Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * verify_nbtree.c
4 : * Verifies the integrity of nbtree indexes based on invariants.
5 : *
6 : * For B-Tree indexes, verification includes checking that each page in the
7 : * target index has items in logical order as reported by an insertion scankey
8 : * (the insertion scankey sort-wise NULL semantics are needed for
9 : * verification).
10 : *
11 : * When index-to-heap verification is requested, a Bloom filter is used to
12 : * fingerprint all tuples in the target index, as the index is traversed to
13 : * verify its structure. A heap scan later uses Bloom filter probes to verify
14 : * that every visible heap tuple has a matching index tuple.
15 : *
16 : *
17 : * Copyright (c) 2017-2025, PostgreSQL Global Development Group
18 : *
19 : * IDENTIFICATION
20 : * contrib/amcheck/verify_nbtree.c
21 : *
22 : *-------------------------------------------------------------------------
23 : */
24 : #include "postgres.h"
25 :
26 : #include "access/heaptoast.h"
27 : #include "access/htup_details.h"
28 : #include "access/nbtree.h"
29 : #include "access/table.h"
30 : #include "access/tableam.h"
31 : #include "access/transam.h"
32 : #include "access/xact.h"
33 : #include "verify_common.h"
34 : #include "catalog/index.h"
35 : #include "catalog/pg_am.h"
36 : #include "catalog/pg_opfamily_d.h"
37 : #include "common/pg_prng.h"
38 : #include "lib/bloomfilter.h"
39 : #include "miscadmin.h"
40 : #include "storage/smgr.h"
41 : #include "utils/guc.h"
42 : #include "utils/memutils.h"
43 : #include "utils/snapmgr.h"
44 :
45 :
46 642 : PG_MODULE_MAGIC_EXT(
47 : .name = "amcheck",
48 : .version = PG_VERSION
49 : );
50 :
51 : /*
52 : * A B-Tree cannot possibly have this many levels, since there must be one
53 : * block per level, which is bound by the range of BlockNumber:
54 : */
55 : #define InvalidBtreeLevel ((uint32) InvalidBlockNumber)
56 : #define BTreeTupleGetNKeyAtts(itup, rel) \
57 : Min(IndexRelationGetNumberOfKeyAttributes(rel), BTreeTupleGetNAtts(itup, rel))
58 :
59 : /*
60 : * State associated with verifying a B-Tree index
61 : *
62 : * target is the point of reference for a verification operation.
63 : *
64 : * Other B-Tree pages may be allocated, but those are always auxiliary (e.g.,
65 : * they are current target's child pages). Conceptually, problems are only
66 : * ever found in the current target page (or for a particular heap tuple during
67 : * heapallindexed verification). Each page found by verification's left/right,
68 : * top/bottom scan becomes the target exactly once.
69 : */
70 : typedef struct BtreeCheckState
71 : {
72 : /*
73 : * Unchanging state, established at start of verification:
74 : */
75 :
76 : /* B-Tree Index Relation and associated heap relation */
77 : Relation rel;
78 : Relation heaprel;
79 : /* rel is heapkeyspace index? */
80 : bool heapkeyspace;
81 : /* ShareLock held on heap/index, rather than AccessShareLock? */
82 : bool readonly;
83 : /* Also verifying heap has no unindexed tuples? */
84 : bool heapallindexed;
85 : /* Also making sure non-pivot tuples can be found by new search? */
86 : bool rootdescend;
87 : /* Also check uniqueness constraint if index is unique */
88 : bool checkunique;
89 : /* Per-page context */
90 : MemoryContext targetcontext;
91 : /* Buffer access strategy */
92 : BufferAccessStrategy checkstrategy;
93 :
94 : /*
95 : * Info for uniqueness checking. Fill these fields once per index check.
96 : */
97 : IndexInfo *indexinfo;
98 : Snapshot snapshot;
99 :
100 : /*
101 : * Mutable state, for verification of particular page:
102 : */
103 :
104 : /* Current target page */
105 : Page target;
106 : /* Target block number */
107 : BlockNumber targetblock;
108 : /* Target page's LSN */
109 : XLogRecPtr targetlsn;
110 :
111 : /*
112 : * Low key: high key of left sibling of target page. Used only for child
113 : * verification. So, 'lowkey' is kept only when 'readonly' is set.
114 : */
115 : IndexTuple lowkey;
116 :
117 : /*
118 : * The rightlink and incomplete split flag of block one level down to the
119 : * target page, which was visited last time via downlink from target page.
120 : * We use it to check for missing downlinks.
121 : */
122 : BlockNumber prevrightlink;
123 : bool previncompletesplit;
124 :
125 : /*
126 : * Mutable state, for optional heapallindexed verification:
127 : */
128 :
129 : /* Bloom filter fingerprints B-Tree index */
130 : bloom_filter *filter;
131 : /* Debug counter */
132 : int64 heaptuplespresent;
133 : } BtreeCheckState;
134 :
135 : /*
136 : * Starting point for verifying an entire B-Tree index level
137 : */
138 : typedef struct BtreeLevel
139 : {
140 : /* Level number (0 is leaf page level). */
141 : uint32 level;
142 :
143 : /* Left most block on level. Scan of level begins here. */
144 : BlockNumber leftmost;
145 :
146 : /* Is this level reported as "true" root level by meta page? */
147 : bool istruerootlevel;
148 : } BtreeLevel;
149 :
150 : /*
151 : * Information about the last visible entry with current B-tree key. Used
152 : * for validation of the unique constraint.
153 : */
154 : typedef struct BtreeLastVisibleEntry
155 : {
156 : BlockNumber blkno; /* Index block */
157 : OffsetNumber offset; /* Offset on index block */
158 : int postingIndex; /* Number in the posting list (-1 for
159 : * non-deduplicated tuples) */
160 : ItemPointer tid; /* Heap tid */
161 : } BtreeLastVisibleEntry;
162 :
163 : /*
164 : * arguments for the bt_index_check_callback callback
165 : */
166 : typedef struct BTCallbackState
167 : {
168 : bool parentcheck;
169 : bool heapallindexed;
170 : bool rootdescend;
171 : bool checkunique;
172 : } BTCallbackState;
173 :
174 174 : PG_FUNCTION_INFO_V1(bt_index_check);
175 120 : PG_FUNCTION_INFO_V1(bt_index_parent_check);
176 :
177 : static void bt_index_check_callback(Relation indrel, Relation heaprel,
178 : void *state, bool readonly);
179 : static void bt_check_every_level(Relation rel, Relation heaprel,
180 : bool heapkeyspace, bool readonly, bool heapallindexed,
181 : bool rootdescend, bool checkunique);
182 : static BtreeLevel bt_check_level_from_leftmost(BtreeCheckState *state,
183 : BtreeLevel level);
184 : static bool bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
185 : BlockNumber start,
186 : BTPageOpaque start_opaque);
187 : static void bt_recheck_sibling_links(BtreeCheckState *state,
188 : BlockNumber btpo_prev_from_target,
189 : BlockNumber leftcurrent);
190 : static bool heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid);
191 : static void bt_report_duplicate(BtreeCheckState *state,
192 : BtreeLastVisibleEntry *lVis,
193 : ItemPointer nexttid,
194 : BlockNumber nblock, OffsetNumber noffset,
195 : int nposting);
196 : static void bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
197 : BlockNumber targetblock, OffsetNumber offset,
198 : BtreeLastVisibleEntry *lVis);
199 : static void bt_target_page_check(BtreeCheckState *state);
200 : static BTScanInsert bt_right_page_check_scankey(BtreeCheckState *state,
201 : OffsetNumber *rightfirstoffset);
202 : static void bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
203 : OffsetNumber downlinkoffnum);
204 : static void bt_child_highkey_check(BtreeCheckState *state,
205 : OffsetNumber target_downlinkoffnum,
206 : Page loaded_child,
207 : uint32 target_level);
208 : static void bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
209 : BlockNumber blkno, Page page);
210 : static void bt_tuple_present_callback(Relation index, ItemPointer tid,
211 : Datum *values, bool *isnull,
212 : bool tupleIsAlive, void *checkstate);
213 : static IndexTuple bt_normalize_tuple(BtreeCheckState *state,
214 : IndexTuple itup);
215 : static inline IndexTuple bt_posting_plain_tuple(IndexTuple itup, int n);
216 : static bool bt_rootdescend(BtreeCheckState *state, IndexTuple itup);
217 : static inline bool offset_is_negative_infinity(BTPageOpaque opaque,
218 : OffsetNumber offset);
219 : static inline bool invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
220 : OffsetNumber upperbound);
221 : static inline bool invariant_leq_offset(BtreeCheckState *state,
222 : BTScanInsert key,
223 : OffsetNumber upperbound);
224 : static inline bool invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
225 : OffsetNumber lowerbound);
226 : static inline bool invariant_l_nontarget_offset(BtreeCheckState *state,
227 : BTScanInsert key,
228 : BlockNumber nontargetblock,
229 : Page nontarget,
230 : OffsetNumber upperbound);
231 : static Page palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum);
232 : static inline BTScanInsert bt_mkscankey_pivotsearch(Relation rel,
233 : IndexTuple itup);
234 : static ItemId PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block,
235 : Page page, OffsetNumber offset);
236 : static inline ItemPointer BTreeTupleGetHeapTIDCareful(BtreeCheckState *state,
237 : IndexTuple itup, bool nonpivot);
238 : static inline ItemPointer BTreeTupleGetPointsToTID(IndexTuple itup);
239 :
240 : /*
241 : * bt_index_check(index regclass, heapallindexed boolean, checkunique boolean)
242 : *
243 : * Verify integrity of B-Tree index.
244 : *
245 : * Acquires AccessShareLock on heap & index relations. Does not consider
246 : * invariants that exist between parent/child pages. Optionally verifies
247 : * that heap does not contain any unindexed or incorrectly indexed tuples.
248 : */
249 : Datum
250 7898 : bt_index_check(PG_FUNCTION_ARGS)
251 : {
252 7898 : Oid indrelid = PG_GETARG_OID(0);
253 : BTCallbackState args;
254 :
255 7898 : args.heapallindexed = false;
256 7898 : args.rootdescend = false;
257 7898 : args.parentcheck = false;
258 7898 : args.checkunique = false;
259 :
260 7898 : if (PG_NARGS() >= 2)
261 7886 : args.heapallindexed = PG_GETARG_BOOL(1);
262 7898 : if (PG_NARGS() >= 3)
263 1372 : args.checkunique = PG_GETARG_BOOL(2);
264 :
265 7898 : amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
266 : bt_index_check_callback,
267 : AccessShareLock, &args);
268 :
269 7846 : PG_RETURN_VOID();
270 : }
271 :
272 : /*
273 : * bt_index_parent_check(index regclass, heapallindexed boolean, rootdescend boolean, checkunique boolean)
274 : *
275 : * Verify integrity of B-Tree index.
276 : *
277 : * Acquires ShareLock on heap & index relations. Verifies that downlinks in
278 : * parent pages are valid lower bounds on child pages. Optionally verifies
279 : * that heap does not contain any unindexed or incorrectly indexed tuples.
280 : */
281 : Datum
282 120 : bt_index_parent_check(PG_FUNCTION_ARGS)
283 : {
284 120 : Oid indrelid = PG_GETARG_OID(0);
285 : BTCallbackState args;
286 :
287 120 : args.heapallindexed = false;
288 120 : args.rootdescend = false;
289 120 : args.parentcheck = true;
290 120 : args.checkunique = false;
291 :
292 120 : if (PG_NARGS() >= 2)
293 110 : args.heapallindexed = PG_GETARG_BOOL(1);
294 120 : if (PG_NARGS() >= 3)
295 104 : args.rootdescend = PG_GETARG_BOOL(2);
296 120 : if (PG_NARGS() >= 4)
297 52 : args.checkunique = PG_GETARG_BOOL(3);
298 :
299 120 : amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
300 : bt_index_check_callback,
301 : ShareLock, &args);
302 :
303 86 : PG_RETURN_VOID();
304 : }
305 :
306 : /*
307 : * Helper for bt_index_[parent_]check, coordinating the bulk of the work.
308 : */
309 : static void
310 8008 : bt_index_check_callback(Relation indrel, Relation heaprel, void *state, bool readonly)
311 : {
312 8008 : BTCallbackState *args = (BTCallbackState *) state;
313 : bool heapkeyspace,
314 : allequalimage;
315 :
316 8008 : if (!smgrexists(RelationGetSmgr(indrel), MAIN_FORKNUM))
317 36 : ereport(ERROR,
318 : (errcode(ERRCODE_INDEX_CORRUPTED),
319 : errmsg("index \"%s\" lacks a main relation fork",
320 : RelationGetRelationName(indrel))));
321 :
322 : /* Extract metadata from metapage, and sanitize it in passing */
323 7972 : _bt_metaversion(indrel, &heapkeyspace, &allequalimage);
324 7972 : if (allequalimage && !heapkeyspace)
325 0 : ereport(ERROR,
326 : (errcode(ERRCODE_INDEX_CORRUPTED),
327 : errmsg("index \"%s\" metapage has equalimage field set on unsupported nbtree version",
328 : RelationGetRelationName(indrel))));
329 7972 : if (allequalimage && !_bt_allequalimage(indrel, false))
330 : {
331 0 : bool has_interval_ops = false;
332 :
333 0 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(indrel); i++)
334 0 : if (indrel->rd_opfamily[i] == INTERVAL_BTREE_FAM_OID)
335 : {
336 0 : has_interval_ops = true;
337 0 : ereport(ERROR,
338 : (errcode(ERRCODE_INDEX_CORRUPTED),
339 : errmsg("index \"%s\" metapage incorrectly indicates that deduplication is safe",
340 : RelationGetRelationName(indrel)),
341 : has_interval_ops
342 : ? errhint("This is known of \"interval\" indexes last built on a version predating 2023-11.")
343 : : 0));
344 : }
345 : }
346 :
347 : /* Check index, possibly against table it is an index on */
348 7972 : bt_check_every_level(indrel, heaprel, heapkeyspace, readonly,
349 7972 : args->heapallindexed, args->rootdescend, args->checkunique);
350 7932 : }
351 :
352 : /*
353 : * Main entry point for B-Tree SQL-callable functions. Walks the B-Tree in
354 : * logical order, verifying invariants as it goes. Optionally, verification
355 : * checks if the heap relation contains any tuples that are not represented in
356 : * the index but should be.
357 : *
358 : * It is the caller's responsibility to acquire appropriate heavyweight lock on
359 : * the index relation, and advise us if extra checks are safe when a ShareLock
360 : * is held. (A lock of the same type must also have been acquired on the heap
361 : * relation.)
362 : *
363 : * A ShareLock is generally assumed to prevent any kind of physical
364 : * modification to the index structure, including modifications that VACUUM may
365 : * make. This does not include setting of the LP_DEAD bit by concurrent index
366 : * scans, although that is just metadata that is not able to directly affect
367 : * any check performed here. Any concurrent process that might act on the
368 : * LP_DEAD bit being set (recycle space) requires a heavyweight lock that
369 : * cannot be held while we hold a ShareLock. (Besides, even if that could
370 : * happen, the ad-hoc recycling when a page might otherwise split is performed
371 : * per-page, and requires an exclusive buffer lock, which wouldn't cause us
372 : * trouble. _bt_delitems_vacuum() may only delete leaf items, and so the extra
373 : * parent/child check cannot be affected.)
374 : */
375 : static void
376 7972 : bt_check_every_level(Relation rel, Relation heaprel, bool heapkeyspace,
377 : bool readonly, bool heapallindexed, bool rootdescend,
378 : bool checkunique)
379 : {
380 : BtreeCheckState *state;
381 : Page metapage;
382 : BTMetaPageData *metad;
383 : uint32 previouslevel;
384 : BtreeLevel current;
385 7972 : Snapshot snapshot = SnapshotAny;
386 :
387 7972 : if (!readonly)
388 7874 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\"",
389 : RelationGetRelationName(rel));
390 : else
391 98 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\" with cross-level checks",
392 : RelationGetRelationName(rel));
393 :
394 : /*
395 : * This assertion matches the one in index_getnext_tid(). See page
396 : * recycling/"visible to everyone" notes in nbtree README.
397 : */
398 : Assert(TransactionIdIsValid(RecentXmin));
399 :
400 : /*
401 : * Initialize state for entire verification operation
402 : */
403 7972 : state = palloc0(sizeof(BtreeCheckState));
404 7972 : state->rel = rel;
405 7972 : state->heaprel = heaprel;
406 7972 : state->heapkeyspace = heapkeyspace;
407 7972 : state->readonly = readonly;
408 7972 : state->heapallindexed = heapallindexed;
409 7972 : state->rootdescend = rootdescend;
410 7972 : state->checkunique = checkunique;
411 7972 : state->snapshot = InvalidSnapshot;
412 :
413 7972 : if (state->heapallindexed)
414 : {
415 : int64 total_pages;
416 : int64 total_elems;
417 : uint64 seed;
418 :
419 : /*
420 : * Size Bloom filter based on estimated number of tuples in index,
421 : * while conservatively assuming that each block must contain at least
422 : * MaxTIDsPerBTreePage / 3 "plain" tuples -- see
423 : * bt_posting_plain_tuple() for definition, and details of how posting
424 : * list tuples are handled.
425 : */
426 128 : total_pages = RelationGetNumberOfBlocks(rel);
427 128 : total_elems = Max(total_pages * (MaxTIDsPerBTreePage / 3),
428 : (int64) state->rel->rd_rel->reltuples);
429 : /* Generate a random seed to avoid repetition */
430 128 : seed = pg_prng_uint64(&pg_global_prng_state);
431 : /* Create Bloom filter to fingerprint index */
432 128 : state->filter = bloom_create(total_elems, maintenance_work_mem, seed);
433 128 : state->heaptuplespresent = 0;
434 :
435 : /*
436 : * Register our own snapshot in !readonly case, rather than asking
437 : * table_index_build_scan() to do this for us later. This needs to
438 : * happen before index fingerprinting begins, so we can later be
439 : * certain that index fingerprinting should have reached all tuples
440 : * returned by table_index_build_scan().
441 : */
442 128 : if (!state->readonly)
443 : {
444 74 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
445 :
446 : /*
447 : * GetTransactionSnapshot() always acquires a new MVCC snapshot in
448 : * READ COMMITTED mode. A new snapshot is guaranteed to have all
449 : * the entries it requires in the index.
450 : *
451 : * We must defend against the possibility that an old xact
452 : * snapshot was returned at higher isolation levels when that
453 : * snapshot is not safe for index scans of the target index. This
454 : * is possible when the snapshot sees tuples that are before the
455 : * index's indcheckxmin horizon. Throwing an error here should be
456 : * very rare. It doesn't seem worth using a secondary snapshot to
457 : * avoid this.
458 : */
459 74 : if (IsolationUsesXactSnapshot() && rel->rd_index->indcheckxmin &&
460 0 : !TransactionIdPrecedes(HeapTupleHeaderGetXmin(rel->rd_indextuple->t_data),
461 : snapshot->xmin))
462 0 : ereport(ERROR,
463 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
464 : errmsg("index \"%s\" cannot be verified using transaction snapshot",
465 : RelationGetRelationName(rel))));
466 : }
467 : }
468 :
469 : /*
470 : * We need a snapshot to check the uniqueness of the index. For better
471 : * performance take it once per index check. If snapshot already taken
472 : * reuse it.
