 : * for a possible recheck.
982 : * ----------------
983 : */
984 : void
985 364600 : index_store_float8_orderby_distances(IndexScanDesc scan, Oid *orderByTypes,
986 : IndexOrderByDistance *distances,
987 : bool recheckOrderBy)
988 : {
989 : int i;
990 :
991 : Assert(distances || !recheckOrderBy);
992 :
993 364600 : scan->xs_recheckorderby = recheckOrderBy;
994 :
995 729218 : for (i = 0; i < scan->numberOfOrderBys; i++)
996 : {
997 364618 : if (orderByTypes[i] == FLOAT8OID)
998 : {
999 364488 : if (distances && !distances[i].isnull)
1000 : {
1001 364428 : scan->xs_orderbyvals[i] = Float8GetDatum(distances[i].value);
1002 364428 : scan->xs_orderbynulls[i] = false;
1003 : }
1004 : else
1005 : {
1006 60 : scan->xs_orderbyvals[i] = (Datum) 0;
1007 60 : scan->xs_orderbynulls[i] = true;
1008 : }
1009 : }
1010 130 : else if (orderByTypes[i] == FLOAT4OID)
1011 : {
1012 : /* convert distance function's result to ORDER BY type */
1013 70 : if (distances && !distances[i].isnull)
1014 : {
1015 70 : scan->xs_orderbyvals[i] = Float4GetDatum((float4) distances[i].value);
1016 70 : scan->xs_orderbynulls[i] = false;
1017 : }
1018 : else
1019 : {
1020 0 : scan->xs_orderbyvals[i] = (Datum) 0;
1021 0 : scan->xs_orderbynulls[i] = true;
1022 : }
1023 : }
1024 : else
1025 : {
1026 : /*
1027 : * If the ordering operator's return value is anything else, we
1028 : * don't know how to convert the float8 bound calculated by the
1029 : * distance function to that. The executor won't actually need
1030 : * the order by values we return here, if there are no lossy
1031 : * results, so only insist on converting if the *recheck flag is
1032 : * set.
1033 : */
1034 60 : if (scan->xs_recheckorderby)
1035 0 : elog(ERROR, "ORDER BY operator must return float8 or float4 if the distance function is lossy");
1036 60 : scan->xs_orderbynulls[i] = true;
1037 : }
1038 : }
1039 364600 : }
1040 :
1041 : /* ----------------
1042 : * index_opclass_options
1043 : *
1044 : * Parse opclass-specific options for index column.
1045 : * ----------------
1046 : */
1047 : bytea *
1048 1172052 : index_opclass_options(Relation indrel, AttrNumber attnum, Datum attoptions,
1049 : bool validate)
1050 : {
1051 1172052 : int amoptsprocnum = indrel->rd_indam->amoptsprocnum;
1052 1172052 : Oid procid = InvalidOid;
1053 : FmgrInfo *procinfo;
1054 : local_relopts relopts;
1055 :
1056 : /* fetch options support procedure if specified */
1057 1172052 : if (amoptsprocnum != 0)
1058 1171992 : procid = index_getprocid(indrel, attnum, amoptsprocnum);
1059 :
1060 1172052 : if (!OidIsValid(procid))
1061 : {
1062 : Oid opclass;
1063 : Datum indclassDatum;
1064 : oidvector *indclass;
1065 :
1066 1169428 : if (!DatumGetPointer(attoptions))
1067 1169422 : return NULL; /* ok, no options, no procedure */
1068 :
1069 : /*
1070 : * Report an error if the opclass's options-parsing procedure does not
1071 : * exist but the opclass options are specified.
1072 : */
1073 6 : indclassDatum = SysCacheGetAttrNotNull(INDEXRELID, indrel->rd_indextuple,
1074 : Anum_pg_index_indclass);
1075 6 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1076 6 : opclass = indclass->values[attnum - 1];
1077 :
1078 6 : ereport(ERROR,
1079 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1080 : errmsg("operator class %s has no options",
1081 : generate_opclass_name(opclass))));
1082 : }
1083 :
1084 2624 : init_local_reloptions(&relopts, 0);
1085 :
1086 2624 : procinfo = index_getprocinfo(indrel, attnum, amoptsprocnum);
1087 :
1088 2624 : (void) FunctionCall1(procinfo, PointerGetDatum(&relopts));
1089 :
1090 2624 : return build_local_reloptions(&relopts, attoptions, validate);
1091 : }
|
