Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * genam.c
4 : * general index access method routines
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/access/index/genam.c
12 : *
13 : * NOTES
14 : * many of the old access method routines have been turned into
15 : * macros and moved to genam.h -cim 4/30/91
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 :
20 : #include "postgres.h"
21 :
22 : #include "access/genam.h"
23 : #include "access/heapam.h"
24 : #include "access/relscan.h"
25 : #include "access/tableam.h"
26 : #include "access/transam.h"
27 : #include "catalog/index.h"
28 : #include "lib/stringinfo.h"
29 : #include "miscadmin.h"
30 : #include "storage/bufmgr.h"
31 : #include "storage/procarray.h"
32 : #include "utils/acl.h"
33 : #include "utils/injection_point.h"
34 : #include "utils/lsyscache.h"
35 : #include "utils/rel.h"
36 : #include "utils/rls.h"
37 : #include "utils/ruleutils.h"
38 : #include "utils/snapmgr.h"
39 :
40 :
41 : /* ----------------------------------------------------------------
42 : * general access method routines
43 : *
44 : * All indexed access methods use an identical scan structure.
45 : * We don't know how the various AMs do locking, however, so we don't
46 : * do anything about that here.
47 : *
48 : * The intent is that an AM implementor will define a beginscan routine
49 : * that calls RelationGetIndexScan, to fill in the scan, and then does
50 : * whatever kind of locking he wants.
51 : *
52 : * At the end of a scan, the AM's endscan routine undoes the locking,
53 : * but does *not* call IndexScanEnd --- the higher-level index_endscan
54 : * routine does that. (We can't do it in the AM because index_endscan
55 : * still needs to touch the IndexScanDesc after calling the AM.)
56 : *
57 : * Because of this, the AM does not have a choice whether to call
58 : * RelationGetIndexScan or not; its beginscan routine must return an
59 : * object made by RelationGetIndexScan. This is kinda ugly but not
60 : * worth cleaning up now.
61 : * ----------------------------------------------------------------
62 : */
63 :
64 : /* ----------------
65 : * RelationGetIndexScan -- Create and fill an IndexScanDesc.
66 : *
67 : * This routine creates an index scan structure and sets up initial
68 : * contents for it.
69 : *
70 : * Parameters:
71 : * indexRelation -- index relation for scan.
72 : * nkeys -- count of scan keys (index qual conditions).
73 : * norderbys -- count of index order-by operators.
74 : *
75 : * Returns:
76 : * An initialized IndexScanDesc.
77 : * ----------------
78 : */
79 : IndexScanDesc
80 9859819 : RelationGetIndexScan(Relation indexRelation, int nkeys, int norderbys)
81 : {
82 : IndexScanDesc scan;
83 :
84 9859819 : scan = palloc_object(IndexScanDescData);
85 :
86 9859819 : scan->heapRelation = NULL; /* may be set later */
87 9859819 : scan->xs_heapfetch = NULL;
88 9859819 : scan->indexRelation = indexRelation;
89 9859819 : scan->xs_snapshot = InvalidSnapshot; /* caller must initialize this */
90 9859819 : scan->numberOfKeys = nkeys;
91 9859819 : scan->numberOfOrderBys = norderbys;
92 :
93 : /*
94 : * We allocate key workspace here, but it won't get filled until amrescan.
95 : */
96 9859819 : if (nkeys > 0)
97 9852051 : scan->keyData = palloc_array(ScanKeyData, nkeys);
98 : else
99 7768 : scan->keyData = NULL;
100 9859819 : if (norderbys > 0)
101 119 : scan->orderByData = palloc_array(ScanKeyData, norderbys);
102 : else
103 9859700 : scan->orderByData = NULL;
104 :
105 9859819 : scan->xs_want_itup = false; /* may be set later */
106 :
107 : /*
108 : * During recovery we ignore killed tuples and don't bother to kill them
109 : * either. We do this because the xmin on the primary node could easily be
110 : * later than the xmin on the standby node, so that what the primary
111 : * thinks is killed is supposed to be visible on standby. So for correct
112 : * MVCC for queries during recovery we must ignore these hints and check
113 : * all tuples. Do *not* set ignore_killed_tuples to true when running in a
114 : * transaction that was started during recovery. xactStartedInRecovery
115 : * should not be altered by index AMs.
