Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * genam.c
4 : * general index access method routines
5 : *
6 : * Portions Copyright (c) 1996-2024, 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 12662750 : RelationGetIndexScan(Relation indexRelation, int nkeys, int norderbys)
81 : {
82 : IndexScanDesc scan;
83 :
84 12662750 : scan = (IndexScanDesc) palloc(sizeof(IndexScanDescData));
85 :
86 12662750 : scan->heapRelation = NULL; /* may be set later */
87 12662750 : scan->xs_heapfetch = NULL;
88 12662750 : scan->indexRelation = indexRelation;
89 12662750 : scan->xs_snapshot = InvalidSnapshot; /* caller must initialize this */
90 12662750 : scan->numberOfKeys = nkeys;
91 12662750 : scan->numberOfOrderBys = norderbys;
92 :
93 : /*
94 : * We allocate key workspace here, but it won't get filled until amrescan.
95 : */
96 12662750 : if (nkeys > 0)
97 12650744 : scan->keyData = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
98 : else
99 12006 : scan->keyData = NULL;
100 12662750 : if (norderbys > 0)
101 192 : scan->orderByData = (ScanKey) palloc(sizeof(ScanKeyData) * norderbys);
102 : else
103 12662558 : scan->orderByData = NULL;
104 :
105 12662750 : 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 12662750 : scan->kill_prior_tuple = false;
118 12662750 : scan->xactStartedInRecovery = TransactionStartedDuringRecovery();
119 12662750 : scan->ignore_killed_tuples = !scan->xactStartedInRecovery;
120 :
121 12662750 : scan->opaque = NULL;
122 :
123 12662750 : scan->xs_itup = NULL;
124 12662750 : scan->xs_itupdesc = NULL;
125 12662750 : scan->xs_hitup = NULL;
126 12662750 : scan->xs_hitupdesc = NULL;
127 :
128 12662750 : return scan;
129 : }
130 :
131 : /* ----------------
132 : * IndexScanEnd -- End an index scan.
133 : *
134 : * This routine just releases the storage acquired by
135 : * RelationGetIndexScan(). Any AM-level resources are
136 : * assumed to already have been released by the AM's
137 : * endscan routine.
138 : *
139 : * Returns:
140 : * None.
141 : * ----------------
142 : */
143 : void
144 12661028 : IndexScanEnd(IndexScanDesc scan)
145 : {
146 12661028 : if (scan->keyData != NULL)
147 12649070 : pfree(scan->keyData);
148 12661028 : if (scan->orderByData != NULL)
149 186 : pfree(scan->orderByData);
150 :
151 12661028 : pfree(scan);
152 12661028 : }
153 :
154 : /*
155 : * BuildIndexValueDescription
156 : *
157 : * Construct a string describing the contents of an index entry, in the
158 : * form "(key_name, ...)=(key_value, ...)". This is currently used
159 : * for building unique-constraint, exclusion-constraint error messages, and
160 : * logical replication conflict error messages so only key columns of the index
161 : * are checked and printed.
162 : *
163 : * Note that if the user does not have permissions to view all of the
164 : * columns involved then a NULL is returned. Returning a partial key seems
165 : * unlikely to be useful and we have no way to know which of the columns the
166 : * user provided (unlike in ExecBuildSlotValueDescription).
167 : *
168 : * The passed-in values/nulls arrays are the "raw" input to the index AM,
169 : * e.g. results of FormIndexDatum --- this is not necessarily what is stored
170 : * in the index, but it's what the user perceives to be stored.
171 : *
172 : * Note: if you change anything here, check whether
173 : * ExecBuildSlotPartitionKeyDescription() in execMain.c needs a similar
174 : * change.
175 : */
176 : char *
177 958 : BuildIndexValueDescription(Relation indexRelation,
178 : const Datum *values, const bool *isnull)
179 : {
180 : StringInfoData buf;
181 : Form_pg_index idxrec;
182 : int indnkeyatts;
183 : int i;
184 : int keyno;
185 958 : Oid indexrelid = RelationGetRelid(indexRelation);
186 : Oid indrelid;
187 : AclResult aclresult;
188 :
189 958 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(indexRelation);
190 :
191 : /*
192 : * Check permissions- if the user does not have access to view all of the
193 : * key columns then return NULL to avoid leaking data.
