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 10105981 : RelationGetIndexScan(Relation indexRelation, int nkeys, int norderbys)
81 : {
82 : IndexScanDesc scan;
83 :
84 10105981 : scan = palloc_object(IndexScanDescData);
85 :
86 10105981 : scan->heapRelation = NULL; /* may be set later */
87 10105981 : scan->xs_heapfetch = NULL;
88 10105981 : scan->indexRelation = indexRelation;
89 10105981 : scan->xs_snapshot = InvalidSnapshot; /* caller must initialize this */
90 10105981 : scan->numberOfKeys = nkeys;
91 10105981 : scan->numberOfOrderBys = norderbys;
92 :
93 : /*
94 : * We allocate key workspace here, but it won't get filled until amrescan.
95 : */
96 10105981 : if (nkeys > 0)
97 10098146 : scan->keyData = palloc_array(ScanKeyData, nkeys);
98 : else
99 7835 : scan->keyData = NULL;
100 10105981 : if (norderbys > 0)
101 119 : scan->orderByData = palloc_array(ScanKeyData, norderbys);
102 : else
103 10105862 : scan->orderByData = NULL;
104 :
105 10105981 : 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 10105981 : scan->kill_prior_tuple = false;
118 10105981 : scan->xactStartedInRecovery = TransactionStartedDuringRecovery();
119 10105981 : scan->ignore_killed_tuples = !scan->xactStartedInRecovery;
120 :
121 10105981 : scan->opaque = NULL;
122 10105981 : scan->instrument = NULL;
123 :
124 10105981 : scan->xs_itup = NULL;
125 10105981 : scan->xs_itupdesc = NULL;
126 10105981 : scan->xs_hitup = NULL;
127 10105981 : scan->xs_hitupdesc = NULL;
128 :
129 10105981 : 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 10104637 : IndexScanEnd(IndexScanDesc scan)
146 : {
147 10104637 : if (scan->keyData != NULL)
148 10096826 : pfree(scan->keyData);
149 10104637 : if (scan->orderByData != NULL)
150 115 : pfree(scan->orderByData);
151 :
152 10104637 : pfree(scan);
153 10104637 : }
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 *
178 754 : 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;
186 754 : Oid indexrelid = RelationGetRelid(indexRelation);
187 : Oid indrelid;
188 : AclResult aclresult;
189 :
190 754 : 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 : */
202 754 : idxrec = indexRelation->rd_index;
203 754 : indrelid = idxrec->indrelid;
204 : Assert(indexrelid == idxrec->indexrelid);
205 :
206 : /* RLS check- if RLS is enabled then we don't return anything. */
207 754 : 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 */
211 746 : aclresult = pg_class_aclcheck(indrelid, GetUserId(), ACL_SELECT);
212 746 : 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 :
238 738 : initStringInfo(&buf);
239 738 : appendStringInfo(&buf, "(%s)=(",
240 : pg_get_indexdef_columns(indexrelid, true));
241 :
242 1754 : for (i = 0; i < indnkeyatts; i++)
243 : {
244 : char *val;
245 :
246 1016 : 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 1004 : getTypeOutputInfo(indexRelation->rd_opcintype[i],
264 : &foutoid, &typisvarlena);
265 1004 : val = OidOutputFunctionCall(foutoid, values[i]);
266 : }
267 :
268 1016 : if (i > 0)
269 278 : appendStringInfoString(&buf, ", ");
270 1016 : appendStringInfoString(&buf, val);
271 : }
272 :
273 738 : appendStringInfoChar(&buf, ')');
274 :
275 738 : 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 10028282 : 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 10028282 : if (indexOK &&
398 9878288 : !IgnoreSystemIndexes &&
399 9802991 : !ReindexIsProcessingIndex(indexId))
400 9794300 : irel = index_open(indexId, AccessShareLock);
401 : else
402 233982 : irel = NULL;
403 :
404 10028279 : sysscan = palloc_object(SysScanDescData);
405 :
406 10028279 : sysscan->heap_rel = heapRelation;
407 10028279 : sysscan->irel = irel;
408 10028279 : sysscan->slot = table_slot_create(heapRelation, NULL);
409 :
410 10028279 : if (snapshot == NULL)
411 : {
412 9320654 : Oid relid = RelationGetRelid(heapRelation);
413 :
414 9320654 : snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
415 9320654 : sysscan->snapshot = snapshot;
416 : }
417 : else
418 : {
419 : /* Caller is responsible for any snapshot. */
420 707625 : sysscan->snapshot = NULL;
421 : }
422 :
423 : /*
424 : * If CheckXidAlive is set then set a flag to indicate that system table
425 : * scan is in-progress. See detailed comments in xact.c where these
426 : * variables are declared.