116 : */
117 9859819 : scan->kill_prior_tuple = false;
118 9859819 : scan->xactStartedInRecovery = TransactionStartedDuringRecovery();
119 9859819 : scan->ignore_killed_tuples = !scan->xactStartedInRecovery;
120 :
121 9859819 : scan->opaque = NULL;
122 9859819 : scan->instrument = NULL;
123 :
124 9859819 : scan->xs_itup = NULL;
125 9859819 : scan->xs_itupdesc = NULL;
126 9859819 : scan->xs_hitup = NULL;
127 9859819 : scan->xs_hitupdesc = NULL;
128 :
129 9859819 : return scan;
130 : }
131 :
132 : /* ----------------
133 : * IndexScanEnd -- End an index scan.
134 : *
135 : * This routine just releases the storage acquired by
136 : * RelationGetIndexScan(). Any AM-level resources are
137 : * assumed to already have been released by the AM's
138 : * endscan routine.
139 : *
140 : * Returns:
141 : * None.
142 : * ----------------
143 : */
144 : void
145 9858497 : IndexScanEnd(IndexScanDesc scan)
146 : {
147 9858497 : if (scan->keyData != NULL)
148 9850752 : pfree(scan->keyData);
149 9858497 : if (scan->orderByData != NULL)
150 115 : pfree(scan->orderByData);
151 :
152 9858497 : pfree(scan);
153 9858497 : }
154 :
155 : /*
156 : * BuildIndexValueDescription
157 : *
158 : * Construct a string describing the contents of an index entry, in the
159 : * form "(key_name, ...)=(key_value, ...)". This is currently used
160 : * for building unique-constraint, exclusion-constraint error messages, and
161 : * logical replication conflict error messages so only key columns of the index
162 : * are checked and printed.
163 : *
164 : * Note that if the user does not have permissions to view all of the
165 : * columns involved then a NULL is returned. Returning a partial key seems
166 : * unlikely to be useful and we have no way to know which of the columns the
167 : * user provided (unlike in ExecBuildSlotValueDescription).
168 : *
169 : * The passed-in values/nulls arrays are the "raw" input to the index AM,
170 : * e.g. results of FormIndexDatum --- this is not necessarily what is stored
171 : * in the index, but it's what the user perceives to be stored.
172 : *
173 : * Note: if you change anything here, check whether
174 : * ExecBuildSlotPartitionKeyDescription() in execMain.c needs a similar
175 : * change.
176 : */
177 : char *
178 725 : BuildIndexValueDescription(Relation indexRelation,
179 : const Datum *values, const bool *isnull)
180 : {
181 : StringInfoData buf;
182 : Form_pg_index idxrec;
183 : int indnkeyatts;
184 : int i;
185 : int keyno;
186 725 : Oid indexrelid = RelationGetRelid(indexRelation);
187 : Oid indrelid;
188 : AclResult aclresult;
189 :
190 725 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(indexRelation);
191 :
192 : /*
193 : * Check permissions- if the user does not have access to view all of the
194 : * key columns then return NULL to avoid leaking data.
195 : *
196 : * First check if RLS is enabled for the relation. If so, return NULL to
197 : * avoid leaking data.
198 : *
199 : * Next we need to check table-level SELECT access and then, if there is
200 : * no access there, check column-level permissions.
201 : */
202 725 : idxrec = indexRelation->rd_index;
203 725 : indrelid = idxrec->indrelid;
204 : Assert(indexrelid == idxrec->indexrelid);
205 :
206 : /* RLS check- if RLS is enabled then we don't return anything. */
207 725 : if (check_enable_rls(indrelid, InvalidOid, true) == RLS_ENABLED)
208 8 : return NULL;
209 :
210 : /* Table-level SELECT is enough, if the user has it */
211 717 : aclresult = pg_class_aclcheck(indrelid, GetUserId(), ACL_SELECT);
212 717 : if (aclresult != ACLCHECK_OK)
213 : {
214 : /*
215 : * No table-level access, so step through the columns in the index and
216 : * make sure the user has SELECT rights on all of them.
217 : */
218 16 : for (keyno = 0; keyno < indnkeyatts; keyno++)
219 : {
220 16 : AttrNumber attnum = idxrec->indkey.values[keyno];
221 :
222 : /*
223 : * Note that if attnum == InvalidAttrNumber, then this is an index
224 : * based on an expression and we return no detail rather than try
225 : * to figure out what column(s) the expression includes and if the
226 : * user has SELECT rights on them.
