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 :
46 650 : 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 :
174 180 : PG_FUNCTION_INFO_V1(bt_index_check);
175 130 : 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 7878 : bt_index_check(PG_FUNCTION_ARGS)
251 : {
252 7878 : Oid indrelid = PG_GETARG_OID(0);
253 : BTCallbackState args;
254 :
255 7878 : args.heapallindexed = false;
256 7878 : args.rootdescend = false;
257 7878 : args.parentcheck = false;
258 7878 : args.checkunique = false;
259 :
260 7878 : if (PG_NARGS() >= 2)
261 7866 : args.heapallindexed = PG_GETARG_BOOL(1);
262 7878 : if (PG_NARGS() >= 3)
263 1364 : args.checkunique = PG_GETARG_BOOL(2);
264 :
265 7878 : amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
266 : bt_index_check_callback,
267 : AccessShareLock, &args);
268 :
269 7826 : 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
282 124 : bt_index_parent_check(PG_FUNCTION_ARGS)
283 : {
284 124 : Oid indrelid = PG_GETARG_OID(0);
285 : BTCallbackState args;
286 :
287 124 : args.heapallindexed = false;
288 124 : args.rootdescend = false;
289 124 : args.parentcheck = true;
290 124 : args.checkunique = false;
291 :
292 124 : if (PG_NARGS() >= 2)
293 112 : args.heapallindexed = PG_GETARG_BOOL(1);
294 124 : if (PG_NARGS() >= 3)
295 104 : args.rootdescend = PG_GETARG_BOOL(2);
296 124 : if (PG_NARGS() >= 4)
297 52 : args.checkunique = PG_GETARG_BOOL(3);
298 :
299 124 : amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
300 : bt_index_check_callback,
301 : ShareLock, &args);
302 :
303 88 : PG_RETURN_VOID();
304 : }
305 :
306 : /*
307 : * Helper for bt_index_[parent_]check, coordinating the bulk of the work.
308 : */
309 : static void
310 7990 : bt_index_check_callback(Relation indrel, Relation heaprel, void *state, bool readonly)
311 : {
312 7990 : BTCallbackState *args = (BTCallbackState *) state;
313 : bool heapkeyspace,
314 : allequalimage;
315 :
316 7990 : 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 7954 : _bt_metaversion(indrel, &heapkeyspace, &allequalimage);
324 7954 : 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 7954 : 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 7954 : bt_check_every_level(indrel, heaprel, heapkeyspace, readonly,
349 7954 : args->heapallindexed, args->rootdescend, args->checkunique);
350 7914 : }
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 7954 : 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 :
386 7954 : if (!readonly)
387 7854 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\"",
388 : RelationGetRelationName(rel));
389 : else
390 100 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\" with cross-level checks",
391 : RelationGetRelationName(rel));
392 :
393 : /*
394 : * This assertion matches the one in index_getnext_tid(). See page
395 : * recycling/"visible to everyone" notes in nbtree README.
396 : */
397 : Assert(TransactionIdIsValid(RecentXmin));
398 :
399 : /*
400 : * Initialize state for entire verification operation
401 : */
402 7954 : state = palloc0_object(BtreeCheckState);
403 7954 : state->rel = rel;
404 7954 : state->heaprel = heaprel;
405 7954 : state->heapkeyspace = heapkeyspace;
406 7954 : state->readonly = readonly;
407 7954 : state->heapallindexed = heapallindexed;
408 7954 : state->rootdescend = rootdescend;
409 7954 : state->checkunique = checkunique;
410 7954 : state->snapshot = InvalidSnapshot;
411 :
412 7954 : if (state->heapallindexed)
413 : {
414 : int64 total_pages;
415 : int64 total_elems;
416 : uint64 seed;
417 :
418 : /*
419 : * Size Bloom filter based on estimated number of tuples in index,
420 : * while conservatively assuming that each block must contain at least
421 : * MaxTIDsPerBTreePage / 3 "plain" tuples -- see
422 : * bt_posting_plain_tuple() for definition, and details of how posting
423 : * list tuples are handled.
424 : */
425 152 : total_pages = RelationGetNumberOfBlocks(rel);
426 152 : total_elems = Max(total_pages * (MaxTIDsPerBTreePage / 3),
427 : (int64) state->rel->rd_rel->reltuples);
428 : /* Generate a random seed to avoid repetition */
429 152 : seed = pg_prng_uint64(&pg_global_prng_state);
430 : /* Create Bloom filter to fingerprint index */
431 152 : state->filter = bloom_create(total_elems, maintenance_work_mem, seed);
432 152 : state->heaptuplespresent = 0;
433 :
434 : /*
435 : * Register our own snapshot for heapallindexed, rather than asking
436 : * table_index_build_scan() to do this for us later. This needs to
437 : * happen before index fingerprinting begins, so we can later be
438 : * certain that index fingerprinting should have reached all tuples
439 : * returned by table_index_build_scan().
440 : */
441 152 : state->snapshot = RegisterSnapshot(GetTransactionSnapshot());
442 :
443 : /*
444 : * GetTransactionSnapshot() always acquires a new MVCC snapshot in
445 : * READ COMMITTED mode. A new snapshot is guaranteed to have all the
446 : * entries it requires in the index.
447 : *
448 : * We must defend against the possibility that an old xact snapshot
449 : * was returned at higher isolation levels when that snapshot is not
450 : * safe for index scans of the target index. This is possible when
451 : * the snapshot sees tuples that are before the index's indcheckxmin
452 : * horizon. Throwing an error here should be very rare. It doesn't
453 : * seem worth using a secondary snapshot to avoid this.
454 : */
455 152 : if (IsolationUsesXactSnapshot() && rel->rd_index->indcheckxmin &&
456 0 : !TransactionIdPrecedes(HeapTupleHeaderGetXmin(rel->rd_indextuple->t_data),
457 0 : state->snapshot->xmin))
458 0 : ereport(ERROR,
459 : errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
460 : errmsg("index \"%s\" cannot be verified using transaction snapshot",
461 : RelationGetRelationName(rel)));
462 : }
463 :
464 : /*
465 : * We need a snapshot to check the uniqueness of the index. For better
466 : * performance, take it once per index check. If one was already taken
467 : * above, use that.
468 : */
469 7954 : if (state->checkunique)
470 : {
471 1408 : state->indexinfo = BuildIndexInfo(state->rel);
472 :
473 1408 : if (state->indexinfo->ii_Unique && state->snapshot == InvalidSnapshot)
474 1242 : state->snapshot = RegisterSnapshot(GetTransactionSnapshot());
475 : }
476 :
477 : Assert(!state->rootdescend || state->readonly);
478 7954 : if (state->rootdescend && !state->heapkeyspace)
479 0 : ereport(ERROR,
480 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
481 : errmsg("cannot verify that tuples from index \"%s\" can each be found by an independent index search",
482 : RelationGetRelationName(rel)),
483 : errhint("Only B-Tree version 4 indexes support rootdescend verification.")));
484 :
485 : /* Create context for page */
486 7954 : state->targetcontext = AllocSetContextCreate(CurrentMemoryContext,
487 : "amcheck context",
488 : ALLOCSET_DEFAULT_SIZES);
489 7954 : state->checkstrategy = GetAccessStrategy(BAS_BULKREAD);
490 :
491 : /* Get true root block from meta-page */
492 7954 : metapage = palloc_btree_page(state, BTREE_METAPAGE);
493 7954 : metad = BTPageGetMeta(metapage);
494 :
495 : /*
496 : * Certain deletion patterns can result in "skinny" B-Tree indexes, where
497 : * the fast root and true root differ.
498 : *
499 : * Start from the true root, not the fast root, unlike conventional index
500 : * scans. This approach is more thorough, and removes the risk of
501 : * following a stale fast root from the meta page.
502 : */
503 7954 : if (metad->btm_fastroot != metad->btm_root)
504 26 : ereport(DEBUG1,
505 : (errcode(ERRCODE_NO_DATA),
506 : errmsg_internal("harmless fast root mismatch in index \"%s\"",
507 : RelationGetRelationName(rel)),
508 : errdetail_internal("Fast root block %u (level %u) differs from true root block %u (level %u).",
509 : metad->btm_fastroot, metad->btm_fastlevel,
510 : metad->btm_root, metad->btm_level)));
511 :
512 : /*
513 : * Starting at the root, verify every level. Move left to right, top to
514 : * bottom. Note that there may be no pages other than the meta page (meta
515 : * page can indicate that root is P_NONE when the index is totally empty).
516 : */
517 7954 : previouslevel = InvalidBtreeLevel;
518 7954 : current.level = metad->btm_level;
519 7954 : current.leftmost = metad->btm_root;
520 7954 : current.istruerootlevel = true;
521 12942 : while (current.leftmost != P_NONE)
522 : {
523 : /*
524 : * Verify this level, and get left most page for next level down, if
525 : * not at leaf level
526 : */
527 5024 : current = bt_check_level_from_leftmost(state, current);
528 :
529 4988 : if (current.leftmost == InvalidBlockNumber)
530 0 : ereport(ERROR,
531 : (errcode(ERRCODE_INDEX_CORRUPTED),
532 : errmsg("index \"%s\" has no valid pages on level below %u or first level",
533 : RelationGetRelationName(rel), previouslevel)));
534 :
535 4988 : previouslevel = current.level;
536 : }
537 :
538 : /*
539 : * * Check whether heap contains unindexed/malformed tuples *
540 : */
541 7918 : if (state->heapallindexed)
542 : {
543 138 : IndexInfo *indexinfo = BuildIndexInfo(state->rel);
544 : TableScanDesc scan;
545 :
546 : /*
547 : * Create our own scan for table_index_build_scan(), rather than
548 : * getting it to do so for us. This is required so that we can
549 : * actually use the MVCC snapshot registered earlier.