427 : */
428 10028279 : if (TransactionIdIsValid(CheckXidAlive))
429 1073 : bsysscan = true;
430 :
431 10028279 : if (irel)
432 : {
433 : int i;
434 : ScanKey idxkey;
435 :
436 9794297 : idxkey = palloc_array(ScanKeyData, nkeys);
437 :
438 : /* Convert attribute numbers to be index column numbers. */
439 25694959 : for (i = 0; i < nkeys; i++)
440 : {
441 : int j;
442 :
443 15900662 : memcpy(&idxkey[i], &key[i], sizeof(ScanKeyData));
444 :
445 23062274 : for (j = 0; j < IndexRelationGetNumberOfAttributes(irel); j++)
446 : {
447 23062274 : if (key[i].sk_attno == irel->rd_index->indkey.values[j])
448 : {
449 15900662 : idxkey[i].sk_attno = j + 1;
450 15900662 : break;
451 : }
452 : }
453 15900662 : if (j == IndexRelationGetNumberOfAttributes(irel))
454 0 : elog(ERROR, "column is not in index");
455 : }
456 :
457 9794297 : sysscan->iscan = index_beginscan(heapRelation, irel,
458 : snapshot, NULL, nkeys, 0,
459 : SO_NONE);
460 9794297 : index_rescan(sysscan->iscan, idxkey, nkeys, NULL, 0);
461 9794297 : sysscan->scan = NULL;
462 :
463 9794297 : pfree(idxkey);
464 : }
465 : else
466 : {
467 : /*
468 : * We disallow synchronized scans when forced to use a heapscan on a
469 : * catalog. In most cases the desired rows are near the front, so
470 : * that the unpredictable start point of a syncscan is a serious
471 : * disadvantage; and there are no compensating advantages, because
472 : * it's unlikely that such scans will occur in parallel.
473 : */
474 233982 : sysscan->scan = table_beginscan_strat(heapRelation, snapshot,
475 : nkeys, key,
476 : true, false);
477 233982 : sysscan->iscan = NULL;
478 : }
479 :
480 10028279 : return sysscan;
481 : }
482 :
483 : /*
484 : * HandleConcurrentAbort - Handle concurrent abort of the CheckXidAlive.
485 : *
486 : * Error out, if CheckXidAlive is aborted. We can't directly use
487 : * TransactionIdDidAbort as after crash such transaction might not have been
488 : * marked as aborted. See detailed comments in xact.c where the variable
489 : * is declared.
490 : */
491 : static inline void
492 20818128 : HandleConcurrentAbort(void)
493 : {
494 20818128 : if (TransactionIdIsValid(CheckXidAlive) &&
495 1736 : !TransactionIdIsInProgress(CheckXidAlive) &&
496 32 : !TransactionIdDidCommit(CheckXidAlive))
497 8 : ereport(ERROR,
498 : (errcode(ERRCODE_TRANSACTION_ROLLBACK),
499 : errmsg("transaction aborted during system catalog scan")));
500 20818120 : }
501 :
502 : /*
503 : * systable_getnext --- get next tuple in a heap-or-index scan
504 : *
505 : * Returns NULL if no more tuples available.
506 : *
507 : * Note that returned tuple is a reference to data in a disk buffer;
508 : * it must not be modified, and should be presumed inaccessible after
509 : * next getnext() or endscan() call.
510 : *
511 : * XXX: It'd probably make sense to offer a slot based interface, at least
512 : * optionally.
513 : */
514 : HeapTuple
515 20512882 : systable_getnext(SysScanDesc sysscan)
516 : {
517 20512882 : HeapTuple htup = NULL;
518 :
519 20512882 : if (sysscan->irel)
520 : {
521 18204290 : if (index_getnext_slot(sysscan->iscan, ForwardScanDirection, sysscan->slot))
522 : {
523 : bool shouldFree;
524 :
525 14106040 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
526 : Assert(!shouldFree);
527 :
528 : /*
529 : * We currently don't need to support lossy index operators for
530 : * any system catalog scan. It could be done here, using the scan
531 : * keys to drive the operator calls, if we arranged to save the
532 : * heap attnums during systable_beginscan(); this is practical
533 : * because we still wouldn't need to support indexes on
534 : * expressions.