194 : *
195 : * First check if RLS is enabled for the relation. If so, return NULL to
196 : * avoid leaking data.
197 : *
198 : * Next we need to check table-level SELECT access and then, if there is
199 : * no access there, check column-level permissions.
200 : */
201 958 : idxrec = indexRelation->rd_index;
202 958 : indrelid = idxrec->indrelid;
203 : Assert(indexrelid == idxrec->indexrelid);
204 :
205 : /* RLS check- if RLS is enabled then we don't return anything. */
206 958 : if (check_enable_rls(indrelid, InvalidOid, true) == RLS_ENABLED)
207 12 : return NULL;
208 :
209 : /* Table-level SELECT is enough, if the user has it */
210 946 : aclresult = pg_class_aclcheck(indrelid, GetUserId(), ACL_SELECT);
211 946 : if (aclresult != ACLCHECK_OK)
212 : {
213 : /*
214 : * No table-level access, so step through the columns in the index and
215 : * make sure the user has SELECT rights on all of them.
216 : */
217 24 : for (keyno = 0; keyno < indnkeyatts; keyno++)
218 : {
219 24 : AttrNumber attnum = idxrec->indkey.values[keyno];
220 :
221 : /*
222 : * Note that if attnum == InvalidAttrNumber, then this is an index
223 : * based on an expression and we return no detail rather than try
224 : * to figure out what column(s) the expression includes and if the
225 : * user has SELECT rights on them.
226 : */
227 48 : if (attnum == InvalidAttrNumber ||
228 24 : pg_attribute_aclcheck(indrelid, attnum, GetUserId(),
229 : ACL_SELECT) != ACLCHECK_OK)
230 : {
231 : /* No access, so clean up and return */
232 12 : return NULL;
233 : }
234 : }
235 : }
236 :
237 934 : initStringInfo(&buf);
238 934 : appendStringInfo(&buf, "(%s)=(",
239 : pg_get_indexdef_columns(indexrelid, true));
240 :
241 2178 : for (i = 0; i < indnkeyatts; i++)
242 : {
243 : char *val;
244 :
245 1244 : if (isnull[i])
246 18 : val = "null";
247 : else
248 : {
249 : Oid foutoid;
250 : bool typisvarlena;
251 :
252 : /*
253 : * The provided data is not necessarily of the type stored in the
254 : * index; rather it is of the index opclass's input type. So look
255 : * at rd_opcintype not the index tupdesc.
256 : *
257 : * Note: this is a bit shaky for opclasses that have pseudotype
258 : * input types such as ANYARRAY or RECORD. Currently, the
259 : * typoutput functions associated with the pseudotypes will work
260 : * okay, but we might have to try harder in future.
261 : */
262 1226 : getTypeOutputInfo(indexRelation->rd_opcintype[i],
263 : &foutoid, &typisvarlena);
264 1226 : val = OidOutputFunctionCall(foutoid, values[i]);
265 : }
266 :
267 1244 : if (i > 0)
268 310 : appendStringInfoString(&buf, ", ");
269 1244 : appendStringInfoString(&buf, val);
270 : }
271 :
272 934 : appendStringInfoChar(&buf, ')');
273 :
274 934 : return buf.data;
275 : }
276 :
277 : /*
278 : * Get the snapshotConflictHorizon from the table entries pointed to by the
279 : * index tuples being deleted using an AM-generic approach.
280 : *
281 : * This is a table_index_delete_tuples() shim used by index AMs that only need
282 : * to consult the tableam to get a snapshotConflictHorizon value, and only
283 : * expect to delete index tuples that are already known deletable (typically
284 : * due to having LP_DEAD bits set). When a snapshotConflictHorizon value
285 : * isn't needed in index AM's deletion WAL record, it is safe for it to skip
286 : * calling here entirely.
287 : *
288 : * We assume that caller index AM uses the standard IndexTuple representation,
289 : * with table TIDs stored in the t_tid field. We also expect (and assert)
290 : * that the line pointers on page for 'itemnos' offsets are already marked
291 : * LP_DEAD.