550 : *
551 : * Note that table_index_build_scan() calls heap_endscan() for us.
552 : */
553 138 : scan = table_beginscan_strat(state->heaprel, /* relation */
554 : state->snapshot, /* snapshot */
555 : 0, /* number of keys */
556 : NULL, /* scan key */
557 : true, /* buffer access strategy OK */
558 : true); /* syncscan OK? */
559 :
560 : /*
561 : * Scan will behave as the first scan of a CREATE INDEX CONCURRENTLY
562 : * behaves.
563 : *
564 : * It's okay that we don't actually use the same lock strength for the
565 : * heap relation as any other ii_Concurrent caller would. We have no
566 : * reason to care about a concurrent VACUUM operation, since there
567 : * isn't going to be a second scan of the heap that needs to be sure
568 : * that there was no concurrent recycling of TIDs.
569 : */
570 134 : indexinfo->ii_Concurrent = true;
571 :
572 : /*
573 : * Don't wait for uncommitted tuple xact commit/abort when index is a
574 : * unique index on a catalog (or an index used by an exclusion
575 : * constraint). This could otherwise happen in the readonly case.
576 : */
577 134 : indexinfo->ii_Unique = false;
578 134 : indexinfo->ii_ExclusionOps = NULL;
579 134 : indexinfo->ii_ExclusionProcs = NULL;
580 134 : indexinfo->ii_ExclusionStrats = NULL;
581 :
582 134 : elog(DEBUG1, "verifying that tuples from index \"%s\" are present in \"%s\"",
583 : RelationGetRelationName(state->rel),
584 : RelationGetRelationName(state->heaprel));
585 :
586 134 : table_index_build_scan(state->heaprel, state->rel, indexinfo, true, false,
587 : bt_tuple_present_callback, state, scan);
588 :
589 134 : ereport(DEBUG1,
590 : (errmsg_internal("finished verifying presence of " INT64_FORMAT " tuples from table \"%s\" with bitset %.2f%% set",
591 : state->heaptuplespresent, RelationGetRelationName(heaprel),
592 : 100.0 * bloom_prop_bits_set(state->filter))));
593 :
594 134 : bloom_free(state->filter);
595 : }
596 :
597 : /* Be tidy: */
598 7914 : if (state->snapshot != InvalidSnapshot)
599 1376 : UnregisterSnapshot(state->snapshot);
600 7914 : MemoryContextDelete(state->targetcontext);
601 7914 : }
602 :
603 : /*
604 : * Given a left-most block at some level, move right, verifying each page
605 : * individually (with more verification across pages for "readonly"
606 : * callers). Caller should pass the true root page as the leftmost initially,
607 : * working their way down by passing what is returned for the last call here
608 : * until level 0 (leaf page level) was reached.
609 : *
610 : * Returns state for next call, if any. This includes left-most block number
611 : * one level lower that should be passed on next level/call, which is set to
612 : * P_NONE on last call here (when leaf level is verified). Level numbers
613 : * follow the nbtree convention: higher levels have higher numbers, because new
614 : * levels are added only due to a root page split. Note that prior to the
615 : * first root page split, the root is also a leaf page, so there is always a
616 : * level 0 (leaf level), and it's always the last level processed.
617 : *
618 : * Note on memory management: State's per-page context is reset here, between
619 : * each call to bt_target_page_check().
620 : */
621 : static BtreeLevel
622 5024 : bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
623 : {
624 : /* State to establish early, concerning entire level */
625 : BTPageOpaque opaque;
626 : MemoryContext oldcontext;
627 : BtreeLevel nextleveldown;
628 :
629 : /* Variables for iterating across level using right links */
630 5024 : BlockNumber leftcurrent = P_NONE;
631 5024 : BlockNumber current = level.leftmost;
632 :
633 : /* Initialize return state */
634 5024 : nextleveldown.leftmost = InvalidBlockNumber;
635 5024 : nextleveldown.level = InvalidBtreeLevel;
636 5024 : nextleveldown.istruerootlevel = false;
637 :
638 : /* Use page-level context for duration of this call */
639 5024 : oldcontext = MemoryContextSwitchTo(state->targetcontext);
640 :
641 5024 : elog(DEBUG1, "verifying level %u%s", level.level,
642 : level.istruerootlevel ?
643 : " (true root level)" : level.level == 0 ? " (leaf level)" : "");
644 :
645 5024 : state->prevrightlink = InvalidBlockNumber;
646 5024 : state->previncompletesplit = false;
647 :
648 : do
649 : {
650 : /* Don't rely on CHECK_FOR_INTERRUPTS() calls at lower level */
651 18252 : CHECK_FOR_INTERRUPTS();
652 :
653 : /* Initialize state for this iteration */
654 18252 : state->targetblock = current;
655 18252 : state->target = palloc_btree_page(state, state->targetblock);
656 18228 : state->targetlsn = PageGetLSN(state->target);
657 :
658 18228 : opaque = BTPageGetOpaque(state->target);
659 :
660 18228 : if (P_IGNORE(opaque))
661 : {
662 : /*
663 : * Since there cannot be a concurrent VACUUM operation in readonly
664 : * mode, and since a page has no links within other pages
665 : * (siblings and parent) once it is marked fully deleted, it
666 : * should be impossible to land on a fully deleted page in
667 : * readonly mode. See bt_child_check() for further details.
668 : *
669 : * The bt_child_check() P_ISDELETED() check is repeated here so
670 : * that pages that are only reachable through sibling links get
671 : * checked.
672 : */
673 0 : if (state->readonly && P_ISDELETED(opaque))
674 0 : ereport(ERROR,
675 : (errcode(ERRCODE_INDEX_CORRUPTED),
676 : errmsg("downlink or sibling link points to deleted block in index \"%s\"",
677 : RelationGetRelationName(state->rel)),
678 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
679 : current, leftcurrent, opaque->btpo_prev)));
680 :
681 0 : if (P_RIGHTMOST(opaque))
682 0 : ereport(ERROR,
683 : (errcode(ERRCODE_INDEX_CORRUPTED),
684 : errmsg("block %u fell off the end of index \"%s\"",
685 : current, RelationGetRelationName(state->rel))));
686 : else
687 0 : ereport(DEBUG1,
688 : (errcode(ERRCODE_NO_DATA),
689 : errmsg_internal("block %u of index \"%s\" concurrently deleted",
690 : current, RelationGetRelationName(state->rel))));
691 0 : goto nextpage;
692 : }
693 18228 : else if (nextleveldown.leftmost == InvalidBlockNumber)
694 : {
695 : /*
696 : * A concurrent page split could make the caller supplied leftmost
697 : * block no longer contain the leftmost page, or no longer be the
698 : * true root, but where that isn't possible due to heavyweight
699 : * locking, check that the first valid page meets caller's
700 : * expectations.
701 : */
702 5000 : if (state->readonly)
703 : {
704 90 : if (!bt_leftmost_ignoring_half_dead(state, current, opaque))
705 0 : ereport(ERROR,
706 : (errcode(ERRCODE_INDEX_CORRUPTED),
707 : errmsg("block %u is not leftmost in index \"%s\"",
708 : current, RelationGetRelationName(state->rel))));
709 :
710 90 : if (level.istruerootlevel && (!P_ISROOT(opaque) && !P_INCOMPLETE_SPLIT(opaque)))
711 0 : ereport(ERROR,
712 : (errcode(ERRCODE_INDEX_CORRUPTED),
713 : errmsg("block %u is not true root in index \"%s\"",
714 : current, RelationGetRelationName(state->rel))));
715 : }
716 :
717 : /*
718 : * Before beginning any non-trivial examination of level, prepare
719 : * state for next bt_check_level_from_leftmost() invocation for
720 : * the next level for the next level down (if any).
721 : *
722 : * There should be at least one non-ignorable page per level,
723 : * unless this is the leaf level, which is assumed by caller to be
724 : * final level.
725 : */
726 5000 : if (!P_ISLEAF(opaque))
727 : {
728 : IndexTuple itup;
729 : ItemId itemid;
730 :
731 : /* Internal page -- downlink gets leftmost on next level */
732 1110 : itemid = PageGetItemIdCareful(state, state->targetblock,
733 : state->target,
734 1110 : P_FIRSTDATAKEY(opaque));
735 1110 : itup = (IndexTuple) PageGetItem(state->target, itemid);
736 1110 : nextleveldown.leftmost = BTreeTupleGetDownLink(itup);
737 1110 : nextleveldown.level = opaque->btpo_level - 1;
738 : }
739 : else
740 : {
741 : /*
742 : * Leaf page -- final level caller must process.
743 : *
744 : * Note that this could also be the root page, if there has
745 : * been no root page split yet.
746 : */
747 3890 : nextleveldown.leftmost = P_NONE;
748 3890 : nextleveldown.level = InvalidBtreeLevel;
749 : }
750 :
751 : /*
752 : * Finished setting up state for this call/level. Control will
753 : * never end up back here in any future loop iteration for this
754 : * level.