473 : */
474 7972 : if (state->checkunique)
475 : {
476 1416 : state->indexinfo = BuildIndexInfo(state->rel);
477 1416 : if (state->indexinfo->ii_Unique)
478 : {
479 1276 : if (snapshot != SnapshotAny)
480 14 : state->snapshot = snapshot;
481 : else
482 1262 : state->snapshot = RegisterSnapshot(GetTransactionSnapshot());
483 : }
484 : }
485 :
486 : Assert(!state->rootdescend || state->readonly);
487 7972 : if (state->rootdescend && !state->heapkeyspace)
488 0 : ereport(ERROR,
489 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
490 : errmsg("cannot verify that tuples from index \"%s\" can each be found by an independent index search",
491 : RelationGetRelationName(rel)),
492 : errhint("Only B-Tree version 4 indexes support rootdescend verification.")));
493 :
494 : /* Create context for page */
495 7972 : state->targetcontext = AllocSetContextCreate(CurrentMemoryContext,
496 : "amcheck context",
497 : ALLOCSET_DEFAULT_SIZES);
498 7972 : state->checkstrategy = GetAccessStrategy(BAS_BULKREAD);
499 :
500 : /* Get true root block from meta-page */
501 7972 : metapage = palloc_btree_page(state, BTREE_METAPAGE);
502 7972 : metad = BTPageGetMeta(metapage);
503 :
504 : /*
505 : * Certain deletion patterns can result in "skinny" B-Tree indexes, where
506 : * the fast root and true root differ.
507 : *
508 : * Start from the true root, not the fast root, unlike conventional index
509 : * scans. This approach is more thorough, and removes the risk of
510 : * following a stale fast root from the meta page.
511 : */
512 7972 : if (metad->btm_fastroot != metad->btm_root)
513 26 : ereport(DEBUG1,
514 : (errcode(ERRCODE_NO_DATA),
515 : errmsg_internal("harmless fast root mismatch in index \"%s\"",
516 : RelationGetRelationName(rel)),
517 : errdetail_internal("Fast root block %u (level %u) differs from true root block %u (level %u).",
518 : metad->btm_fastroot, metad->btm_fastlevel,
519 : metad->btm_root, metad->btm_level)));
520 :
521 : /*
522 : * Starting at the root, verify every level. Move left to right, top to
523 : * bottom. Note that there may be no pages other than the meta page (meta
524 : * page can indicate that root is P_NONE when the index is totally empty).
525 : */
526 7972 : previouslevel = InvalidBtreeLevel;
527 7972 : current.level = metad->btm_level;
528 7972 : current.leftmost = metad->btm_root;
529 7972 : current.istruerootlevel = true;
530 12936 : while (current.leftmost != P_NONE)
531 : {
532 : /*
533 : * Verify this level, and get left most page for next level down, if
534 : * not at leaf level
535 : */
536 5000 : current = bt_check_level_from_leftmost(state, current);
537 :
538 4964 : if (current.leftmost == InvalidBlockNumber)
539 0 : ereport(ERROR,
540 : (errcode(ERRCODE_INDEX_CORRUPTED),
541 : errmsg("index \"%s\" has no valid pages on level below %u or first level",
542 : RelationGetRelationName(rel), previouslevel)));
543 :
544 4964 : previouslevel = current.level;
545 : }
546 :
547 : /*
548 : * * Check whether heap contains unindexed/malformed tuples *
549 : */
550 7936 : if (state->heapallindexed)
551 : {
552 114 : IndexInfo *indexinfo = BuildIndexInfo(state->rel);
553 : TableScanDesc scan;
554 :
555 : /*
556 : * Create our own scan for table_index_build_scan(), rather than
557 : * getting it to do so for us. This is required so that we can
558 : * actually use the MVCC snapshot registered earlier in !readonly
559 : * case.
560 : *
561 : * Note that table_index_build_scan() calls heap_endscan() for us.
562 : */
563 114 : scan = table_beginscan_strat(state->heaprel, /* relation */
564 : snapshot, /* snapshot */
565 : 0, /* number of keys */
566 : NULL, /* scan key */
567 : true, /* buffer access strategy OK */
568 : true); /* syncscan OK? */
569 :
570 : /*
571 : * Scan will behave as the first scan of a CREATE INDEX CONCURRENTLY
572 : * behaves in !readonly case.
573 : *
574 : * It's okay that we don't actually use the same lock strength for the
575 : * heap relation as any other ii_Concurrent caller would in !readonly
576 : * case. We have no reason to care about a concurrent VACUUM
577 : * operation, since there isn't going to be a second scan of the heap
578 : * that needs to be sure that there was no concurrent recycling of
579 : * TIDs.
580 : */
581 110 : indexinfo->ii_Concurrent = !state->readonly;
582 :
583 : /*
584 : * Don't wait for uncommitted tuple xact commit/abort when index is a
585 : * unique index on a catalog (or an index used by an exclusion
586 : * constraint). This could otherwise happen in the readonly case.
587 : */
588 110 : indexinfo->ii_Unique = false;
589 110 : indexinfo->ii_ExclusionOps = NULL;
590 110 : indexinfo->ii_ExclusionProcs = NULL;
591 110 : indexinfo->ii_ExclusionStrats = NULL;
592 :
593 110 : elog(DEBUG1, "verifying that tuples from index \"%s\" are present in \"%s\"",
594 : RelationGetRelationName(state->rel),
595 : RelationGetRelationName(state->heaprel));
596 :
597 110 : table_index_build_scan(state->heaprel, state->rel, indexinfo, true, false,
598 : bt_tuple_present_callback, state, scan);
599 :
600 110 : ereport(DEBUG1,
601 : (errmsg_internal("finished verifying presence of " INT64_FORMAT " tuples from table \"%s\" with bitset %.2f%% set",
602 : state->heaptuplespresent, RelationGetRelationName(heaprel),
603 : 100.0 * bloom_prop_bits_set(state->filter))));
604 :
605 110 : if (snapshot != SnapshotAny)
606 64 : UnregisterSnapshot(snapshot);
607 :
608 110 : bloom_free(state->filter);
609 : }
610 :
611 : /* Be tidy: */
612 7932 : if (snapshot == SnapshotAny && state->snapshot != InvalidSnapshot)
613 1262 : UnregisterSnapshot(state->snapshot);
614 7932 : MemoryContextDelete(state->targetcontext);
615 7932 : }
616 :
617 : /*
618 : * Given a left-most block at some level, move right, verifying each page
619 : * individually (with more verification across pages for "readonly"
620 : * callers). Caller should pass the true root page as the leftmost initially,
621 : * working their way down by passing what is returned for the last call here
622 : * until level 0 (leaf page level) was reached.
623 : *
624 : * Returns state for next call, if any. This includes left-most block number
625 : * one level lower that should be passed on next level/call, which is set to
626 : * P_NONE on last call here (when leaf level is verified). Level numbers
627 : * follow the nbtree convention: higher levels have higher numbers, because new
628 : * levels are added only due to a root page split. Note that prior to the
629 : * first root page split, the root is also a leaf page, so there is always a
630 : * level 0 (leaf level), and it's always the last level processed.
631 : *
632 : * Note on memory management: State's per-page context is reset here, between
633 : * each call to bt_target_page_check().
634 : */
635 : static BtreeLevel
636 5000 : bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
637 : {
638 : /* State to establish early, concerning entire level */
639 : BTPageOpaque opaque;
640 : MemoryContext oldcontext;
641 : BtreeLevel nextleveldown;
642 :
643 : /* Variables for iterating across level using right links */
644 5000 : BlockNumber leftcurrent = P_NONE;
645 5000 : BlockNumber current = level.leftmost;
646 :
647 : /* Initialize return state */
648 5000 : nextleveldown.leftmost = InvalidBlockNumber;
649 5000 : nextleveldown.level = InvalidBtreeLevel;
650 5000 : nextleveldown.istruerootlevel = false;
651 :
652 : /* Use page-level context for duration of this call */
653 5000 : oldcontext = MemoryContextSwitchTo(state->targetcontext);
654 :
655 5000 : elog(DEBUG1, "verifying level %u%s", level.level,
656 : level.istruerootlevel ?
657 : " (true root level)" : level.level == 0 ? " (leaf level)" : "");
658 :
659 5000 : state->prevrightlink = InvalidBlockNumber;
660 5000 : state->previncompletesplit = false;
661 :
662 : do
663 : {
664 : /* Don't rely on CHECK_FOR_INTERRUPTS() calls at lower level */
665 18204 : CHECK_FOR_INTERRUPTS();
666 :
667 : /* Initialize state for this iteration */
668 18204 : state->targetblock = current;
669 18204 : state->target = palloc_btree_page(state, state->targetblock);
670 18180 : state->targetlsn = PageGetLSN(state->target);
671 :
672 18180 : opaque = BTPageGetOpaque(state->target);
673 :
674 18180 : if (P_IGNORE(opaque))
675 : {
676 : /*
677 : * Since there cannot be a concurrent VACUUM operation in readonly
678 : * mode, and since a page has no links within other pages
679 : * (siblings and parent) once it is marked fully deleted, it
680 : * should be impossible to land on a fully deleted page in
681 : * readonly mode. See bt_child_check() for further details.
682 : *
683 : * The bt_child_check() P_ISDELETED() check is repeated here so
684 : * that pages that are only reachable through sibling links get
685 : * checked.
686 : */
687 0 : if (state->readonly && P_ISDELETED(opaque))
688 0 : ereport(ERROR,
689 : (errcode(ERRCODE_INDEX_CORRUPTED),
690 : errmsg("downlink or sibling link points to deleted block in index \"%s\"",
691 : RelationGetRelationName(state->rel)),
692 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
693 : current, leftcurrent, opaque->btpo_prev)));
694 :
695 0 : if (P_RIGHTMOST(opaque))
696 0 : ereport(ERROR,
697 : (errcode(ERRCODE_INDEX_CORRUPTED),
698 : errmsg("block %u fell off the end of index \"%s\"",
699 : current, RelationGetRelationName(state->rel))));
700 : else
701 0 : ereport(DEBUG1,
702 : (errcode(ERRCODE_NO_DATA),
703 : errmsg_internal("block %u of index \"%s\" concurrently deleted",
704 : current, RelationGetRelationName(state->rel))));
705 0 : goto nextpage;
706 : }
707 18180 : else if (nextleveldown.leftmost == InvalidBlockNumber)
708 : {
709 : /*
710 : * A concurrent page split could make the caller supplied leftmost
711 : * block no longer contain the leftmost page, or no longer be the
712 : * true root, but where that isn't possible due to heavyweight
713 : * locking, check that the first valid page meets caller's
714 : * expectations.
715 : */
716 4976 : if (state->readonly)
717 : {
718 88 : if (!bt_leftmost_ignoring_half_dead(state, current, opaque))
719 0 : ereport(ERROR,
720 : (errcode(ERRCODE_INDEX_CORRUPTED),
721 : errmsg("block %u is not leftmost in index \"%s\"",
722 : current, RelationGetRelationName(state->rel))));
723 :
724 88 : if (level.istruerootlevel && !P_ISROOT(opaque))
725 0 : ereport(ERROR,
726 : (errcode(ERRCODE_INDEX_CORRUPTED),
727 : errmsg("block %u is not true root in index \"%s\"",
728 : current, RelationGetRelationName(state->rel))));
729 : }
730 :
731 : /*
732 : * Before beginning any non-trivial examination of level, prepare
733 : * state for next bt_check_level_from_leftmost() invocation for
734 : * the next level for the next level down (if any).
735 : *
736 : * There should be at least one non-ignorable page per level,
737 : * unless this is the leaf level, which is assumed by caller to be
738 : * final level.
739 : */
740 4976 : if (!P_ISLEAF(opaque))
741 : {
742 : IndexTuple itup;
743 : ItemId itemid;
744 :
745 : /* Internal page -- downlink gets leftmost on next level */
746 1110 : itemid = PageGetItemIdCareful(state, state->targetblock,
747 : state->target,
748 1110 : P_FIRSTDATAKEY(opaque));
749 1110 : itup = (IndexTuple) PageGetItem(state->target, itemid);
750 1110 : nextleveldown.leftmost = BTreeTupleGetDownLink(itup);
751 1110 : nextleveldown.level = opaque->btpo_level - 1;
752 : }
753 : else
754 : {
755 : /*
756 : * Leaf page -- final level caller must process.
757 : *
758 : * Note that this could also be the root page, if there has
759 : * been no root page split yet.
760 : */
761 3866 : nextleveldown.leftmost = P_NONE;
762 3866 : nextleveldown.level = InvalidBtreeLevel;
763 : }
764 :
765 : /*
766 : * Finished setting up state for this call/level. Control will
767 : * never end up back here in any future loop iteration for this
768 : * level.
769 : */
770 : }
771 :
772 : /*
773 : * Sibling links should be in mutual agreement. There arises
774 : * leftcurrent == P_NONE && btpo_prev != P_NONE when the left sibling
775 : * of the parent's low-key downlink is half-dead. (A half-dead page
776 : * has no downlink from its parent.) Under heavyweight locking, the
777 : * last bt_leftmost_ignoring_half_dead() validated this btpo_prev.
778 : * Without heavyweight locking, validation of the P_NONE case remains
779 : * unimplemented.
780 : */
781 18180 : if (opaque->btpo_prev != leftcurrent && leftcurrent != P_NONE)
782 0 : bt_recheck_sibling_links(state, opaque->btpo_prev, leftcurrent);
783 :
784 : /* Check level */
785 18180 : if (level.level != opaque->btpo_level)
786 0 : ereport(ERROR,
787 : (errcode(ERRCODE_INDEX_CORRUPTED),
788 : errmsg("leftmost down link for level points to block in index \"%s\" whose level is not one level down",
789 : RelationGetRelationName(state->rel)),
790 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
791 : current, level.level, opaque->btpo_level)));
792 :
793 : /* Verify invariants for page */
794 18180 : bt_target_page_check(state);
795 :
796 18168 : nextpage:
797 :
798 : /* Try to detect circular links */
799 18168 : if (current == leftcurrent || current == opaque->btpo_prev)
800 0 : ereport(ERROR,
801 : (errcode(ERRCODE_INDEX_CORRUPTED),
802 : errmsg("circular link chain found in block %u of index \"%s\"",
803 : current, RelationGetRelationName(state->rel))));
804 :
805 18168 : leftcurrent = current;
806 18168 : current = opaque->btpo_next;
807 :
808 18168 : if (state->lowkey)
809 : {
810 : Assert(state->readonly);
811 3724 : pfree(state->lowkey);
812 3724 : state->lowkey = NULL;
813 : }
814 :
815 : /*
816 : * Copy current target high key as the low key of right sibling.
817 : * Allocate memory in upper level context, so it would be cleared
818 : * after reset of target context.
819 : *
820 : * We only need the low key in corner cases of checking child high
821 : * keys. We use high key only when incomplete split on the child level
822 : * falls to the boundary of pages on the target level. See
823 : * bt_child_highkey_check() for details. So, typically we won't end
824 : * up doing anything with low key, but it's simpler for general case
825 : * high key verification to always have it available.
826 : *
827 : * The correctness of managing low key in the case of concurrent
828 : * splits wasn't investigated yet. Thankfully we only need low key
829 : * for readonly verification and concurrent splits won't happen.
830 : */
831 18168 : if (state->readonly && !P_RIGHTMOST(opaque))
832 : {
833 : IndexTuple itup;
834 : ItemId itemid;
835 :
836 3724 : itemid = PageGetItemIdCareful(state, state->targetblock,
837 : state->target, P_HIKEY);
838 3724 : itup = (IndexTuple) PageGetItem(state->target, itemid);
839 :
840 3724 : state->lowkey = MemoryContextAlloc(oldcontext, IndexTupleSize(itup));
841 3724 : memcpy(state->lowkey, itup, IndexTupleSize(itup));
842 : }
843 :
844 : /* Free page and associated memory for this iteration */
845 18168 : MemoryContextReset(state->targetcontext);
846 : }
847 18168 : while (current != P_NONE);
848 :
849 4964 : if (state->lowkey)
850 : {
851 : Assert(state->readonly);
852 0 : pfree(state->lowkey);
853 0 : state->lowkey = NULL;
854 : }
855 :
856 : /* Don't change context for caller */
857 4964 : MemoryContextSwitchTo(oldcontext);
858 :
859 4964 : return nextleveldown;
860 : }
861 :
862 : /* Check visibility of the table entry referenced by nbtree index */
863 : static bool
864 676 : heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid)
865 : {
866 : bool tid_visible;
867 :
868 676 : TupleTableSlot *slot = table_slot_create(state->heaprel, NULL);
869 :
870 676 : tid_visible = table_tuple_fetch_row_version(state->heaprel,
871 : tid, state->snapshot, slot);
872 676 : if (slot != NULL)
873 676 : ExecDropSingleTupleTableSlot(slot);
874 :
875 676 : return tid_visible;
876 : }
877 :
878 : /*
879 : * Prepare an error message for unique constrain violation in
880 : * a btree index and report ERROR.
881 : */
882 : static void
883 6 : bt_report_duplicate(BtreeCheckState *state,
884 : BtreeLastVisibleEntry *lVis,
885 : ItemPointer nexttid, BlockNumber nblock, OffsetNumber noffset,
886 : int nposting)
887 : {
888 : char *htid,
889 : *nhtid,
890 : *itid,
891 6 : *nitid = "",
892 6 : *pposting = "",
893 6 : *pnposting = "";
894 :
895 6 : htid = psprintf("tid=(%u,%u)",
896 6 : ItemPointerGetBlockNumberNoCheck(lVis->tid),
897 6 : ItemPointerGetOffsetNumberNoCheck(lVis->tid));
898 6 : nhtid = psprintf("tid=(%u,%u)",
899 : ItemPointerGetBlockNumberNoCheck(nexttid),
900 6 : ItemPointerGetOffsetNumberNoCheck(nexttid));
901 6 : itid = psprintf("tid=(%u,%u)", lVis->blkno, lVis->offset);
902 :
903 6 : if (nblock != lVis->blkno || noffset != lVis->offset)
904 6 : nitid = psprintf(" tid=(%u,%u)", nblock, noffset);
905 :
906 6 : if (lVis->postingIndex >= 0)
907 0 : pposting = psprintf(" posting %u", lVis->postingIndex);
908 :
909 6 : if (nposting >= 0)
910 0 : pnposting = psprintf(" posting %u", nposting);
911 :
912 6 : ereport(ERROR,
913 : (errcode(ERRCODE_INDEX_CORRUPTED),
914 : errmsg("index uniqueness is violated for index \"%s\"",
915 : RelationGetRelationName(state->rel)),
916 : errdetail("Index %s%s and%s%s (point to heap %s and %s) page lsn=%X/%X.",
917 : itid, pposting, nitid, pnposting, htid, nhtid,
918 : LSN_FORMAT_ARGS(state->targetlsn))));
919 : }
920 :
921 : /* Check if current nbtree leaf entry complies with UNIQUE constraint */
922 : static void
923 676 : bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
924 : BlockNumber targetblock, OffsetNumber offset,
925 : BtreeLastVisibleEntry *lVis)
926 : {
927 : ItemPointer tid;
928 676 : bool has_visible_entry = false;
929 :
930 : Assert(targetblock != P_NONE);
931 :
932 : /*
933 : * Current tuple has posting list. Report duplicate if TID of any posting
934 : * list entry is visible and lVis->tid is valid.
935 : */
936 676 : if (BTreeTupleIsPosting(itup))
937 : {
938 0 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
939 : {
940 0 : tid = BTreeTupleGetPostingN(itup, i);
941 0 : if (heap_entry_is_visible(state, tid))
942 : {
943 0 : has_visible_entry = true;
944 0 : if (ItemPointerIsValid(lVis->tid))
945 : {
946 0 : bt_report_duplicate(state,
947 : lVis,
948 : tid, targetblock,
949 : offset, i);
950 : }
951 :
952 : /*
953 : * Prevent double reporting unique constraint violation
954 : * between the posting list entries of the first tuple on the
955 : * page after cross-page check.
956 : */
957 0 : if (lVis->blkno != targetblock && ItemPointerIsValid(lVis->tid))
958 0 : return;
959 :
960 0 : lVis->blkno = targetblock;
961 0 : lVis->offset = offset;
962 0 : lVis->postingIndex = i;
963 0 : lVis->tid = tid;
964 : }
965 : }
966 : }
967 :
968 : /*
969 : * Current tuple has no posting list. If TID is visible save info about it
970 : * for the next comparisons in the loop in bt_target_page_check(). Report
971 : * duplicate if lVis->tid is already valid.