292 : */
293 : TransactionId
294 0 : index_compute_xid_horizon_for_tuples(Relation irel,
295 : Relation hrel,
296 : Buffer ibuf,
297 : OffsetNumber *itemnos,
298 : int nitems)
299 : {
300 : TM_IndexDeleteOp delstate;
301 0 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
302 0 : Page ipage = BufferGetPage(ibuf);
303 : IndexTuple itup;
304 :
305 : Assert(nitems > 0);
306 :
307 0 : delstate.irel = irel;
308 0 : delstate.iblknum = BufferGetBlockNumber(ibuf);
309 0 : delstate.bottomup = false;
310 0 : delstate.bottomupfreespace = 0;
311 0 : delstate.ndeltids = 0;
312 0 : delstate.deltids = palloc(nitems * sizeof(TM_IndexDelete));
313 0 : delstate.status = palloc(nitems * sizeof(TM_IndexStatus));
314 :
315 : /* identify what the index tuples about to be deleted point to */
316 0 : for (int i = 0; i < nitems; i++)
317 : {
318 0 : OffsetNumber offnum = itemnos[i];
319 : ItemId iitemid;
320 :
321 0 : iitemid = PageGetItemId(ipage, offnum);
322 0 : itup = (IndexTuple) PageGetItem(ipage, iitemid);
323 :
324 : Assert(ItemIdIsDead(iitemid));
325 :
326 0 : ItemPointerCopy(&itup->t_tid, &delstate.deltids[i].tid);
327 0 : delstate.deltids[i].id = delstate.ndeltids;
328 0 : delstate.status[i].idxoffnum = offnum;
329 0 : delstate.status[i].knowndeletable = true; /* LP_DEAD-marked */
330 0 : delstate.status[i].promising = false; /* unused */
331 0 : delstate.status[i].freespace = 0; /* unused */
332 :
333 0 : delstate.ndeltids++;
334 : }
335 :
336 : /* determine the actual xid horizon */
337 0 : snapshotConflictHorizon = table_index_delete_tuples(hrel, &delstate);
338 :
339 : /* assert tableam agrees that all items are deletable */
340 : Assert(delstate.ndeltids == nitems);
341 :
342 0 : pfree(delstate.deltids);
343 0 : pfree(delstate.status);
344 :
345 0 : return snapshotConflictHorizon;
346 : }
347 :
348 :
349 : /* ----------------------------------------------------------------
350 : * heap-or-index-scan access to system catalogs
351 : *
352 : * These functions support system catalog accesses that normally use
353 : * an index but need to be capable of being switched to heap scans
354 : * if the system indexes are unavailable.
355 : *
356 : * The specified scan keys must be compatible with the named index.
357 : * Generally this means that they must constrain either all columns
358 : * of the index, or the first K columns of an N-column index.
359 : *
360 : * These routines could work with non-system tables, actually,
361 : * but they're only useful when there is a known index to use with
362 : * the given scan keys; so in practice they're only good for
363 : * predetermined types of scans of system catalogs.
364 : * ----------------------------------------------------------------
365 : */
366 :
367 : /*
368 : * systable_beginscan --- set up for heap-or-index scan
369 : *
370 : * rel: catalog to scan, already opened and suitably locked
371 : * indexId: OID of index to conditionally use
372 : * indexOK: if false, forces a heap scan (see notes below)
373 : * snapshot: time qual to use (NULL for a recent catalog snapshot)
374 : * nkeys, key: scan keys
375 : *
376 : * The attribute numbers in the scan key should be set for the heap case.
377 : * If we choose to index, we convert them to 1..n to reference the index
378 : * columns. Note this means there must be one scankey qualification per
379 : * index column! This is checked by the Asserts in the normal, index-using
380 : * case, but won't be checked if the heapscan path is taken.
381 : *
382 : * The routine checks the normal cases for whether an indexscan is safe,
383 : * but caller can make additional checks and pass indexOK=false if needed.
384 : * In standard case indexOK can simply be constant TRUE.