755 : */
756 : }
757 :
758 : /*
759 : * Sibling links should be in mutual agreement. There arises
760 : * leftcurrent == P_NONE && btpo_prev != P_NONE when the left sibling
761 : * of the parent's low-key downlink is half-dead. (A half-dead page
762 : * has no downlink from its parent.) Under heavyweight locking, the
763 : * last bt_leftmost_ignoring_half_dead() validated this btpo_prev.
764 : * Without heavyweight locking, validation of the P_NONE case remains
765 : * unimplemented.
766 : */
767 18228 : if (opaque->btpo_prev != leftcurrent && leftcurrent != P_NONE)
768 0 : bt_recheck_sibling_links(state, opaque->btpo_prev, leftcurrent);
769 :
770 : /* Check level */
771 18228 : if (level.level != opaque->btpo_level)
772 0 : ereport(ERROR,
773 : (errcode(ERRCODE_INDEX_CORRUPTED),
774 : errmsg("leftmost down link for level points to block in index \"%s\" whose level is not one level down",
775 : RelationGetRelationName(state->rel)),
776 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
777 : current, level.level, opaque->btpo_level)));
778 :
779 : /* Verify invariants for page */
780 18228 : bt_target_page_check(state);
781 :
782 18216 : nextpage:
783 :
784 : /* Try to detect circular links */
785 18216 : if (current == leftcurrent || current == opaque->btpo_prev)
786 0 : ereport(ERROR,
787 : (errcode(ERRCODE_INDEX_CORRUPTED),
788 : errmsg("circular link chain found in block %u of index \"%s\"",
789 : current, RelationGetRelationName(state->rel))));
790 :
791 18216 : leftcurrent = current;
792 18216 : current = opaque->btpo_next;
793 :
794 18216 : if (state->lowkey)
795 : {
796 : Assert(state->readonly);
797 3724 : pfree(state->lowkey);
798 3724 : state->lowkey = NULL;
799 : }
800 :
801 : /*
802 : * Copy current target high key as the low key of right sibling.
803 : * Allocate memory in upper level context, so it would be cleared
804 : * after reset of target context.
805 : *
806 : * We only need the low key in corner cases of checking child high
807 : * keys. We use high key only when incomplete split on the child level
808 : * falls to the boundary of pages on the target level. See
809 : * bt_child_highkey_check() for details. So, typically we won't end
810 : * up doing anything with low key, but it's simpler for general case
811 : * high key verification to always have it available.
812 : *
813 : * The correctness of managing low key in the case of concurrent
814 : * splits wasn't investigated yet. Thankfully we only need low key
815 : * for readonly verification and concurrent splits won't happen.
816 : */
817 18216 : if (state->readonly && !P_RIGHTMOST(opaque))
818 : {
819 : IndexTuple itup;
820 : ItemId itemid;
821 :
822 3724 : itemid = PageGetItemIdCareful(state, state->targetblock,
823 : state->target, P_HIKEY);
824 3724 : itup = (IndexTuple) PageGetItem(state->target, itemid);
825 :
826 3724 : state->lowkey = MemoryContextAlloc(oldcontext, IndexTupleSize(itup));
827 3724 : memcpy(state->lowkey, itup, IndexTupleSize(itup));
828 : }
829 :
830 : /* Free page and associated memory for this iteration */
831 18216 : MemoryContextReset(state->targetcontext);
832 : }
833 18216 : while (current != P_NONE);
834 :
835 4988 : if (state->lowkey)
836 : {
837 : Assert(state->readonly);
838 0 : pfree(state->lowkey);
839 0 : state->lowkey = NULL;
840 : }
841 :
842 : /* Don't change context for caller */
843 4988 : MemoryContextSwitchTo(oldcontext);
844 :
845 4988 : return nextleveldown;
846 : }
847 :
848 : /* Check visibility of the table entry referenced by nbtree index */
849 : static bool
850 798 : heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid)
851 : {
852 : bool tid_visible;
853 :
854 798 : TupleTableSlot *slot = table_slot_create(state->heaprel, NULL);
855 :
856 798 : tid_visible = table_tuple_fetch_row_version(state->heaprel,
857 : tid, state->snapshot, slot);
858 798 : if (slot != NULL)
859 798 : ExecDropSingleTupleTableSlot(slot);
860 :
861 798 : return tid_visible;
862 : }
863 :
864 : /*
865 : * Prepare an error message for unique constrain violation in
866 : * a btree index and report ERROR.
867 : */
868 : static void
869 6 : bt_report_duplicate(BtreeCheckState *state,
870 : BtreeLastVisibleEntry *lVis,
871 : ItemPointer nexttid, BlockNumber nblock, OffsetNumber noffset,
872 : int nposting)
873 : {
874 : char *htid,
875 : *nhtid,
876 : *itid,
877 6 : *nitid = "",
878 6 : *pposting = "",
879 6 : *pnposting = "";
880 :
881 6 : htid = psprintf("tid=(%u,%u)",
882 6 : ItemPointerGetBlockNumberNoCheck(lVis->tid),
883 6 : ItemPointerGetOffsetNumberNoCheck(lVis->tid));
884 6 : nhtid = psprintf("tid=(%u,%u)",
885 : ItemPointerGetBlockNumberNoCheck(nexttid),
886 6 : ItemPointerGetOffsetNumberNoCheck(nexttid));
887 6 : itid = psprintf("tid=(%u,%u)", lVis->blkno, lVis->offset);
888 :
889 6 : if (nblock != lVis->blkno || noffset != lVis->offset)
890 6 : nitid = psprintf(" tid=(%u,%u)", nblock, noffset);
891 :
892 6 : if (lVis->postingIndex >= 0)
893 0 : pposting = psprintf(" posting %u", lVis->postingIndex);
894 :
895 6 : if (nposting >= 0)
896 0 : pnposting = psprintf(" posting %u", nposting);
897 :
898 6 : ereport(ERROR,
899 : (errcode(ERRCODE_INDEX_CORRUPTED),
900 : errmsg("index uniqueness is violated for index \"%s\"",
901 : RelationGetRelationName(state->rel)),
902 : errdetail("Index %s%s and%s%s (point to heap %s and %s) page lsn=%X/%08X.",
903 : itid, pposting, nitid, pnposting, htid, nhtid,
904 : LSN_FORMAT_ARGS(state->targetlsn))));
905 : }
906 :
907 : /* Check if current nbtree leaf entry complies with UNIQUE constraint */
908 : static void
909 766 : bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
910 : BlockNumber targetblock, OffsetNumber offset,
911 : BtreeLastVisibleEntry *lVis)
912 : {
913 : ItemPointer tid;
914 766 : bool has_visible_entry = false;
915 :
916 : Assert(targetblock != P_NONE);
917 :
918 : /*
919 : * Current tuple has posting list. Report duplicate if TID of any posting
920 : * list entry is visible and lVis->tid is valid.
921 : */
922 766 : if (BTreeTupleIsPosting(itup))
923 : {
924 96 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
925 : {
926 64 : tid = BTreeTupleGetPostingN(itup, i);
927 64 : if (heap_entry_is_visible(state, tid))
928 : {
929 32 : has_visible_entry = true;
930 32 : if (ItemPointerIsValid(lVis->tid))
931 : {
932 0 : bt_report_duplicate(state,
933 : lVis,
934 : tid, targetblock,
935 : offset, i);
936 : }
937 :
938 : /*
939 : * Prevent double reporting unique constraint violation
940 : * between the posting list entries of the first tuple on the
941 : * page after cross-page check.
942 : */
943 32 : if (lVis->blkno != targetblock && ItemPointerIsValid(lVis->tid))
944 0 : return;
945 :
946 32 : lVis->blkno = targetblock;
947 32 : lVis->offset = offset;
948 32 : lVis->postingIndex = i;
949 32 : lVis->tid = tid;
950 : }
951 : }
952 : }
953 :
954 : /*
955 : * Current tuple has no posting list. If TID is visible save info about it
956 : * for the next comparisons in the loop in bt_target_page_check(). Report
957 : * duplicate if lVis->tid is already valid.
958 : */
959 : else
960 : {
961 734 : tid = BTreeTupleGetHeapTID(itup);
962 734 : if (heap_entry_is_visible(state, tid))
963 : {
964 54 : has_visible_entry = true;
965 54 : if (ItemPointerIsValid(lVis->tid))
966 : {
967 6 : bt_report_duplicate(state,
968 : lVis,
969 : tid, targetblock,
970 : offset, -1);
971 : }
972 :
973 48 : lVis->blkno = targetblock;
974 48 : lVis->offset = offset;
975 48 : lVis->tid = tid;
976 48 : lVis->postingIndex = -1;
977 : }
978 : }
979 :
980 760 : if (!has_visible_entry &&
981 680 : lVis->blkno != InvalidBlockNumber &&
982 42 : lVis->blkno != targetblock)
983 : {
984 0 : char *posting = "";
985 :
986 0 : if (lVis->postingIndex >= 0)
987 0 : posting = psprintf(" posting %u", lVis->postingIndex);
988 0 : ereport(DEBUG1,
989 : (errcode(ERRCODE_NO_DATA),
990 : errmsg("index uniqueness can not be checked for index tid=(%u,%u) in index \"%s\"",
991 : targetblock, offset,
992 : RelationGetRelationName(state->rel)),
993 : errdetail("It doesn't have visible heap tids and key is equal to the tid=(%u,%u)%s (points to heap tid=(%u,%u)).",
994 : lVis->blkno, lVis->offset, posting,
995 : ItemPointerGetBlockNumberNoCheck(lVis->tid),
996 : ItemPointerGetOffsetNumberNoCheck(lVis->tid)),
997 : errhint("VACUUM the table and repeat the check.")));
998 : }
999 : }
1000 :
1001 : /*
1002 : * Like P_LEFTMOST(start_opaque), but accept an arbitrarily-long chain of
1003 : * half-dead, sibling-linked pages to the left. If a half-dead page appears
1004 : * under state->readonly, the database exited recovery between the first-stage
1005 : * and second-stage WAL records of a deletion.