227 : */
228 32 : if (attnum == InvalidAttrNumber ||
229 16 : pg_attribute_aclcheck(indrelid, attnum, GetUserId(),
230 : ACL_SELECT) != ACLCHECK_OK)
231 : {
232 : /* No access, so clean up and return */
233 8 : return NULL;
234 : }
235 : }
236 : }
237 :
238 709 : initStringInfo(&buf);
239 709 : appendStringInfo(&buf, "(%s)=(",
240 : pg_get_indexdef_columns(indexrelid, true));
241 :
242 1674 : for (i = 0; i < indnkeyatts; i++)
243 : {
244 : char *val;
245 :
246 965 : if (isnull[i])
247 12 : val = "null";
248 : else
249 : {
250 : Oid foutoid;
251 : bool typisvarlena;
252 :
253 : /*
254 : * The provided data is not necessarily of the type stored in the
255 : * index; rather it is of the index opclass's input type. So look
256 : * at rd_opcintype not the index tupdesc.
257 : *
258 : * Note: this is a bit shaky for opclasses that have pseudotype
259 : * input types such as ANYARRAY or RECORD. Currently, the
260 : * typoutput functions associated with the pseudotypes will work
261 : * okay, but we might have to try harder in future.
262 : */
263 953 : getTypeOutputInfo(indexRelation->rd_opcintype[i],
264 : &foutoid, &typisvarlena);
265 953 : val = OidOutputFunctionCall(foutoid, values[i]);
266 : }
267 :
268 965 : if (i > 0)
269 256 : appendStringInfoString(&buf, ", ");
270 965 : appendStringInfoString(&buf, val);
271 : }
272 :
273 709 : appendStringInfoChar(&buf, ')');
274 :
275 709 : return buf.data;
276 : }
277 :
278 : /*
279 : * Get the snapshotConflictHorizon from the table entries pointed to by the
280 : * index tuples being deleted using an AM-generic approach.
281 : *
282 : * This is a table_index_delete_tuples() shim used by index AMs that only need
283 : * to consult the tableam to get a snapshotConflictHorizon value, and only
284 : * expect to delete index tuples that are already known deletable (typically
285 : * due to having LP_DEAD bits set). When a snapshotConflictHorizon value
286 : * isn't needed in index AM's deletion WAL record, it is safe for it to skip
287 : * calling here entirely.
288 : *
289 : * We assume that caller index AM uses the standard IndexTuple representation,
290 : * with table TIDs stored in the t_tid field. We also expect (and assert)
291 : * that the line pointers on page for 'itemnos' offsets are already marked
292 : * LP_DEAD.
293 : */
294 : TransactionId
295 4 : index_compute_xid_horizon_for_tuples(Relation irel,
296 : Relation hrel,
297 : Buffer ibuf,
298 : OffsetNumber *itemnos,
299 : int nitems)
300 : {
301 : TM_IndexDeleteOp delstate;
302 4 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
303 4 : Page ipage = BufferGetPage(ibuf);
304 : IndexTuple itup;
305 :
306 : Assert(nitems > 0);
307 :
308 4 : delstate.irel = irel;
309 4 : delstate.iblknum = BufferGetBlockNumber(ibuf);
310 4 : delstate.bottomup = false;
311 4 : delstate.bottomupfreespace = 0;
312 4 : delstate.ndeltids = 0;
313 4 : delstate.deltids = palloc_array(TM_IndexDelete, nitems);
314 4 : delstate.status = palloc_array(TM_IndexStatus, nitems);
315 :
316 : /* identify what the index tuples about to be deleted point to */
317 888 : for (int i = 0; i < nitems; i++)
318 : {
319 884 : OffsetNumber offnum = itemnos[i];
320 : ItemId iitemid;
321 :
322 884 : iitemid = PageGetItemId(ipage, offnum);
323 884 : itup = (IndexTuple) PageGetItem(ipage, iitemid);
324 :
325 : Assert(ItemIdIsDead(iitemid));
326 :
327 884 : ItemPointerCopy(&itup->t_tid, &delstate.deltids[i].tid);
328 884 : delstate.deltids[i].id = delstate.ndeltids;
329 884 : delstate.status[i].idxoffnum = offnum;
330 884 : delstate.status[i].knowndeletable = true; /* LP_DEAD-marked */
331 884 : delstate.status[i].promising = false; /* unused */
332 884 : delstate.status[i].freespace = 0; /* unused */
333 :
334 884 : delstate.ndeltids++;
335 : }
336 :
337 : /* determine the actual xid horizon */
338 4 : snapshotConflictHorizon = table_index_delete_tuples(hrel, &delstate);
339 :
340 : /* assert tableam agrees that all items are deletable */
341 : Assert(delstate.ndeltids == nitems);
342 :
343 4 : pfree(delstate.deltids);
344 4 : pfree(delstate.status);
345 :
346 4 : return snapshotConflictHorizon;
347 : }
348 :
349 :
350 : /* ----------------------------------------------------------------
351 : * heap-or-index-scan access to system catalogs
352 : *
353 : * These functions support system catalog accesses that normally use
354 : * an index but need to be capable of being switched to heap scans
355 : * if the system indexes are unavailable.