972 : */
973 : else
974 : {
975 676 : tid = BTreeTupleGetHeapTID(itup);
976 676 : if (heap_entry_is_visible(state, tid))
977 : {
978 30 : has_visible_entry = true;
979 30 : if (ItemPointerIsValid(lVis->tid))
980 : {
981 6 : bt_report_duplicate(state,
982 : lVis,
983 : tid, targetblock,
984 : offset, -1);
985 : }
986 :
987 24 : lVis->blkno = targetblock;
988 24 : lVis->offset = offset;
989 24 : lVis->tid = tid;
990 24 : lVis->postingIndex = -1;
991 : }
992 : }
993 :
994 670 : if (!has_visible_entry &&
995 646 : lVis->blkno != InvalidBlockNumber &&
996 18 : lVis->blkno != targetblock)
997 : {
998 0 : char *posting = "";
999 :
1000 0 : if (lVis->postingIndex >= 0)
1001 0 : posting = psprintf(" posting %u", lVis->postingIndex);
1002 0 : ereport(DEBUG1,
1003 : (errcode(ERRCODE_NO_DATA),
1004 : errmsg("index uniqueness can not be checked for index tid=(%u,%u) in index \"%s\"",
1005 : targetblock, offset,
1006 : RelationGetRelationName(state->rel)),
1007 : errdetail("It doesn't have visible heap tids and key is equal to the tid=(%u,%u)%s (points to heap tid=(%u,%u)).",
1008 : lVis->blkno, lVis->offset, posting,
1009 : ItemPointerGetBlockNumberNoCheck(lVis->tid),
1010 : ItemPointerGetOffsetNumberNoCheck(lVis->tid)),
1011 : errhint("VACUUM the table and repeat the check.")));
1012 : }
1013 : }
1014 :
1015 : /*
1016 : * Like P_LEFTMOST(start_opaque), but accept an arbitrarily-long chain of
1017 : * half-dead, sibling-linked pages to the left. If a half-dead page appears
1018 : * under state->readonly, the database exited recovery between the first-stage
1019 : * and second-stage WAL records of a deletion.
1020 : */
1021 : static bool
1022 110 : bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
1023 : BlockNumber start,
1024 : BTPageOpaque start_opaque)
1025 : {
1026 110 : BlockNumber reached = start_opaque->btpo_prev,
1027 110 : reached_from = start;
1028 110 : bool all_half_dead = true;
1029 :
1030 : /*
1031 : * To handle the !readonly case, we'd need to accept BTP_DELETED pages and
1032 : * potentially observe nbtree/README "Page deletion and backwards scans".
1033 : */
1034 : Assert(state->readonly);
1035 :
1036 114 : while (reached != P_NONE && all_half_dead)
1037 : {
1038 4 : Page page = palloc_btree_page(state, reached);
1039 4 : BTPageOpaque reached_opaque = BTPageGetOpaque(page);
1040 :
1041 4 : CHECK_FOR_INTERRUPTS();
1042 :
1043 : /*
1044 : * Try to detect btpo_prev circular links. _bt_unlink_halfdead_page()
1045 : * writes that side-links will continue to point to the siblings.
1046 : * Check btpo_next for that property.
1047 : */
1048 4 : all_half_dead = P_ISHALFDEAD(reached_opaque) &&
1049 4 : reached != start &&
1050 8 : reached != reached_from &&
1051 4 : reached_opaque->btpo_next == reached_from;
1052 4 : if (all_half_dead)
1053 : {
1054 4 : XLogRecPtr pagelsn = PageGetLSN(page);
1055 :
1056 : /* pagelsn should point to an XLOG_BTREE_MARK_PAGE_HALFDEAD */
1057 4 : ereport(DEBUG1,
1058 : (errcode(ERRCODE_NO_DATA),
1059 : errmsg_internal("harmless interrupted page deletion detected in index \"%s\"",
1060 : RelationGetRelationName(state->rel)),
1061 : errdetail_internal("Block=%u right block=%u page lsn=%X/%X.",
1062 : reached, reached_from,
1063 : LSN_FORMAT_ARGS(pagelsn))));
1064 :
1065 4 : reached_from = reached;
1066 4 : reached = reached_opaque->btpo_prev;
1067 : }
1068 :
1069 4 : pfree(page);
1070 : }
1071 :
1072 110 : return all_half_dead;
1073 : }
1074 :
1075 : /*
1076 : * Raise an error when target page's left link does not point back to the
1077 : * previous target page, called leftcurrent here. The leftcurrent page's
1078 : * right link was followed to get to the current target page, and we expect
1079 : * mutual agreement among leftcurrent and the current target page. Make sure
1080 : * that this condition has definitely been violated in the !readonly case,
1081 : * where concurrent page splits are something that we need to deal with.
1082 : *
1083 : * Cross-page inconsistencies involving pages that don't agree about being
1084 : * siblings are known to be a particularly good indicator of corruption
1085 : * involving partial writes/lost updates. The bt_right_page_check_scankey
1086 : * check also provides a way of detecting cross-page inconsistencies for
1087 : * !readonly callers, but it can only detect sibling pages that have an
1088 : * out-of-order keyspace, which can't catch many of the problems that we
1089 : * expect to catch here.
1090 : *
1091 : * The classic example of the kind of inconsistency that we can only catch
1092 : * with this check (when in !readonly mode) involves three sibling pages that
1093 : * were affected by a faulty page split at some point in the past. The
1094 : * effects of the split are reflected in the original page and its new right
1095 : * sibling page, with a lack of any accompanying changes for the _original_
1096 : * right sibling page. The original right sibling page's left link fails to
1097 : * point to the new right sibling page (its left link still points to the
1098 : * original page), even though the first phase of a page split is supposed to
1099 : * work as a single atomic action. This subtle inconsistency will probably
1100 : * only break backwards scans in practice.
1101 : *
1102 : * Note that this is the only place where amcheck will "couple" buffer locks
1103 : * (and only for !readonly callers). In general we prefer to avoid more
1104 : * thorough cross-page checks in !readonly mode, but it seems worth the
1105 : * complexity here. Also, the performance overhead of performing lock
1106 : * coupling here is negligible in practice. Control only reaches here with a
1107 : * non-corrupt index when there is a concurrent page split at the instant
1108 : * caller crossed over to target page from leftcurrent page.
1109 : */
1110 : static void
1111 0 : bt_recheck_sibling_links(BtreeCheckState *state,
1112 : BlockNumber btpo_prev_from_target,
1113 : BlockNumber leftcurrent)
1114 : {
1115 : /* passing metapage to BTPageGetOpaque() would give irrelevant findings */
1116 : Assert(leftcurrent != P_NONE);
1117 :
1118 0 : if (!state->readonly)
1119 : {
1120 : Buffer lbuf;
1121 : Buffer newtargetbuf;
1122 : Page page;
1123 : BTPageOpaque opaque;
1124 : BlockNumber newtargetblock;
1125 :
1126 : /* Couple locks in the usual order for nbtree: Left to right */
1127 0 : lbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM, leftcurrent,
1128 : RBM_NORMAL, state->checkstrategy);
1129 0 : LockBuffer(lbuf, BT_READ);
1130 0 : _bt_checkpage(state->rel, lbuf);
1131 0 : page = BufferGetPage(lbuf);
1132 0 : opaque = BTPageGetOpaque(page);
1133 0 : if (P_ISDELETED(opaque))
1134 : {
1135 : /*
1136 : * Cannot reason about concurrently deleted page -- the left link
1137 : * in the page to the right is expected to point to some other
1138 : * page to the left (not leftcurrent page).
1139 : *
1140 : * Note that we deliberately don't give up with a half-dead page.
1141 : */
1142 0 : UnlockReleaseBuffer(lbuf);
1143 0 : return;
1144 : }
1145 :
1146 0 : newtargetblock = opaque->btpo_next;
1147 : /* Avoid self-deadlock when newtargetblock == leftcurrent */
1148 0 : if (newtargetblock != leftcurrent)
1149 : {
1150 0 : newtargetbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM,
1151 : newtargetblock, RBM_NORMAL,
1152 : state->checkstrategy);
1153 0 : LockBuffer(newtargetbuf, BT_READ);
1154 0 : _bt_checkpage(state->rel, newtargetbuf);
1155 0 : page = BufferGetPage(newtargetbuf);
1156 0 : opaque = BTPageGetOpaque(page);
1157 : /* btpo_prev_from_target may have changed; update it */
1158 0 : btpo_prev_from_target = opaque->btpo_prev;
1159 : }
1160 : else
1161 : {
1162 : /*
1163 : * leftcurrent right sibling points back to leftcurrent block.
1164 : * Index is corrupt. Easiest way to handle this is to pretend
1165 : * that we actually read from a distinct page that has an invalid
1166 : * block number in its btpo_prev.
1167 : */
1168 0 : newtargetbuf = InvalidBuffer;
1169 0 : btpo_prev_from_target = InvalidBlockNumber;
1170 : }
1171 :
1172 : /*
1173 : * No need to check P_ISDELETED here, since new target block cannot be
1174 : * marked deleted as long as we hold a lock on lbuf
1175 : */
1176 0 : if (BufferIsValid(newtargetbuf))
1177 0 : UnlockReleaseBuffer(newtargetbuf);
1178 0 : UnlockReleaseBuffer(lbuf);
1179 :
1180 0 : if (btpo_prev_from_target == leftcurrent)
1181 : {
1182 : /* Report split in left sibling, not target (or new target) */
1183 0 : ereport(DEBUG1,
1184 : (errcode(ERRCODE_INTERNAL_ERROR),
1185 : errmsg_internal("harmless concurrent page split detected in index \"%s\"",
1186 : RelationGetRelationName(state->rel)),
1187 : errdetail_internal("Block=%u new right sibling=%u original right sibling=%u.",
1188 : leftcurrent, newtargetblock,
1189 : state->targetblock)));
1190 0 : return;
1191 : }
1192 :
1193 : /*
1194 : * Index is corrupt. Make sure that we report correct target page.
1195 : *
1196 : * This could have changed in cases where there was a concurrent page
1197 : * split, as well as index corruption (at least in theory). Note that
1198 : * btpo_prev_from_target was already updated above.
1199 : */
1200 0 : state->targetblock = newtargetblock;
1201 : }
1202 :
1203 0 : ereport(ERROR,
1204 : (errcode(ERRCODE_INDEX_CORRUPTED),
1205 : errmsg("left link/right link pair in index \"%s\" not in agreement",
1206 : RelationGetRelationName(state->rel)),
1207 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
1208 : state->targetblock, leftcurrent,
1209 : btpo_prev_from_target)));
1210 : }
1211 :
1212 : /*
1213 : * Function performs the following checks on target page, or pages ancillary to
1214 : * target page:
1215 : *
1216 : * - That every "real" data item is less than or equal to the high key, which
1217 : * is an upper bound on the items on the page. Data items should be
1218 : * strictly less than the high key when the page is an internal page.
1219 : *
1220 : * - That within the page, every data item is strictly less than the item
1221 : * immediately to its right, if any (i.e., that the items are in order
1222 : * within the page, so that the binary searches performed by index scans are
1223 : * sane).
1224 : *
1225 : * - That the last data item stored on the page is strictly less than the
1226 : * first data item on the page to the right (when such a first item is
1227 : * available).
1228 : *
1229 : * - Various checks on the structure of tuples themselves. For example, check
1230 : * that non-pivot tuples have no truncated attributes.
1231 : *
1232 : * - For index with unique constraint make sure that only one of table entries
1233 : * for equal keys is visible.
1234 : *
1235 : * Furthermore, when state passed shows ShareLock held, function also checks:
1236 : *
1237 : * - That all child pages respect strict lower bound from parent's pivot
1238 : * tuple.
1239 : *
1240 : * - That downlink to block was encountered in parent where that's expected.
1241 : *
1242 : * - That high keys of child pages matches corresponding pivot keys in parent.
1243 : *
1244 : * This is also where heapallindexed callers use their Bloom filter to
1245 : * fingerprint IndexTuples for later table_index_build_scan() verification.
1246 : *
1247 : * Note: Memory allocated in this routine is expected to be released by caller
1248 : * resetting state->targetcontext.
1249 : */
1250 : static void
1251 18180 : bt_target_page_check(BtreeCheckState *state)
1252 : {
1253 : OffsetNumber offset;
1254 : OffsetNumber max;
1255 : BTPageOpaque topaque;
1256 :
1257 : /* Last visible entry info for checking indexes with unique constraint */
1258 18180 : BtreeLastVisibleEntry lVis = {InvalidBlockNumber, InvalidOffsetNumber, -1, NULL};
1259 :
1260 18180 : topaque = BTPageGetOpaque(state->target);
1261 18180 : max = PageGetMaxOffsetNumber(state->target);
1262 :
1263 18180 : elog(DEBUG2, "verifying %u items on %s block %u", max,
1264 : P_ISLEAF(topaque) ? "leaf" : "internal", state->targetblock);
1265 :
1266 : /*
1267 : * Check the number of attributes in high key. Note, rightmost page
1268 : * doesn't contain a high key, so nothing to check
1269 : */
1270 18180 : if (!P_RIGHTMOST(topaque))
1271 : {
1272 : ItemId itemid;
1273 : IndexTuple itup;
1274 :
1275 : /* Verify line pointer before checking tuple */
1276 13212 : itemid = PageGetItemIdCareful(state, state->targetblock,
1277 : state->target, P_HIKEY);
1278 13212 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1279 : P_HIKEY))
1280 : {
1281 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1282 0 : ereport(ERROR,
1283 : (errcode(ERRCODE_INDEX_CORRUPTED),
1284 : errmsg("wrong number of high key index tuple attributes in index \"%s\"",
1285 : RelationGetRelationName(state->rel)),
1286 : errdetail_internal("Index block=%u natts=%u block type=%s page lsn=%X/%X.",
1287 : state->targetblock,
1288 : BTreeTupleGetNAtts(itup, state->rel),
1289 : P_ISLEAF(topaque) ? "heap" : "index",
1290 : LSN_FORMAT_ARGS(state->targetlsn))));
1291 : }
1292 : }
1293 :
1294 : /*
1295 : * Loop over page items, starting from first non-highkey item, not high
1296 : * key (if any). Most tests are not performed for the "negative infinity"
1297 : * real item (if any).
1298 : */
1299 4051186 : for (offset = P_FIRSTDATAKEY(topaque);
1300 : offset <= max;
1301 4033006 : offset = OffsetNumberNext(offset))
1302 : {
1303 : ItemId itemid;
1304 : IndexTuple itup;
1305 : size_t tupsize;
1306 : BTScanInsert skey;
1307 : bool lowersizelimit;
1308 : ItemPointer scantid;
1309 :
1310 : /*
1311 : * True if we already called bt_entry_unique_check() for the current
1312 : * item. This helps to avoid visiting the heap for keys, which are
1313 : * anyway presented only once and can't comprise a unique violation.
1314 : */
1315 4033018 : bool unique_checked = false;
1316 :
1317 4033018 : CHECK_FOR_INTERRUPTS();
1318 :
1319 4033018 : itemid = PageGetItemIdCareful(state, state->targetblock,
1320 : state->target, offset);
1321 4033018 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1322 4033018 : tupsize = IndexTupleSize(itup);
1323 :
1324 : /*
1325 : * lp_len should match the IndexTuple reported length exactly, since
1326 : * lp_len is completely redundant in indexes, and both sources of
1327 : * tuple length are MAXALIGN()'d. nbtree does not use lp_len all that
1328 : * frequently, and is surprisingly tolerant of corrupt lp_len fields.
1329 : */
1330 4033018 : if (tupsize != ItemIdGetLength(itemid))
1331 0 : ereport(ERROR,
1332 : (errcode(ERRCODE_INDEX_CORRUPTED),
1333 : errmsg("index tuple size does not equal lp_len in index \"%s\"",
1334 : RelationGetRelationName(state->rel)),
1335 : errdetail_internal("Index tid=(%u,%u) tuple size=%zu lp_len=%u page lsn=%X/%X.",
1336 : state->targetblock, offset,
1337 : tupsize, ItemIdGetLength(itemid),
1338 : LSN_FORMAT_ARGS(state->targetlsn)),
1339 : errhint("This could be a torn page problem.")));
1340 :
1341 : /* Check the number of index tuple attributes */
1342 4033018 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1343 : offset))
1344 : {
1345 : ItemPointer tid;
1346 : char *itid,
1347 : *htid;
1348 :
1349 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1350 0 : tid = BTreeTupleGetPointsToTID(itup);
1351 0 : htid = psprintf("(%u,%u)",
1352 : ItemPointerGetBlockNumberNoCheck(tid),
1353 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1354 :
1355 0 : ereport(ERROR,
1356 : (errcode(ERRCODE_INDEX_CORRUPTED),
1357 : errmsg("wrong number of index tuple attributes in index \"%s\"",
1358 : RelationGetRelationName(state->rel)),
1359 : errdetail_internal("Index tid=%s natts=%u points to %s tid=%s page lsn=%X/%X.",
1360 : itid,
1361 : BTreeTupleGetNAtts(itup, state->rel),
1362 : P_ISLEAF(topaque) ? "heap" : "index",
1363 : htid,
1364 : LSN_FORMAT_ARGS(state->targetlsn))));
1365 : }
1366 :
1367 : /*
1368 : * Don't try to generate scankey using "negative infinity" item on
1369 : * internal pages. They are always truncated to zero attributes.
1370 : */
1371 4033018 : if (offset_is_negative_infinity(topaque, offset))
1372 : {
1373 : /*
1374 : * We don't call bt_child_check() for "negative infinity" items.
1375 : * But if we're performing downlink connectivity check, we do it
1376 : * for every item including "negative infinity" one.
1377 : */
1378 1114 : if (!P_ISLEAF(topaque) && state->readonly)
1379 : {
1380 24 : bt_child_highkey_check(state,
1381 : offset,
1382 : NULL,
1383 : topaque->btpo_level);
1384 : }
1385 1114 : continue;
1386 : }
1387 :
1388 : /*
1389 : * Readonly callers may optionally verify that non-pivot tuples can
1390 : * each be found by an independent search that starts from the root.
1391 : * Note that we deliberately don't do individual searches for each
1392 : * TID, since the posting list itself is validated by other checks.
1393 : */
1394 4031904 : if (state->rootdescend && P_ISLEAF(topaque) &&
1395 402196 : !bt_rootdescend(state, itup))
1396 : {
1397 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1398 : char *itid,
1399 : *htid;
1400 :
1401 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1402 0 : htid = psprintf("(%u,%u)", ItemPointerGetBlockNumber(tid),
1403 0 : ItemPointerGetOffsetNumber(tid));
1404 :
1405 0 : ereport(ERROR,
1406 : (errcode(ERRCODE_INDEX_CORRUPTED),
1407 : errmsg("could not find tuple using search from root page in index \"%s\"",
1408 : RelationGetRelationName(state->rel)),
1409 : errdetail_internal("Index tid=%s points to heap tid=%s page lsn=%X/%X.",
1410 : itid, htid,
1411 : LSN_FORMAT_ARGS(state->targetlsn))));
1412 : }
1413 :
1414 : /*
1415 : * If tuple is a posting list tuple, make sure posting list TIDs are
1416 : * in order
1417 : */
1418 4031904 : if (BTreeTupleIsPosting(itup))
1419 : {
1420 : ItemPointerData last;
1421 : ItemPointer current;
1422 :
1423 21808 : ItemPointerCopy(BTreeTupleGetHeapTID(itup), &last);
1424 :
1425 157670 : for (int i = 1; i < BTreeTupleGetNPosting(itup); i++)
1426 : {
1427 :
1428 135862 : current = BTreeTupleGetPostingN(itup, i);
1429 :
1430 135862 : if (ItemPointerCompare(current, &last) <= 0)
1431 : {
1432 0 : char *itid = psprintf("(%u,%u)", state->targetblock, offset);
1433 :
1434 0 : ereport(ERROR,
1435 : (errcode(ERRCODE_INDEX_CORRUPTED),
1436 : errmsg_internal("posting list contains misplaced TID in index \"%s\"",
1437 : RelationGetRelationName(state->rel)),
1438 : errdetail_internal("Index tid=%s posting list offset=%d page lsn=%X/%X.",
1439 : itid, i,
1440 : LSN_FORMAT_ARGS(state->targetlsn))));
1441 : }
1442 :
1443 135862 : ItemPointerCopy(current, &last);
1444 : }
1445 : }
1446 :
1447 : /* Build insertion scankey for current page offset */
1448 4031904 : skey = bt_mkscankey_pivotsearch(state->rel, itup);
1449 :
1450 : /*
1451 : * Make sure tuple size does not exceed the relevant BTREE_VERSION
1452 : * specific limit.