1006 : */
1007 : static bool
1008 112 : bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
1009 : BlockNumber start,
1010 : BTPageOpaque start_opaque)
1011 : {
1012 112 : BlockNumber reached = start_opaque->btpo_prev,
1013 112 : reached_from = start;
1014 112 : bool all_half_dead = true;
1015 :
1016 : /*
1017 : * To handle the !readonly case, we'd need to accept BTP_DELETED pages and
1018 : * potentially observe nbtree/README "Page deletion and backwards scans".
1019 : */
1020 : Assert(state->readonly);
1021 :
1022 116 : while (reached != P_NONE && all_half_dead)
1023 : {
1024 4 : Page page = palloc_btree_page(state, reached);
1025 4 : BTPageOpaque reached_opaque = BTPageGetOpaque(page);
1026 :
1027 4 : CHECK_FOR_INTERRUPTS();
1028 :
1029 : /*
1030 : * Try to detect btpo_prev circular links. _bt_unlink_halfdead_page()
1031 : * writes that side-links will continue to point to the siblings.
1032 : * Check btpo_next for that property.
1033 : */
1034 4 : all_half_dead = P_ISHALFDEAD(reached_opaque) &&
1035 4 : reached != start &&
1036 8 : reached != reached_from &&
1037 4 : reached_opaque->btpo_next == reached_from;
1038 4 : if (all_half_dead)
1039 : {
1040 4 : XLogRecPtr pagelsn = PageGetLSN(page);
1041 :
1042 : /* pagelsn should point to an XLOG_BTREE_MARK_PAGE_HALFDEAD */
1043 4 : ereport(DEBUG1,
1044 : (errcode(ERRCODE_NO_DATA),
1045 : errmsg_internal("harmless interrupted page deletion detected in index \"%s\"",
1046 : RelationGetRelationName(state->rel)),
1047 : errdetail_internal("Block=%u right block=%u page lsn=%X/%08X.",
1048 : reached, reached_from,
1049 : LSN_FORMAT_ARGS(pagelsn))));
1050 :
1051 4 : reached_from = reached;
1052 4 : reached = reached_opaque->btpo_prev;
1053 : }
1054 :
1055 4 : pfree(page);
1056 : }
1057 :
1058 112 : return all_half_dead;
1059 : }
1060 :
1061 : /*
1062 : * Raise an error when target page's left link does not point back to the
1063 : * previous target page, called leftcurrent here. The leftcurrent page's
1064 : * right link was followed to get to the current target page, and we expect
1065 : * mutual agreement among leftcurrent and the current target page. Make sure
1066 : * that this condition has definitely been violated in the !readonly case,
1067 : * where concurrent page splits are something that we need to deal with.
1068 : *
1069 : * Cross-page inconsistencies involving pages that don't agree about being
1070 : * siblings are known to be a particularly good indicator of corruption
1071 : * involving partial writes/lost updates. The bt_right_page_check_scankey
1072 : * check also provides a way of detecting cross-page inconsistencies for
1073 : * !readonly callers, but it can only detect sibling pages that have an
1074 : * out-of-order keyspace, which can't catch many of the problems that we
1075 : * expect to catch here.
1076 : *
1077 : * The classic example of the kind of inconsistency that we can only catch
1078 : * with this check (when in !readonly mode) involves three sibling pages that
1079 : * were affected by a faulty page split at some point in the past. The
1080 : * effects of the split are reflected in the original page and its new right
1081 : * sibling page, with a lack of any accompanying changes for the _original_
1082 : * right sibling page. The original right sibling page's left link fails to
1083 : * point to the new right sibling page (its left link still points to the
1084 : * original page), even though the first phase of a page split is supposed to
1085 : * work as a single atomic action. This subtle inconsistency will probably
1086 : * only break backwards scans in practice.
1087 : *
1088 : * Note that this is the only place where amcheck will "couple" buffer locks
1089 : * (and only for !readonly callers). In general we prefer to avoid more
1090 : * thorough cross-page checks in !readonly mode, but it seems worth the
1091 : * complexity here. Also, the performance overhead of performing lock
1092 : * coupling here is negligible in practice. Control only reaches here with a
1093 : * non-corrupt index when there is a concurrent page split at the instant
1094 : * caller crossed over to target page from leftcurrent page.
1095 : */
1096 : static void
1097 0 : bt_recheck_sibling_links(BtreeCheckState *state,
1098 : BlockNumber btpo_prev_from_target,
1099 : BlockNumber leftcurrent)
1100 : {
1101 : /* passing metapage to BTPageGetOpaque() would give irrelevant findings */
1102 : Assert(leftcurrent != P_NONE);
1103 :
1104 0 : if (!state->readonly)
1105 : {
1106 : Buffer lbuf;
1107 : Buffer newtargetbuf;
1108 : Page page;
1109 : BTPageOpaque opaque;
1110 : BlockNumber newtargetblock;
1111 :
1112 : /* Couple locks in the usual order for nbtree: Left to right */
1113 0 : lbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM, leftcurrent,
1114 : RBM_NORMAL, state->checkstrategy);
1115 0 : LockBuffer(lbuf, BT_READ);
1116 0 : _bt_checkpage(state->rel, lbuf);
1117 0 : page = BufferGetPage(lbuf);
1118 0 : opaque = BTPageGetOpaque(page);
1119 0 : if (P_ISDELETED(opaque))
1120 : {
1121 : /*
1122 : * Cannot reason about concurrently deleted page -- the left link
1123 : * in the page to the right is expected to point to some other
1124 : * page to the left (not leftcurrent page).
1125 : *
1126 : * Note that we deliberately don't give up with a half-dead page.
1127 : */
1128 0 : UnlockReleaseBuffer(lbuf);
1129 0 : return;
1130 : }
1131 :
1132 0 : newtargetblock = opaque->btpo_next;
1133 : /* Avoid self-deadlock when newtargetblock == leftcurrent */
1134 0 : if (newtargetblock != leftcurrent)
1135 : {
1136 0 : newtargetbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM,
1137 : newtargetblock, RBM_NORMAL,
1138 : state->checkstrategy);
1139 0 : LockBuffer(newtargetbuf, BT_READ);
1140 0 : _bt_checkpage(state->rel, newtargetbuf);
1141 0 : page = BufferGetPage(newtargetbuf);
1142 0 : opaque = BTPageGetOpaque(page);
1143 : /* btpo_prev_from_target may have changed; update it */
1144 0 : btpo_prev_from_target = opaque->btpo_prev;
1145 : }
1146 : else
1147 : {
1148 : /*
1149 : * leftcurrent right sibling points back to leftcurrent block.
1150 : * Index is corrupt. Easiest way to handle this is to pretend
1151 : * that we actually read from a distinct page that has an invalid
1152 : * block number in its btpo_prev.
1153 : */
1154 0 : newtargetbuf = InvalidBuffer;
1155 0 : btpo_prev_from_target = InvalidBlockNumber;
1156 : }
1157 :
1158 : /*
1159 : * No need to check P_ISDELETED here, since new target block cannot be
1160 : * marked deleted as long as we hold a lock on lbuf
1161 : */
1162 0 : if (BufferIsValid(newtargetbuf))
1163 0 : UnlockReleaseBuffer(newtargetbuf);
1164 0 : UnlockReleaseBuffer(lbuf);
1165 :
1166 0 : if (btpo_prev_from_target == leftcurrent)
1167 : {
1168 : /* Report split in left sibling, not target (or new target) */
1169 0 : ereport(DEBUG1,
1170 : (errcode(ERRCODE_INTERNAL_ERROR),
1171 : errmsg_internal("harmless concurrent page split detected in index \"%s\"",
1172 : RelationGetRelationName(state->rel)),
1173 : errdetail_internal("Block=%u new right sibling=%u original right sibling=%u.",
1174 : leftcurrent, newtargetblock,
1175 : state->targetblock)));
1176 0 : return;
1177 : }
1178 :
1179 : /*
1180 : * Index is corrupt. Make sure that we report correct target page.
1181 : *
1182 : * This could have changed in cases where there was a concurrent page
1183 : * split, as well as index corruption (at least in theory). Note that
1184 : * btpo_prev_from_target was already updated above.
1185 : */
1186 0 : state->targetblock = newtargetblock;
1187 : }
1188 :
1189 0 : ereport(ERROR,
1190 : (errcode(ERRCODE_INDEX_CORRUPTED),
1191 : errmsg("left link/right link pair in index \"%s\" not in agreement",
1192 : RelationGetRelationName(state->rel)),
1193 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
1194 : state->targetblock, leftcurrent,
1195 : btpo_prev_from_target)));
1196 : }
1197 :
1198 : /*
1199 : * Function performs the following checks on target page, or pages ancillary to
1200 : * target page:
1201 : *
1202 : * - That every "real" data item is less than or equal to the high key, which
1203 : * is an upper bound on the items on the page. Data items should be
1204 : * strictly less than the high key when the page is an internal page.