385 : */
386 : SysScanDesc
387 12529524 : systable_beginscan(Relation heapRelation,
388 : Oid indexId,
389 : bool indexOK,
390 : Snapshot snapshot,
391 : int nkeys, ScanKey key)
392 : {
393 : SysScanDesc sysscan;
394 : Relation irel;
395 :
396 12529524 : if (indexOK &&
397 12316766 : !IgnoreSystemIndexes &&
398 12205346 : !ReindexIsProcessingIndex(indexId))
399 12195388 : irel = index_open(indexId, AccessShareLock);
400 : else
401 334136 : irel = NULL;
402 :
403 12529516 : sysscan = (SysScanDesc) palloc(sizeof(SysScanDescData));
404 :
405 12529516 : sysscan->heap_rel = heapRelation;
406 12529516 : sysscan->irel = irel;
407 12529516 : sysscan->slot = table_slot_create(heapRelation, NULL);
408 :
409 12529516 : if (snapshot == NULL)
410 : {
411 11485976 : Oid relid = RelationGetRelid(heapRelation);
412 :
413 11485976 : snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
414 11485976 : sysscan->snapshot = snapshot;
415 : }
416 : else
417 : {
418 : /* Caller is responsible for any snapshot. */
419 1043540 : sysscan->snapshot = NULL;
420 : }
421 :
422 12529516 : if (irel)
423 : {
424 : int i;
425 : ScanKey idxkey;
426 :
427 12195380 : idxkey = palloc_array(ScanKeyData, nkeys);
428 :
429 : /* Convert attribute numbers to be index column numbers. */
430 32122724 : for (i = 0; i < nkeys; i++)
431 : {
432 : int j;
433 :
434 19927344 : memcpy(&idxkey[i], &key[i], sizeof(ScanKeyData));
435 :
436 29127450 : for (j = 0; j < IndexRelationGetNumberOfAttributes(irel); j++)
437 : {
438 29127450 : if (key[i].sk_attno == irel->rd_index->indkey.values[j])
439 : {
440 19927344 : idxkey[i].sk_attno = j + 1;
441 19927344 : break;
442 : }
443 : }
444 19927344 : if (j == IndexRelationGetNumberOfAttributes(irel))
445 0 : elog(ERROR, "column is not in index");
446 : }
447 :
448 12195380 : sysscan->iscan = index_beginscan(heapRelation, irel,
449 : snapshot, nkeys, 0);
450 12195380 : index_rescan(sysscan->iscan, idxkey, nkeys, NULL, 0);
451 12195380 : sysscan->scan = NULL;
452 : }
453 : else
454 : {
455 : /*
456 : * We disallow synchronized scans when forced to use a heapscan on a
457 : * catalog. In most cases the desired rows are near the front, so
458 : * that the unpredictable start point of a syncscan is a serious
459 : * disadvantage; and there are no compensating advantages, because
460 : * it's unlikely that such scans will occur in parallel.
461 : */
462 334136 : sysscan->scan = table_beginscan_strat(heapRelation, snapshot,
463 : nkeys, key,
464 : true, false);
465 334136 : sysscan->iscan = NULL;
466 : }
467 :
468 : /*
469 : * If CheckXidAlive is set then set a flag to indicate that system table
470 : * scan is in-progress. See detailed comments in xact.c where these
471 : * variables are declared.
472 : */
473 12529516 : if (TransactionIdIsValid(CheckXidAlive))
474 1664 : bsysscan = true;
475 :
476 12529516 : return sysscan;
477 : }
478 :
479 : /*
480 : * HandleConcurrentAbort - Handle concurrent abort of the CheckXidAlive.
481 : *
482 : * Error out, if CheckXidAlive is aborted. We can't directly use
483 : * TransactionIdDidAbort as after crash such transaction might not have been
484 : * marked as aborted. See detailed comments in xact.c where the variable
485 : * is declared.
486 : */
487 : static inline void
488 25920370 : HandleConcurrentAbort()
489 : {
490 25920370 : if (TransactionIdIsValid(CheckXidAlive) &&
491 2418 : !TransactionIdIsInProgress(CheckXidAlive) &&
492 16 : !TransactionIdDidCommit(CheckXidAlive))
493 16 : ereport(ERROR,
494 : (errcode(ERRCODE_TRANSACTION_ROLLBACK),
495 : errmsg("transaction aborted during system catalog scan")));
496 25920354 : }
497 :
498 : /*
499 : * systable_getnext --- get next tuple in a heap-or-index scan
500 : *
501 : * Returns NULL if no more tuples available.
