1/*-------------------------------------------------------------------------
2 *
3 * genam.c
4 * general index access method routines
5 *
6 * Portions Copyright (c) 1996-2019, 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 "utils/acl.h"
32#include "utils/builtins.h"
33#include "utils/lsyscache.h"
34#include "utils/rel.h"
35#include "utils/rls.h"
36#include "utils/ruleutils.h"
37#include "utils/snapmgr.h"
38#include "utils/syscache.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 */
79IndexScanDesc
80RelationGetIndexScan(Relation indexRelation, int nkeys, int norderbys)
81{
82 IndexScanDesc scan;
83
84 scan = (IndexScanDesc) palloc(sizeof(IndexScanDescData));
85
86 scan->heapRelation = NULL; /* may be set later */
87 scan->xs_heapfetch = NULL;
88 scan->indexRelation = indexRelation;
89 scan->xs_snapshot = InvalidSnapshot; /* caller must initialize this */
90 scan->numberOfKeys = nkeys;
91 scan->numberOfOrderBys = norderbys;
92
93 /*
94 * We allocate key workspace here, but it won't get filled until amrescan.
95 */
96 if (nkeys > 0)
97 scan->keyData = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
98 else
99 scan->keyData = NULL;
100 if (norderbys > 0)
101 scan->orderByData = (ScanKey) palloc(sizeof(ScanKeyData) * norderbys);
102 else
103 scan->orderByData = NULL;
104
105 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 scan->kill_prior_tuple = false;
118 scan->xactStartedInRecovery = TransactionStartedDuringRecovery();
119 scan->ignore_killed_tuples = !scan->xactStartedInRecovery;
120
121 scan->opaque = NULL;
122
123 scan->xs_itup = NULL;
124 scan->xs_itupdesc = NULL;
125 scan->xs_hitup = NULL;
126 scan->xs_hitupdesc = NULL;
127
128 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 */
143void
144IndexScanEnd(IndexScanDesc scan)
145{
146 if (scan->keyData != NULL)
147 pfree(scan->keyData);
148 if (scan->orderByData != NULL)
149 pfree(scan->orderByData);
150
151 pfree(scan);
152}
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 and exclusion-constraint error messages,
160 * so only key columns of the index are checked and printed.
161 *
162 * Note that if the user does not have permissions to view all of the
163 * columns involved then a NULL is returned. Returning a partial key seems
164 * unlikely to be useful and we have no way to know which of the columns the
165 * user provided (unlike in ExecBuildSlotValueDescription).
166 *
167 * The passed-in values/nulls arrays are the "raw" input to the index AM,
168 * e.g. results of FormIndexDatum --- this is not necessarily what is stored
169 * in the index, but it's what the user perceives to be stored.
170 *
171 * Note: if you change anything here, check whether
172 * ExecBuildSlotPartitionKeyDescription() in execMain.c needs a similar
173 * change.
174 */
175char *
176BuildIndexValueDescription(Relation indexRelation,
177 Datum *values, bool *isnull)
178{
179 StringInfoData buf;
180 Form_pg_index idxrec;
181 int indnkeyatts;
182 int i;
183 int keyno;
184 Oid indexrelid = RelationGetRelid(indexRelation);
185 Oid indrelid;
186 AclResult aclresult;
187
188 indnkeyatts = IndexRelationGetNumberOfKeyAttributes(indexRelation);
189
190 /*
191 * Check permissions- if the user does not have access to view all of the
192 * key columns then return NULL to avoid leaking data.
193 *
194 * First check if RLS is enabled for the relation. If so, return NULL to
195 * avoid leaking data.
196 *
197 * Next we need to check table-level SELECT access and then, if there is
198 * no access there, check column-level permissions.
199 */
200 idxrec = indexRelation->rd_index;
201 indrelid = idxrec->indrelid;
202 Assert(indexrelid == idxrec->indexrelid);
203
204 /* RLS check- if RLS is enabled then we don't return anything. */
205 if (check_enable_rls(indrelid, InvalidOid, true) == RLS_ENABLED)
206 return NULL;
207
208 /* Table-level SELECT is enough, if the user has it */
209 aclresult = pg_class_aclcheck(indrelid, GetUserId(), ACL_SELECT);
210 if (aclresult != ACLCHECK_OK)
211 {
212 /*
213 * No table-level access, so step through the columns in the index and
214 * make sure the user has SELECT rights on all of them.
215 */
216 for (keyno = 0; keyno < indnkeyatts; keyno++)
217 {
218 AttrNumber attnum = idxrec->indkey.values[keyno];
219
220 /*
221 * Note that if attnum == InvalidAttrNumber, then this is an index
222 * based on an expression and we return no detail rather than try
223 * to figure out what column(s) the expression includes and if the
224 * user has SELECT rights on them.
