1 /*-------------------------------------------------------------------------
2 *
3 * genam.c
4 * general index access method routines
5 *
6 * Portions Copyright (c) 1996-2017, 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/relscan.h"
23 #include "access/transam.h"
24 #include "catalog/index.h"
25 #include "lib/stringinfo.h"
26 #include "miscadmin.h"
27 #include "storage/bufmgr.h"
28 #include "utils/acl.h"
29 #include "utils/builtins.h"
30 #include "utils/lsyscache.h"
31 #include "utils/rel.h"
32 #include "utils/rls.h"
33 #include "utils/ruleutils.h"
34 #include "utils/snapmgr.h"
35 #include "utils/syscache.h"
36 #include "utils/tqual.h"
37
38
39 /* ----------------------------------------------------------------
40 * general access method routines
41 *
42 * All indexed access methods use an identical scan structure.
43 * We don't know how the various AMs do locking, however, so we don't
44 * do anything about that here.
45 *
46 * The intent is that an AM implementor will define a beginscan routine
47 * that calls RelationGetIndexScan, to fill in the scan, and then does
48 * whatever kind of locking he wants.
49 *
50 * At the end of a scan, the AM's endscan routine undoes the locking,
51 * but does *not* call IndexScanEnd --- the higher-level index_endscan
52 * routine does that. (We can't do it in the AM because index_endscan
53 * still needs to touch the IndexScanDesc after calling the AM.)
54 *
55 * Because of this, the AM does not have a choice whether to call
56 * RelationGetIndexScan or not; its beginscan routine must return an
57 * object made by RelationGetIndexScan. This is kinda ugly but not
58 * worth cleaning up now.
59 * ----------------------------------------------------------------
60 */
61
62 /* ----------------
63 * RelationGetIndexScan -- Create and fill an IndexScanDesc.
64 *
65 * This routine creates an index scan structure and sets up initial
66 * contents for it.
67 *
68 * Parameters:
69 * indexRelation -- index relation for scan.
70 * nkeys -- count of scan keys (index qual conditions).
71 * norderbys -- count of index order-by operators.
72 *
73 * Returns:
74 * An initialized IndexScanDesc.
75 * ----------------
76 */
77 IndexScanDesc
RelationGetIndexScan(Relation indexRelation,int nkeys,int norderbys)78 RelationGetIndexScan(Relation indexRelation, int nkeys, int norderbys)
79 {
80 IndexScanDesc scan;
81
82 scan = (IndexScanDesc) palloc(sizeof(IndexScanDescData));
83
84 scan->heapRelation = NULL; /* may be set later */
85 scan->indexRelation = indexRelation;
86 scan->xs_snapshot = InvalidSnapshot; /* caller must initialize this */
87 scan->numberOfKeys = nkeys;
88 scan->numberOfOrderBys = norderbys;
89
90 /*
91 * We allocate key workspace here, but it won't get filled until amrescan.
92 */
93 if (nkeys > 0)
94 scan->keyData = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
95 else
96 scan->keyData = NULL;
97 if (norderbys > 0)
98 scan->orderByData = (ScanKey) palloc(sizeof(ScanKeyData) * norderbys);
99 else
100 scan->orderByData = NULL;
101
102 scan->xs_want_itup = false; /* may be set later */
103
104 /*
105 * During recovery we ignore killed tuples and don't bother to kill them
106 * either. We do this because the xmin on the primary node could easily be
107 * later than the xmin on the standby node, so that what the primary
108 * thinks is killed is supposed to be visible on standby. So for correct
109 * MVCC for queries during recovery we must ignore these hints and check
110 * all tuples. Do *not* set ignore_killed_tuples to true when running in a
111 * transaction that was started during recovery. xactStartedInRecovery
112 * should not be altered by index AMs.
113 */
114 scan->kill_prior_tuple = false;
115 scan->xactStartedInRecovery = TransactionStartedDuringRecovery();
116 scan->ignore_killed_tuples = !scan->xactStartedInRecovery;
117
118 scan->opaque = NULL;
119
120 scan->xs_itup = NULL;
121 scan->xs_itupdesc = NULL;
122 scan->xs_hitup = NULL;
123 scan->xs_hitupdesc = NULL;
124
125 ItemPointerSetInvalid(&scan->xs_ctup.t_self);
126 scan->xs_ctup.t_data = NULL;
127 scan->xs_cbuf = InvalidBuffer;
128 scan->xs_continue_hot = false;
129
130 return scan;
131 }
132
133 /* ----------------
134 * IndexScanEnd -- End an index scan.