356 : *
357 : * The specified scan keys must be compatible with the named index.
358 : * Generally this means that they must constrain either all columns
359 : * of the index, or the first K columns of an N-column index.
360 : *
361 : * These routines could work with non-system tables, actually,
362 : * but they're only useful when there is a known index to use with
363 : * the given scan keys; so in practice they're only good for
364 : * predetermined types of scans of system catalogs.
365 : * ----------------------------------------------------------------
366 : */
367 :
368 : /*
369 : * systable_beginscan --- set up for heap-or-index scan
370 : *
371 : * rel: catalog to scan, already opened and suitably locked
372 : * indexId: OID of index to conditionally use
373 : * indexOK: if false, forces a heap scan (see notes below)
374 : * snapshot: time qual to use (NULL for a recent catalog snapshot)
375 : * nkeys, key: scan keys
376 : *
377 : * The attribute numbers in the scan key should be set for the heap case.
378 : * If we choose to index, we convert them to 1..n to reference the index
379 : * columns. Note this means there must be one scankey qualification per
380 : * index column! This is checked by the Asserts in the normal, index-using
381 : * case, but won't be checked if the heapscan path is taken.
382 : *
383 : * The routine checks the normal cases for whether an indexscan is safe,
384 : * but caller can make additional checks and pass indexOK=false if needed.
385 : * In standard case indexOK can simply be constant TRUE.
386 : */
387 : SysScanDesc
388 9783139 : systable_beginscan(Relation heapRelation,
389 : Oid indexId,
390 : bool indexOK,
391 : Snapshot snapshot,
392 : int nkeys, ScanKey key)
393 : {
394 : SysScanDesc sysscan;
395 : Relation irel;
396 :
397 : /*
398 : * If this backend promised that it won't access shared catalogs during
399 : * logical decoding, this it the right place to verify.
400 : */
401 : Assert(!HistoricSnapshotActive() ||
402 : accessSharedCatalogsInDecoding ||
403 : !heapRelation->rd_rel->relisshared);
404 :
405 9783139 : if (indexOK &&
406 9633640 : !IgnoreSystemIndexes &&
407 9558343 : !ReindexIsProcessingIndex(indexId))
408 9549743 : irel = index_open(indexId, AccessShareLock);
409 : else
410 233396 : irel = NULL;
411 :
412 9783135 : sysscan = palloc_object(SysScanDescData);
413 :
414 9783135 : sysscan->heap_rel = heapRelation;
415 9783135 : sysscan->irel = irel;
416 9783135 : sysscan->slot = table_slot_create(heapRelation, NULL);
417 :
418 9783135 : if (snapshot == NULL)
419 : {
420 9133173 : Oid relid = RelationGetRelid(heapRelation);
421 :
422 9133173 : snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
423 9133173 : sysscan->snapshot = snapshot;
424 : }
425 : else
426 : {
427 : /* Caller is responsible for any snapshot. */
428 649962 : sysscan->snapshot = NULL;
429 : }
430 :
431 : /*
432 : * If CheckXidAlive is set then set a flag to indicate that system table
433 : * scan is in-progress. See detailed comments in xact.c where these
434 : * variables are declared.
435 : */
436 9783135 : if (TransactionIdIsValid(CheckXidAlive))
437 1044 : bsysscan = true;
438 :
439 9783135 : if (irel)
440 : {
441 : int i;
442 : ScanKey idxkey;
443 :
444 9549739 : idxkey = palloc_array(ScanKeyData, nkeys);
445 :
446 : /* Convert attribute numbers to be index column numbers. */
447 25145293 : for (i = 0; i < nkeys; i++)
448 : {
449 : int j;
450 :
451 15595554 : memcpy(&idxkey[i], &key[i], sizeof(ScanKeyData));
452 :
453 22691562 : for (j = 0; j < IndexRelationGetNumberOfAttributes(irel); j++)
454 : {
455 22691562 : if (key[i].sk_attno == irel->rd_index->indkey.values[j])
456 : {
457 15595554 : idxkey[i].sk_attno = j + 1;
458 15595554 : break;
459 : }
460 : }
461 15595554 : if (j == IndexRelationGetNumberOfAttributes(irel))
462 0 : elog(ERROR, "column is not in index");
463 : }
464 :
465 9549739 : sysscan->iscan = index_beginscan(heapRelation, irel,
466 : snapshot, NULL, nkeys, 0,
467 : SO_NONE);
468 9549739 : index_rescan(sysscan->iscan, idxkey, nkeys, NULL, 0);
469 9549739 : sysscan->scan = NULL;
470 :
471 9549739 : pfree(idxkey);
472 : }
473 : else
474 : {
475 : /*
476 : * We disallow synchronized scans when forced to use a heapscan on a
477 : * catalog. In most cases the desired rows are near the front, so
478 : * that the unpredictable start point of a syncscan is a serious
479 : * disadvantage; and there are no compensating advantages, because
480 : * it's unlikely that such scans will occur in parallel.