225 */
226 if (attnum == InvalidAttrNumber ||
227 pg_attribute_aclcheck(indrelid, attnum, GetUserId(),
228 ACL_SELECT) != ACLCHECK_OK)
229 {
230 /* No access, so clean up and return */
231 return NULL;
232 }
233 }
234 }
235
236 initStringInfo(&buf);
237 appendStringInfo(&buf, "(%s)=(",
238 pg_get_indexdef_columns(indexrelid, true));
239
240 for (i = 0; i < indnkeyatts; i++)
241 {
242 char *val;
243
244 if (isnull[i])
245 val = "null";
246 else
247 {
248 Oid foutoid;
249 bool typisvarlena;
250
251 /*
252 * The provided data is not necessarily of the type stored in the
253 * index; rather it is of the index opclass's input type. So look
254 * at rd_opcintype not the index tupdesc.
255 *
256 * Note: this is a bit shaky for opclasses that have pseudotype
257 * input types such as ANYARRAY or RECORD. Currently, the
258 * typoutput functions associated with the pseudotypes will work
259 * okay, but we might have to try harder in future.
260 */
261 getTypeOutputInfo(indexRelation->rd_opcintype[i],
262 &foutoid, &typisvarlena);
263 val = OidOutputFunctionCall(foutoid, values[i]);
264 }
265
266 if (i > 0)
267 appendStringInfoString(&buf, ", ");
268 appendStringInfoString(&buf, val);
269 }
270
271 appendStringInfoChar(&buf, ')');
272
273 return buf.data;
274}
275
276/*
277 * Get the latestRemovedXid from the table entries pointed at by the index
278 * tuples being deleted.
279 */
280TransactionId
281index_compute_xid_horizon_for_tuples(Relation irel,
282 Relation hrel,
283 Buffer ibuf,
284 OffsetNumber *itemnos,
285 int nitems)
286{
287 ItemPointerData *ttids =
288 (ItemPointerData *) palloc(sizeof(ItemPointerData) * nitems);
289 TransactionId latestRemovedXid = InvalidTransactionId;
290 Page ipage = BufferGetPage(ibuf);
291 IndexTuple itup;
292
293 /* identify what the index tuples about to be deleted point to */
294 for (int i = 0; i < nitems; i++)
295 {
296 ItemId iitemid;
297
298 iitemid = PageGetItemId(ipage, itemnos[i]);
299 itup = (IndexTuple) PageGetItem(ipage, iitemid);
300
301 ItemPointerCopy(&itup->t_tid, &ttids[i]);
302 }
303
304 /* determine the actual xid horizon */
305 latestRemovedXid =
306 table_compute_xid_horizon_for_tuples(hrel, ttids, nitems);
307
308 pfree(ttids);
309
310 return latestRemovedXid;
311}
312
313
314/* ----------------------------------------------------------------
315 * heap-or-index-scan access to system catalogs
316 *
317 * These functions support system catalog accesses that normally use
318 * an index but need to be capable of being switched to heap scans
319 * if the system indexes are unavailable.
320 *
321 * The specified scan keys must be compatible with the named index.
322 * Generally this means that they must constrain either all columns
323 * of the index, or the first K columns of an N-column index.
324 *
325 * These routines could work with non-system tables, actually,
326 * but they're only useful when there is a known index to use with
327 * the given scan keys; so in practice they're only good for
328 * predetermined types of scans of system catalogs.
329 * ----------------------------------------------------------------
330 */
331
332/*
333 * systable_beginscan --- set up for heap-or-index scan
334 *
335 * rel: catalog to scan, already opened and suitably locked
336 * indexId: OID of index to conditionally use
337 * indexOK: if false, forces a heap scan (see notes below)
338 * snapshot: time qual to use (NULL for a recent catalog snapshot)
339 * nkeys, key: scan keys
340 *
341 * The attribute numbers in the scan key should be set for the heap case.
342 * If we choose to index, we reset them to 1..n to reference the index
343 * columns. Note this means there must be one scankey qualification per
344 * index column! This is checked by the Asserts in the normal, index-using
345 * case, but won't be checked if the heapscan path is taken.
346 *
347 * The routine checks the normal cases for whether an indexscan is safe,
348 * but caller can make additional checks and pass indexOK=false if needed.
349 * In standard case indexOK can simply be constant TRUE.