1205 : *
1206 : * - That within the page, every data item is strictly less than the item
1207 : * immediately to its right, if any (i.e., that the items are in order
1208 : * within the page, so that the binary searches performed by index scans are
1209 : * sane).
1210 : *
1211 : * - That the last data item stored on the page is strictly less than the
1212 : * first data item on the page to the right (when such a first item is
1213 : * available).
1214 : *
1215 : * - Various checks on the structure of tuples themselves. For example, check
1216 : * that non-pivot tuples have no truncated attributes.
1217 : *
1218 : * - For index with unique constraint make sure that only one of table entries
1219 : * for equal keys is visible.
1220 : *
1221 : * Furthermore, when state passed shows ShareLock held, function also checks:
1222 : *
1223 : * - That all child pages respect strict lower bound from parent's pivot
1224 : * tuple.
1225 : *
1226 : * - That downlink to block was encountered in parent where that's expected.
1227 : *
1228 : * - That high keys of child pages matches corresponding pivot keys in parent.
1229 : *
1230 : * This is also where heapallindexed callers use their Bloom filter to
1231 : * fingerprint IndexTuples for later table_index_build_scan() verification.
1232 : *
1233 : * Note: Memory allocated in this routine is expected to be released by caller
1234 : * resetting state->targetcontext.
1235 : */
1236 : static void
1237 18228 : bt_target_page_check(BtreeCheckState *state)
1238 : {
1239 : OffsetNumber offset;
1240 : OffsetNumber max;
1241 : BTPageOpaque topaque;
1242 :
1243 : /* Last visible entry info for checking indexes with unique constraint */
1244 18228 : BtreeLastVisibleEntry lVis = {InvalidBlockNumber, InvalidOffsetNumber, -1, NULL};
1245 :
1246 18228 : topaque = BTPageGetOpaque(state->target);
1247 18228 : max = PageGetMaxOffsetNumber(state->target);
1248 :
1249 18228 : elog(DEBUG2, "verifying %u items on %s block %u", max,
1250 : P_ISLEAF(topaque) ? "leaf" : "internal", state->targetblock);
1251 :
1252 : /*
1253 : * Check the number of attributes in high key. Note, rightmost page
1254 : * doesn't contain a high key, so nothing to check
1255 : */
1256 18228 : if (!P_RIGHTMOST(topaque))
1257 : {
1258 : ItemId itemid;
1259 : IndexTuple itup;
1260 :
1261 : /* Verify line pointer before checking tuple */
1262 13236 : itemid = PageGetItemIdCareful(state, state->targetblock,
1263 : state->target, P_HIKEY);
1264 13236 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1265 : P_HIKEY))
1266 : {
1267 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1268 0 : ereport(ERROR,
1269 : (errcode(ERRCODE_INDEX_CORRUPTED),
1270 : errmsg("wrong number of high key index tuple attributes in index \"%s\"",
1271 : RelationGetRelationName(state->rel)),
1272 : errdetail_internal("Index block=%u natts=%u block type=%s page lsn=%X/%08X.",
1273 : state->targetblock,
1274 : BTreeTupleGetNAtts(itup, state->rel),
1275 : P_ISLEAF(topaque) ? "heap" : "index",
1276 : LSN_FORMAT_ARGS(state->targetlsn))));
1277 : }
1278 : }
1279 :
1280 : /*
1281 : * Loop over page items, starting from first non-highkey item, not high
1282 : * key (if any). Most tests are not performed for the "negative infinity"
1283 : * real item (if any).
1284 : */
1285 18228 : for (offset = P_FIRSTDATAKEY(topaque);
1286 4067624 : offset <= max;
1287 4049396 : offset = OffsetNumberNext(offset))
1288 : {
1289 : ItemId itemid;
1290 : IndexTuple itup;
1291 : size_t tupsize;
1292 : BTScanInsert skey;
1293 : bool lowersizelimit;
1294 : ItemPointer scantid;
1295 :
1296 : /*
1297 : * True if we already called bt_entry_unique_check() for the current
1298 : * item. This helps to avoid visiting the heap for keys, which are
1299 : * anyway presented only once and can't comprise a unique violation.
1300 : */
1301 4049408 : bool unique_checked = false;
1302 :
1303 4049408 : CHECK_FOR_INTERRUPTS();
1304 :
1305 4049408 : itemid = PageGetItemIdCareful(state, state->targetblock,
1306 : state->target, offset);
1307 4049408 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1308 4049408 : tupsize = IndexTupleSize(itup);
1309 :
1310 : /*
1311 : * lp_len should match the IndexTuple reported length exactly, since
1312 : * lp_len is completely redundant in indexes, and both sources of
1313 : * tuple length are MAXALIGN()'d. nbtree does not use lp_len all that
1314 : * frequently, and is surprisingly tolerant of corrupt lp_len fields.
1315 : */
1316 4049408 : if (tupsize != ItemIdGetLength(itemid))
1317 0 : ereport(ERROR,
1318 : (errcode(ERRCODE_INDEX_CORRUPTED),
1319 : errmsg("index tuple size does not equal lp_len in index \"%s\"",
1320 : RelationGetRelationName(state->rel)),
1321 : errdetail_internal("Index tid=(%u,%u) tuple size=%zu lp_len=%u page lsn=%X/%08X.",
1322 : state->targetblock, offset,
1323 : tupsize, ItemIdGetLength(itemid),
1324 : LSN_FORMAT_ARGS(state->targetlsn)),
1325 : errhint("This could be a torn page problem.")));
1326 :
1327 : /* Check the number of index tuple attributes */
1328 4049408 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1329 : offset))
1330 : {
1331 : ItemPointer tid;
1332 : char *itid,
1333 : *htid;
1334 :
1335 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1336 0 : tid = BTreeTupleGetPointsToTID(itup);
1337 0 : htid = psprintf("(%u,%u)",
1338 : ItemPointerGetBlockNumberNoCheck(tid),
1339 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1340 :
1341 0 : ereport(ERROR,
1342 : (errcode(ERRCODE_INDEX_CORRUPTED),
1343 : errmsg("wrong number of index tuple attributes in index \"%s\"",
1344 : RelationGetRelationName(state->rel)),
1345 : errdetail_internal("Index tid=%s natts=%u points to %s tid=%s page lsn=%X/%08X.",
1346 : itid,
1347 : BTreeTupleGetNAtts(itup, state->rel),
1348 : P_ISLEAF(topaque) ? "heap" : "index",
1349 : htid,
1350 : LSN_FORMAT_ARGS(state->targetlsn))));
1351 : }
1352 :
1353 : /*
1354 : * Don't try to generate scankey using "negative infinity" item on
1355 : * internal pages. They are always truncated to zero attributes.
1356 : */
1357 4049408 : if (offset_is_negative_infinity(topaque, offset))
1358 : {
1359 : /*
1360 : * We don't call bt_child_check() for "negative infinity" items.
1361 : * But if we're performing downlink connectivity check, we do it
1362 : * for every item including "negative infinity" one.
1363 : */
1364 1114 : if (!P_ISLEAF(topaque) && state->readonly)
1365 : {
1366 24 : bt_child_highkey_check(state,
1367 : offset,
1368 : NULL,
1369 : topaque->btpo_level);
1370 : }
1371 1114 : continue;
1372 : }
1373 :
1374 : /*
1375 : * Readonly callers may optionally verify that non-pivot tuples can
1376 : * each be found by an independent search that starts from the root.
1377 : * Note that we deliberately don't do individual searches for each
1378 : * TID, since the posting list itself is validated by other checks.
1379 : */
1380 4048294 : if (state->rootdescend && P_ISLEAF(topaque) &&
1381 402196 : !bt_rootdescend(state, itup))
1382 : {
1383 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1384 : char *itid,
1385 : *htid;
1386 :
1387 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1388 0 : htid = psprintf("(%u,%u)", ItemPointerGetBlockNumber(tid),
1389 0 : ItemPointerGetOffsetNumber(tid));
1390 :
1391 0 : ereport(ERROR,
1392 : (errcode(ERRCODE_INDEX_CORRUPTED),
1393 : errmsg("could not find tuple using search from root page in index \"%s\"",
1394 : RelationGetRelationName(state->rel)),
1395 : errdetail_internal("Index tid=%s points to heap tid=%s page lsn=%X/%08X.",
1396 : itid, htid,
1397 : LSN_FORMAT_ARGS(state->targetlsn))));
1398 : }
1399 :
1400 : /*
1401 : * If tuple is a posting list tuple, make sure posting list TIDs are
1402 : * in order
1403 : */
1404 4048294 : if (BTreeTupleIsPosting(itup))
1405 : {
1406 : ItemPointerData last;
1407 : ItemPointer current;
1408 :
1409 22338 : ItemPointerCopy(BTreeTupleGetHeapTID(itup), &last);
1410 :
1411 164668 : for (int i = 1; i < BTreeTupleGetNPosting(itup); i++)
1412 : {
1413 :
1414 142330 : current = BTreeTupleGetPostingN(itup, i);
1415 :
1416 142330 : if (ItemPointerCompare(current, &last) <= 0)
1417 : {
1418 0 : char *itid = psprintf("(%u,%u)", state->targetblock, offset);
1419 :
1420 0 : ereport(ERROR,
1421 : (errcode(ERRCODE_INDEX_CORRUPTED),
1422 : errmsg_internal("posting list contains misplaced TID in index \"%s\"",
1423 : RelationGetRelationName(state->rel)),
1424 : errdetail_internal("Index tid=%s posting list offset=%d page lsn=%X/%08X.",
1425 : itid, i,
1426 : LSN_FORMAT_ARGS(state->targetlsn))));
1427 : }
1428 :
1429 142330 : ItemPointerCopy(current, &last);
1430 : }
1431 : }
1432 :
1433 : /* Build insertion scankey for current page offset */
1434 4048294 : skey = bt_mkscankey_pivotsearch(state->rel, itup);
1435 :
1436 : /*
1437 : * Make sure tuple size does not exceed the relevant BTREE_VERSION
1438 : * specific limit.