481 : */
482 233396 : sysscan->scan = table_beginscan_strat(heapRelation, snapshot,
483 : nkeys, key,
484 : true, false);
485 233396 : sysscan->iscan = NULL;
486 : }
487 :
488 9783135 : return sysscan;
489 : }
490 :
491 : /*
492 : * HandleConcurrentAbort - Handle concurrent abort of the CheckXidAlive.
493 : *
494 : * Error out, if CheckXidAlive is aborted. We can't directly use
495 : * TransactionIdDidAbort as after crash such transaction might not have been
496 : * marked as aborted. See detailed comments in xact.c where the variable
497 : * is declared.
498 : */
499 : static inline void
500 20446447 : HandleConcurrentAbort(void)
501 : {
502 20446447 : if (TransactionIdIsValid(CheckXidAlive) &&
503 1699 : !TransactionIdIsInProgress(CheckXidAlive) &&
504 8 : !TransactionIdDidCommit(CheckXidAlive))
505 8 : ereport(ERROR,
506 : (errcode(ERRCODE_TRANSACTION_ROLLBACK),
507 : errmsg("transaction aborted during system catalog scan")));
508 20446439 : }
509 :
510 : /*
511 : * systable_getnext --- get next tuple in a heap-or-index scan
512 : *
513 : * Returns NULL if no more tuples available.
514 : *
515 : * Note that returned tuple is a reference to data in a disk buffer;
516 : * it must not be modified, and should be presumed inaccessible after
517 : * next getnext() or endscan() call.
518 : *
519 : * XXX: It'd probably make sense to offer a slot based interface, at least
520 : * optionally.
521 : */
522 : HeapTuple
523 20143488 : systable_getnext(SysScanDesc sysscan)
524 : {
525 20143488 : HeapTuple htup = NULL;
526 :
527 20143488 : if (sysscan->irel)
528 : {
529 17886084 : if (index_getnext_slot(sysscan->iscan, ForwardScanDirection, sysscan->slot))
530 : {
531 : bool shouldFree;
532 :
533 13878841 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
534 : Assert(!shouldFree);
535 :
536 : /*
537 : * We currently don't need to support lossy index operators for
538 : * any system catalog scan. It could be done here, using the scan
539 : * keys to drive the operator calls, if we arranged to save the
540 : * heap attnums during systable_beginscan(); this is practical
541 : * because we still wouldn't need to support indexes on
542 : * expressions.
543 : */
544 13878841 : if (sysscan->iscan->xs_recheck)
545 0 : elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
546 : }
547 : }
548 : else
549 : {
550 2257404 : if (table_scan_getnextslot(sysscan->scan, ForwardScanDirection, sysscan->slot))
551 : {
552 : bool shouldFree;
553 :
554 2201318 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
555 : Assert(!shouldFree);
556 : }
557 : }
558 :
559 : /*
560 : * Handle the concurrent abort while fetching the catalog tuple during
561 : * logical streaming of a transaction.
562 : */
563 20143485 : HandleConcurrentAbort();
564 :
565 20143477 : return htup;
566 : }
567 :
568 : /*
569 : * systable_recheck_tuple --- recheck visibility of most-recently-fetched tuple
570 : *
571 : * In particular, determine if this tuple would be visible to a catalog scan
572 : * that started now. We don't handle the case of a non-MVCC scan snapshot,
573 : * because no caller needs that yet.
574 : *
575 : * This is useful to test whether an object was deleted while we waited to
576 : * acquire lock on it.
577 : *
578 : * Note: we don't actually *need* the tuple to be passed in, but it's a
579 : * good crosscheck that the caller is interested in the right tuple.