350 */
351SysScanDesc
352systable_beginscan(Relation heapRelation,
353 Oid indexId,
354 bool indexOK,
355 Snapshot snapshot,
356 int nkeys, ScanKey key)
357{
358 SysScanDesc sysscan;
359 Relation irel;
360
361 if (indexOK &&
362 !IgnoreSystemIndexes &&
363 !ReindexIsProcessingIndex(indexId))
364 irel = index_open(indexId, AccessShareLock);
365 else
366 irel = NULL;
367
368 sysscan = (SysScanDesc) palloc(sizeof(SysScanDescData));
369
370 sysscan->heap_rel = heapRelation;
371 sysscan->irel = irel;
372 sysscan->slot = table_slot_create(heapRelation, NULL);
373
374 if (snapshot == NULL)
375 {
376 Oid relid = RelationGetRelid(heapRelation);
377
378 snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
379 sysscan->snapshot = snapshot;
380 }
381 else
382 {
383 /* Caller is responsible for any snapshot. */
384 sysscan->snapshot = NULL;
385 }
386
387 if (irel)
388 {
389 int i;
390
391 /* Change attribute numbers to be index column numbers. */
392 for (i = 0; i < nkeys; i++)
393 {
394 int j;
395
396 for (j = 0; j < IndexRelationGetNumberOfAttributes(irel); j++)
397 {
398 if (key[i].sk_attno == irel->rd_index->indkey.values[j])
399 {
400 key[i].sk_attno = j + 1;
401 break;
402 }
403 }
404 if (j == IndexRelationGetNumberOfAttributes(irel))
405 elog(ERROR, "column is not in index");
406 }
407
408 sysscan->iscan = index_beginscan(heapRelation, irel,
409 snapshot, nkeys, 0);
410 index_rescan(sysscan->iscan, key, nkeys, NULL, 0);
411 sysscan->scan = NULL;
412 }
413 else
414 {
415 /*
416 * We disallow synchronized scans when forced to use a heapscan on a
417 * catalog. In most cases the desired rows are near the front, so
418 * that the unpredictable start point of a syncscan is a serious
419 * disadvantage; and there are no compensating advantages, because
420 * it's unlikely that such scans will occur in parallel.
421 */
422 sysscan->scan = table_beginscan_strat(heapRelation, snapshot,
423 nkeys, key,
424 true, false);
425 sysscan->iscan = NULL;
426 }
427
428 return sysscan;
429}
430
431/*
432 * systable_getnext --- get next tuple in a heap-or-index scan
433 *
434 * Returns NULL if no more tuples available.
435 *
436 * Note that returned tuple is a reference to data in a disk buffer;
437 * it must not be modified, and should be presumed inaccessible after
438 * next getnext() or endscan() call.
439 *
440 * XXX: It'd probably make sense to offer a slot based interface, at least
441 * optionally.
442 */
443HeapTuple
444systable_getnext(SysScanDesc sysscan)
445{
446 HeapTuple htup = NULL;
447
448 if (sysscan->irel)
449 {
450 if (index_getnext_slot(sysscan->iscan, ForwardScanDirection, sysscan->slot))
451 {
452 bool shouldFree;
453
454 htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
455 Assert(!shouldFree);
456
457 /*
458 * We currently don't need to support lossy index operators for
459 * any system catalog scan. It could be done here, using the scan
460 * keys to drive the operator calls, if we arranged to save the
461 * heap attnums during systable_beginscan(); this is practical
462 * because we still wouldn't need to support indexes on
463 * expressions.
464 */
465 if (sysscan->iscan->xs_recheck)
466 elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
467 }
468 }
469 else
470 {
471 if (table_scan_getnextslot(sysscan->scan, ForwardScanDirection, sysscan->slot))
472 {
473 bool shouldFree;
474
475 htup = ExecFetchSlotHeapTuple(sysscan->slot, false, &shouldFree);
476 Assert(!shouldFree);
477 }
478 }
479
480 return htup;
481}
482
483/*
484 * systable_recheck_tuple --- recheck visibility of most-recently-fetched tuple
485 *
486 * In particular, determine if this tuple would be visible to a catalog scan
487 * that started now. We don't handle the case of a non-MVCC scan snapshot,
488 * because no caller needs that yet.
489 *
490 * This is useful to test whether an object was deleted while we waited to
491 * acquire lock on it.
492 *
493 * Note: we don't actually *need* the tuple to be passed in, but it's a
494 * good crosscheck that the caller is interested in the right tuple.
495 */
496bool
497systable_recheck_tuple(SysScanDesc sysscan, HeapTuple tup)
498{
499 Snapshot freshsnap;
500 bool result;
501
502 Assert(tup == ExecFetchSlotHeapTuple(sysscan->slot, false, NULL));
503
504 /*
505 * Trust that table_tuple_satisfies_snapshot() and its subsidiaries
506 * (commonly LockBuffer() and HeapTupleSatisfiesMVCC()) do not themselves
507 * acquire snapshots, so we need not register the snapshot. Those
508 * facilities are too low-level to have any business scanning tables.
509 */
510 freshsnap = GetCatalogSnapshot(RelationGetRelid(sysscan->heap_rel));
511
512 result = table_tuple_satisfies_snapshot(sysscan->heap_rel,
513 sysscan->slot,
514 freshsnap);
515
516 return result;
517}
518
519/*
520 * systable_endscan --- close scan, release resources
521 *
522 * Note that it's still up to the caller to close the heap relation.