502 : *
503 : * Note that returned tuple is a reference to data in a disk buffer;
504 : * it must not be modified, and should be presumed inaccessible after
505 : * next getnext() or endscan() call.
506 : *
507 : * XXX: It'd probably make sense to offer a slot based interface, at least
508 : * optionally.
509 : */
510 : HeapTuple
511 25487612 : systable_getnext(SysScanDesc sysscan)
512 : {
513 25487612 : HeapTuple htup = NULL;
514 :
515 25487612 : if (sysscan->irel)
516 : {
517 22996016 : if (index_getnext_slot(sysscan->iscan, ForwardScanDirection, sysscan->slot))
518 : {
519 : bool shouldFree;
520 :
521 17583610 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
522 : Assert(!shouldFree);
523 :
524 : /*
525 : * We currently don't need to support lossy index operators for
526 : * any system catalog scan. It could be done here, using the scan
527 : * keys to drive the operator calls, if we arranged to save the
528 : * heap attnums during systable_beginscan(); this is practical
529 : * because we still wouldn't need to support indexes on
530 : * expressions.
531 : */
532 17583610 : if (sysscan->iscan->xs_recheck)
533 0 : elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
534 : }
535 : }
536 : else
537 : {
538 2491596 : if (table_scan_getnextslot(sysscan->scan, ForwardScanDirection, sysscan->slot))
539 : {
540 : bool shouldFree;
541 :
542 2413900 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
543 : Assert(!shouldFree);
544 : }
545 : }
546 :
547 : /*
548 : * Handle the concurrent abort while fetching the catalog tuple during
549 : * logical streaming of a transaction.
550 : */
551 25487610 : HandleConcurrentAbort();
552 :
553 25487594 : return htup;
554 : }
555 :
556 : /*
557 : * systable_recheck_tuple --- recheck visibility of most-recently-fetched tuple
558 : *
559 : * In particular, determine if this tuple would be visible to a catalog scan
560 : * that started now. We don't handle the case of a non-MVCC scan snapshot,
561 : * because no caller needs that yet.
562 : *
563 : * This is useful to test whether an object was deleted while we waited to
564 : * acquire lock on it.
565 : *
566 : * Note: we don't actually *need* the tuple to be passed in, but it's a
567 : * good crosscheck that the caller is interested in the right tuple.
568 : */
569 : bool
570 215122 : systable_recheck_tuple(SysScanDesc sysscan, HeapTuple tup)
571 : {
572 : Snapshot freshsnap;
573 : bool result;
574 :
575 : Assert(tup == ExecFetchSlotHeapTuple(sysscan->slot, false, NULL));
576 :
577 : /*
578 : * Trust that table_tuple_satisfies_snapshot() and its subsidiaries
579 : * (commonly LockBuffer() and HeapTupleSatisfiesMVCC()) do not themselves
580 : * acquire snapshots, so we need not register the snapshot. Those
581 : * facilities are too low-level to have any business scanning tables.
582 : */
583 215122 : freshsnap = GetCatalogSnapshot(RelationGetRelid(sysscan->heap_rel));
584 :
585 215122 : result = table_tuple_satisfies_snapshot(sysscan->heap_rel,
586 215122 : sysscan->slot,
587 : freshsnap);
588 :
589 : /*
590 : * Handle the concurrent abort while fetching the catalog tuple during
591 : * logical streaming of a transaction.
592 : */
593 215122 : HandleConcurrentAbort();
594 :
595 215122 : return result;
596 : }
597 :
598 : /*
599 : * systable_endscan --- close scan, release resources
600 : *
601 : * Note that it's still up to the caller to close the heap relation.