135 *
136 * This routine just releases the storage acquired by
137 * RelationGetIndexScan(). Any AM-level resources are
138 * assumed to already have been released by the AM's
139 * endscan routine.
140 *
141 * Returns:
142 * None.
143 * ----------------
144 */
145 void
IndexScanEnd(IndexScanDesc scan)146 IndexScanEnd(IndexScanDesc scan)
147 {
148 if (scan->keyData != NULL)
149 pfree(scan->keyData);
150 if (scan->orderByData != NULL)
151 pfree(scan->orderByData);
152
153 pfree(scan);
154 }
155
156 /*
157 * BuildIndexValueDescription
158 *
159 * Construct a string describing the contents of an index entry, in the
160 * form "(key_name, ...)=(key_value, ...)". This is currently used
161 * for building unique-constraint and exclusion-constraint error messages.
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 *
BuildIndexValueDescription(Relation indexRelation,Datum * values,bool * isnull)177 BuildIndexValueDescription(Relation indexRelation,
178 Datum *values, bool *isnull)
179 {
180 StringInfoData buf;
181 Form_pg_index idxrec;
182 HeapTuple ht_idx;
183 int natts = indexRelation->rd_rel->relnatts;
184 int i;
185 int keyno;
186 Oid indexrelid = RelationGetRelid(indexRelation);
187 Oid indrelid;
188 AclResult aclresult;
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
201 /*
202 * Fetch the pg_index tuple by the Oid of the index
203 */
204 ht_idx = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexrelid));
205 if (!HeapTupleIsValid(ht_idx))
206 elog(ERROR, "cache lookup failed for index %u", indexrelid);
207 idxrec = (Form_pg_index) GETSTRUCT(ht_idx);
208
209 indrelid = idxrec->indrelid;
210 Assert(indexrelid == idxrec->indexrelid);
211
212 /* RLS check- if RLS is enabled then we don't return anything. */
213 if (check_enable_rls(indrelid, InvalidOid, true) == RLS_ENABLED)
214 {
215 ReleaseSysCache(ht_idx);
216 return NULL;
217 }
218
219 /* Table-level SELECT is enough, if the user has it */
220 aclresult = pg_class_aclcheck(indrelid, GetUserId(), ACL_SELECT);
221 if (aclresult != ACLCHECK_OK)
222 {
223 /*
224 * No table-level access, so step through the columns in the index and
225 * make sure the user has SELECT rights on all of them.
226 */
227 for (keyno = 0; keyno < idxrec->indnatts; keyno++)
228 {
229 AttrNumber attnum = idxrec->indkey.values[keyno];
230
231 /*
232 * Note that if attnum == InvalidAttrNumber, then this is an index
233 * based on an expression and we return no detail rather than try
234 * to figure out what column(s) the expression includes and if the
235 * user has SELECT rights on them.
236 */
237 if (attnum == InvalidAttrNumber ||
238 pg_attribute_aclcheck(indrelid, attnum, GetUserId(),
239 ACL_SELECT) != ACLCHECK_OK)
240 {
241 /* No access, so clean up and return */
242 ReleaseSysCache(ht_idx);
243 return NULL;
244 }
245 }
246 }
247 ReleaseSysCache(ht_idx);
248
249 initStringInfo(&buf);
250 appendStringInfo(&buf, "(%s)=(",
251 pg_get_indexdef_columns(indexrelid, true));
252
253 for (i = 0; i < natts; i++)
254 {
255 char *val;
256
257 if (isnull[i])
258 val = "null";
259 else
260 {
261 Oid foutoid;
262 bool typisvarlena;
263
264 /*
265 * The provided data is not necessarily of the type stored in the
266 * index; rather it is of the index opclass's input type. So look
267 * at rd_opcintype not the index tupdesc.
268 *
269 * Note: this is a bit shaky for opclasses that have pseudotype
270 * input types such as ANYARRAY or RECORD. Currently, the
271 * typoutput functions associated with the pseudotypes will work
272 * okay, but we might have to try harder in future.