1453 : *
1454 : * BTREE_VERSION 4 (which introduced heapkeyspace rules) requisitioned
1455 : * a small amount of space from BTMaxItemSize() in order to ensure
1456 : * that suffix truncation always has enough space to add an explicit
1457 : * heap TID back to a tuple -- we pessimistically assume that every
1458 : * newly inserted tuple will eventually need to have a heap TID
1459 : * appended during a future leaf page split, when the tuple becomes
1460 : * the basis of the new high key (pivot tuple) for the leaf page.
1461 : *
1462 : * Since the reclaimed space is reserved for that purpose, we must not
1463 : * enforce the slightly lower limit when the extra space has been used
1464 : * as intended. In other words, there is only a cross-version
1465 : * difference in the limit on tuple size within leaf pages.
1466 : *
1467 : * Still, we're particular about the details within BTREE_VERSION 4
1468 : * internal pages. Pivot tuples may only use the extra space for its
1469 : * designated purpose. Enforce the lower limit for pivot tuples when
1470 : * an explicit heap TID isn't actually present. (In all other cases
1471 : * suffix truncation is guaranteed to generate a pivot tuple that's no
1472 : * larger than the firstright tuple provided to it by its caller.)
1473 : */
1474 8063808 : lowersizelimit = skey->heapkeyspace &&
1475 4031904 : (P_ISLEAF(topaque) || BTreeTupleGetHeapTID(itup) == NULL);
1476 4031904 : if (tupsize > (lowersizelimit ? BTMaxItemSize : BTMaxItemSizeNoHeapTid))
1477 : {
1478 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1479 : char *itid,
1480 : *htid;
1481 :
1482 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1483 0 : htid = psprintf("(%u,%u)",
1484 : ItemPointerGetBlockNumberNoCheck(tid),
1485 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1486 :
1487 0 : ereport(ERROR,
1488 : (errcode(ERRCODE_INDEX_CORRUPTED),
1489 : errmsg("index row size %zu exceeds maximum for index \"%s\"",
1490 : tupsize, RelationGetRelationName(state->rel)),
1491 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%X.",
1492 : itid,
1493 : P_ISLEAF(topaque) ? "heap" : "index",
1494 : htid,
1495 : LSN_FORMAT_ARGS(state->targetlsn))));
1496 : }
1497 :
1498 : /* Fingerprint leaf page tuples (those that point to the heap) */
1499 4031904 : if (state->heapallindexed && P_ISLEAF(topaque) && !ItemIdIsDead(itemid))
1500 : {
1501 : IndexTuple norm;
1502 :
1503 1013828 : if (BTreeTupleIsPosting(itup))
1504 : {
1505 : /* Fingerprint all elements as distinct "plain" tuples */
1506 52106 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
1507 : {
1508 : IndexTuple logtuple;
1509 :
1510 51750 : logtuple = bt_posting_plain_tuple(itup, i);
1511 51750 : norm = bt_normalize_tuple(state, logtuple);
1512 51750 : bloom_add_element(state->filter, (unsigned char *) norm,
1513 : IndexTupleSize(norm));
1514 : /* Be tidy */
1515 51750 : if (norm != logtuple)
1516 4 : pfree(norm);
1517 51750 : pfree(logtuple);
1518 : }
1519 : }
1520 : else
1521 : {
1522 1013472 : norm = bt_normalize_tuple(state, itup);
1523 1013472 : bloom_add_element(state->filter, (unsigned char *) norm,
1524 : IndexTupleSize(norm));
1525 : /* Be tidy */
1526 1013472 : if (norm != itup)
1527 2 : pfree(norm);
1528 : }
1529 : }
1530 :
1531 : /*
1532 : * * High key check *
1533 : *
1534 : * If there is a high key (if this is not the rightmost page on its
1535 : * entire level), check that high key actually is upper bound on all
1536 : * page items. If this is a posting list tuple, we'll need to set
1537 : * scantid to be highest TID in posting list.
1538 : *
1539 : * We prefer to check all items against high key rather than checking
1540 : * just the last and trusting that the operator class obeys the
1541 : * transitive law (which implies that all previous items also
1542 : * respected the high key invariant if they pass the item order
1543 : * check).
1544 : *
1545 : * Ideally, we'd compare every item in the index against every other
1546 : * item in the index, and not trust opclass obedience of the
1547 : * transitive law to bridge the gap between children and their
1548 : * grandparents (as well as great-grandparents, and so on). We don't
1549 : * go to those lengths because that would be prohibitively expensive,
1550 : * and probably not markedly more effective in practice.
1551 : *
1552 : * On the leaf level, we check that the key is <= the highkey.
1553 : * However, on non-leaf levels we check that the key is < the highkey,
1554 : * because the high key is "just another separator" rather than a copy
1555 : * of some existing key item; we expect it to be unique among all keys
1556 : * on the same level. (Suffix truncation will sometimes produce a
1557 : * leaf highkey that is an untruncated copy of the lastleft item, but
1558 : * never any other item, which necessitates weakening the leaf level
1559 : * check to <=.)
1560 : *
1561 : * Full explanation for why a highkey is never truly a copy of another
1562 : * item from the same level on internal levels:
1563 : *
1564 : * While the new left page's high key is copied from the first offset
1565 : * on the right page during an internal page split, that's not the
1566 : * full story. In effect, internal pages are split in the middle of
1567 : * the firstright tuple, not between the would-be lastleft and
1568 : * firstright tuples: the firstright key ends up on the left side as
1569 : * left's new highkey, and the firstright downlink ends up on the
1570 : * right side as right's new "negative infinity" item. The negative
1571 : * infinity tuple is truncated to zero attributes, so we're only left
1572 : * with the downlink. In other words, the copying is just an
1573 : * implementation detail of splitting in the middle of a (pivot)
1574 : * tuple. (See also: "Notes About Data Representation" in the nbtree
1575 : * README.)
1576 : */
1577 4031904 : scantid = skey->scantid;
1578 4031904 : if (state->heapkeyspace && BTreeTupleIsPosting(itup))
1579 21808 : skey->scantid = BTreeTupleGetMaxHeapTID(itup);
1580 :
1581 7731452 : if (!P_RIGHTMOST(topaque) &&
1582 3699548 : !(P_ISLEAF(topaque) ? invariant_leq_offset(state, skey, P_HIKEY) :
1583 1132 : invariant_l_offset(state, skey, P_HIKEY)))
1584 : {
1585 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1586 : char *itid,
1587 : *htid;
1588 :
1589 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1590 0 : htid = psprintf("(%u,%u)",
1591 : ItemPointerGetBlockNumberNoCheck(tid),
1592 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1593 :
1594 0 : ereport(ERROR,
1595 : (errcode(ERRCODE_INDEX_CORRUPTED),
1596 : errmsg("high key invariant violated for index \"%s\"",
1597 : RelationGetRelationName(state->rel)),
1598 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%X.",
1599 : itid,
1600 : P_ISLEAF(topaque) ? "heap" : "index",
1601 : htid,
1602 : LSN_FORMAT_ARGS(state->targetlsn))));
1603 : }
1604 : /* Reset, in case scantid was set to (itup) posting tuple's max TID */
1605 4031904 : skey->scantid = scantid;
1606 :
1607 : /*
1608 : * * Item order check *
1609 : *
1610 : * Check that items are stored on page in logical order, by checking
1611 : * current item is strictly less than next item (if any).
1612 : */
1613 4031904 : if (OffsetNumberNext(offset) <= max &&
1614 4013740 : !invariant_l_offset(state, skey, OffsetNumberNext(offset)))
1615 : {
1616 : ItemPointer tid;
1617 : char *itid,
1618 : *htid,
1619 : *nitid,
1620 : *nhtid;
1621 :
1622 6 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1623 6 : tid = BTreeTupleGetPointsToTID(itup);
1624 6 : htid = psprintf("(%u,%u)",
1625 : ItemPointerGetBlockNumberNoCheck(tid),
1626 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1627 6 : nitid = psprintf("(%u,%u)", state->targetblock,
1628 6 : OffsetNumberNext(offset));
1629 :
1630 : /* Reuse itup to get pointed-to heap location of second item */
1631 6 : itemid = PageGetItemIdCareful(state, state->targetblock,
1632 : state->target,
1633 6 : OffsetNumberNext(offset));
1634 6 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1635 6 : tid = BTreeTupleGetPointsToTID(itup);
1636 6 : nhtid = psprintf("(%u,%u)",
1637 : ItemPointerGetBlockNumberNoCheck(tid),
1638 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1639 :
1640 6 : ereport(ERROR,
1641 : (errcode(ERRCODE_INDEX_CORRUPTED),
1642 : errmsg("item order invariant violated for index \"%s\"",
1643 : RelationGetRelationName(state->rel)),
1644 : errdetail_internal("Lower index tid=%s (points to %s tid=%s) "
1645 : "higher index tid=%s (points to %s tid=%s) "
1646 : "page lsn=%X/%X.",
1647 : itid,
1648 : P_ISLEAF(topaque) ? "heap" : "index",
1649 : htid,
1650 : nitid,
1651 : P_ISLEAF(topaque) ? "heap" : "index",
1652 : nhtid,
1653 : LSN_FORMAT_ARGS(state->targetlsn))));
1654 : }
1655 :
1656 : /*
1657 : * If the index is unique verify entries uniqueness by checking the
1658 : * heap tuples visibility. Immediately check posting tuples and
1659 : * tuples with repeated keys. Postpone check for keys, which have the
1660 : * first appearance.
1661 : */
1662 4031898 : if (state->checkunique && state->indexinfo->ii_Unique &&
1663 573908 : P_ISLEAF(topaque) && !skey->anynullkeys &&
1664 572720 : (BTreeTupleIsPosting(itup) || ItemPointerIsValid(lVis.tid)))
1665 : {
1666 24 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1667 : &lVis);
1668 18 : unique_checked = true;
1669 : }
1670 :
1671 4031892 : if (state->checkunique && state->indexinfo->ii_Unique &&
1672 287542 : P_ISLEAF(topaque) && OffsetNumberNext(offset) <= max)
1673 : {
1674 : /* Save current scankey tid */
1675 284914 : scantid = skey->scantid;
1676 :
1677 : /*
1678 : * Invalidate scankey tid to make _bt_compare compare only keys in
1679 : * the item to report equality even if heap TIDs are different
1680 : */
1681 284914 : skey->scantid = NULL;
1682 :
1683 : /*
1684 : * If next key tuple is different, invalidate last visible entry
1685 : * data (whole index tuple or last posting in index tuple). Key
1686 : * containing null value does not violate unique constraint and
1687 : * treated as different to any other key.
1688 : *
1689 : * If the next key is the same as the previous one, do the
1690 : * bt_entry_unique_check() call if it was postponed.
1691 : */
1692 284914 : if (_bt_compare(state->rel, skey, state->target,
1693 285706 : OffsetNumberNext(offset)) != 0 || skey->anynullkeys)
1694 : {
1695 284262 : lVis.blkno = InvalidBlockNumber;
1696 284262 : lVis.offset = InvalidOffsetNumber;
1697 284262 : lVis.postingIndex = -1;
1698 284262 : lVis.tid = NULL;
1699 : }
1700 652 : else if (!unique_checked)
1701 : {
1702 652 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1703 : &lVis);
1704 : }
1705 284914 : skey->scantid = scantid; /* Restore saved scan key state */
1706 : }
1707 :
1708 : /*
1709 : * * Last item check *
1710 : *
1711 : * Check last item against next/right page's first data item's when
1712 : * last item on page is reached. This additional check will detect
1713 : * transposed pages iff the supposed right sibling page happens to
1714 : * belong before target in the key space. (Otherwise, a subsequent
1715 : * heap verification will probably detect the problem.)
1716 : *
1717 : * This check is similar to the item order check that will have
1718 : * already been performed for every other "real" item on target page
1719 : * when last item is checked. The difference is that the next item
1720 : * (the item that is compared to target's last item) needs to come
1721 : * from the next/sibling page. There may not be such an item
1722 : * available from sibling for various reasons, though (e.g., target is
1723 : * the rightmost page on level).
1724 : */
1725 4031892 : if (offset == max)
1726 : {
1727 : BTScanInsert rightkey;
1728 :
1729 : /* first offset on a right index page (log only) */
1730 18164 : OffsetNumber rightfirstoffset = InvalidOffsetNumber;
1731 :
1732 : /* Get item in next/right page */
1733 18164 : rightkey = bt_right_page_check_scankey(state, &rightfirstoffset);
1734 :
1735 18164 : if (rightkey &&
1736 13204 : !invariant_g_offset(state, rightkey, max))
1737 : {
1738 : /*
1739 : * As explained at length in bt_right_page_check_scankey(),
1740 : * there is a known !readonly race that could account for
1741 : * apparent violation of invariant, which we must check for
1742 : * before actually proceeding with raising error. Our canary
1743 : * condition is that target page was deleted.
1744 : */
1745 0 : if (!state->readonly)
1746 : {
1747 : /* Get fresh copy of target page */
1748 0 : state->target = palloc_btree_page(state, state->targetblock);
1749 : /* Note that we deliberately do not update target LSN */
1750 0 : topaque = BTPageGetOpaque(state->target);
1751 :
1752 : /*
1753 : * All !readonly checks now performed; just return
1754 : */
1755 0 : if (P_IGNORE(topaque))
1756 0 : return;
1757 : }
1758 :
1759 0 : ereport(ERROR,
1760 : (errcode(ERRCODE_INDEX_CORRUPTED),
1761 : errmsg("cross page item order invariant violated for index \"%s\"",
1762 : RelationGetRelationName(state->rel)),
1763 : errdetail_internal("Last item on page tid=(%u,%u) page lsn=%X/%X.",
1764 : state->targetblock, offset,
1765 : LSN_FORMAT_ARGS(state->targetlsn))));
1766 : }
1767 :
1768 : /*
1769 : * If index has unique constraint make sure that no more than one
1770 : * found equal items is visible.
1771 : */
1772 18164 : if (state->checkunique && state->indexinfo->ii_Unique &&
1773 1024 : rightkey && P_ISLEAF(topaque) && !P_RIGHTMOST(topaque))
1774 : {
1775 1024 : BlockNumber rightblock_number = topaque->btpo_next;
1776 :
1777 1024 : elog(DEBUG2, "check cross page unique condition");
1778 :
1779 : /*
1780 : * Make _bt_compare compare only index keys without heap TIDs.
1781 : * rightkey->scantid is modified destructively but it is ok
1782 : * for it is not used later.
1783 : */
1784 1024 : rightkey->scantid = NULL;
1785 :
1786 : /* The first key on the next page is the same */
1787 1024 : if (_bt_compare(state->rel, rightkey, state->target, max) == 0 &&
1788 14 : !rightkey->anynullkeys)
1789 : {
1790 : Page rightpage;
1791 :
1792 : /*
1793 : * Do the bt_entry_unique_check() call if it was
1794 : * postponed.
1795 : */
1796 0 : if (!unique_checked)
1797 0 : bt_entry_unique_check(state, itup, state->targetblock,
1798 : offset, &lVis);
1799 :
1800 0 : elog(DEBUG2, "cross page equal keys");
1801 0 : rightpage = palloc_btree_page(state,
1802 : rightblock_number);
1803 0 : topaque = BTPageGetOpaque(rightpage);
1804 :
1805 0 : if (P_IGNORE(topaque))
1806 : {
1807 0 : pfree(rightpage);
1808 0 : break;
1809 : }
1810 :
1811 0 : if (unlikely(!P_ISLEAF(topaque)))
1812 0 : ereport(ERROR,
1813 : (errcode(ERRCODE_INDEX_CORRUPTED),
1814 : errmsg("right block of leaf block is non-leaf for index \"%s\"",
1815 : RelationGetRelationName(state->rel)),
1816 : errdetail_internal("Block=%u page lsn=%X/%X.",
1817 : state->targetblock,
1818 : LSN_FORMAT_ARGS(state->targetlsn))));
1819 :
1820 0 : itemid = PageGetItemIdCareful(state, rightblock_number,
1821 : rightpage,
1822 : rightfirstoffset);
1823 0 : itup = (IndexTuple) PageGetItem(rightpage, itemid);
1824 :
1825 0 : bt_entry_unique_check(state, itup, rightblock_number, rightfirstoffset, &lVis);
1826 :
1827 0 : pfree(rightpage);
1828 : }
1829 : }
1830 : }
1831 :
1832 : /*
1833 : * * Downlink check *
1834 : *
1835 : * Additional check of child items iff this is an internal page and
1836 : * caller holds a ShareLock. This happens for every downlink (item)
1837 : * in target excluding the negative-infinity downlink (again, this is
1838 : * because it has no useful value to compare).
1839 : */
1840 4031892 : if (!P_ISLEAF(topaque) && state->readonly)
1841 3722 : bt_child_check(state, skey, offset);
1842 : }
1843 :
1844 : /*
1845 : * Special case bt_child_highkey_check() call
1846 : *
1847 : * We don't pass a real downlink, but we've to finish the level
1848 : * processing. If condition is satisfied, we've already processed all the
1849 : * downlinks from the target level. But there still might be pages to the
1850 : * right of the child page pointer to by our rightmost downlink. And they
1851 : * might have missing downlinks. This final call checks for them.
1852 : */
1853 18168 : if (!P_ISLEAF(topaque) && P_RIGHTMOST(topaque) && state->readonly)
1854 : {
1855 22 : bt_child_highkey_check(state, InvalidOffsetNumber,
1856 : NULL, topaque->btpo_level);
1857 : }
1858 : }
1859 :
1860 : /*
1861 : * Return a scankey for an item on page to right of current target (or the
1862 : * first non-ignorable page), sufficient to check ordering invariant on last
1863 : * item in current target page. Returned scankey relies on local memory
1864 : * allocated for the child page, which caller cannot pfree(). Caller's memory
1865 : * context should be reset between calls here.
1866 : *
1867 : * This is the first data item, and so all adjacent items are checked against
1868 : * their immediate sibling item (which may be on a sibling page, or even a
1869 : * "cousin" page at parent boundaries where target's rightlink points to page
1870 : * with different parent page). If no such valid item is available, return
1871 : * NULL instead.
1872 : *
1873 : * Note that !readonly callers must reverify that target page has not
1874 : * been concurrently deleted.
1875 : *
1876 : * Save rightfirstoffset for detailed error message.
1877 : */
1878 : static BTScanInsert
1879 18164 : bt_right_page_check_scankey(BtreeCheckState *state, OffsetNumber *rightfirstoffset)
1880 : {
1881 : BTPageOpaque opaque;
1882 : ItemId rightitem;
1883 : IndexTuple firstitup;
1884 : BlockNumber targetnext;
1885 : Page rightpage;
1886 : OffsetNumber nline;
1887 :
1888 : /* Determine target's next block number */
1889 18164 : opaque = BTPageGetOpaque(state->target);
1890 :
1891 : /* If target is already rightmost, no right sibling; nothing to do here */
1892 18164 : if (P_RIGHTMOST(opaque))
1893 4960 : return NULL;
1894 :
1895 : /*
1896 : * General notes on concurrent page splits and page deletion:
1897 : *
1898 : * Routines like _bt_search() don't require *any* page split interlock
1899 : * when descending the tree, including something very light like a buffer
1900 : * pin. That's why it's okay that we don't either. This avoidance of any
1901 : * need to "couple" buffer locks is the raison d' etre of the Lehman & Yao
1902 : * algorithm, in fact.