1439 : *
1440 : * BTREE_VERSION 4 (which introduced heapkeyspace rules) requisitioned
1441 : * a small amount of space from BTMaxItemSize() in order to ensure
1442 : * that suffix truncation always has enough space to add an explicit
1443 : * heap TID back to a tuple -- we pessimistically assume that every
1444 : * newly inserted tuple will eventually need to have a heap TID
1445 : * appended during a future leaf page split, when the tuple becomes
1446 : * the basis of the new high key (pivot tuple) for the leaf page.
1447 : *
1448 : * Since the reclaimed space is reserved for that purpose, we must not
1449 : * enforce the slightly lower limit when the extra space has been used
1450 : * as intended. In other words, there is only a cross-version
1451 : * difference in the limit on tuple size within leaf pages.
1452 : *
1453 : * Still, we're particular about the details within BTREE_VERSION 4
1454 : * internal pages. Pivot tuples may only use the extra space for its
1455 : * designated purpose. Enforce the lower limit for pivot tuples when
1456 : * an explicit heap TID isn't actually present. (In all other cases
1457 : * suffix truncation is guaranteed to generate a pivot tuple that's no
1458 : * larger than the firstright tuple provided to it by its caller.)
1459 : */
1460 8096588 : lowersizelimit = skey->heapkeyspace &&
1461 4048294 : (P_ISLEAF(topaque) || BTreeTupleGetHeapTID(itup) == NULL);
1462 4048294 : if (tupsize > (lowersizelimit ? BTMaxItemSize : BTMaxItemSizeNoHeapTid))
1463 : {
1464 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1465 : char *itid,
1466 : *htid;
1467 :
1468 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1469 0 : htid = psprintf("(%u,%u)",
1470 : ItemPointerGetBlockNumberNoCheck(tid),
1471 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1472 :
1473 0 : ereport(ERROR,
1474 : (errcode(ERRCODE_INDEX_CORRUPTED),
1475 : errmsg("index row size %zu exceeds maximum for index \"%s\"",
1476 : tupsize, RelationGetRelationName(state->rel)),
1477 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%08X.",
1478 : itid,
1479 : P_ISLEAF(topaque) ? "heap" : "index",
1480 : htid,
1481 : LSN_FORMAT_ARGS(state->targetlsn))));
1482 : }
1483 :
1484 : /* Fingerprint leaf page tuples (those that point to the heap) */
1485 4048294 : if (state->heapallindexed && P_ISLEAF(topaque) && !ItemIdIsDead(itemid))
1486 : {
1487 : IndexTuple norm;
1488 :
1489 1014108 : if (BTreeTupleIsPosting(itup))
1490 : {
1491 : /* Fingerprint all elements as distinct "plain" tuples */
1492 58222 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
1493 : {
1494 : IndexTuple logtuple;
1495 :
1496 57804 : logtuple = bt_posting_plain_tuple(itup, i);
1497 57804 : norm = bt_normalize_tuple(state, logtuple);
1498 57804 : bloom_add_element(state->filter, (unsigned char *) norm,
1499 : IndexTupleSize(norm));
1500 : /* Be tidy */
1501 57804 : if (norm != logtuple)
1502 4 : pfree(norm);
1503 57804 : pfree(logtuple);
1504 : }
1505 : }
1506 : else
1507 : {
1508 1013690 : norm = bt_normalize_tuple(state, itup);
1509 1013690 : bloom_add_element(state->filter, (unsigned char *) norm,
1510 : IndexTupleSize(norm));
1511 : /* Be tidy */
1512 1013690 : if (norm != itup)
1513 2 : pfree(norm);
1514 : }
1515 : }
1516 :
1517 : /*
1518 : * * High key check *
1519 : *
1520 : * If there is a high key (if this is not the rightmost page on its
1521 : * entire level), check that high key actually is upper bound on all
1522 : * page items. If this is a posting list tuple, we'll need to set
1523 : * scantid to be highest TID in posting list.
1524 : *
1525 : * We prefer to check all items against high key rather than checking
1526 : * just the last and trusting that the operator class obeys the
1527 : * transitive law (which implies that all previous items also
1528 : * respected the high key invariant if they pass the item order
1529 : * check).
1530 : *
1531 : * Ideally, we'd compare every item in the index against every other
1532 : * item in the index, and not trust opclass obedience of the
1533 : * transitive law to bridge the gap between children and their
1534 : * grandparents (as well as great-grandparents, and so on). We don't
1535 : * go to those lengths because that would be prohibitively expensive,
1536 : * and probably not markedly more effective in practice.
1537 : *
1538 : * On the leaf level, we check that the key is <= the highkey.
1539 : * However, on non-leaf levels we check that the key is < the highkey,
1540 : * because the high key is "just another separator" rather than a copy
1541 : * of some existing key item; we expect it to be unique among all keys
1542 : * on the same level. (Suffix truncation will sometimes produce a
1543 : * leaf highkey that is an untruncated copy of the lastleft item, but
1544 : * never any other item, which necessitates weakening the leaf level
1545 : * check to <=.)
1546 : *
1547 : * Full explanation for why a highkey is never truly a copy of another
1548 : * item from the same level on internal levels:
1549 : *
1550 : * While the new left page's high key is copied from the first offset
1551 : * on the right page during an internal page split, that's not the
1552 : * full story. In effect, internal pages are split in the middle of
1553 : * the firstright tuple, not between the would-be lastleft and
1554 : * firstright tuples: the firstright key ends up on the left side as
1555 : * left's new highkey, and the firstright downlink ends up on the
1556 : * right side as right's new "negative infinity" item. The negative
1557 : * infinity tuple is truncated to zero attributes, so we're only left
1558 : * with the downlink. In other words, the copying is just an
1559 : * implementation detail of splitting in the middle of a (pivot)
1560 : * tuple. (See also: "Notes About Data Representation" in the nbtree
1561 : * README.)
1562 : */
1563 4048294 : scantid = skey->scantid;
1564 4048294 : if (state->heapkeyspace && BTreeTupleIsPosting(itup))
1565 22338 : skey->scantid = BTreeTupleGetMaxHeapTID(itup);
1566 :
1567 7750028 : if (!P_RIGHTMOST(topaque) &&
1568 3701734 : !(P_ISLEAF(topaque) ? invariant_leq_offset(state, skey, P_HIKEY) :
1569 1132 : invariant_l_offset(state, skey, P_HIKEY)))
1570 : {
1571 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1572 : char *itid,
1573 : *htid;
1574 :
1575 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1576 0 : htid = psprintf("(%u,%u)",
1577 : ItemPointerGetBlockNumberNoCheck(tid),
1578 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1579 :
1580 0 : ereport(ERROR,
1581 : (errcode(ERRCODE_INDEX_CORRUPTED),
1582 : errmsg("high key invariant violated for index \"%s\"",
1583 : RelationGetRelationName(state->rel)),
1584 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%08X.",
1585 : itid,
1586 : P_ISLEAF(topaque) ? "heap" : "index",
1587 : htid,
1588 : LSN_FORMAT_ARGS(state->targetlsn))));
1589 : }
1590 : /* Reset, in case scantid was set to (itup) posting tuple's max TID */
1591 4048294 : skey->scantid = scantid;
1592 :
1593 : /*
1594 : * * Item order check *
1595 : *
1596 : * Check that items are stored on page in logical order, by checking
1597 : * current item is strictly less than next item (if any).
1598 : */
1599 4048294 : if (OffsetNumberNext(offset) <= max &&
1600 4030082 : !invariant_l_offset(state, skey, OffsetNumberNext(offset)))
1601 : {
1602 : ItemPointer tid;
1603 : char *itid,
1604 : *htid,
1605 : *nitid,
1606 : *nhtid;
1607 :
1608 6 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1609 6 : tid = BTreeTupleGetPointsToTID(itup);
1610 6 : htid = psprintf("(%u,%u)",
1611 : ItemPointerGetBlockNumberNoCheck(tid),
1612 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1613 6 : nitid = psprintf("(%u,%u)", state->targetblock,
1614 6 : OffsetNumberNext(offset));
1615 :
1616 : /* Reuse itup to get pointed-to heap location of second item */
1617 6 : itemid = PageGetItemIdCareful(state, state->targetblock,
1618 : state->target,
1619 6 : OffsetNumberNext(offset));
1620 6 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1621 6 : tid = BTreeTupleGetPointsToTID(itup);
1622 6 : nhtid = psprintf("(%u,%u)",
1623 : ItemPointerGetBlockNumberNoCheck(tid),
1624 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1625 :
1626 6 : ereport(ERROR,
1627 : (errcode(ERRCODE_INDEX_CORRUPTED),
1628 : errmsg("item order invariant violated for index \"%s\"",
1629 : RelationGetRelationName(state->rel)),
1630 : errdetail_internal("Lower index tid=%s (points to %s tid=%s) higher index tid=%s (points to %s tid=%s) page lsn=%X/%08X.",
1631 : itid,
1632 : P_ISLEAF(topaque) ? "heap" : "index",
1633 : htid,
1634 : nitid,
1635 : P_ISLEAF(topaque) ? "heap" : "index",
1636 : nhtid,
1637 : LSN_FORMAT_ARGS(state->targetlsn))));
1638 : }
1639 :
1640 : /*
1641 : * If the index is unique verify entries uniqueness by checking the
1642 : * heap tuples visibility. Immediately check posting tuples and
1643 : * tuples with repeated keys. Postpone check for keys, which have the
1644 : * first appearance.