273 */
274 getTypeOutputInfo(indexRelation->rd_opcintype[i],
275 &foutoid, &typisvarlena);
276 val = OidOutputFunctionCall(foutoid, values[i]);
277 }
278
279 if (i > 0)
280 appendStringInfoString(&buf, ", ");
281 appendStringInfoString(&buf, val);
282 }
283
284 appendStringInfoChar(&buf, ')');
285
286 return buf.data;
287 }
288
289
290 /* ----------------------------------------------------------------
291 * heap-or-index-scan access to system catalogs
292 *
293 * These functions support system catalog accesses that normally use
294 * an index but need to be capable of being switched to heap scans
295 * if the system indexes are unavailable.
296 *
297 * The specified scan keys must be compatible with the named index.
298 * Generally this means that they must constrain either all columns
299 * of the index, or the first K columns of an N-column index.
300 *
301 * These routines could work with non-system tables, actually,
302 * but they're only useful when there is a known index to use with
303 * the given scan keys; so in practice they're only good for
304 * predetermined types of scans of system catalogs.
305 * ----------------------------------------------------------------
306 */
307
308 /*
309 * systable_beginscan --- set up for heap-or-index scan
310 *
311 * rel: catalog to scan, already opened and suitably locked
312 * indexId: OID of index to conditionally use
313 * indexOK: if false, forces a heap scan (see notes below)
314 * snapshot: time qual to use (NULL for a recent catalog snapshot)
315 * nkeys, key: scan keys
316 *
317 * The attribute numbers in the scan key should be set for the heap case.
318 * If we choose to index, we reset them to 1..n to reference the index
319 * columns. Note this means there must be one scankey qualification per
320 * index column! This is checked by the Asserts in the normal, index-using
321 * case, but won't be checked if the heapscan path is taken.
322 *
323 * The routine checks the normal cases for whether an indexscan is safe,
324 * but caller can make additional checks and pass indexOK=false if needed.
325 * In standard case indexOK can simply be constant TRUE.
326 */
327 SysScanDesc
systable_beginscan(Relation heapRelation,Oid indexId,bool indexOK,Snapshot snapshot,int nkeys,ScanKey key)328 systable_beginscan(Relation heapRelation,
329 Oid indexId,
330 bool indexOK,
331 Snapshot snapshot,
332 int nkeys, ScanKey key)
333 {
334 SysScanDesc sysscan;
335 Relation irel;
336
337 if (indexOK &&
338 !IgnoreSystemIndexes &&
339 !ReindexIsProcessingIndex(indexId))
340 irel = index_open(indexId, AccessShareLock);
341 else
342 irel = NULL;
343
344 sysscan = (SysScanDesc) palloc(sizeof(SysScanDescData));
345
346 sysscan->heap_rel = heapRelation;
347 sysscan->irel = irel;
348
349 if (snapshot == NULL)
350 {
351 Oid relid = RelationGetRelid(heapRelation);
352
353 snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
354 sysscan->snapshot = snapshot;
355 }
356 else
357 {
358 /* Caller is responsible for any snapshot. */
359 sysscan->snapshot = NULL;
360 }
361
362 if (irel)
363 {
364 int i;
365
366 /* Change attribute numbers to be index column numbers. */
367 for (i = 0; i < nkeys; i++)
368 {
369 int j;
370
371 for (j = 0; j < irel->rd_index->indnatts; j++)
372 {
373 if (key[i].sk_attno == irel->rd_index->indkey.values[j])
374 {
375 key[i].sk_attno = j + 1;
376 break;
377 }
378 }
379 if (j == irel->rd_index->indnatts)
380 elog(ERROR, "column is not in index");
381 }
382
383 sysscan->iscan = index_beginscan(heapRelation, irel,
384 snapshot, nkeys, 0);
385 index_rescan(sysscan->iscan, key, nkeys, NULL, 0);
386 sysscan->scan = NULL;
387 }
388 else
389 {
390 /*
391 * We disallow synchronized scans when forced to use a heapscan on a
392 * catalog. In most cases the desired rows are near the front, so
393 * that the unpredictable start point of a syncscan is a serious
394 * disadvantage; and there are no compensating advantages, because
395 * it's unlikely that such scans will occur in parallel.