602 : */
603 : void
604 12528934 : systable_endscan(SysScanDesc sysscan)
605 : {
606 12528934 : if (sysscan->slot)
607 : {
608 12528934 : ExecDropSingleTupleTableSlot(sysscan->slot);
609 12528934 : sysscan->slot = NULL;
610 : }
611 :
612 12528934 : if (sysscan->irel)
613 : {
614 12194812 : index_endscan(sysscan->iscan);
615 12194812 : index_close(sysscan->irel, AccessShareLock);
616 : }
617 : else
618 334122 : table_endscan(sysscan->scan);
619 :
620 12528934 : if (sysscan->snapshot)
621 11485394 : UnregisterSnapshot(sysscan->snapshot);
622 :
623 : /*
624 : * Reset the bsysscan flag at the end of the systable scan. See detailed
625 : * comments in xact.c where these variables are declared.
626 : */
627 12528934 : if (TransactionIdIsValid(CheckXidAlive))
628 1648 : bsysscan = false;
629 :
630 12528934 : pfree(sysscan);
631 12528934 : }
632 :
633 :
634 : /*
635 : * systable_beginscan_ordered --- set up for ordered catalog scan
636 : *
637 : * These routines have essentially the same API as systable_beginscan etc,
638 : * except that they guarantee to return multiple matching tuples in
639 : * index order. Also, for largely historical reasons, the index to use
640 : * is opened and locked by the caller, not here.
641 : *
642 : * Currently we do not support non-index-based scans here. (In principle
643 : * we could do a heapscan and sort, but the uses are in places that
644 : * probably don't need to still work with corrupted catalog indexes.)
645 : * For the moment, therefore, these functions are merely the thinest of
646 : * wrappers around index_beginscan/index_getnext_slot. The main reason for
647 : * their existence is to centralize possible future support of lossy operators
648 : * in catalog scans.
649 : */
650 : SysScanDesc
651 54076 : systable_beginscan_ordered(Relation heapRelation,
652 : Relation indexRelation,
653 : Snapshot snapshot,
654 : int nkeys, ScanKey key)
655 : {
656 : SysScanDesc sysscan;
657 : int i;
658 : ScanKey idxkey;
659 :
660 : /* REINDEX can probably be a hard error here ... */
661 54076 : if (ReindexIsProcessingIndex(RelationGetRelid(indexRelation)))
662 0 : ereport(ERROR,
663 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
664 : errmsg("cannot access index \"%s\" while it is being reindexed",
665 : RelationGetRelationName(indexRelation))));
666 : /* ... but we only throw a warning about violating IgnoreSystemIndexes */
667 54076 : if (IgnoreSystemIndexes)
668 0 : elog(WARNING, "using index \"%s\" despite IgnoreSystemIndexes",
669 : RelationGetRelationName(indexRelation));
670 :
671 54076 : sysscan = (SysScanDesc) palloc(sizeof(SysScanDescData));
672 :
673 54076 : sysscan->heap_rel = heapRelation;
674 54076 : sysscan->irel = indexRelation;
675 54076 : sysscan->slot = table_slot_create(heapRelation, NULL);
676 :
677 54076 : if (snapshot == NULL)
678 : {
679 8018 : Oid relid = RelationGetRelid(heapRelation);
680 :
681 8018 : snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
682 8018 : sysscan->snapshot = snapshot;
683 : }
684 : else
685 : {
686 : /* Caller is responsible for any snapshot. */
687 46058 : sysscan->snapshot = NULL;
688 : }
689 :
690 54076 : idxkey = palloc_array(ScanKeyData, nkeys);
691 :
692 : /* Convert attribute numbers to be index column numbers. */
693 105224 : for (i = 0; i < nkeys; i++)
694 : {
695 : int j;
696 :
697 51148 : memcpy(&idxkey[i], &key[i], sizeof(ScanKeyData));
698 :
699 54378 : for (j = 0; j < IndexRelationGetNumberOfAttributes(indexRelation); j++)
700 : {
701 54378 : if (key[i].sk_attno == indexRelation->rd_index->indkey.values[j])
702 : {
703 51148 : idxkey[i].sk_attno = j + 1;
704 51148 : break;
705 : }
706 : }
707 51148 : if (j == IndexRelationGetNumberOfAttributes(indexRelation))
708 0 : elog(ERROR, "column is not in index");
709 : }
710 :
711 54076 : sysscan->iscan = index_beginscan(heapRelation, indexRelation,
712 : snapshot, nkeys, 0);
713 54076 : index_rescan(sysscan->iscan, idxkey, nkeys, NULL, 0);
714 54076 : sysscan->scan = NULL;
715 :
716 : /*
717 : * If CheckXidAlive is set then set a flag to indicate that system table
718 : * scan is in-progress. See detailed comments in xact.c where these
719 : * variables are declared.