535 : */
536 14106040 : if (sysscan->iscan->xs_recheck)
537 0 : elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
538 : }
539 : }
540 : else
541 : {
542 2308592 : if (table_scan_getnextslot(sysscan->scan, ForwardScanDirection, sysscan->slot))
543 : {
544 : bool shouldFree;
545 :
546 2252292 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
547 : Assert(!shouldFree);
548 : }
549 : }
550 :
551 : /*
552 : * Handle the concurrent abort while fetching the catalog tuple during
553 : * logical streaming of a transaction.
554 : */
555 20512879 : HandleConcurrentAbort();
556 :
557 20512871 : return htup;
558 : }
559 :
560 : /*
561 : * systable_recheck_tuple --- recheck visibility of most-recently-fetched tuple
562 : *
563 : * In particular, determine if this tuple would be visible to a catalog scan
564 : * that started now. We don't handle the case of a non-MVCC scan snapshot,
565 : * because no caller needs that yet.
566 : *
567 : * This is useful to test whether an object was deleted while we waited to
568 : * acquire lock on it.
569 : *
570 : * Note: we don't actually *need* the tuple to be passed in, but it's a
571 : * good crosscheck that the caller is interested in the right tuple.
572 : */
573 : bool
574 163938 : systable_recheck_tuple(SysScanDesc sysscan, HeapTuple tup)
575 : {
576 : Snapshot freshsnap;
577 : bool result;
578 :
579 : Assert(tup == ExecFetchSlotHeapTuple(sysscan->slot, false, NULL));
580 :
581 163938 : freshsnap = GetCatalogSnapshot(RelationGetRelid(sysscan->heap_rel));
582 163938 : freshsnap = RegisterSnapshot(freshsnap);
583 :
584 163938 : result = table_tuple_satisfies_snapshot(sysscan->heap_rel,
585 163938 : sysscan->slot,
586 : freshsnap);
587 163938 : UnregisterSnapshot(freshsnap);
588 :
589 : /*
590 : * Handle the concurrent abort while fetching the catalog tuple during
591 : * logical streaming of a transaction.
592 : */
593 163938 : HandleConcurrentAbort();
594 :
595 163938 : 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 10027719 : systable_endscan(SysScanDesc sysscan)
605 : {
606 10027719 : if (sysscan->slot)
607 : {
608 10027719 : ExecDropSingleTupleTableSlot(sysscan->slot);
609 10027719 : sysscan->slot = NULL;
610 : }
611 :
612 10027719 : if (sysscan->irel)
613 : {
614 9793747 : index_endscan(sysscan->iscan);
615 9793747 : index_close(sysscan->irel, AccessShareLock);
616 : }
617 : else
618 233972 : table_endscan(sysscan->scan);
619 :
620 10027719 : if (sysscan->snapshot)
621 9320102 : 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 10027719 : if (TransactionIdIsValid(CheckXidAlive))
628 1065 : bsysscan = false;
629 :
630 10027719 : pfree(sysscan);
631 10027719 : }
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 37873 : 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 37873 : 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 37873 : if (IgnoreSystemIndexes)
668 0 : elog(WARNING, "using index \"%s\" despite IgnoreSystemIndexes",
669 : RelationGetRelationName(indexRelation));
670 :
671 37873 : sysscan = palloc_object(SysScanDescData);
672 :
673 37873 : sysscan->heap_rel = heapRelation;
674 37873 : sysscan->irel = indexRelation;
675 37873 : sysscan->slot = table_slot_create(heapRelation, NULL);
676 :
677 37873 : if (snapshot == NULL)
678 : {
679 5336 : Oid relid = RelationGetRelid(heapRelation);
680 :
681 5336 : snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
682 5336 : sysscan->snapshot = snapshot;
683 : }
684 : else
685 : {
686 : /* Caller is responsible for any snapshot. */
687 32537 : sysscan->snapshot = NULL;
688 : }
689 :
690 37873 : idxkey = palloc_array(ScanKeyData, nkeys);
691 :
692 : /* Convert attribute numbers to be index column numbers. */
693 73721 : for (i = 0; i < nkeys; i++)
694 : {
695 : int j;
696 :
697 35848 : memcpy(&idxkey[i], &key[i], sizeof(ScanKeyData));
698 :
699 37889 : for (j = 0; j < IndexRelationGetNumberOfAttributes(indexRelation); j++)
700 : {
701 37889 : if (key[i].sk_attno == indexRelation->rd_index->indkey.values[j])
702 : {
703 35848 : idxkey[i].sk_attno = j + 1;
704 35848 : break;
705 : }
706 : }
707 35848 : if (j == IndexRelationGetNumberOfAttributes(indexRelation))
708 0 : elog(ERROR, "column is not in index");
709 : }
710 :
711 : /*
712 : * If CheckXidAlive is set then set a flag to indicate that system table
713 : * scan is in-progress. See detailed comments in xact.c where these
714 : * variables are declared.