1645 : */
1646 4048288 : if (state->checkunique && state->indexinfo->ii_Unique &&
1647 579550 : P_ISLEAF(topaque) && !skey->anynullkeys &&
1648 578328 : (BTreeTupleIsPosting(itup) || ItemPointerIsValid(lVis.tid)))
1649 : {
1650 80 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1651 : &lVis);
1652 74 : unique_checked = true;
1653 : }
1654 :
1655 4048282 : if (state->checkunique && state->indexinfo->ii_Unique &&
1656 290364 : P_ISLEAF(topaque) && OffsetNumberNext(offset) <= max)
1657 : {
1658 : /* Save current scankey tid */
1659 287732 : scantid = skey->scantid;
1660 :
1661 : /*
1662 : * Invalidate scankey tid to make _bt_compare compare only keys in
1663 : * the item to report equality even if heap TIDs are different
1664 : */
1665 287732 : skey->scantid = NULL;
1666 :
1667 : /*
1668 : * If next key tuple is different, invalidate last visible entry
1669 : * data (whole index tuple or last posting in index tuple). Key
1670 : * containing null value does not violate unique constraint and
1671 : * treated as different to any other key.
1672 : *
1673 : * If the next key is the same as the previous one, do the
1674 : * bt_entry_unique_check() call if it was postponed.
1675 : */
1676 287732 : if (_bt_compare(state->rel, skey, state->target,
1677 288558 : OffsetNumberNext(offset)) != 0 || skey->anynullkeys)
1678 : {
1679 287046 : lVis.blkno = InvalidBlockNumber;
1680 287046 : lVis.offset = InvalidOffsetNumber;
1681 287046 : lVis.postingIndex = -1;
1682 287046 : lVis.tid = NULL;
1683 : }
1684 686 : else if (!unique_checked)
1685 : {
1686 686 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1687 : &lVis);
1688 : }
1689 287732 : skey->scantid = scantid; /* Restore saved scan key state */
1690 : }
1691 :
1692 : /*
1693 : * * Last item check *
1694 : *
1695 : * Check last item against next/right page's first data item's when
1696 : * last item on page is reached. This additional check will detect
1697 : * transposed pages iff the supposed right sibling page happens to
1698 : * belong before target in the key space. (Otherwise, a subsequent
1699 : * heap verification will probably detect the problem.)
1700 : *
1701 : * This check is similar to the item order check that will have
1702 : * already been performed for every other "real" item on target page
1703 : * when last item is checked. The difference is that the next item
1704 : * (the item that is compared to target's last item) needs to come
1705 : * from the next/sibling page. There may not be such an item
1706 : * available from sibling for various reasons, though (e.g., target is
1707 : * the rightmost page on level).
1708 : */
1709 4048282 : if (offset == max)
1710 : {
1711 : BTScanInsert rightkey;
1712 :
1713 : /* first offset on a right index page (log only) */
1714 18212 : OffsetNumber rightfirstoffset = InvalidOffsetNumber;
1715 :
1716 : /* Get item in next/right page */
1717 18212 : rightkey = bt_right_page_check_scankey(state, &rightfirstoffset);
1718 :
1719 18212 : if (rightkey &&
1720 13228 : !invariant_g_offset(state, rightkey, max))
1721 : {
1722 : /*
1723 : * As explained at length in bt_right_page_check_scankey(),
1724 : * there is a known !readonly race that could account for
1725 : * apparent violation of invariant, which we must check for
1726 : * before actually proceeding with raising error. Our canary
1727 : * condition is that target page was deleted.
1728 : */
1729 0 : if (!state->readonly)
1730 : {
1731 : /* Get fresh copy of target page */
1732 0 : state->target = palloc_btree_page(state, state->targetblock);
1733 : /* Note that we deliberately do not update target LSN */
1734 0 : topaque = BTPageGetOpaque(state->target);
1735 :
1736 : /*
1737 : * All !readonly checks now performed; just return
1738 : */
1739 0 : if (P_IGNORE(topaque))
1740 0 : return;
1741 : }
1742 :
1743 0 : ereport(ERROR,
1744 : (errcode(ERRCODE_INDEX_CORRUPTED),
1745 : errmsg("cross page item order invariant violated for index \"%s\"",
1746 : RelationGetRelationName(state->rel)),
1747 : errdetail_internal("Last item on page tid=(%u,%u) page lsn=%X/%08X.",
1748 : state->targetblock, offset,
1749 : LSN_FORMAT_ARGS(state->targetlsn))));
1750 : }
1751 :
1752 : /*
1753 : * If index has unique constraint make sure that no more than one
1754 : * found equal items is visible.
1755 : */
1756 18212 : if (state->checkunique && state->indexinfo->ii_Unique &&
1757 1026 : rightkey && P_ISLEAF(topaque) && !P_RIGHTMOST(topaque))
1758 : {
1759 1026 : BlockNumber rightblock_number = topaque->btpo_next;
1760 :
1761 1026 : elog(DEBUG2, "check cross page unique condition");
1762 :
1763 : /*
1764 : * Make _bt_compare compare only index keys without heap TIDs.
1765 : * rightkey->scantid is modified destructively but it is ok
1766 : * for it is not used later.
1767 : */
1768 1026 : rightkey->scantid = NULL;
1769 :
1770 : /* The first key on the next page is the same */
1771 1026 : if (_bt_compare(state->rel, rightkey, state->target, max) == 0 &&
1772 14 : !rightkey->anynullkeys)
1773 : {
1774 : Page rightpage;
1775 :
1776 : /*
1777 : * Do the bt_entry_unique_check() call if it was
1778 : * postponed.
1779 : */
1780 0 : if (!unique_checked)
1781 0 : bt_entry_unique_check(state, itup, state->targetblock,
1782 : offset, &lVis);
1783 :
1784 0 : elog(DEBUG2, "cross page equal keys");
1785 0 : rightpage = palloc_btree_page(state,
1786 : rightblock_number);
1787 0 : topaque = BTPageGetOpaque(rightpage);
1788 :
1789 0 : if (P_IGNORE(topaque))
1790 : {
1791 0 : pfree(rightpage);
1792 0 : break;
1793 : }
1794 :
1795 0 : if (unlikely(!P_ISLEAF(topaque)))
1796 0 : ereport(ERROR,
1797 : (errcode(ERRCODE_INDEX_CORRUPTED),
1798 : errmsg("right block of leaf block is non-leaf for index \"%s\"",
1799 : RelationGetRelationName(state->rel)),
1800 : errdetail_internal("Block=%u page lsn=%X/%08X.",
1801 : state->targetblock,
1802 : LSN_FORMAT_ARGS(state->targetlsn))));
1803 :
1804 0 : itemid = PageGetItemIdCareful(state, rightblock_number,
1805 : rightpage,
1806 : rightfirstoffset);
1807 0 : itup = (IndexTuple) PageGetItem(rightpage, itemid);
1808 :
1809 0 : bt_entry_unique_check(state, itup, rightblock_number, rightfirstoffset, &lVis);
1810 :
1811 0 : pfree(rightpage);
1812 : }
1813 : }
1814 : }
1815 :
1816 : /*
1817 : * * Downlink check *
1818 : *
1819 : * Additional check of child items iff this is an internal page and
1820 : * caller holds a ShareLock. This happens for every downlink (item)
1821 : * in target excluding the negative-infinity downlink (again, this is
1822 : * because it has no useful value to compare).
1823 : */
1824 4048282 : if (!P_ISLEAF(topaque) && state->readonly)
1825 3722 : bt_child_check(state, skey, offset);
1826 : }
1827 :
1828 : /*
1829 : * Special case bt_child_highkey_check() call
1830 : *
1831 : * We don't pass a real downlink, but we've to finish the level
1832 : * processing. If condition is satisfied, we've already processed all the
1833 : * downlinks from the target level. But there still might be pages to the
1834 : * right of the child page pointer to by our rightmost downlink. And they
1835 : * might have missing downlinks. This final call checks for them.
1836 : */
1837 18216 : if (!P_ISLEAF(topaque) && P_RIGHTMOST(topaque) && state->readonly)
1838 : {
1839 22 : bt_child_highkey_check(state, InvalidOffsetNumber,
1840 : NULL, topaque->btpo_level);
1841 : }
1842 : }
1843 :
1844 : /*
1845 : * Return a scankey for an item on page to right of current target (or the
1846 : * first non-ignorable page), sufficient to check ordering invariant on last
1847 : * item in current target page. Returned scankey relies on local memory
1848 : * allocated for the child page, which caller cannot pfree(). Caller's memory
1849 : * context should be reset between calls here.
1850 : *
1851 : * This is the first data item, and so all adjacent items are checked against
1852 : * their immediate sibling item (which may be on a sibling page, or even a
1853 : * "cousin" page at parent boundaries where target's rightlink points to page
1854 : * with different parent page). If no such valid item is available, return
1855 : * NULL instead.