720 : */
721 54076 : if (TransactionIdIsValid(CheckXidAlive))
722 2 : bsysscan = true;
723 :
724 54076 : return sysscan;
725 : }
726 :
727 : /*
728 : * systable_getnext_ordered --- get next tuple in an ordered catalog scan
729 : */
730 : HeapTuple
731 217644 : systable_getnext_ordered(SysScanDesc sysscan, ScanDirection direction)
732 : {
733 217644 : HeapTuple htup = NULL;
734 :
735 : Assert(sysscan->irel);
736 217644 : if (index_getnext_slot(sysscan->iscan, direction, sysscan->slot))
737 164798 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, NULL);
738 :
739 : /* See notes in systable_getnext */
740 217638 : if (htup && sysscan->iscan->xs_recheck)
741 0 : elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
742 :
743 : /*
744 : * Handle the concurrent abort while fetching the catalog tuple during
745 : * logical streaming of a transaction.
746 : */
747 217638 : HandleConcurrentAbort();
748 :
749 217638 : return htup;
750 : }
751 :
752 : /*
753 : * systable_endscan_ordered --- close scan, release resources
754 : */
755 : void
756 54058 : systable_endscan_ordered(SysScanDesc sysscan)
757 : {
758 54058 : if (sysscan->slot)
759 : {
760 54058 : ExecDropSingleTupleTableSlot(sysscan->slot);
761 54058 : sysscan->slot = NULL;
762 : }
763 :
764 : Assert(sysscan->irel);
765 54058 : index_endscan(sysscan->iscan);
766 54058 : if (sysscan->snapshot)
767 8006 : UnregisterSnapshot(sysscan->snapshot);
768 :
769 : /*
770 : * Reset the bsysscan flag at the end of the systable scan. See detailed
771 : * comments in xact.c where these variables are declared.
772 : */
773 54058 : if (TransactionIdIsValid(CheckXidAlive))
774 2 : bsysscan = false;
775 :
776 54058 : pfree(sysscan);
777 54058 : }
778 :
779 : /*
780 : * systable_inplace_update_begin --- update a row "in place" (overwrite it)
781 : *
782 : * Overwriting violates both MVCC and transactional safety, so the uses of
783 : * this function in Postgres are extremely limited. Nonetheless we find some
784 : * places to use it. See README.tuplock section "Locking to write
785 : * inplace-updated tables" and later sections for expectations of readers and
786 : * writers of a table that gets inplace updates. Standard flow:
787 : *
788 : * ... [any slow preparation not requiring oldtup] ...
789 : * systable_inplace_update_begin([...], &tup, &inplace_state);
790 : * if (!HeapTupleIsValid(tup))
791 : * elog(ERROR, [...]);
792 : * ... [buffer is exclusive-locked; mutate "tup"] ...
793 : * if (dirty)
794 : * systable_inplace_update_finish(inplace_state, tup);
795 : * else
796 : * systable_inplace_update_cancel(inplace_state);
797 : *
798 : * The first several params duplicate the systable_beginscan() param list.
799 : * "oldtupcopy" is an output parameter, assigned NULL if the key ceases to
800 : * find a live tuple. (In PROC_IN_VACUUM, that is a low-probability transient
801 : * condition.) If "oldtupcopy" gets non-NULL, you must pass output parameter
802 : * "state" to systable_inplace_update_finish() or
803 : * systable_inplace_update_cancel().