580 : */
581 : bool
582 162185 : systable_recheck_tuple(SysScanDesc sysscan, HeapTuple tup)
583 : {
584 : Snapshot freshsnap;
585 : bool result;
586 :
587 : Assert(tup == ExecFetchSlotHeapTuple(sysscan->slot, false, NULL));
588 :
589 162185 : freshsnap = GetCatalogSnapshot(RelationGetRelid(sysscan->heap_rel));
590 162185 : freshsnap = RegisterSnapshot(freshsnap);
591 :
592 162185 : result = table_tuple_satisfies_snapshot(sysscan->heap_rel,
593 162185 : sysscan->slot,
594 : freshsnap);
595 162185 : UnregisterSnapshot(freshsnap);
596 :
597 : /*
598 : * Handle the concurrent abort while fetching the catalog tuple during
599 : * logical streaming of a transaction.
600 : */
601 162185 : HandleConcurrentAbort();
602 :
603 162185 : return result;
604 : }
605 :
606 : /*
607 : * systable_endscan --- close scan, release resources
608 : *
609 : * Note that it's still up to the caller to close the heap relation.
610 : */
611 : void
612 9782587 : systable_endscan(SysScanDesc sysscan)
613 : {
614 9782587 : if (sysscan->slot)
615 : {
616 9782587 : ExecDropSingleTupleTableSlot(sysscan->slot);
617 9782587 : sysscan->slot = NULL;
618 : }
619 :
620 9782587 : if (sysscan->irel)
621 : {
622 9549201 : index_endscan(sysscan->iscan);
623 9549201 : index_close(sysscan->irel, AccessShareLock);
624 : }
625 : else
626 233386 : table_endscan(sysscan->scan);
627 :
628 9782587 : if (sysscan->snapshot)
629 9132625 : UnregisterSnapshot(sysscan->snapshot);
630 :
631 : /*
632 : * Reset the bsysscan flag at the end of the systable scan. See detailed
633 : * comments in xact.c where these variables are declared.
634 : */
635 9782587 : if (TransactionIdIsValid(CheckXidAlive))
636 1036 : bsysscan = false;
637 :
638 9782587 : pfree(sysscan);
639 9782587 : }
640 :
641 :
642 : /*
643 : * systable_beginscan_ordered --- set up for ordered catalog scan
644 : *
645 : * These routines have essentially the same API as systable_beginscan etc,
646 : * except that they guarantee to return multiple matching tuples in
647 : * index order. Also, for largely historical reasons, the index to use
648 : * is opened and locked by the caller, not here.
649 : *
650 : * Currently we do not support non-index-based scans here. (In principle
651 : * we could do a heapscan and sort, but the uses are in places that
652 : * probably don't need to still work with corrupted catalog indexes.)
653 : * For the moment, therefore, these functions are merely the thinest of
654 : * wrappers around index_beginscan/index_getnext_slot. The main reason for
655 : * their existence is to centralize possible future support of lossy operators
656 : * in catalog scans.
657 : */
658 : SysScanDesc
659 37636 : systable_beginscan_ordered(Relation heapRelation,
660 : Relation indexRelation,
661 : Snapshot snapshot,
662 : int nkeys, ScanKey key)
663 : {
664 : SysScanDesc sysscan;
665 : int i;
666 : ScanKey idxkey;
667 :
668 : /* REINDEX can probably be a hard error here ... */
669 37636 : if (ReindexIsProcessingIndex(RelationGetRelid(indexRelation)))
670 0 : ereport(ERROR,
671 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
672 : errmsg("cannot access index \"%s\" while it is being reindexed",
673 : RelationGetRelationName(indexRelation))));
674 : /* ... but we only throw a warning about violating IgnoreSystemIndexes */
675 37636 : if (IgnoreSystemIndexes)
676 0 : elog(WARNING, "using index \"%s\" despite IgnoreSystemIndexes",
677 : RelationGetRelationName(indexRelation));
678 :
679 37636 : sysscan = palloc_object(SysScanDescData);
680 :
681 37636 : sysscan->heap_rel = heapRelation;
682 37636 : sysscan->irel = indexRelation;
683 37636 : sysscan->slot = table_slot_create(heapRelation, NULL);
684 :
685 37636 : if (snapshot == NULL)
686 : {
687 5289 : Oid relid = RelationGetRelid(heapRelation);
688 :
689 5289 : snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
690 5289 : sysscan->snapshot = snapshot;
691 : }
692 : else
693 : {
694 : /* Caller is responsible for any snapshot. */
695 32347 : sysscan->snapshot = NULL;
696 : }
697 :
698 37636 : idxkey = palloc_array(ScanKeyData, nkeys);
699 :
700 : /* Convert attribute numbers to be index column numbers. */
701 73294 : for (i = 0; i < nkeys; i++)
702 : {
703 : int j;
704 :
705 35658 : memcpy(&idxkey[i], &key[i], sizeof(ScanKeyData));
706 :
707 37699 : for (j = 0; j < IndexRelationGetNumberOfAttributes(indexRelation); j++)
708 : {
709 37699 : if (key[i].sk_attno == indexRelation->rd_index->indkey.values[j])
710 : {
711 35658 : idxkey[i].sk_attno = j + 1;
712 35658 : break;
713 : }
714 : }
715 35658 : if (j == IndexRelationGetNumberOfAttributes(indexRelation))
716 0 : elog(ERROR, "column is not in index");
717 : }
718 :
719 : /*
720 : * If CheckXidAlive is set then set a flag to indicate that system table
721 : * scan is in-progress. See detailed comments in xact.c where these
722 : * variables are declared.