1903 : *
1904 : * That leaves deletion. A deleted page won't actually be recycled by
1905 : * VACUUM early enough for us to fail to at least follow its right link
1906 : * (or left link, or downlink) and find its sibling, because recycling
1907 : * does not occur until no possible index scan could land on the page.
1908 : * Index scans can follow links with nothing more than their snapshot as
1909 : * an interlock and be sure of at least that much. (See page
1910 : * recycling/"visible to everyone" notes in nbtree README.)
1911 : *
1912 : * Furthermore, it's okay if we follow a rightlink and find a half-dead or
1913 : * dead (ignorable) page one or more times. There will either be a
1914 : * further right link to follow that leads to a live page before too long
1915 : * (before passing by parent's rightmost child), or we will find the end
1916 : * of the entire level instead (possible when parent page is itself the
1917 : * rightmost on its level).
1918 : */
1919 13204 : targetnext = opaque->btpo_next;
1920 : for (;;)
1921 : {
1922 13204 : CHECK_FOR_INTERRUPTS();
1923 :
1924 13204 : rightpage = palloc_btree_page(state, targetnext);
1925 13204 : opaque = BTPageGetOpaque(rightpage);
1926 :
1927 13204 : if (!P_IGNORE(opaque) || P_RIGHTMOST(opaque))
1928 : break;
1929 :
1930 : /*
1931 : * We landed on a deleted or half-dead sibling page. Step right until
1932 : * we locate a live sibling page.
1933 : */
1934 0 : ereport(DEBUG2,
1935 : (errcode(ERRCODE_NO_DATA),
1936 : errmsg_internal("level %u sibling page in block %u of index \"%s\" was found deleted or half dead",
1937 : opaque->btpo_level, targetnext, RelationGetRelationName(state->rel)),
1938 : errdetail_internal("Deleted page found when building scankey from right sibling.")));
1939 :
1940 0 : targetnext = opaque->btpo_next;
1941 :
1942 : /* Be slightly more pro-active in freeing this memory, just in case */
1943 0 : pfree(rightpage);
1944 : }
1945 :
1946 : /*
1947 : * No ShareLock held case -- why it's safe to proceed.
1948 : *
1949 : * Problem:
1950 : *
1951 : * We must avoid false positive reports of corruption when caller treats
1952 : * item returned here as an upper bound on target's last item. In
1953 : * general, false positives are disallowed. Avoiding them here when
1954 : * caller is !readonly is subtle.
1955 : *
1956 : * A concurrent page deletion by VACUUM of the target page can result in
1957 : * the insertion of items on to this right sibling page that would
1958 : * previously have been inserted on our target page. There might have
1959 : * been insertions that followed the target's downlink after it was made
1960 : * to point to right sibling instead of target by page deletion's first
1961 : * phase. The inserters insert items that would belong on target page.
1962 : * This race is very tight, but it's possible. This is our only problem.
1963 : *
1964 : * Non-problems:
1965 : *
1966 : * We are not hindered by a concurrent page split of the target; we'll
1967 : * never land on the second half of the page anyway. A concurrent split
1968 : * of the right page will also not matter, because the first data item
1969 : * remains the same within the left half, which we'll reliably land on. If
1970 : * we had to skip over ignorable/deleted pages, it cannot matter because
1971 : * their key space has already been atomically merged with the first
1972 : * non-ignorable page we eventually find (doesn't matter whether the page
1973 : * we eventually find is a true sibling or a cousin of target, which we go
1974 : * into below).
1975 : *
1976 : * Solution:
1977 : *
1978 : * Caller knows that it should reverify that target is not ignorable
1979 : * (half-dead or deleted) when cross-page sibling item comparison appears
1980 : * to indicate corruption (invariant fails). This detects the single race
1981 : * condition that exists for caller. This is correct because the
1982 : * continued existence of target block as non-ignorable (not half-dead or
1983 : * deleted) implies that target page was not merged into from the right by
1984 : * deletion; the key space at or after target never moved left. Target's
1985 : * parent either has the same downlink to target as before, or a <
1986 : * downlink due to deletion at the left of target. Target either has the
1987 : * same highkey as before, or a highkey < before when there is a page
1988 : * split. (The rightmost concurrently-split-from-target-page page will
1989 : * still have the same highkey as target was originally found to have,
1990 : * which for our purposes is equivalent to target's highkey itself never
1991 : * changing, since we reliably skip over
1992 : * concurrently-split-from-target-page pages.)
1993 : *
1994 : * In simpler terms, we allow that the key space of the target may expand
1995 : * left (the key space can move left on the left side of target only), but
1996 : * the target key space cannot expand right and get ahead of us without
1997 : * our detecting it. The key space of the target cannot shrink, unless it
1998 : * shrinks to zero due to the deletion of the original page, our canary
1999 : * condition. (To be very precise, we're a bit stricter than that because
2000 : * it might just have been that the target page split and only the
2001 : * original target page was deleted. We can be more strict, just not more
2002 : * lax.)
2003 : *
2004 : * Top level tree walk caller moves on to next page (makes it the new
2005 : * target) following recovery from this race. (cf. The rationale for
2006 : * child/downlink verification needing a ShareLock within
2007 : * bt_child_check(), where page deletion is also the main source of
2008 : * trouble.)
2009 : *
2010 : * Note that it doesn't matter if right sibling page here is actually a
2011 : * cousin page, because in order for the key space to be readjusted in a
2012 : * way that causes us issues in next level up (guiding problematic
2013 : * concurrent insertions to the cousin from the grandparent rather than to
2014 : * the sibling from the parent), there'd have to be page deletion of
2015 : * target's parent page (affecting target's parent's downlink in target's
2016 : * grandparent page). Internal page deletion only occurs when there are
2017 : * no child pages (they were all fully deleted), and caller is checking
2018 : * that the target's parent has at least one non-deleted (so
2019 : * non-ignorable) child: the target page. (Note that the first phase of
2020 : * deletion atomically marks the page to be deleted half-dead/ignorable at
2021 : * the same time downlink in its parent is removed, so caller will
2022 : * definitely not fail to detect that this happened.)
2023 : *
2024 : * This trick is inspired by the method backward scans use for dealing
2025 : * with concurrent page splits; concurrent page deletion is a problem that
2026 : * similarly receives special consideration sometimes (it's possible that
2027 : * the backwards scan will re-read its "original" block after failing to
2028 : * find a right-link to it, having already moved in the opposite direction
2029 : * (right/"forwards") a few times to try to locate one). Just like us,
2030 : * that happens only to determine if there was a concurrent page deletion
2031 : * of a reference page, and just like us if there was a page deletion of
2032 : * that reference page it means we can move on from caring about the
2033 : * reference page. See the nbtree README for a full description of how
2034 : * that works.
2035 : */
2036 13204 : nline = PageGetMaxOffsetNumber(rightpage);
2037 :
2038 : /*
2039 : * Get first data item, if any
2040 : */
2041 13204 : if (P_ISLEAF(opaque) && nline >= P_FIRSTDATAKEY(opaque))
2042 : {
2043 : /* Return first data item (if any) */
2044 13200 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2045 13200 : P_FIRSTDATAKEY(opaque));
2046 13200 : *rightfirstoffset = P_FIRSTDATAKEY(opaque);
2047 : }
2048 8 : else if (!P_ISLEAF(opaque) &&
2049 4 : nline >= OffsetNumberNext(P_FIRSTDATAKEY(opaque)))
2050 : {
2051 : /*
2052 : * Return first item after the internal page's "negative infinity"
2053 : * item
2054 : */
2055 4 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2056 4 : OffsetNumberNext(P_FIRSTDATAKEY(opaque)));
2057 : }
2058 : else
2059 : {
2060 : /*
2061 : * No first item. Page is probably empty leaf page, but it's also
2062 : * possible that it's an internal page with only a negative infinity
2063 : * item.
2064 : */
2065 0 : ereport(DEBUG2,
2066 : (errcode(ERRCODE_NO_DATA),
2067 : errmsg_internal("%s block %u of index \"%s\" has no first data item",
2068 : P_ISLEAF(opaque) ? "leaf" : "internal", targetnext,
2069 : RelationGetRelationName(state->rel))));
2070 0 : return NULL;
2071 : }
2072 :
2073 : /*
2074 : * Return first real item scankey. Note that this relies on right page
2075 : * memory remaining allocated.
2076 : */
2077 13204 : firstitup = (IndexTuple) PageGetItem(rightpage, rightitem);
2078 13204 : return bt_mkscankey_pivotsearch(state->rel, firstitup);
2079 : }
2080 :
2081 : /*
2082 : * Check if two tuples are binary identical except the block number. So,
2083 : * this function is capable to compare pivot keys on different levels.
2084 : */
2085 : static bool
2086 3724 : bt_pivot_tuple_identical(bool heapkeyspace, IndexTuple itup1, IndexTuple itup2)
2087 : {
2088 3724 : if (IndexTupleSize(itup1) != IndexTupleSize(itup2))
2089 0 : return false;
2090 :
2091 3724 : if (heapkeyspace)
2092 : {
2093 : /*
2094 : * Offset number will contain important information in heapkeyspace
2095 : * indexes: the number of attributes left in the pivot tuple following
2096 : * suffix truncation. Don't skip over it (compare it too).
2097 : */
2098 3724 : if (memcmp(&itup1->t_tid.ip_posid, &itup2->t_tid.ip_posid,
2099 3724 : IndexTupleSize(itup1) -
2100 : offsetof(ItemPointerData, ip_posid)) != 0)
2101 0 : return false;
2102 : }
2103 : else
2104 : {
2105 : /*
2106 : * Cannot rely on offset number field having consistent value across
2107 : * levels on pg_upgrade'd !heapkeyspace indexes. Compare contents of
2108 : * tuple starting from just after item pointer (i.e. after block
2109 : * number and offset number).
2110 : */
2111 0 : if (memcmp(&itup1->t_info, &itup2->t_info,
2112 0 : IndexTupleSize(itup1) -
2113 : offsetof(IndexTupleData, t_info)) != 0)
2114 0 : return false;
2115 : }
2116 :
2117 3724 : return true;
2118 : }
2119 :
2120 : /*---
2121 : * Check high keys on the child level. Traverse rightlinks from previous
2122 : * downlink to the current one. Check that there are no intermediate pages
2123 : * with missing downlinks.
2124 : *
2125 : * If 'loaded_child' is given, it's assumed to be the page pointed to by the
2126 : * downlink referenced by 'downlinkoffnum' of the target page.
2127 : *
2128 : * Basically this function is called for each target downlink and checks two
2129 : * invariants:
2130 : *
2131 : * 1) You can reach the next child from previous one via rightlinks;
2132 : * 2) Each child high key have matching pivot key on target level.
2133 : *
2134 : * Consider the sample tree picture.
2135 : *
2136 : * 1
2137 : * / \
2138 : * 2 <-> 3
2139 : * / \ / \
2140 : * 4 <> 5 <> 6 <> 7 <> 8
2141 : *
2142 : * This function will be called for blocks 4, 5, 6 and 8. Consider what is
2143 : * happening for each function call.
2144 : *
2145 : * - The function call for block 4 initializes data structure and matches high
2146 : * key of block 4 to downlink's pivot key of block 2.
2147 : * - The high key of block 5 is matched to the high key of block 2.
2148 : * - The block 6 has an incomplete split flag set, so its high key isn't
2149 : * matched to anything.
2150 : * - The function call for block 8 checks that block 8 can be found while
2151 : * following rightlinks from block 6. The high key of block 7 will be
2152 : * matched to downlink's pivot key in block 3.
2153 : *
2154 : * There is also final call of this function, which checks that there is no
2155 : * missing downlinks for children to the right of the child referenced by
2156 : * rightmost downlink in target level.
2157 : */
2158 : static void
2159 3768 : bt_child_highkey_check(BtreeCheckState *state,
2160 : OffsetNumber target_downlinkoffnum,
2161 : Page loaded_child,
2162 : uint32 target_level)
2163 : {
2164 3768 : BlockNumber blkno = state->prevrightlink;
2165 : Page page;
2166 : BTPageOpaque opaque;
2167 3768 : bool rightsplit = state->previncompletesplit;
2168 3768 : bool first = true;
2169 : ItemId itemid;
2170 : IndexTuple itup;
2171 : BlockNumber downlink;
2172 :
2173 3768 : if (OffsetNumberIsValid(target_downlinkoffnum))
2174 : {
2175 3746 : itemid = PageGetItemIdCareful(state, state->targetblock,
2176 : state->target, target_downlinkoffnum);
2177 3746 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2178 3746 : downlink = BTreeTupleGetDownLink(itup);
2179 : }
2180 : else
2181 : {
2182 22 : downlink = P_NONE;
2183 : }
2184 :
2185 : /*
2186 : * If no previous rightlink is memorized for current level just below
2187 : * target page's level, we are about to start from the leftmost page. We
2188 : * can't follow rightlinks from previous page, because there is no
2189 : * previous page. But we still can match high key.
2190 : *
2191 : * So we initialize variables for the loop above like there is previous
2192 : * page referencing current child. Also we imply previous page to not
2193 : * have incomplete split flag, that would make us require downlink for
2194 : * current child. That's correct, because leftmost page on the level
2195 : * should always have parent downlink.
2196 : */
2197 3768 : if (!BlockNumberIsValid(blkno))
2198 : {
2199 22 : blkno = downlink;
2200 22 : rightsplit = false;
2201 : }
2202 :
2203 : /* Move to the right on the child level */
2204 : while (true)
2205 : {
2206 : /*
2207 : * Did we traverse the whole tree level and this is check for pages to
2208 : * the right of rightmost downlink?
2209 : */
2210 3768 : if (blkno == P_NONE && downlink == P_NONE)
2211 : {
2212 22 : state->prevrightlink = InvalidBlockNumber;
2213 22 : state->previncompletesplit = false;
2214 22 : return;
2215 : }
2216 :
2217 : /* Did we traverse the whole tree level and don't find next downlink? */
2218 3746 : if (blkno == P_NONE)
2219 0 : ereport(ERROR,
2220 : (errcode(ERRCODE_INDEX_CORRUPTED),
2221 : errmsg("can't traverse from downlink %u to downlink %u of index \"%s\"",
2222 : state->prevrightlink, downlink,
2223 : RelationGetRelationName(state->rel))));
2224 :
2225 : /* Load page contents */
2226 3746 : if (blkno == downlink && loaded_child)
2227 3722 : page = loaded_child;
2228 : else
2229 24 : page = palloc_btree_page(state, blkno);
2230 :
2231 3746 : opaque = BTPageGetOpaque(page);
2232 :
2233 : /* The first page we visit at the level should be leftmost */
2234 3746 : if (first && !BlockNumberIsValid(state->prevrightlink) &&
2235 22 : !bt_leftmost_ignoring_half_dead(state, blkno, opaque))
2236 0 : ereport(ERROR,
2237 : (errcode(ERRCODE_INDEX_CORRUPTED),
2238 : errmsg("the first child of leftmost target page is not leftmost of its level in index \"%s\"",
2239 : RelationGetRelationName(state->rel)),
2240 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2241 : state->targetblock, blkno,
2242 : LSN_FORMAT_ARGS(state->targetlsn))));
2243 :
2244 : /* Do level sanity check */
2245 3746 : if ((!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque)) &&
2246 3746 : opaque->btpo_level != target_level - 1)
2247 0 : ereport(ERROR,
2248 : (errcode(ERRCODE_INDEX_CORRUPTED),
2249 : errmsg("block found while following rightlinks from child of index \"%s\" has invalid level",
2250 : RelationGetRelationName(state->rel)),
2251 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
2252 : blkno, target_level - 1, opaque->btpo_level)));
2253 :
2254 : /* Try to detect circular links */
2255 3746 : if ((!first && blkno == state->prevrightlink) || blkno == opaque->btpo_prev)
2256 0 : ereport(ERROR,
2257 : (errcode(ERRCODE_INDEX_CORRUPTED),
2258 : errmsg("circular link chain found in block %u of index \"%s\"",
2259 : blkno, RelationGetRelationName(state->rel))));
2260 :
2261 3746 : if (blkno != downlink && !P_IGNORE(opaque))
2262 : {
2263 : /* blkno probably has missing parent downlink */
2264 0 : bt_downlink_missing_check(state, rightsplit, blkno, page);
2265 : }
2266 :
2267 3746 : rightsplit = P_INCOMPLETE_SPLIT(opaque);
2268 :
2269 : /*
2270 : * If we visit page with high key, check that it is equal to the
2271 : * target key next to corresponding downlink.
2272 : */
2273 3746 : if (!rightsplit && !P_RIGHTMOST(opaque))
2274 : {
2275 : BTPageOpaque topaque;
2276 : IndexTuple highkey;
2277 : OffsetNumber pivotkey_offset;
2278 :
2279 : /* Get high key */
2280 3724 : itemid = PageGetItemIdCareful(state, blkno, page, P_HIKEY);
2281 3724 : highkey = (IndexTuple) PageGetItem(page, itemid);
2282 :
2283 : /*
2284 : * There might be two situations when we examine high key. If
2285 : * current child page is referenced by given target downlink, we
2286 : * should look to the next offset number for matching key from
2287 : * target page.
2288 : *
2289 : * Alternatively, we're following rightlinks somewhere in the
2290 : * middle between page referenced by previous target's downlink
2291 : * and the page referenced by current target's downlink. If
2292 : * current child page hasn't incomplete split flag set, then its
2293 : * high key should match to the target's key of current offset
2294 : * number. This happens when a previous call here (to
2295 : * bt_child_highkey_check()) found an incomplete split, and we
2296 : * reach a right sibling page without a downlink -- the right
2297 : * sibling page's high key still needs to be matched to a
2298 : * separator key on the parent/target level.
2299 : *
2300 : * Don't apply OffsetNumberNext() to target_downlinkoffnum when we
2301 : * already had to step right on the child level. Our traversal of
2302 : * the child level must try to move in perfect lockstep behind (to
2303 : * the left of) the target/parent level traversal.
2304 : */
2305 3724 : if (blkno == downlink)
2306 3724 : pivotkey_offset = OffsetNumberNext(target_downlinkoffnum);
2307 : else
2308 0 : pivotkey_offset = target_downlinkoffnum;
2309 :
2310 3724 : topaque = BTPageGetOpaque(state->target);
2311 :
2312 3724 : if (!offset_is_negative_infinity(topaque, pivotkey_offset))
2313 : {
2314 : /*
2315 : * If we're looking for the next pivot tuple in target page,
2316 : * but there is no more pivot tuples, then we should match to
2317 : * high key instead.
2318 : */
2319 3724 : if (pivotkey_offset > PageGetMaxOffsetNumber(state->target))
2320 : {
2321 2 : if (P_RIGHTMOST(topaque))
2322 0 : ereport(ERROR,
2323 : (errcode(ERRCODE_INDEX_CORRUPTED),
2324 : errmsg("child high key is greater than rightmost pivot key on target level in index \"%s\"",
2325 : RelationGetRelationName(state->rel)),
2326 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2327 : state->targetblock, blkno,
2328 : LSN_FORMAT_ARGS(state->targetlsn))));
2329 2 : pivotkey_offset = P_HIKEY;
2330 : }
2331 3724 : itemid = PageGetItemIdCareful(state, state->targetblock,
2332 : state->target, pivotkey_offset);
2333 3724 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2334 : }
2335 : else
2336 : {
2337 : /*
2338 : * We cannot try to match child's high key to a negative
2339 : * infinity key in target, since there is nothing to compare.