715 : */
716 37873 : if (TransactionIdIsValid(CheckXidAlive))
717 0 : bsysscan = true;
718 :
719 37873 : sysscan->iscan = index_beginscan(heapRelation, indexRelation,
720 : snapshot, NULL, nkeys, 0,
721 : SO_NONE);
722 37873 : index_rescan(sysscan->iscan, idxkey, nkeys, NULL, 0);
723 37873 : sysscan->scan = NULL;
724 :
725 37873 : pfree(idxkey);
726 :
727 37873 : return sysscan;
728 : }
729 :
730 : /*
731 : * systable_getnext_ordered --- get next tuple in an ordered catalog scan
732 : */
733 : HeapTuple
734 141314 : systable_getnext_ordered(SysScanDesc sysscan, ScanDirection direction)
735 : {
736 141314 : HeapTuple htup = NULL;
737 :
738 : Assert(sysscan->irel);
739 141314 : if (index_getnext_slot(sysscan->iscan, direction, sysscan->slot))
740 104293 : htup = ExecFetchSlotHeapTuple(sysscan->slot, false, NULL);
741 :
742 : /* See notes in systable_getnext */
743 141311 : if (htup && sysscan->iscan->xs_recheck)
744 0 : elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
745 :
746 : /*
747 : * Handle the concurrent abort while fetching the catalog tuple during
748 : * logical streaming of a transaction.
749 : */
750 141311 : HandleConcurrentAbort();
751 :
752 141311 : return htup;
753 : }
754 :
755 : /*
756 : * systable_endscan_ordered --- close scan, release resources
757 : */
758 : void
759 37862 : systable_endscan_ordered(SysScanDesc sysscan)
760 : {
761 37862 : if (sysscan->slot)
762 : {
763 37862 : ExecDropSingleTupleTableSlot(sysscan->slot);
764 37862 : sysscan->slot = NULL;
765 : }
766 :
767 : Assert(sysscan->irel);
768 37862 : index_endscan(sysscan->iscan);
769 37862 : if (sysscan->snapshot)
770 5328 : UnregisterSnapshot(sysscan->snapshot);
771 :
772 : /*
773 : * Reset the bsysscan flag at the end of the systable scan. See detailed
774 : * comments in xact.c where these variables are declared.
775 : */
776 37862 : if (TransactionIdIsValid(CheckXidAlive))
777 0 : bsysscan = false;
778 :
779 37862 : pfree(sysscan);
780 37862 : }
781 :
782 : /*
783 : * systable_inplace_update_begin --- update a row "in place" (overwrite it)
784 : *
785 : * Overwriting violates both MVCC and transactional safety, so the uses of
786 : * this function in Postgres are extremely limited. This makes no effort to
787 : * support updating cache key columns or other indexed columns. Nonetheless
788 : * we find some places to use it. See README.tuplock section "Locking to
789 : * write inplace-updated tables" and later sections for expectations of
790 : * readers and writers of a table that gets inplace updates. Standard flow:
791 : *
792 : * ... [any slow preparation not requiring oldtup] ...
793 : * systable_inplace_update_begin([...], &tup, &inplace_state);
794 : * if (!HeapTupleIsValid(tup))
795 : * elog(ERROR, [...]);
796 : * ... [buffer is exclusive-locked; mutate "tup"] ...
797 : * if (dirty)
798 : * systable_inplace_update_finish(inplace_state, tup);
799 : * else
800 : * systable_inplace_update_cancel(inplace_state);
801 : *
802 : * The first several params duplicate the systable_beginscan() param list.
803 : * "oldtupcopy" is an output parameter, assigned NULL if the key ceases to
804 : * find a live tuple. (In PROC_IN_VACUUM, that is a low-probability transient
805 : * condition.) If "oldtupcopy" gets non-NULL, you must pass output parameter
806 : * "state" to systable_inplace_update_finish() or
807 : * systable_inplace_update_cancel().