723 : */
724 37636 : if (TransactionIdIsValid(CheckXidAlive))
725 0 : bsysscan = true;
726 :
727 37636 : sysscan->iscan = index_beginscan(heapRelation, indexRelation,
728 : snapshot, NULL, nkeys, 0,
729 : SO_NONE);
730 37636 : index_rescan(sysscan->iscan, idxkey, nkeys, NULL, 0);
731 37636 : sysscan->scan = NULL;
732 :
733 37636 : pfree(idxkey);
734 :
735 37636 : return sysscan;
736 : }
737 :
738 : /*
739 : * systable_getnext_ordered --- get next tuple in an ordered catalog scan
740 : */
741 : HeapTuple
742 140780 : systable_getnext_ordered(SysScanDesc sysscan, ScanDirection direction)
743 : {
744 140780 : HeapTuple htup = NULL;
745 :
746 : Assert(sysscan->irel);
747 140780 : if (index_getnext_slot(sysscan->iscan, direction, sysscan->slot))
748 103996 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, NULL);
749 :
750 : /* See notes in systable_getnext */
751 140777 : if (htup && sysscan->iscan->xs_recheck)
752 0 : elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
753 :
754 : /*
755 : * Handle the concurrent abort while fetching the catalog tuple during
756 : * logical streaming of a transaction.
757 : */
758 140777 : HandleConcurrentAbort();
759 :
760 140777 : return htup;
761 : }
762 :
763 : /*
764 : * systable_endscan_ordered --- close scan, release resources
765 : */
766 : void
767 37625 : systable_endscan_ordered(SysScanDesc sysscan)
768 : {
769 37625 : if (sysscan->slot)
770 : {
771 37625 : ExecDropSingleTupleTableSlot(sysscan->slot);
772 37625 : sysscan->slot = NULL;
773 : }
774 :
775 : Assert(sysscan->irel);
776 37625 : index_endscan(sysscan->iscan);
777 37625 : if (sysscan->snapshot)
778 5281 : UnregisterSnapshot(sysscan->snapshot);
779 :
780 : /*
781 : * Reset the bsysscan flag at the end of the systable scan. See detailed
782 : * comments in xact.c where these variables are declared.
783 : */
784 37625 : if (TransactionIdIsValid(CheckXidAlive))
785 0 : bsysscan = false;
786 :
787 37625 : pfree(sysscan);
788 37625 : }
789 :
790 : /*
791 : * systable_inplace_update_begin --- update a row "in place" (overwrite it)
792 : *
793 : * Overwriting violates both MVCC and transactional safety, so the uses of
794 : * this function in Postgres are extremely limited. This makes no effort to
795 : * support updating cache key columns or other indexed columns. Nonetheless
796 : * we find some places to use it. See README.tuplock section "Locking to
797 : * write inplace-updated tables" and later sections for expectations of
798 : * readers and writers of a table that gets inplace updates. Standard flow:
799 : *
800 : * ... [any slow preparation not requiring oldtup] ...
801 : * systable_inplace_update_begin([...], &tup, &inplace_state);
802 : * if (!HeapTupleIsValid(tup))
803 : * elog(ERROR, [...]);
804 : * ... [buffer is exclusive-locked; mutate "tup"] ...
805 : * if (dirty)
806 : * systable_inplace_update_finish(inplace_state, tup);
807 : * else
808 : * systable_inplace_update_cancel(inplace_state);
809 : *
810 : * The first several params duplicate the systable_beginscan() param list.
811 : * "oldtupcopy" is an output parameter, assigned NULL if the key ceases to
812 : * find a live tuple. (In PROC_IN_VACUUM, that is a low-probability transient
813 : * condition.) If "oldtupcopy" gets non-NULL, you must pass output parameter
814 : * "state" to systable_inplace_update_finish() or
815 : * systable_inplace_update_cancel().