2340 : * However, it's still possible to match child's high key
2341 : * outside of target page. The reason why we're are is that
2342 : * bt_child_highkey_check() was previously called for the
2343 : * cousin page of 'loaded_child', which is incomplete split.
2344 : * So, now we traverse to the right of that cousin page and
2345 : * current child level page under consideration still belongs
2346 : * to the subtree of target's left sibling. Thus, we need to
2347 : * match child's high key to its left uncle page high key.
2348 : * Thankfully we saved it, it's called a "low key" of target
2349 : * page.
2350 : */
2351 0 : if (!state->lowkey)
2352 0 : ereport(ERROR,
2353 : (errcode(ERRCODE_INDEX_CORRUPTED),
2354 : errmsg("can't find left sibling high key in index \"%s\"",
2355 : RelationGetRelationName(state->rel)),
2356 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2357 : state->targetblock, blkno,
2358 : LSN_FORMAT_ARGS(state->targetlsn))));
2359 0 : itup = state->lowkey;
2360 : }
2361 :
2362 3724 : if (!bt_pivot_tuple_identical(state->heapkeyspace, highkey, itup))
2363 : {
2364 0 : ereport(ERROR,
2365 : (errcode(ERRCODE_INDEX_CORRUPTED),
2366 : errmsg("mismatch between parent key and child high key in index \"%s\"",
2367 : RelationGetRelationName(state->rel)),
2368 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2369 : state->targetblock, blkno,
2370 : LSN_FORMAT_ARGS(state->targetlsn))));
2371 : }
2372 : }
2373 :
2374 : /* Exit if we already found next downlink */
2375 3746 : if (blkno == downlink)
2376 : {
2377 3746 : state->prevrightlink = opaque->btpo_next;
2378 3746 : state->previncompletesplit = rightsplit;
2379 3746 : return;
2380 : }
2381 :
2382 : /* Traverse to the next page using rightlink */
2383 0 : blkno = opaque->btpo_next;
2384 :
2385 : /* Free page contents if it's allocated by us */
2386 0 : if (page != loaded_child)
2387 0 : pfree(page);
2388 0 : first = false;
2389 : }
2390 : }
2391 :
2392 : /*
2393 : * Checks one of target's downlink against its child page.
2394 : *
2395 : * Conceptually, the target page continues to be what is checked here. The
2396 : * target block is still blamed in the event of finding an invariant violation.
2397 : * The downlink insertion into the target is probably where any problem raised
2398 : * here arises, and there is no such thing as a parent link, so doing the
2399 : * verification this way around is much more practical.
2400 : *
2401 : * This function visits child page and it's sequentially called for each
2402 : * downlink of target page. Assuming this we also check downlink connectivity
2403 : * here in order to save child page visits.
2404 : */
2405 : static void
2406 3722 : bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
2407 : OffsetNumber downlinkoffnum)
2408 : {
2409 : ItemId itemid;
2410 : IndexTuple itup;
2411 : BlockNumber childblock;
2412 : OffsetNumber offset;
2413 : OffsetNumber maxoffset;
2414 : Page child;
2415 : BTPageOpaque copaque;
2416 : BTPageOpaque topaque;
2417 :
2418 3722 : itemid = PageGetItemIdCareful(state, state->targetblock,
2419 : state->target, downlinkoffnum);
2420 3722 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2421 3722 : childblock = BTreeTupleGetDownLink(itup);
2422 :
2423 : /*
2424 : * Caller must have ShareLock on target relation, because of
2425 : * considerations around page deletion by VACUUM.
2426 : *
2427 : * NB: In general, page deletion deletes the right sibling's downlink, not
2428 : * the downlink of the page being deleted; the deleted page's downlink is
2429 : * reused for its sibling. The key space is thereby consolidated between
2430 : * the deleted page and its right sibling. (We cannot delete a parent
2431 : * page's rightmost child unless it is the last child page, and we intend
2432 : * to also delete the parent itself.)
2433 : *
2434 : * If this verification happened without a ShareLock, the following race
2435 : * condition could cause false positives:
2436 : *
2437 : * In general, concurrent page deletion might occur, including deletion of
2438 : * the left sibling of the child page that is examined here. If such a
2439 : * page deletion were to occur, closely followed by an insertion into the
2440 : * newly expanded key space of the child, a window for the false positive
2441 : * opens up: the stale parent/target downlink originally followed to get
2442 : * to the child legitimately ceases to be a lower bound on all items in
2443 : * the page, since the key space was concurrently expanded "left".
2444 : * (Insertion followed the "new" downlink for the child, not our now-stale
2445 : * downlink, which was concurrently physically removed in target/parent as
2446 : * part of deletion's first phase.)
2447 : *
2448 : * While we use various techniques elsewhere to perform cross-page
2449 : * verification for !readonly callers, a similar trick seems difficult
2450 : * here. The tricks used by bt_recheck_sibling_links and by
2451 : * bt_right_page_check_scankey both involve verification of a same-level,
2452 : * cross-sibling invariant. Cross-level invariants are far more squishy,
2453 : * though. The nbtree REDO routines do not actually couple buffer locks
2454 : * across levels during page splits, so making any cross-level check work
2455 : * reliably in !readonly mode may be impossible.
2456 : */
2457 : Assert(state->readonly);
2458 :
2459 : /*
2460 : * Verify child page has the downlink key from target page (its parent) as
2461 : * a lower bound; downlink must be strictly less than all keys on the
2462 : * page.
2463 : *
2464 : * Check all items, rather than checking just the first and trusting that
2465 : * the operator class obeys the transitive law.
2466 : */
2467 3722 : topaque = BTPageGetOpaque(state->target);
2468 3722 : child = palloc_btree_page(state, childblock);
2469 3722 : copaque = BTPageGetOpaque(child);
2470 3722 : maxoffset = PageGetMaxOffsetNumber(child);
2471 :
2472 : /*
2473 : * Since we've already loaded the child block, combine this check with
2474 : * check for downlink connectivity.
2475 : */
2476 3722 : bt_child_highkey_check(state, downlinkoffnum,
2477 : child, topaque->btpo_level);
2478 :
2479 : /*
2480 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2481 : * and since a page has no links within other pages (siblings and parent)
2482 : * once it is marked fully deleted, it should be impossible to land on a
2483 : * fully deleted page.
2484 : *
2485 : * It does not quite make sense to enforce that the page cannot even be
2486 : * half-dead, despite the fact the downlink is modified at the same stage
2487 : * that the child leaf page is marked half-dead. That's incorrect because
2488 : * there may occasionally be multiple downlinks from a chain of pages
2489 : * undergoing deletion, where multiple successive calls are made to
2490 : * _bt_unlink_halfdead_page() by VACUUM before it can finally safely mark
2491 : * the leaf page as fully dead. While _bt_mark_page_halfdead() usually
2492 : * removes the downlink to the leaf page that is marked half-dead, that's
2493 : * not guaranteed, so it's possible we'll land on a half-dead page with a
2494 : * downlink due to an interrupted multi-level page deletion.
2495 : *
2496 : * We go ahead with our checks if the child page is half-dead. It's safe
2497 : * to do so because we do not test the child's high key, so it does not
2498 : * matter that the original high key will have been replaced by a dummy
2499 : * truncated high key within _bt_mark_page_halfdead(). All other page
2500 : * items are left intact on a half-dead page, so there is still something
2501 : * to test.
2502 : */
2503 3722 : if (P_ISDELETED(copaque))
2504 0 : ereport(ERROR,
2505 : (errcode(ERRCODE_INDEX_CORRUPTED),
2506 : errmsg("downlink to deleted page found in index \"%s\"",
2507 : RelationGetRelationName(state->rel)),
2508 : errdetail_internal("Parent block=%u child block=%u parent page lsn=%X/%X.",
2509 : state->targetblock, childblock,
2510 : LSN_FORMAT_ARGS(state->targetlsn))));
2511 :
2512 1200700 : for (offset = P_FIRSTDATAKEY(copaque);
2513 : offset <= maxoffset;
2514 1196978 : offset = OffsetNumberNext(offset))
2515 : {
2516 : /*
2517 : * Skip comparison of target page key against "negative infinity"
2518 : * item, if any. Checking it would indicate that it's not a strict
2519 : * lower bound, but that's only because of the hard-coding for
2520 : * negative infinity items within _bt_compare().
2521 : *
2522 : * If nbtree didn't truncate negative infinity tuples during internal
2523 : * page splits then we'd expect child's negative infinity key to be
2524 : * equal to the scankey/downlink from target/parent (it would be a
2525 : * "low key" in this hypothetical scenario, and so it would still need
2526 : * to be treated as a special case here).
2527 : *
2528 : * Negative infinity items can be thought of as a strict lower bound
2529 : * that works transitively, with the last non-negative-infinity pivot
2530 : * followed during a descent from the root as its "true" strict lower
2531 : * bound. Only a small number of negative infinity items are truly
2532 : * negative infinity; those that are the first items of leftmost
2533 : * internal pages. In more general terms, a negative infinity item is
2534 : * only negative infinity with respect to the subtree that the page is
2535 : * at the root of.
2536 : *
2537 : * See also: bt_rootdescend(), which can even detect transitive
2538 : * inconsistencies on cousin leaf pages.
2539 : */
2540 1196978 : if (offset_is_negative_infinity(copaque, offset))
2541 2 : continue;
2542 :
2543 1196976 : if (!invariant_l_nontarget_offset(state, targetkey, childblock, child,
2544 : offset))
2545 0 : ereport(ERROR,
2546 : (errcode(ERRCODE_INDEX_CORRUPTED),
2547 : errmsg("down-link lower bound invariant violated for index \"%s\"",
2548 : RelationGetRelationName(state->rel)),
2549 : errdetail_internal("Parent block=%u child index tid=(%u,%u) parent page lsn=%X/%X.",
2550 : state->targetblock, childblock, offset,
2551 : LSN_FORMAT_ARGS(state->targetlsn))));
2552 : }
2553 :
2554 3722 : pfree(child);
2555 3722 : }
2556 :
2557 : /*
2558 : * Checks if page is missing a downlink that it should have.
2559 : *
2560 : * A page that lacks a downlink/parent may indicate corruption. However, we
2561 : * must account for the fact that a missing downlink can occasionally be
2562 : * encountered in a non-corrupt index. This can be due to an interrupted page
2563 : * split, or an interrupted multi-level page deletion (i.e. there was a hard
2564 : * crash or an error during a page split, or while VACUUM was deleting a
2565 : * multi-level chain of pages).
2566 : *
2567 : * Note that this can only be called in readonly mode, so there is no need to
2568 : * be concerned about concurrent page splits or page deletions.
2569 : */
2570 : static void
2571 0 : bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
2572 : BlockNumber blkno, Page page)
2573 : {
2574 0 : BTPageOpaque opaque = BTPageGetOpaque(page);
2575 : ItemId itemid;
2576 : IndexTuple itup;
2577 : Page child;
2578 : BTPageOpaque copaque;
2579 : uint32 level;
2580 : BlockNumber childblk;
2581 : XLogRecPtr pagelsn;
2582 :
2583 : Assert(state->readonly);
2584 : Assert(!P_IGNORE(opaque));
2585 :
2586 : /* No next level up with downlinks to fingerprint from the true root */
2587 0 : if (P_ISROOT(opaque))
2588 0 : return;
2589 :
2590 0 : pagelsn = PageGetLSN(page);
2591 :
2592 : /*
2593 : * Incomplete (interrupted) page splits can account for the lack of a
2594 : * downlink. Some inserting transaction should eventually complete the
2595 : * page split in passing, when it notices that the left sibling page is
2596 : * P_INCOMPLETE_SPLIT().
2597 : *
2598 : * In general, VACUUM is not prepared for there to be no downlink to a
2599 : * page that it deletes. This is the main reason why the lack of a
2600 : * downlink can be reported as corruption here. It's not obvious that an
2601 : * invalid missing downlink can result in wrong answers to queries,
2602 : * though, since index scans that land on the child may end up
2603 : * consistently moving right. The handling of concurrent page splits (and
2604 : * page deletions) within _bt_moveright() cannot distinguish
2605 : * inconsistencies that last for a moment from inconsistencies that are
2606 : * permanent and irrecoverable.
2607 : *
2608 : * VACUUM isn't even prepared to delete pages that have no downlink due to
2609 : * an incomplete page split, but it can detect and reason about that case
2610 : * by design, so it shouldn't be taken to indicate corruption. See
2611 : * _bt_pagedel() for full details.
2612 : */
2613 0 : if (rightsplit)
2614 : {
2615 0 : ereport(DEBUG1,
2616 : (errcode(ERRCODE_NO_DATA),
2617 : errmsg_internal("harmless interrupted page split detected in index \"%s\"",
2618 : RelationGetRelationName(state->rel)),
2619 : errdetail_internal("Block=%u level=%u left sibling=%u page lsn=%X/%X.",
2620 : blkno, opaque->btpo_level,
2621 : opaque->btpo_prev,
2622 : LSN_FORMAT_ARGS(pagelsn))));
2623 0 : return;
2624 : }
2625 :
2626 : /*
2627 : * Page under check is probably the "top parent" of a multi-level page
2628 : * deletion. We'll need to descend the subtree to make sure that
2629 : * descendant pages are consistent with that, though.
2630 : *
2631 : * If the page (which must be non-ignorable) is a leaf page, then clearly
2632 : * it can't be the top parent. The lack of a downlink is probably a
2633 : * symptom of a broad problem that could just as easily cause
2634 : * inconsistencies anywhere else.
2635 : */
2636 0 : if (P_ISLEAF(opaque))
2637 0 : ereport(ERROR,
2638 : (errcode(ERRCODE_INDEX_CORRUPTED),
2639 : errmsg("leaf index block lacks downlink in index \"%s\"",
2640 : RelationGetRelationName(state->rel)),
2641 : errdetail_internal("Block=%u page lsn=%X/%X.",
2642 : blkno,
2643 : LSN_FORMAT_ARGS(pagelsn))));
2644 :
2645 : /* Descend from the given page, which is an internal page */
2646 0 : elog(DEBUG1, "checking for interrupted multi-level deletion due to missing downlink in index \"%s\"",
2647 : RelationGetRelationName(state->rel));
2648 :
2649 0 : level = opaque->btpo_level;
2650 0 : itemid = PageGetItemIdCareful(state, blkno, page, P_FIRSTDATAKEY(opaque));
2651 0 : itup = (IndexTuple) PageGetItem(page, itemid);
2652 0 : childblk = BTreeTupleGetDownLink(itup);
2653 : for (;;)
2654 : {
2655 0 : CHECK_FOR_INTERRUPTS();
2656 :
2657 0 : child = palloc_btree_page(state, childblk);
2658 0 : copaque = BTPageGetOpaque(child);
2659 :
2660 0 : if (P_ISLEAF(copaque))
2661 0 : break;
2662 :
2663 : /* Do an extra sanity check in passing on internal pages */
2664 0 : if (copaque->btpo_level != level - 1)
2665 0 : ereport(ERROR,
2666 : (errcode(ERRCODE_INDEX_CORRUPTED),
2667 : errmsg_internal("downlink points to block in index \"%s\" whose level is not one level down",
2668 : RelationGetRelationName(state->rel)),
2669 : errdetail_internal("Top parent/under check block=%u block pointed to=%u expected level=%u level in pointed to block=%u.",
2670 : blkno, childblk,
2671 : level - 1, copaque->btpo_level)));
2672 :
2673 0 : level = copaque->btpo_level;
2674 0 : itemid = PageGetItemIdCareful(state, childblk, child,
2675 0 : P_FIRSTDATAKEY(copaque));
2676 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2677 0 : childblk = BTreeTupleGetDownLink(itup);
2678 : /* Be slightly more pro-active in freeing this memory, just in case */
2679 0 : pfree(child);
2680 : }
2681 :
2682 : /*
2683 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2684 : * and since a page has no links within other pages (siblings and parent)
2685 : * once it is marked fully deleted, it should be impossible to land on a
2686 : * fully deleted page. See bt_child_check() for further details.
2687 : *
2688 : * The bt_child_check() P_ISDELETED() check is repeated here because
2689 : * bt_child_check() does not visit pages reachable through negative
2690 : * infinity items. Besides, bt_child_check() is unwilling to descend
2691 : * multiple levels. (The similar bt_child_check() P_ISDELETED() check
2692 : * within bt_check_level_from_leftmost() won't reach the page either,
2693 : * since the leaf's live siblings should have their sibling links updated
2694 : * to bypass the deletion target page when it is marked fully dead.)
2695 : *
2696 : * If this error is raised, it might be due to a previous multi-level page
2697 : * deletion that failed to realize that it wasn't yet safe to mark the
2698 : * leaf page as fully dead. A "dangling downlink" will still remain when
2699 : * this happens. The fact that the dangling downlink's page (the leaf's
2700 : * parent/ancestor page) lacked a downlink is incidental.
2701 : */
2702 0 : if (P_ISDELETED(copaque))
2703 0 : ereport(ERROR,
2704 : (errcode(ERRCODE_INDEX_CORRUPTED),
2705 : errmsg_internal("downlink to deleted leaf page found in index \"%s\"",
2706 : RelationGetRelationName(state->rel)),
2707 : errdetail_internal("Top parent/target block=%u leaf block=%u top parent/under check lsn=%X/%X.",
2708 : blkno, childblk,
2709 : LSN_FORMAT_ARGS(pagelsn))));
2710 :
2711 : /*
2712 : * Iff leaf page is half-dead, its high key top parent link should point
2713 : * to what VACUUM considered to be the top parent page at the instant it
2714 : * was interrupted. Provided the high key link actually points to the
2715 : * page under check, the missing downlink we detected is consistent with
2716 : * there having been an interrupted multi-level page deletion. This means
2717 : * that the subtree with the page under check at its root (a page deletion
2718 : * chain) is in a consistent state, enabling VACUUM to resume deleting the
2719 : * entire chain the next time it encounters the half-dead leaf page.
2720 : */
2721 0 : if (P_ISHALFDEAD(copaque) && !P_RIGHTMOST(copaque))
2722 : {
2723 0 : itemid = PageGetItemIdCareful(state, childblk, child, P_HIKEY);
2724 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2725 0 : if (BTreeTupleGetTopParent(itup) == blkno)
2726 0 : return;
2727 : }
2728 :
2729 0 : ereport(ERROR,
2730 : (errcode(ERRCODE_INDEX_CORRUPTED),
2731 : errmsg("internal index block lacks downlink in index \"%s\"",
2732 : RelationGetRelationName(state->rel)),
2733 : errdetail_internal("Block=%u level=%u page lsn=%X/%X.",
2734 : blkno, opaque->btpo_level,
2735 : LSN_FORMAT_ARGS(pagelsn))));
2736 : }
2737 :
2738 : /*
2739 : * Per-tuple callback from table_index_build_scan, used to determine if index has
2740 : * all the entries that definitely should have been observed in leaf pages of
2741 : * the target index (that is, all IndexTuples that were fingerprinted by our
2742 : * Bloom filter). All heapallindexed checks occur here.
2743 : *
2744 : * The redundancy between an index and the table it indexes provides a good
2745 : * opportunity to detect corruption, especially corruption within the table.
2746 : * The high level principle behind the verification performed here is that any
2747 : * IndexTuple that should be in an index following a fresh CREATE INDEX (based
2748 : * on the same index definition) should also have been in the original,
2749 : * existing index, which should have used exactly the same representation
2750 : *
2751 : * Since the overall structure of the index has already been verified, the most
2752 : * likely explanation for error here is a corrupt heap page (could be logical
2753 : * or physical corruption). Index corruption may still be detected here,
2754 : * though. Only readonly callers will have verified that left links and right
2755 : * links are in agreement, and so it's possible that a leaf page transposition
2756 : * within index is actually the source of corruption detected here (for
2757 : * !readonly callers). The checks performed only for readonly callers might
2758 : * more accurately frame the problem as a cross-page invariant issue (this
2759 : * could even be due to recovery not replaying all WAL records). The !readonly
2760 : * ERROR message raised here includes a HINT about retrying with readonly
2761 : * verification, just in case it's a cross-page invariant issue, though that
2762 : * isn't particularly likely.