396 */
397 sysscan->scan = heap_beginscan_strat(heapRelation, snapshot,
398 nkeys, key,
399 true, false);
400 sysscan->iscan = NULL;
401 }
402
403 return sysscan;
404 }
405
406 /*
407 * systable_getnext --- get next tuple in a heap-or-index scan
408 *
409 * Returns NULL if no more tuples available.
410 *
411 * Note that returned tuple is a reference to data in a disk buffer;
412 * it must not be modified, and should be presumed inaccessible after
413 * next getnext() or endscan() call.
414 */
415 HeapTuple
systable_getnext(SysScanDesc sysscan)416 systable_getnext(SysScanDesc sysscan)
417 {
418 HeapTuple htup;
419
420 if (sysscan->irel)
421 {
422 htup = index_getnext(sysscan->iscan, ForwardScanDirection);
423
424 /*
425 * We currently don't need to support lossy index operators for any
426 * system catalog scan. It could be done here, using the scan keys to
427 * drive the operator calls, if we arranged to save the heap attnums
428 * during systable_beginscan(); this is practical because we still
429 * wouldn't need to support indexes on expressions.
430 */
431 if (htup && sysscan->iscan->xs_recheck)
432 elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
433 }
434 else
435 htup = heap_getnext(sysscan->scan, ForwardScanDirection);
436
437 return htup;
438 }
439
440 /*
441 * systable_recheck_tuple --- recheck visibility of most-recently-fetched tuple
442 *
443 * In particular, determine if this tuple would be visible to a catalog scan
444 * that started now. We don't handle the case of a non-MVCC scan snapshot,
445 * because no caller needs that yet.
446 *
447 * This is useful to test whether an object was deleted while we waited to
448 * acquire lock on it.
449 *
450 * Note: we don't actually *need* the tuple to be passed in, but it's a
451 * good crosscheck that the caller is interested in the right tuple.
452 */
453 bool
systable_recheck_tuple(SysScanDesc sysscan,HeapTuple tup)454 systable_recheck_tuple(SysScanDesc sysscan, HeapTuple tup)
455 {
456 Snapshot freshsnap;
457 bool result;
458
459 /*
460 * Trust that LockBuffer() and HeapTupleSatisfiesMVCC() do not themselves
461 * acquire snapshots, so we need not register the snapshot. Those
462 * facilities are too low-level to have any business scanning tables.
463 */
464 freshsnap = GetCatalogSnapshot(RelationGetRelid(sysscan->heap_rel));
465
466 if (sysscan->irel)
467 {
468 IndexScanDesc scan = sysscan->iscan;
469
470 Assert(IsMVCCSnapshot(scan->xs_snapshot));
471 Assert(tup == &scan->xs_ctup);
472 Assert(BufferIsValid(scan->xs_cbuf));
473 /* must hold a buffer lock to call HeapTupleSatisfiesVisibility */
474 LockBuffer(scan->xs_cbuf, BUFFER_LOCK_SHARE);
475 result = HeapTupleSatisfiesVisibility(tup, freshsnap, scan->xs_cbuf);
476 LockBuffer(scan->xs_cbuf, BUFFER_LOCK_UNLOCK);
477 }
478 else
479 {
480 HeapScanDesc scan = sysscan->scan;
481
482 Assert(IsMVCCSnapshot(scan->rs_snapshot));
483 Assert(tup == &scan->rs_ctup);
484 Assert(BufferIsValid(scan->rs_cbuf));
485 /* must hold a buffer lock to call HeapTupleSatisfiesVisibility */
486 LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
487 result = HeapTupleSatisfiesVisibility(tup, freshsnap, scan->rs_cbuf);
488 LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
489 }
490 return result;
491 }
492
493 /*
494 * systable_endscan --- close scan, release resources
495 *
496 * Note that it's still up to the caller to close the heap relation.