816 : */
817 : void
818 217417 : systable_inplace_update_begin(Relation relation,
819 : Oid indexId,
820 : bool indexOK,
821 : Snapshot snapshot,
822 : int nkeys, const ScanKeyData *key,
823 : HeapTuple *oldtupcopy,
824 : void **state)
825 : {
826 217417 : int retries = 0;
827 : SysScanDesc scan;
828 : HeapTuple oldtup;
829 : BufferHeapTupleTableSlot *bslot;
830 :
831 : /*
832 : * For now, we don't allow parallel updates. Unlike a regular update,
833 : * this should never create a combo CID, so it might be possible to relax
834 : * this restriction, but not without more thought and testing. It's not
835 : * clear that it would be useful, anyway.
836 : */
837 217417 : if (IsInParallelMode())
838 0 : ereport(ERROR,
839 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
840 : errmsg("cannot update tuples during a parallel operation")));
841 :
842 : /*
843 : * Accept a snapshot argument, for symmetry, but this function advances
844 : * its snapshot as needed to reach the tail of the updated tuple chain.
845 : */
846 : Assert(snapshot == NULL);
847 :
848 : Assert(IsInplaceUpdateRelation(relation) || !IsSystemRelation(relation));
849 :
850 : /* Loop for an exclusive-locked buffer of a non-updated tuple. */
851 : do
852 : {
853 : TupleTableSlot *slot;
854 :
855 217441 : CHECK_FOR_INTERRUPTS();
856 :
857 : /*
858 : * Processes issuing heap_update (e.g. GRANT) at maximum speed could
859 : * drive us to this error. A hostile table owner has stronger ways to
860 : * damage their own table, so that's minor.
861 : */
862 217441 : if (retries++ > 10000)
863 0 : elog(ERROR, "giving up after too many tries to overwrite row");
864 :
865 217441 : INJECTION_POINT("inplace-before-pin", NULL);
866 217441 : scan = systable_beginscan(relation, indexId, indexOK, snapshot,
867 : nkeys, unconstify(ScanKeyData *, key));
868 217441 : oldtup = systable_getnext(scan);
869 217441 : if (!HeapTupleIsValid(oldtup))
870 : {
871 0 : systable_endscan(scan);
872 0 : *oldtupcopy = NULL;
873 0 : return;
874 : }
875 :
876 217441 : slot = scan->slot;
877 : Assert(TTS_IS_BUFFERTUPLE(slot));
878 217441 : bslot = (BufferHeapTupleTableSlot *) slot;
879 217441 : } while (!heap_inplace_lock(scan->heap_rel,
880 : bslot->base.tuple, bslot->buffer,
881 : (void (*) (void *)) systable_endscan, scan));
882 :
883 217417 : *oldtupcopy = heap_copytuple(oldtup);
884 217417 : *state = scan;
885 : }
886 :
887 : /*
888 : * systable_inplace_update_finish --- second phase of inplace update
889 : *
890 : * The tuple cannot change size, and therefore its header fields and null
891 : * bitmap (if any) don't change either.
892 : */
893 : void
894 100019 : systable_inplace_update_finish(void *state, HeapTuple tuple)
895 : {
896 100019 : SysScanDesc scan = (SysScanDesc) state;
897 100019 : Relation relation = scan->heap_rel;
898 100019 : TupleTableSlot *slot = scan->slot;
899 100019 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
900 100019 : HeapTuple oldtup = bslot->base.tuple;
901 100019 : Buffer buffer = bslot->buffer;
902 :
903 100019 : heap_inplace_update_and_unlock(relation, oldtup, tuple, buffer);
904 100019 : systable_endscan(scan);
905 100019 : }
906 :
907 : /*
908 : * systable_inplace_update_cancel --- abandon inplace update
909 : *
910 : * This is an alternative to making a no-op update.
911 : */
912 : void
913 117398 : systable_inplace_update_cancel(void *state)
914 : {
915 117398 : SysScanDesc scan = (SysScanDesc) state;
916 117398 : Relation relation = scan->heap_rel;
917 117398 : TupleTableSlot *slot = scan->slot;
918 117398 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
919 117398 : HeapTuple oldtup = bslot->base.tuple;
920 117398 : Buffer buffer = bslot->buffer;
921 :
922 117398 : heap_inplace_unlock(relation, oldtup, buffer);
923 117398 : systable_endscan(scan);
924 117398 : }
|