2763 : *
2764 : * table_index_build_scan() expects to be able to find the root tuple when a
2765 : * heap-only tuple (the live tuple at the end of some HOT chain) needs to be
2766 : * indexed, in order to replace the actual tuple's TID with the root tuple's
2767 : * TID (which is what we're actually passed back here). The index build heap
2768 : * scan code will raise an error when a tuple that claims to be the root of the
2769 : * heap-only tuple's HOT chain cannot be located. This catches cases where the
2770 : * original root item offset/root tuple for a HOT chain indicates (for whatever
2771 : * reason) that the entire HOT chain is dead, despite the fact that the latest
2772 : * heap-only tuple should be indexed. When this happens, sequential scans may
2773 : * always give correct answers, and all indexes may be considered structurally
2774 : * consistent (i.e. the nbtree structural checks would not detect corruption).
2775 : * It may be the case that only index scans give wrong answers, and yet heap or
2776 : * SLRU corruption is the real culprit. (While it's true that LP_DEAD bit
2777 : * setting will probably also leave the index in a corrupt state before too
2778 : * long, the problem is nonetheless that there is heap corruption.)
2779 : *
2780 : * Heap-only tuple handling within table_index_build_scan() works in a way that
2781 : * helps us to detect index tuples that contain the wrong values (values that
2782 : * don't match the latest tuple in the HOT chain). This can happen when there
2783 : * is no superseding index tuple due to a faulty assessment of HOT safety,
2784 : * perhaps during the original CREATE INDEX. Because the latest tuple's
2785 : * contents are used with the root TID, an error will be raised when a tuple
2786 : * with the same TID but non-matching attribute values is passed back to us.
2787 : * Faulty assessment of HOT-safety was behind at least two distinct CREATE
2788 : * INDEX CONCURRENTLY bugs that made it into stable releases, one of which was
2789 : * undetected for many years. In short, the same principle that allows a
2790 : * REINDEX to repair corruption when there was an (undetected) broken HOT chain
2791 : * also allows us to detect the corruption in many cases.
2792 : */
2793 : static void
2794 1060426 : bt_tuple_present_callback(Relation index, ItemPointer tid, Datum *values,
2795 : bool *isnull, bool tupleIsAlive, void *checkstate)
2796 : {
2797 1060426 : BtreeCheckState *state = (BtreeCheckState *) checkstate;
2798 : IndexTuple itup,
2799 : norm;
2800 :
2801 : Assert(state->heapallindexed);
2802 :
2803 : /* Generate a normalized index tuple for fingerprinting */
2804 1060426 : itup = index_form_tuple(RelationGetDescr(index), values, isnull);
2805 1060426 : itup->t_tid = *tid;
2806 1060426 : norm = bt_normalize_tuple(state, itup);
2807 :
2808 : /* Probe Bloom filter -- tuple should be present */
2809 1060426 : if (bloom_lacks_element(state->filter, (unsigned char *) norm,
2810 : IndexTupleSize(norm)))
2811 0 : ereport(ERROR,
2812 : (errcode(ERRCODE_DATA_CORRUPTED),
2813 : errmsg("heap tuple (%u,%u) from table \"%s\" lacks matching index tuple within index \"%s\"",
2814 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2815 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2816 : RelationGetRelationName(state->heaprel),
2817 : RelationGetRelationName(state->rel)),
2818 : !state->readonly
2819 : ? errhint("Retrying verification using the function bt_index_parent_check() might provide a more specific error.")
2820 : : 0));
2821 :
2822 1060426 : state->heaptuplespresent++;
2823 1060426 : pfree(itup);
2824 : /* Cannot leak memory here */
2825 1060426 : if (norm != itup)
2826 10 : pfree(norm);
2827 1060426 : }
2828 :
2829 : /*
2830 : * Normalize an index tuple for fingerprinting.
2831 : *
2832 : * In general, index tuple formation is assumed to be deterministic by
2833 : * heapallindexed verification, and IndexTuples are assumed immutable. While
2834 : * the LP_DEAD bit is mutable in leaf pages, that's ItemId metadata, which is
2835 : * not fingerprinted. Normalization is required to compensate for corner
2836 : * cases where the determinism assumption doesn't quite work.
2837 : *
2838 : * There is currently one such case: index_form_tuple() does not try to hide
2839 : * the source TOAST state of input datums. The executor applies TOAST
2840 : * compression for heap tuples based on different criteria to the compression
2841 : * applied within btinsert()'s call to index_form_tuple(): it sometimes
2842 : * compresses more aggressively, resulting in compressed heap tuple datums but
2843 : * uncompressed corresponding index tuple datums. A subsequent heapallindexed
2844 : * verification will get a logically equivalent though bitwise unequal tuple
2845 : * from index_form_tuple(). False positive heapallindexed corruption reports
2846 : * could occur without normalizing away the inconsistency.
2847 : *
2848 : * Returned tuple is often caller's own original tuple. Otherwise, it is a
2849 : * new representation of caller's original index tuple, palloc()'d in caller's
2850 : * memory context.
2851 : *
2852 : * Note: This routine is not concerned with distinctions about the
2853 : * representation of tuples beyond those that might break heapallindexed
2854 : * verification. In particular, it won't try to normalize opclass-equal
2855 : * datums with potentially distinct representations (e.g., btree/numeric_ops
2856 : * index datums will not get their display scale normalized-away here).
2857 : * Caller does normalization for non-pivot tuples that have a posting list,
2858 : * since dummy CREATE INDEX callback code generates new tuples with the same
2859 : * normalized representation.
2860 : */
2861 : static IndexTuple
2862 2125648 : bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
2863 : {
2864 2125648 : TupleDesc tupleDescriptor = RelationGetDescr(state->rel);
2865 : Datum normalized[INDEX_MAX_KEYS];
2866 : bool isnull[INDEX_MAX_KEYS];
2867 : bool need_free[INDEX_MAX_KEYS];
2868 2125648 : bool formnewtup = false;
2869 : IndexTuple reformed;
2870 : int i;
2871 :
2872 : /* Caller should only pass "logical" non-pivot tuples here */
2873 : Assert(!BTreeTupleIsPosting(itup) && !BTreeTupleIsPivot(itup));
2874 :
2875 : /* Easy case: It's immediately clear that tuple has no varlena datums */
2876 2125648 : if (!IndexTupleHasVarwidths(itup))
2877 2125600 : return itup;
2878 :
2879 96 : for (i = 0; i < tupleDescriptor->natts; i++)
2880 : {
2881 : Form_pg_attribute att;
2882 :
2883 48 : att = TupleDescAttr(tupleDescriptor, i);
2884 :
2885 : /* Assume untoasted/already normalized datum initially */
2886 48 : need_free[i] = false;
2887 48 : normalized[i] = index_getattr(itup, att->attnum,
2888 : tupleDescriptor,
2889 : &isnull[i]);
2890 48 : if (att->attbyval || att->attlen != -1 || isnull[i])
2891 0 : continue;
2892 :
2893 : /*
2894 : * Callers always pass a tuple that could safely be inserted into the
2895 : * index without further processing, so an external varlena header
2896 : * should never be encountered here
2897 : */
2898 48 : if (VARATT_IS_EXTERNAL(DatumGetPointer(normalized[i])))
2899 0 : ereport(ERROR,
2900 : (errcode(ERRCODE_INDEX_CORRUPTED),
2901 : errmsg("external varlena datum in tuple that references heap row (%u,%u) in index \"%s\"",
2902 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2903 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2904 : RelationGetRelationName(state->rel))));
2905 48 : else if (!VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])) &&
2906 44 : VARSIZE(DatumGetPointer(normalized[i])) > TOAST_INDEX_TARGET &&
2907 42 : (att->attstorage == TYPSTORAGE_EXTENDED ||
2908 32 : att->attstorage == TYPSTORAGE_MAIN))
2909 : {
2910 : /*
2911 : * This value will be compressed by index_form_tuple() with the
2912 : * current storage settings. We may be here because this tuple
2913 : * was formed with different storage settings. So, force forming.
2914 : */
2915 10 : formnewtup = true;
2916 : }
2917 38 : else if (VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])))
2918 : {
2919 4 : formnewtup = true;
2920 4 : normalized[i] = PointerGetDatum(PG_DETOAST_DATUM(normalized[i]));
2921 4 : need_free[i] = true;
2922 : }
2923 :
2924 : /*
2925 : * Short tuples may have 1B or 4B header. Convert 4B header of short
2926 : * tuples to 1B
2927 : */
2928 34 : else if (VARATT_CAN_MAKE_SHORT(DatumGetPointer(normalized[i])))
2929 : {
2930 : /* convert to short varlena */
2931 2 : Size len = VARATT_CONVERTED_SHORT_SIZE(DatumGetPointer(normalized[i]));
2932 2 : char *data = palloc(len);
2933 :
2934 2 : SET_VARSIZE_SHORT(data, len);
2935 2 : memcpy(data + 1, VARDATA(DatumGetPointer(normalized[i])), len - 1);
2936 :
2937 2 : formnewtup = true;
2938 2 : normalized[i] = PointerGetDatum(data);
2939 2 : need_free[i] = true;
2940 : }
2941 : }
2942 :
2943 : /*
2944 : * Easier case: Tuple has varlena datums, none of which are compressed or
2945 : * short with 4B header
2946 : */
2947 48 : if (!formnewtup)
2948 32 : return itup;
2949 :
2950 : /*
2951 : * Hard case: Tuple had compressed varlena datums that necessitate
2952 : * creating normalized version of the tuple from uncompressed input datums
2953 : * (normalized input datums). This is rather naive, but shouldn't be
2954 : * necessary too often.
2955 : *
2956 : * In the heap, tuples may contain short varlena datums with both 1B
2957 : * header and 4B headers. But the corresponding index tuple should always
2958 : * have such varlena's with 1B headers. So, if there is a short varlena
2959 : * with 4B header, we need to convert it for fingerprinting.
2960 : *
2961 : * Note that we rely on deterministic index_form_tuple() TOAST compression
2962 : * of normalized input.
2963 : */
2964 16 : reformed = index_form_tuple(tupleDescriptor, normalized, isnull);
2965 16 : reformed->t_tid = itup->t_tid;
2966 :
2967 : /* Cannot leak memory here */
2968 32 : for (i = 0; i < tupleDescriptor->natts; i++)
2969 16 : if (need_free[i])
2970 6 : pfree(DatumGetPointer(normalized[i]));
2971 :
2972 16 : return reformed;
2973 : }
2974 :
2975 : /*
2976 : * Produce palloc()'d "plain" tuple for nth posting list entry/TID.
2977 : *
2978 : * In general, deduplication is not supposed to change the logical contents of
2979 : * an index. Multiple index tuples are merged together into one equivalent
2980 : * posting list index tuple when convenient.
2981 : *
2982 : * heapallindexed verification must normalize-away this variation in
2983 : * representation by converting posting list tuples into two or more "plain"
2984 : * tuples. Each tuple must be fingerprinted separately -- there must be one
2985 : * tuple for each corresponding Bloom filter probe during the heap scan.
2986 : *
2987 : * Note: Caller still needs to call bt_normalize_tuple() with returned tuple.
2988 : */
2989 : static inline IndexTuple
2990 51750 : bt_posting_plain_tuple(IndexTuple itup, int n)
2991 : {
2992 : Assert(BTreeTupleIsPosting(itup));
2993 :
2994 : /* Returns non-posting-list tuple */
2995 51750 : return _bt_form_posting(itup, BTreeTupleGetPostingN(itup, n), 1);
2996 : }
2997 :
2998 : /*
2999 : * Search for itup in index, starting from fast root page. itup must be a
3000 : * non-pivot tuple. This is only supported with heapkeyspace indexes, since
3001 : * we rely on having fully unique keys to find a match with only a single
3002 : * visit to a leaf page, barring an interrupted page split, where we may have
3003 : * to move right. (A concurrent page split is impossible because caller must
3004 : * be readonly caller.)
3005 : *
3006 : * This routine can detect very subtle transitive consistency issues across
3007 : * more than one level of the tree. Leaf pages all have a high key (even the
3008 : * rightmost page has a conceptual positive infinity high key), but not a low
3009 : * key. Their downlink in parent is a lower bound, which along with the high
3010 : * key is almost enough to detect every possible inconsistency. A downlink
3011 : * separator key value won't always be available from parent, though, because
3012 : * the first items of internal pages are negative infinity items, truncated
3013 : * down to zero attributes during internal page splits. While it's true that
3014 : * bt_child_check() and the high key check can detect most imaginable key
3015 : * space problems, there are remaining problems it won't detect with non-pivot
3016 : * tuples in cousin leaf pages. Starting a search from the root for every
3017 : * existing leaf tuple detects small inconsistencies in upper levels of the
3018 : * tree that cannot be detected any other way. (Besides all this, this is
3019 : * probably also useful as a direct test of the code used by index scans
3020 : * themselves.)
3021 : */
3022 : static bool
3023 402196 : bt_rootdescend(BtreeCheckState *state, IndexTuple itup)
3024 : {
3025 : BTScanInsert key;
3026 : BTStack stack;
3027 : Buffer lbuf;
3028 : bool exists;
3029 :
3030 402196 : key = _bt_mkscankey(state->rel, itup);
3031 : Assert(key->heapkeyspace && key->scantid != NULL);
3032 :
3033 : /*
3034 : * Search from root.
3035 : *
3036 : * Ideally, we would arrange to only move right within _bt_search() when
3037 : * an interrupted page split is detected (i.e. when the incomplete split
3038 : * bit is found to be set), but for now we accept the possibility that
3039 : * that could conceal an inconsistency.
3040 : */
3041 : Assert(state->readonly && state->rootdescend);
3042 402196 : exists = false;
3043 402196 : stack = _bt_search(state->rel, NULL, key, &lbuf, BT_READ);
3044 :
3045 402196 : if (BufferIsValid(lbuf))
3046 : {
3047 : BTInsertStateData insertstate;
3048 : OffsetNumber offnum;
3049 : Page page;
3050 :
3051 402196 : insertstate.itup = itup;
3052 402196 : insertstate.itemsz = MAXALIGN(IndexTupleSize(itup));
3053 402196 : insertstate.itup_key = key;
3054 402196 : insertstate.postingoff = 0;
3055 402196 : insertstate.bounds_valid = false;
3056 402196 : insertstate.buf = lbuf;
3057 :
3058 : /* Get matching tuple on leaf page */
3059 402196 : offnum = _bt_binsrch_insert(state->rel, &insertstate);
3060 : /* Compare first >= matching item on leaf page, if any */
3061 402196 : page = BufferGetPage(lbuf);
3062 : /* Should match on first heap TID when tuple has a posting list */
3063 402196 : if (offnum <= PageGetMaxOffsetNumber(page) &&
3064 804392 : insertstate.postingoff <= 0 &&
3065 402196 : _bt_compare(state->rel, key, page, offnum) == 0)
3066 402196 : exists = true;
3067 402196 : _bt_relbuf(state->rel, lbuf);
3068 : }
3069 :
3070 402196 : _bt_freestack(stack);
3071 402196 : pfree(key);
3072 :
3073 402196 : return exists;
3074 : }
3075 :
3076 : /*
3077 : * Is particular offset within page (whose special state is passed by caller)
3078 : * the page negative-infinity item?
3079 : *
3080 : * As noted in comments above _bt_compare(), there is special handling of the
3081 : * first data item as a "negative infinity" item. The hard-coding within
3082 : * _bt_compare() makes comparing this item for the purposes of verification
3083 : * pointless at best, since the IndexTuple only contains a valid TID (a
3084 : * reference TID to child page).
3085 : */
3086 : static inline bool
3087 5233720 : offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
3088 : {
3089 : /*
3090 : * For internal pages only, the first item after high key, if any, is
3091 : * negative infinity item. Internal pages always have a negative infinity
3092 : * item, whereas leaf pages never have one. This implies that negative
3093 : * infinity item is either first or second line item, or there is none
3094 : * within page.
3095 : *
3096 : * Negative infinity items are a special case among pivot tuples. They
3097 : * always have zero attributes, while all other pivot tuples always have
3098 : * nkeyatts attributes.
3099 : *
3100 : * Right-most pages don't have a high key, but could be said to
3101 : * conceptually have a "positive infinity" high key. Thus, there is a
3102 : * symmetry between down link items in parent pages, and high keys in
3103 : * children. Together, they represent the part of the key space that
3104 : * belongs to each page in the index. For example, all children of the
3105 : * root page will have negative infinity as a lower bound from root
3106 : * negative infinity downlink, and positive infinity as an upper bound
3107 : * (implicitly, from "imaginary" positive infinity high key in root).
3108 : */
3109 5233720 : return !P_ISLEAF(opaque) && offset == P_FIRSTDATAKEY(opaque);
3110 : }
3111 :
3112 : /*
3113 : * Does the invariant hold that the key is strictly less than a given upper
3114 : * bound offset item?
3115 : *
3116 : * Verifies line pointer on behalf of caller.
3117 : *
3118 : * If this function returns false, convention is that caller throws error due
3119 : * to corruption.
3120 : */
3121 : static inline bool
3122 4014872 : invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
3123 : OffsetNumber upperbound)
3124 : {
3125 : ItemId itemid;
3126 : int32 cmp;
3127 :
3128 : Assert(!key->nextkey && key->backward);
3129 :
3130 : /* Verify line pointer before checking tuple */
3131 4014872 : itemid = PageGetItemIdCareful(state, state->targetblock, state->target,
3132 : upperbound);
3133 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3134 4014872 : if (!key->heapkeyspace)
3135 0 : return invariant_leq_offset(state, key, upperbound);
3136 :
3137 4014872 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3138 :
3139 : /*
3140 : * _bt_compare() is capable of determining that a scankey with a
3141 : * filled-out attribute is greater than pivot tuples where the comparison
3142 : * is resolved at a truncated attribute (value of attribute in pivot is
3143 : * minus infinity). However, it is not capable of determining that a
3144 : * scankey is _less than_ a tuple on the basis of a comparison resolved at
3145 : * _scankey_ minus infinity attribute. Complete an extra step to simulate
3146 : * having minus infinity values for omitted scankey attribute(s).
3147 : */
3148 4014872 : if (cmp == 0)
3149 : {
3150 : BTPageOpaque topaque;
3151 : IndexTuple ritup;
3152 : int uppnkeyatts;
3153 : ItemPointer rheaptid;
3154 : bool nonpivot;
3155 :
3156 0 : ritup = (IndexTuple) PageGetItem(state->target, itemid);
3157 0 : topaque = BTPageGetOpaque(state->target);
3158 0 : nonpivot = P_ISLEAF(topaque) && upperbound >= P_FIRSTDATAKEY(topaque);
3159 :
3160 : /* Get number of keys + heap TID for item to the right */
3161 0 : uppnkeyatts = BTreeTupleGetNKeyAtts(ritup, state->rel);
3162 0 : rheaptid = BTreeTupleGetHeapTIDCareful(state, ritup, nonpivot);
3163 :
3164 : /* Heap TID is tiebreaker key attribute */
3165 0 : if (key->keysz == uppnkeyatts)
3166 0 : return key->scantid == NULL && rheaptid != NULL;
3167 :
3168 0 : return key->keysz < uppnkeyatts;
3169 : }
3170 :
3171 4014872 : return cmp < 0;
3172 : }
3173 :
3174 : /*
3175 : * Does the invariant hold that the key is less than or equal to a given upper
3176 : * bound offset item?
3177 : *
3178 : * Caller should have verified that upperbound's line pointer is consistent
3179 : * using PageGetItemIdCareful() call.