804 : */
805 : void
806 240200 : systable_inplace_update_begin(Relation relation,
807 : Oid indexId,
808 : bool indexOK,
809 : Snapshot snapshot,
810 : int nkeys, const ScanKeyData *key,
811 : HeapTuple *oldtupcopy,
812 : void **state)
813 : {
814 240200 : int retries = 0;
815 : SysScanDesc scan;
816 : HeapTuple oldtup;
817 : BufferHeapTupleTableSlot *bslot;
818 :
819 : /*
820 : * For now, we don't allow parallel updates. Unlike a regular update,
821 : * this should never create a combo CID, so it might be possible to relax
822 : * this restriction, but not without more thought and testing. It's not
823 : * clear that it would be useful, anyway.
824 : */
825 240200 : if (IsInParallelMode())
826 0 : ereport(ERROR,
827 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
828 : errmsg("cannot update tuples during a parallel operation")));
829 :
830 : /*
831 : * Accept a snapshot argument, for symmetry, but this function advances
832 : * its snapshot as needed to reach the tail of the updated tuple chain.
833 : */
834 : Assert(snapshot == NULL);
835 :
836 : Assert(IsInplaceUpdateRelation(relation) || !IsSystemRelation(relation));
837 :
838 : /* Loop for an exclusive-locked buffer of a non-updated tuple. */
839 : do
840 : {
841 : TupleTableSlot *slot;
842 :
843 240248 : CHECK_FOR_INTERRUPTS();
844 :
845 : /*
846 : * Processes issuing heap_update (e.g. GRANT) at maximum speed could
847 : * drive us to this error. A hostile table owner has stronger ways to
848 : * damage their own table, so that's minor.
849 : */
850 240248 : if (retries++ > 10000)
851 0 : elog(ERROR, "giving up after too many tries to overwrite row");
852 :
853 240248 : INJECTION_POINT("inplace-before-pin");
854 240248 : scan = systable_beginscan(relation, indexId, indexOK, snapshot,
855 240248 : nkeys, unconstify(ScanKeyData *, key));
856 240248 : oldtup = systable_getnext(scan);
857 240248 : if (!HeapTupleIsValid(oldtup))
858 : {
859 0 : systable_endscan(scan);
860 0 : *oldtupcopy = NULL;
861 0 : return;
862 : }
863 :
864 240248 : slot = scan->slot;
865 : Assert(TTS_IS_BUFFERTUPLE(slot));
866 240248 : bslot = (BufferHeapTupleTableSlot *) slot;
867 240248 : } while (!heap_inplace_lock(scan->heap_rel,
868 : bslot->base.tuple, bslot->buffer,
869 240248 : (void (*) (void *)) systable_endscan, scan));
870 :
871 240200 : *oldtupcopy = heap_copytuple(oldtup);
872 240200 : *state = scan;
873 : }
874 :
875 : /*
876 : * systable_inplace_update_finish --- second phase of inplace update
877 : *
878 : * The tuple cannot change size, and therefore its header fields and null
879 : * bitmap (if any) don't change either.
880 : */
881 : void
882 147032 : systable_inplace_update_finish(void *state, HeapTuple tuple)
883 : {
884 147032 : SysScanDesc scan = (SysScanDesc) state;
885 147032 : Relation relation = scan->heap_rel;
886 147032 : TupleTableSlot *slot = scan->slot;
887 147032 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
888 147032 : HeapTuple oldtup = bslot->base.tuple;
889 147032 : Buffer buffer = bslot->buffer;
890 :
891 147032 : heap_inplace_update_and_unlock(relation, oldtup, tuple, buffer);
892 147032 : systable_endscan(scan);
893 147032 : }
894 :
895 : /*
896 : * systable_inplace_update_cancel --- abandon inplace update
897 : *
898 : * This is an alternative to making a no-op update.
899 : */
900 : void
901 93168 : systable_inplace_update_cancel(void *state)
902 : {
903 93168 : SysScanDesc scan = (SysScanDesc) state;
904 93168 : Relation relation = scan->heap_rel;
905 93168 : TupleTableSlot *slot = scan->slot;
906 93168 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
907 93168 : HeapTuple oldtup = bslot->base.tuple;
908 93168 : Buffer buffer = bslot->buffer;
909 :
910 93168 : heap_inplace_unlock(relation, oldtup, buffer);
911 93168 : systable_endscan(scan);
912 93168 : }
|