523 */
524void
525systable_endscan(SysScanDesc sysscan)
526{
527 if (sysscan->slot)
528 {
529 ExecDropSingleTupleTableSlot(sysscan->slot);
530 sysscan->slot = NULL;
531 }
532
533 if (sysscan->irel)
534 {
535 index_endscan(sysscan->iscan);
536 index_close(sysscan->irel, AccessShareLock);
537 }
538 else
539 table_endscan(sysscan->scan);
540
541 if (sysscan->snapshot)
542 UnregisterSnapshot(sysscan->snapshot);
543
544 pfree(sysscan);
545}
546
547
548/*
549 * systable_beginscan_ordered --- set up for ordered catalog scan
550 *
551 * These routines have essentially the same API as systable_beginscan etc,
552 * except that they guarantee to return multiple matching tuples in
553 * index order. Also, for largely historical reasons, the index to use
554 * is opened and locked by the caller, not here.
555 *
556 * Currently we do not support non-index-based scans here. (In principle
557 * we could do a heapscan and sort, but the uses are in places that
558 * probably don't need to still work with corrupted catalog indexes.)
559 * For the moment, therefore, these functions are merely the thinnest of
560 * wrappers around index_beginscan/index_getnext. The main reason for their
561 * existence is to centralize possible future support of lossy operators
562 * in catalog scans.
563 */
564SysScanDesc
565systable_beginscan_ordered(Relation heapRelation,
566 Relation indexRelation,
567 Snapshot snapshot,
568 int nkeys, ScanKey key)
569{
570 SysScanDesc sysscan;
571 int i;
572
573 /* REINDEX can probably be a hard error here ... */
574 if (ReindexIsProcessingIndex(RelationGetRelid(indexRelation)))
575 elog(ERROR, "cannot do ordered scan on index \"%s\", because it is being reindexed",
576 RelationGetRelationName(indexRelation));
577 /* ... but we only throw a warning about violating IgnoreSystemIndexes */
578 if (IgnoreSystemIndexes)
579 elog(WARNING, "using index \"%s\" despite IgnoreSystemIndexes",
580 RelationGetRelationName(indexRelation));
581
582 sysscan = (SysScanDesc) palloc(sizeof(SysScanDescData));
583
584 sysscan->heap_rel = heapRelation;
585 sysscan->irel = indexRelation;
586 sysscan->slot = table_slot_create(heapRelation, NULL);
587
588 if (snapshot == NULL)
589 {
590 Oid relid = RelationGetRelid(heapRelation);
591
592 snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
593 sysscan->snapshot = snapshot;
594 }
595 else
596 {
597 /* Caller is responsible for any snapshot. */
598 sysscan->snapshot = NULL;
599 }
600
601 /* Change attribute numbers to be index column numbers. */
602 for (i = 0; i < nkeys; i++)
603 {
604 int j;
605
606 for (j = 0; j < IndexRelationGetNumberOfAttributes(indexRelation); j++)
607 {
608 if (key[i].sk_attno == indexRelation->rd_index->indkey.values[j])
609 {
610 key[i].sk_attno = j + 1;
611 break;
612 }
613 }
614 if (j == IndexRelationGetNumberOfAttributes(indexRelation))
615 elog(ERROR, "column is not in index");
616 }
617
618 sysscan->iscan = index_beginscan(heapRelation, indexRelation,
619 snapshot, nkeys, 0);
620 index_rescan(sysscan->iscan, key, nkeys, NULL, 0);
621 sysscan->scan = NULL;
622
623 return sysscan;
624}
625
626/*
627 * systable_getnext_ordered --- get next tuple in an ordered catalog scan
628 */
629HeapTuple
630systable_getnext_ordered(SysScanDesc sysscan, ScanDirection direction)
631{
632 HeapTuple htup = NULL;
633
634 Assert(sysscan->irel);
635 if (index_getnext_slot(sysscan->iscan, direction, sysscan->slot))
636 htup = ExecFetchSlotHeapTuple(sysscan->slot, false, NULL);
637
638 /* See notes in systable_getnext */
639 if (htup && sysscan->iscan->xs_recheck)
640 elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
641
642 return htup;
643}
644
645/*
646 * systable_endscan_ordered --- close scan, release resources
647 */
648void
649systable_endscan_ordered(SysScanDesc sysscan)
650{
651 if (sysscan->slot)
652 {
653 ExecDropSingleTupleTableSlot(sysscan->slot);
654 sysscan->slot = NULL;
655 }
656
657 Assert(sysscan->irel);
658 index_endscan(sysscan->iscan);
659 if (sysscan->snapshot)
660 UnregisterSnapshot(sysscan->snapshot);
661 pfree(sysscan);
662}
663