3180 : *
3181 : * If this function returns false, convention is that caller throws error due
3182 : * to corruption.
3183 : */
3184 : static inline bool
3185 3698416 : invariant_leq_offset(BtreeCheckState *state, BTScanInsert key,
3186 : OffsetNumber upperbound)
3187 : {
3188 : int32 cmp;
3189 :
3190 : Assert(!key->nextkey && key->backward);
3191 :
3192 3698416 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3193 :
3194 3698416 : return cmp <= 0;
3195 : }
3196 :
3197 : /*
3198 : * Does the invariant hold that the key is strictly greater than a given lower
3199 : * bound offset item?
3200 : *
3201 : * Caller should have verified that lowerbound's line pointer is consistent
3202 : * using PageGetItemIdCareful() call.
3203 : *
3204 : * If this function returns false, convention is that caller throws error due
3205 : * to corruption.
3206 : */
3207 : static inline bool
3208 13204 : invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
3209 : OffsetNumber lowerbound)
3210 : {
3211 : int32 cmp;
3212 :
3213 : Assert(!key->nextkey && key->backward);
3214 :
3215 13204 : cmp = _bt_compare(state->rel, key, state->target, lowerbound);
3216 :
3217 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3218 13204 : if (!key->heapkeyspace)
3219 0 : return cmp >= 0;
3220 :
3221 : /*
3222 : * No need to consider the possibility that scankey has attributes that we
3223 : * need to force to be interpreted as negative infinity. _bt_compare() is
3224 : * able to determine that scankey is greater than negative infinity. The
3225 : * distinction between "==" and "<" isn't interesting here, since
3226 : * corruption is indicated either way.
3227 : */
3228 13204 : return cmp > 0;
3229 : }
3230 :
3231 : /*
3232 : * Does the invariant hold that the key is strictly less than a given upper
3233 : * bound offset item, with the offset relating to a caller-supplied page that
3234 : * is not the current target page?
3235 : *
3236 : * Caller's non-target page is a child page of the target, checked as part of
3237 : * checking a property of the target page (i.e. the key comes from the
3238 : * target). Verifies line pointer on behalf of caller.
3239 : *
3240 : * If this function returns false, convention is that caller throws error due
3241 : * to corruption.
3242 : */
3243 : static inline bool
3244 1196976 : invariant_l_nontarget_offset(BtreeCheckState *state, BTScanInsert key,
3245 : BlockNumber nontargetblock, Page nontarget,
3246 : OffsetNumber upperbound)
3247 : {
3248 : ItemId itemid;
3249 : int32 cmp;
3250 :
3251 : Assert(!key->nextkey && key->backward);
3252 :
3253 : /* Verify line pointer before checking tuple */
3254 1196976 : itemid = PageGetItemIdCareful(state, nontargetblock, nontarget,
3255 : upperbound);
3256 1196976 : cmp = _bt_compare(state->rel, key, nontarget, upperbound);
3257 :
3258 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3259 1196976 : if (!key->heapkeyspace)
3260 0 : return cmp <= 0;
3261 :
3262 : /* See invariant_l_offset() for an explanation of this extra step */
3263 1196976 : if (cmp == 0)
3264 : {
3265 : IndexTuple child;
3266 : int uppnkeyatts;
3267 : ItemPointer childheaptid;
3268 : BTPageOpaque copaque;
3269 : bool nonpivot;
3270 :
3271 3720 : child = (IndexTuple) PageGetItem(nontarget, itemid);
3272 3720 : copaque = BTPageGetOpaque(nontarget);
3273 3720 : nonpivot = P_ISLEAF(copaque) && upperbound >= P_FIRSTDATAKEY(copaque);
3274 :
3275 : /* Get number of keys + heap TID for child/non-target item */
3276 3720 : uppnkeyatts = BTreeTupleGetNKeyAtts(child, state->rel);
3277 3720 : childheaptid = BTreeTupleGetHeapTIDCareful(state, child, nonpivot);
3278 :
3279 : /* Heap TID is tiebreaker key attribute */
3280 3720 : if (key->keysz == uppnkeyatts)
3281 3720 : return key->scantid == NULL && childheaptid != NULL;
3282 :
3283 0 : return key->keysz < uppnkeyatts;
3284 : }
3285 :
3286 1193256 : return cmp < 0;
3287 : }
3288 :
3289 : /*
3290 : * Given a block number of a B-Tree page, return page in palloc()'d memory.
3291 : * While at it, perform some basic checks of the page.
3292 : *
3293 : * There is never an attempt to get a consistent view of multiple pages using
3294 : * multiple concurrent buffer locks; in general, we only acquire a single pin
3295 : * and buffer lock at a time, which is often all that the nbtree code requires.
3296 : * (Actually, bt_recheck_sibling_links couples buffer locks, which is the only
3297 : * exception to this general rule.)
3298 : *
3299 : * Operating on a copy of the page is useful because it prevents control
3300 : * getting stuck in an uninterruptible state when an underlying operator class
3301 : * misbehaves.
3302 : */
3303 : static Page
3304 43130 : palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum)
3305 : {
3306 : Buffer buffer;
3307 : Page page;
3308 : BTPageOpaque opaque;
3309 : OffsetNumber maxoffset;
3310 :
3311 43130 : page = palloc(BLCKSZ);
3312 :
3313 : /*
3314 : * We copy the page into local storage to avoid holding pin on the buffer
3315 : * longer than we must.
3316 : */
3317 43130 : buffer = ReadBufferExtended(state->rel, MAIN_FORKNUM, blocknum, RBM_NORMAL,
3318 : state->checkstrategy);
3319 43106 : LockBuffer(buffer, BT_READ);
3320 :
3321 : /*
3322 : * Perform the same basic sanity checking that nbtree itself performs for
3323 : * every page:
3324 : */
3325 43106 : _bt_checkpage(state->rel, buffer);
3326 :
3327 : /* Only use copy of page in palloc()'d memory */
3328 43106 : memcpy(page, BufferGetPage(buffer), BLCKSZ);
3329 43106 : UnlockReleaseBuffer(buffer);
3330 :
3331 43106 : opaque = BTPageGetOpaque(page);
3332 :
3333 43106 : if (P_ISMETA(opaque) && blocknum != BTREE_METAPAGE)
3334 0 : ereport(ERROR,
3335 : (errcode(ERRCODE_INDEX_CORRUPTED),
3336 : errmsg("invalid meta page found at block %u in index \"%s\"",
3337 : blocknum, RelationGetRelationName(state->rel))));
3338 :
3339 : /* Check page from block that ought to be meta page */
3340 43106 : if (blocknum == BTREE_METAPAGE)
3341 : {
3342 7972 : BTMetaPageData *metad = BTPageGetMeta(page);
3343 :
3344 7972 : if (!P_ISMETA(opaque) ||
3345 7972 : metad->btm_magic != BTREE_MAGIC)
3346 0 : ereport(ERROR,
3347 : (errcode(ERRCODE_INDEX_CORRUPTED),
3348 : errmsg("index \"%s\" meta page is corrupt",
3349 : RelationGetRelationName(state->rel))));
3350 :
3351 7972 : if (metad->btm_version < BTREE_MIN_VERSION ||
3352 7972 : metad->btm_version > BTREE_VERSION)
3353 0 : ereport(ERROR,
3354 : (errcode(ERRCODE_INDEX_CORRUPTED),
3355 : errmsg("version mismatch in index \"%s\": file version %d, "
3356 : "current version %d, minimum supported version %d",
3357 : RelationGetRelationName(state->rel),
3358 : metad->btm_version, BTREE_VERSION,
3359 : BTREE_MIN_VERSION)));
3360 :
3361 : /* Finished with metapage checks */
3362 7972 : return page;
3363 : }
3364 :
3365 : /*
3366 : * Deleted pages that still use the old 32-bit XID representation have no
3367 : * sane "level" field because they type pun the field, but all other pages
3368 : * (including pages deleted on Postgres 14+) have a valid value.
3369 : */
3370 35134 : if (!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque))
3371 : {
3372 : /* Okay, no reason not to trust btpo_level field from page */
3373 :
3374 35134 : if (P_ISLEAF(opaque) && opaque->btpo_level != 0)
3375 0 : ereport(ERROR,
3376 : (errcode(ERRCODE_INDEX_CORRUPTED),
3377 : errmsg_internal("invalid leaf page level %u for block %u in index \"%s\"",
3378 : opaque->btpo_level, blocknum,
3379 : RelationGetRelationName(state->rel))));
3380 :
3381 35134 : if (!P_ISLEAF(opaque) && opaque->btpo_level == 0)
3382 0 : ereport(ERROR,
3383 : (errcode(ERRCODE_INDEX_CORRUPTED),
3384 : errmsg_internal("invalid internal page level 0 for block %u in index \"%s\"",
3385 : blocknum,
3386 : RelationGetRelationName(state->rel))));
3387 : }
3388 :
3389 : /*
3390 : * Sanity checks for number of items on page.
3391 : *
3392 : * As noted at the beginning of _bt_binsrch(), an internal page must have
3393 : * children, since there must always be a negative infinity downlink
3394 : * (there may also be a highkey). In the case of non-rightmost leaf
3395 : * pages, there must be at least a highkey. The exceptions are deleted
3396 : * pages, which contain no items.
3397 : *
3398 : * This is correct when pages are half-dead, since internal pages are
3399 : * never half-dead, and leaf pages must have a high key when half-dead
3400 : * (the rightmost page can never be deleted). It's also correct with
3401 : * fully deleted pages: _bt_unlink_halfdead_page() doesn't change anything
3402 : * about the target page other than setting the page as fully dead, and
3403 : * setting its xact field. In particular, it doesn't change the sibling
3404 : * links in the deletion target itself, since they're required when index
3405 : * scans land on the deletion target, and then need to move right (or need
3406 : * to move left, in the case of backward index scans).
3407 : */
3408 35134 : maxoffset = PageGetMaxOffsetNumber(page);
3409 35134 : if (maxoffset > MaxIndexTuplesPerPage)
3410 0 : ereport(ERROR,
3411 : (errcode(ERRCODE_INDEX_CORRUPTED),
3412 : errmsg("Number of items on block %u of index \"%s\" exceeds MaxIndexTuplesPerPage (%u)",
3413 : blocknum, RelationGetRelationName(state->rel),
3414 : MaxIndexTuplesPerPage)));
3415 :
3416 35134 : if (!P_ISLEAF(opaque) && !P_ISDELETED(opaque) && maxoffset < P_FIRSTDATAKEY(opaque))
3417 0 : ereport(ERROR,
3418 : (errcode(ERRCODE_INDEX_CORRUPTED),
3419 : errmsg("internal block %u in index \"%s\" lacks high key and/or at least one downlink",
3420 : blocknum, RelationGetRelationName(state->rel))));
3421 :
3422 35134 : if (P_ISLEAF(opaque) && !P_ISDELETED(opaque) && !P_RIGHTMOST(opaque) && maxoffset < P_HIKEY)
3423 0 : ereport(ERROR,
3424 : (errcode(ERRCODE_INDEX_CORRUPTED),
3425 : errmsg("non-rightmost leaf block %u in index \"%s\" lacks high key item",
3426 : blocknum, RelationGetRelationName(state->rel))));
3427 :
3428 : /*
3429 : * In general, internal pages are never marked half-dead, except on
3430 : * versions of Postgres prior to 9.4, where it can be valid transient
3431 : * state. This state is nonetheless treated as corruption by VACUUM on
3432 : * from version 9.4 on, so do the same here. See _bt_pagedel() for full
3433 : * details.
3434 : */
3435 35134 : if (!P_ISLEAF(opaque) && P_ISHALFDEAD(opaque))
3436 0 : ereport(ERROR,
3437 : (errcode(ERRCODE_INDEX_CORRUPTED),
3438 : errmsg("internal page block %u in index \"%s\" is half-dead",
3439 : blocknum, RelationGetRelationName(state->rel)),
3440 : errhint("This can be caused by an interrupted VACUUM in version 9.3 or older, before upgrade. Please REINDEX it.")));
3441 :
3442 : /*
3443 : * Check that internal pages have no garbage items, and that no page has
3444 : * an invalid combination of deletion-related page level flags
3445 : */
3446 35134 : if (!P_ISLEAF(opaque) && P_HAS_GARBAGE(opaque))
3447 0 : ereport(ERROR,
3448 : (errcode(ERRCODE_INDEX_CORRUPTED),
3449 : errmsg_internal("internal page block %u in index \"%s\" has garbage items",
3450 : blocknum, RelationGetRelationName(state->rel))));
3451 :
3452 35134 : if (P_HAS_FULLXID(opaque) && !P_ISDELETED(opaque))
3453 0 : ereport(ERROR,
3454 : (errcode(ERRCODE_INDEX_CORRUPTED),
3455 : errmsg_internal("full transaction id page flag appears in non-deleted block %u in index \"%s\"",
3456 : blocknum, RelationGetRelationName(state->rel))));
3457 :
3458 35134 : if (P_ISDELETED(opaque) && P_ISHALFDEAD(opaque))
3459 0 : ereport(ERROR,
3460 : (errcode(ERRCODE_INDEX_CORRUPTED),
3461 : errmsg_internal("deleted page block %u in index \"%s\" is half-dead",
3462 : blocknum, RelationGetRelationName(state->rel))));
3463 :
3464 35134 : return page;
3465 : }
3466 :
3467 : /*
3468 : * _bt_mkscankey() wrapper that automatically prevents insertion scankey from
3469 : * being considered greater than the pivot tuple that its values originated
3470 : * from (or some other identical pivot tuple) in the common case where there
3471 : * are truncated/minus infinity attributes. Without this extra step, there
3472 : * are forms of corruption that amcheck could theoretically fail to report.
3473 : *
3474 : * For example, invariant_g_offset() might miss a cross-page invariant failure
3475 : * on an internal level if the scankey built from the first item on the
3476 : * target's right sibling page happened to be equal to (not greater than) the
3477 : * last item on target page. The !backward tiebreaker in _bt_compare() might
3478 : * otherwise cause amcheck to assume (rather than actually verify) that the
3479 : * scankey is greater.
3480 : */
3481 : static inline BTScanInsert
3482 4045108 : bt_mkscankey_pivotsearch(Relation rel, IndexTuple itup)
3483 : {
3484 : BTScanInsert skey;
3485 :
3486 4045108 : skey = _bt_mkscankey(rel, itup);
3487 4045108 : skey->backward = true;
3488 :
3489 4045108 : return skey;
3490 : }
3491 :
3492 : /*
3493 : * PageGetItemId() wrapper that validates returned line pointer.
3494 : *
3495 : * Buffer page/page item access macros generally trust that line pointers are
3496 : * not corrupt, which might cause problems for verification itself. For
3497 : * example, there is no bounds checking in PageGetItem(). Passing it a
3498 : * corrupt line pointer can cause it to return a tuple/pointer that is unsafe
3499 : * to dereference.
3500 : *
3501 : * Validating line pointers before tuples avoids undefined behavior and
3502 : * assertion failures with corrupt indexes, making the verification process
3503 : * more robust and predictable.
3504 : */
3505 : static ItemId
3506 9291038 : PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block, Page page,
3507 : OffsetNumber offset)
3508 : {
3509 9291038 : ItemId itemid = PageGetItemId(page, offset);
3510 :
3511 9291038 : if (ItemIdGetOffset(itemid) + ItemIdGetLength(itemid) >
3512 : BLCKSZ - MAXALIGN(sizeof(BTPageOpaqueData)))
3513 0 : ereport(ERROR,
3514 : (errcode(ERRCODE_INDEX_CORRUPTED),
3515 : errmsg("line pointer points past end of tuple space in index \"%s\"",
3516 : RelationGetRelationName(state->rel)),
3517 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3518 : block, offset, ItemIdGetOffset(itemid),
3519 : ItemIdGetLength(itemid),
3520 : ItemIdGetFlags(itemid))));
3521 :
3522 : /*
3523 : * Verify that line pointer isn't LP_REDIRECT or LP_UNUSED, since nbtree
3524 : * never uses either. Verify that line pointer has storage, too, since
3525 : * even LP_DEAD items should within nbtree.
3526 : */
3527 9291038 : if (ItemIdIsRedirected(itemid) || !ItemIdIsUsed(itemid) ||
3528 9291038 : ItemIdGetLength(itemid) == 0)
3529 0 : ereport(ERROR,
3530 : (errcode(ERRCODE_INDEX_CORRUPTED),
3531 : errmsg("invalid line pointer storage in index \"%s\"",
3532 : RelationGetRelationName(state->rel)),
3533 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3534 : block, offset, ItemIdGetOffset(itemid),
3535 : ItemIdGetLength(itemid),
3536 : ItemIdGetFlags(itemid))));
3537 :
3538 9291038 : return itemid;
3539 : }
3540 :
3541 : /*
3542 : * BTreeTupleGetHeapTID() wrapper that enforces that a heap TID is present in
3543 : * cases where that is mandatory (i.e. for non-pivot tuples)
3544 : */
3545 : static inline ItemPointer
3546 3720 : BTreeTupleGetHeapTIDCareful(BtreeCheckState *state, IndexTuple itup,
3547 : bool nonpivot)
3548 : {
3549 : ItemPointer htid;
3550 :
3551 : /*
3552 : * Caller determines whether this is supposed to be a pivot or non-pivot
3553 : * tuple using page type and item offset number. Verify that tuple
3554 : * metadata agrees with this.
3555 : */
3556 : Assert(state->heapkeyspace);
3557 3720 : if (BTreeTupleIsPivot(itup) && nonpivot)
3558 0 : ereport(ERROR,
3559 : (errcode(ERRCODE_INDEX_CORRUPTED),
3560 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected pivot tuple",
3561 : state->targetblock,
3562 : RelationGetRelationName(state->rel))));
3563 :
3564 3720 : if (!BTreeTupleIsPivot(itup) && !nonpivot)
3565 0 : ereport(ERROR,
3566 : (errcode(ERRCODE_INDEX_CORRUPTED),
3567 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected non-pivot tuple",
3568 : state->targetblock,
3569 : RelationGetRelationName(state->rel))));
3570 :
3571 3720 : htid = BTreeTupleGetHeapTID(itup);
3572 3720 : if (!ItemPointerIsValid(htid) && nonpivot)
3573 0 : ereport(ERROR,
3574 : (errcode(ERRCODE_INDEX_CORRUPTED),
3575 : errmsg("block %u or its right sibling block or child block in index \"%s\" contains non-pivot tuple that lacks a heap TID",
3576 : state->targetblock,
3577 : RelationGetRelationName(state->rel))));
3578 :
3579 3720 : return htid;
3580 : }
3581 :
3582 : /*
3583 : * Return the "pointed to" TID for itup, which is used to generate a
3584 : * descriptive error message. itup must be a "data item" tuple (it wouldn't
3585 : * make much sense to call here with a high key tuple, since there won't be a
3586 : * valid downlink/block number to display).
3587 : *
3588 : * Returns either a heap TID (which will be the first heap TID in posting list
3589 : * if itup is posting list tuple), or a TID that contains downlink block
3590 : * number, plus some encoded metadata (e.g., the number of attributes present
3591 : * in itup).
3592 : */
3593 : static inline ItemPointer
3594 12 : BTreeTupleGetPointsToTID(IndexTuple itup)
3595 : {
3596 : /*
3597 : * Rely on the assumption that !heapkeyspace internal page data items will
3598 : * correctly return TID with downlink here -- BTreeTupleGetHeapTID() won't
3599 : * recognize it as a pivot tuple, but everything still works out because
3600 : * the t_tid field is still returned
3601 : */
3602 12 : if (!BTreeTupleIsPivot(itup))
3603 8 : return BTreeTupleGetHeapTID(itup);
3604 :
3605 : /* Pivot tuple returns TID with downlink block (heapkeyspace variant) */
3606 4 : return &itup->t_tid;
3607 : }
|