808 : */
809 : void
810 228257 : systable_inplace_update_begin(Relation relation,
811 : Oid indexId,
812 : bool indexOK,
813 : Snapshot snapshot,
814 : int nkeys, const ScanKeyData *key,
815 : HeapTuple *oldtupcopy,
816 : void **state)
817 : {
818 228257 : int retries = 0;
819 : SysScanDesc scan;
820 : HeapTuple oldtup;
821 : BufferHeapTupleTableSlot *bslot;
822 :
823 : /*
824 : * For now, we don't allow parallel updates. Unlike a regular update,
825 : * this should never create a combo CID, so it might be possible to relax
826 : * this restriction, but not without more thought and testing. It's not
827 : * clear that it would be useful, anyway.
828 : */
829 228257 : if (IsInParallelMode())
830 0 : ereport(ERROR,
831 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
832 : errmsg("cannot update tuples during a parallel operation")));
833 :
834 : /*
835 : * Accept a snapshot argument, for symmetry, but this function advances
836 : * its snapshot as needed to reach the tail of the updated tuple chain.
837 : */
838 : Assert(snapshot == NULL);
839 :
840 : Assert(IsInplaceUpdateRelation(relation) || !IsSystemRelation(relation));
841 :
842 : /* Loop for an exclusive-locked buffer of a non-updated tuple. */
843 : do
844 : {
845 : TupleTableSlot *slot;
846 :
847 228288 : CHECK_FOR_INTERRUPTS();
848 :
849 : /*
850 : * Processes issuing heap_update (e.g. GRANT) at maximum speed could
851 : * drive us to this error. A hostile table owner has stronger ways to
852 : * damage their own table, so that's minor.
853 : */
854 228288 : if (retries++ > 10000)
855 0 : elog(ERROR, "giving up after too many tries to overwrite row");
856 :
857 228288 : INJECTION_POINT("inplace-before-pin", NULL);
858 228288 : scan = systable_beginscan(relation, indexId, indexOK, snapshot,
859 : nkeys, unconstify(ScanKeyData *, key));
860 228288 : oldtup = systable_getnext(scan);
861 228288 : if (!HeapTupleIsValid(oldtup))
862 : {
863 0 : systable_endscan(scan);
864 0 : *oldtupcopy = NULL;
865 0 : return;
866 : }
867 :
868 228288 : slot = scan->slot;
869 : Assert(TTS_IS_BUFFERTUPLE(slot));
870 228288 : bslot = (BufferHeapTupleTableSlot *) slot;
871 228288 : } while (!heap_inplace_lock(scan->heap_rel,
872 : bslot->base.tuple, bslot->buffer,
873 : (void (*) (void *)) systable_endscan, scan));
874 :
875 228257 : *oldtupcopy = heap_copytuple(oldtup);
876 228257 : *state = scan;
877 : }
878 :
879 : /*
880 : * systable_inplace_update_finish --- second phase of inplace update
881 : *
882 : * The tuple cannot change size, and therefore its header fields and null
883 : * bitmap (if any) don't change either.
884 : */
885 : void
886 100503 : systable_inplace_update_finish(void *state, HeapTuple tuple)
887 : {
888 100503 : SysScanDesc scan = (SysScanDesc) state;
889 100503 : Relation relation = scan->heap_rel;
890 100503 : TupleTableSlot *slot = scan->slot;
891 100503 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
892 100503 : HeapTuple oldtup = bslot->base.tuple;
893 100503 : Buffer buffer = bslot->buffer;
894 :
895 100503 : heap_inplace_update_and_unlock(relation, oldtup, tuple, buffer);
896 100503 : systable_endscan(scan);
897 100503 : }
898 :
899 : /*
900 : * systable_inplace_update_cancel --- abandon inplace update
901 : *
902 : * This is an alternative to making a no-op update.
903 : */
904 : void
905 127754 : systable_inplace_update_cancel(void *state)
906 : {
907 127754 : SysScanDesc scan = (SysScanDesc) state;
908 127754 : Relation relation = scan->heap_rel;
909 127754 : TupleTableSlot *slot = scan->slot;
910 127754 : BufferHeapTupleTableSlot *bslot = (BufferHeapTupleTableSlot *) slot;
911 127754 : HeapTuple oldtup = bslot->base.tuple;
912 127754 : Buffer buffer = bslot->buffer;
913 :
914 127754 : heap_inplace_unlock(relation, oldtup, buffer);
915 127754 : systable_endscan(scan);
916 127754 : }
|