497 */
498 void
systable_endscan(SysScanDesc sysscan)499 systable_endscan(SysScanDesc sysscan)
500 {
501 if (sysscan->irel)
502 {
503 index_endscan(sysscan->iscan);
504 index_close(sysscan->irel, AccessShareLock);
505 }
506 else
507 heap_endscan(sysscan->scan);
508
509 if (sysscan->snapshot)
510 UnregisterSnapshot(sysscan->snapshot);
511
512 pfree(sysscan);
513 }
514
515
516 /*
517 * systable_beginscan_ordered --- set up for ordered catalog scan
518 *
519 * These routines have essentially the same API as systable_beginscan etc,
520 * except that they guarantee to return multiple matching tuples in
521 * index order. Also, for largely historical reasons, the index to use
522 * is opened and locked by the caller, not here.
523 *
524 * Currently we do not support non-index-based scans here. (In principle
525 * we could do a heapscan and sort, but the uses are in places that
526 * probably don't need to still work with corrupted catalog indexes.)
527 * For the moment, therefore, these functions are merely the thinnest of
528 * wrappers around index_beginscan/index_getnext. The main reason for their
529 * existence is to centralize possible future support of lossy operators
530 * in catalog scans.
531 */
532 SysScanDesc
systable_beginscan_ordered(Relation heapRelation,Relation indexRelation,Snapshot snapshot,int nkeys,ScanKey key)533 systable_beginscan_ordered(Relation heapRelation,
534 Relation indexRelation,
535 Snapshot snapshot,
536 int nkeys, ScanKey key)
537 {
538 SysScanDesc sysscan;
539 int i;
540
541 /* REINDEX can probably be a hard error here ... */
542 if (ReindexIsProcessingIndex(RelationGetRelid(indexRelation)))
543 elog(ERROR, "cannot do ordered scan on index \"%s\", because it is being reindexed",
544 RelationGetRelationName(indexRelation));
545 /* ... but we only throw a warning about violating IgnoreSystemIndexes */
546 if (IgnoreSystemIndexes)
547 elog(WARNING, "using index \"%s\" despite IgnoreSystemIndexes",
548 RelationGetRelationName(indexRelation));
549
550 sysscan = (SysScanDesc) palloc(sizeof(SysScanDescData));
551
552 sysscan->heap_rel = heapRelation;
553 sysscan->irel = indexRelation;
554
555 if (snapshot == NULL)
556 {
557 Oid relid = RelationGetRelid(heapRelation);
558
559 snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
560 sysscan->snapshot = snapshot;
561 }
562 else
563 {
564 /* Caller is responsible for any snapshot. */
565 sysscan->snapshot = NULL;
566 }
567
568 /* Change attribute numbers to be index column numbers. */
569 for (i = 0; i < nkeys; i++)
570 {
571 int j;
572
573 for (j = 0; j < indexRelation->rd_index->indnatts; j++)
574 {
575 if (key[i].sk_attno == indexRelation->rd_index->indkey.values[j])
576 {
577 key[i].sk_attno = j + 1;
578 break;
579 }
580 }
581 if (j == indexRelation->rd_index->indnatts)
582 elog(ERROR, "column is not in index");
583 }
584
585 sysscan->iscan = index_beginscan(heapRelation, indexRelation,
586 snapshot, nkeys, 0);
587 index_rescan(sysscan->iscan, key, nkeys, NULL, 0);
588 sysscan->scan = NULL;
589
590 return sysscan;
591 }
592
593 /*
594 * systable_getnext_ordered --- get next tuple in an ordered catalog scan
595 */
596 HeapTuple
systable_getnext_ordered(SysScanDesc sysscan,ScanDirection direction)597 systable_getnext_ordered(SysScanDesc sysscan, ScanDirection direction)
598 {
599 HeapTuple htup;
600
601 Assert(sysscan->irel);
602 htup = index_getnext(sysscan->iscan, direction);
603 /* See notes in systable_getnext */
604 if (htup && sysscan->iscan->xs_recheck)
605 elog(ERROR, "system catalog scans with lossy index conditions are not implemented");
606
607 return htup;
608 }
609
610 /*
611 * systable_endscan_ordered --- close scan, release resources
612 */
613 void
systable_endscan_ordered(SysScanDesc sysscan)614 systable_endscan_ordered(SysScanDesc sysscan)
615 {
616 Assert(sysscan->irel);
617 index_endscan(sysscan->iscan);
618 if (sysscan->snapshot)
619 UnregisterSnapshot(sysscan->snapshot);
620 pfree(sysscan);
621 }
622