1856 : *
1857 : * Note that !readonly callers must reverify that target page has not
1858 : * been concurrently deleted.
1859 : *
1860 : * Save rightfirstoffset for detailed error message.
1861 : */
1862 : static BTScanInsert
1863 18212 : bt_right_page_check_scankey(BtreeCheckState *state, OffsetNumber *rightfirstoffset)
1864 : {
1865 : BTPageOpaque opaque;
1866 : ItemId rightitem;
1867 : IndexTuple firstitup;
1868 : BlockNumber targetnext;
1869 : Page rightpage;
1870 : OffsetNumber nline;
1871 :
1872 : /* Determine target's next block number */
1873 18212 : opaque = BTPageGetOpaque(state->target);
1874 :
1875 : /* If target is already rightmost, no right sibling; nothing to do here */
1876 18212 : if (P_RIGHTMOST(opaque))
1877 4984 : return NULL;
1878 :
1879 : /*
1880 : * General notes on concurrent page splits and page deletion:
1881 : *
1882 : * Routines like _bt_search() don't require *any* page split interlock
1883 : * when descending the tree, including something very light like a buffer
1884 : * pin. That's why it's okay that we don't either. This avoidance of any
1885 : * need to "couple" buffer locks is the raison d' etre of the Lehman & Yao
1886 : * algorithm, in fact.
1887 : *
1888 : * That leaves deletion. A deleted page won't actually be recycled by
1889 : * VACUUM early enough for us to fail to at least follow its right link
1890 : * (or left link, or downlink) and find its sibling, because recycling
1891 : * does not occur until no possible index scan could land on the page.
1892 : * Index scans can follow links with nothing more than their snapshot as
1893 : * an interlock and be sure of at least that much. (See page
1894 : * recycling/"visible to everyone" notes in nbtree README.)
1895 : *
1896 : * Furthermore, it's okay if we follow a rightlink and find a half-dead or
1897 : * dead (ignorable) page one or more times. There will either be a
1898 : * further right link to follow that leads to a live page before too long
1899 : * (before passing by parent's rightmost child), or we will find the end
1900 : * of the entire level instead (possible when parent page is itself the
1901 : * rightmost on its level).
1902 : */
1903 13228 : targetnext = opaque->btpo_next;
1904 : for (;;)
1905 : {
1906 13228 : CHECK_FOR_INTERRUPTS();
1907 :
1908 13228 : rightpage = palloc_btree_page(state, targetnext);
1909 13228 : opaque = BTPageGetOpaque(rightpage);
1910 :
1911 13228 : if (!P_IGNORE(opaque) || P_RIGHTMOST(opaque))
1912 : break;
1913 :
1914 : /*
1915 : * We landed on a deleted or half-dead sibling page. Step right until
1916 : * we locate a live sibling page.
1917 : */
1918 0 : ereport(DEBUG2,
1919 : (errcode(ERRCODE_NO_DATA),
1920 : errmsg_internal("level %u sibling page in block %u of index \"%s\" was found deleted or half dead",
1921 : opaque->btpo_level, targetnext, RelationGetRelationName(state->rel)),
1922 : errdetail_internal("Deleted page found when building scankey from right sibling.")));
1923 :
1924 0 : targetnext = opaque->btpo_next;
1925 :
1926 : /* Be slightly more pro-active in freeing this memory, just in case */
1927 0 : pfree(rightpage);
1928 : }
1929 :
1930 : /*
1931 : * No ShareLock held case -- why it's safe to proceed.
1932 : *
1933 : * Problem:
1934 : *
1935 : * We must avoid false positive reports of corruption when caller treats
1936 : * item returned here as an upper bound on target's last item. In
1937 : * general, false positives are disallowed. Avoiding them here when
1938 : * caller is !readonly is subtle.
1939 : *
1940 : * A concurrent page deletion by VACUUM of the target page can result in
1941 : * the insertion of items on to this right sibling page that would
1942 : * previously have been inserted on our target page. There might have
1943 : * been insertions that followed the target's downlink after it was made
1944 : * to point to right sibling instead of target by page deletion's first
1945 : * phase. The inserters insert items that would belong on target page.
1946 : * This race is very tight, but it's possible. This is our only problem.
1947 : *
1948 : * Non-problems:
1949 : *
1950 : * We are not hindered by a concurrent page split of the target; we'll
1951 : * never land on the second half of the page anyway. A concurrent split
1952 : * of the right page will also not matter, because the first data item
1953 : * remains the same within the left half, which we'll reliably land on. If
1954 : * we had to skip over ignorable/deleted pages, it cannot matter because
1955 : * their key space has already been atomically merged with the first
1956 : * non-ignorable page we eventually find (doesn't matter whether the page
1957 : * we eventually find is a true sibling or a cousin of target, which we go
1958 : * into below).
1959 : *
1960 : * Solution:
1961 : *
1962 : * Caller knows that it should reverify that target is not ignorable
1963 : * (half-dead or deleted) when cross-page sibling item comparison appears
1964 : * to indicate corruption (invariant fails). This detects the single race
1965 : * condition that exists for caller. This is correct because the
1966 : * continued existence of target block as non-ignorable (not half-dead or
1967 : * deleted) implies that target page was not merged into from the right by
1968 : * deletion; the key space at or after target never moved left. Target's
1969 : * parent either has the same downlink to target as before, or a <
1970 : * downlink due to deletion at the left of target. Target either has the
1971 : * same highkey as before, or a highkey < before when there is a page
1972 : * split. (The rightmost concurrently-split-from-target-page page will
1973 : * still have the same highkey as target was originally found to have,
1974 : * which for our purposes is equivalent to target's highkey itself never
1975 : * changing, since we reliably skip over
1976 : * concurrently-split-from-target-page pages.)
1977 : *
1978 : * In simpler terms, we allow that the key space of the target may expand
1979 : * left (the key space can move left on the left side of target only), but
1980 : * the target key space cannot expand right and get ahead of us without
1981 : * our detecting it. The key space of the target cannot shrink, unless it
1982 : * shrinks to zero due to the deletion of the original page, our canary
1983 : * condition. (To be very precise, we're a bit stricter than that because
1984 : * it might just have been that the target page split and only the
1985 : * original target page was deleted. We can be more strict, just not more
1986 : * lax.)
1987 : *
1988 : * Top level tree walk caller moves on to next page (makes it the new
1989 : * target) following recovery from this race. (cf. The rationale for
1990 : * child/downlink verification needing a ShareLock within
1991 : * bt_child_check(), where page deletion is also the main source of
1992 : * trouble.)
1993 : *
1994 : * Note that it doesn't matter if right sibling page here is actually a
1995 : * cousin page, because in order for the key space to be readjusted in a
1996 : * way that causes us issues in next level up (guiding problematic
1997 : * concurrent insertions to the cousin from the grandparent rather than to
1998 : * the sibling from the parent), there'd have to be page deletion of
1999 : * target's parent page (affecting target's parent's downlink in target's
2000 : * grandparent page). Internal page deletion only occurs when there are
2001 : * no child pages (they were all fully deleted), and caller is checking
2002 : * that the target's parent has at least one non-deleted (so
2003 : * non-ignorable) child: the target page. (Note that the first phase of
2004 : * deletion atomically marks the page to be deleted half-dead/ignorable at
2005 : * the same time downlink in its parent is removed, so caller will
2006 : * definitely not fail to detect that this happened.)
2007 : *
2008 : * This trick is inspired by the method backward scans use for dealing
2009 : * with concurrent page splits; concurrent page deletion is a problem that
2010 : * similarly receives special consideration sometimes (it's possible that
2011 : * the backwards scan will re-read its "original" block after failing to
2012 : * find a right-link to it, having already moved in the opposite direction
2013 : * (right/"forwards") a few times to try to locate one). Just like us,
2014 : * that happens only to determine if there was a concurrent page deletion
2015 : * of a reference page, and just like us if there was a page deletion of
2016 : * that reference page it means we can move on from caring about the
2017 : * reference page. See the nbtree README for a full description of how
2018 : * that works.
2019 : */
2020 13228 : nline = PageGetMaxOffsetNumber(rightpage);
2021 :
2022 : /*
2023 : * Get first data item, if any
2024 : */
2025 13228 : if (P_ISLEAF(opaque) && nline >= P_FIRSTDATAKEY(opaque))
2026 : {
2027 : /* Return first data item (if any) */
2028 13224 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2029 13224 : P_FIRSTDATAKEY(opaque));
2030 13224 : *rightfirstoffset = P_FIRSTDATAKEY(opaque);
2031 : }
2032 8 : else if (!P_ISLEAF(opaque) &&
2033 4 : nline >= OffsetNumberNext(P_FIRSTDATAKEY(opaque)))
2034 : {
2035 : /*
2036 : * Return first item after the internal page's "negative infinity"
2037 : * item
2038 : */
2039 4 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2040 4 : OffsetNumberNext(P_FIRSTDATAKEY(opaque)));
2041 : }
2042 : else
2043 : {
2044 : /*
2045 : * No first item. Page is probably empty leaf page, but it's also
2046 : * possible that it's an internal page with only a negative infinity
2047 : * item.
2048 : */
2049 0 : ereport(DEBUG2,
2050 : (errcode(ERRCODE_NO_DATA),
2051 : errmsg_internal("%s block %u of index \"%s\" has no first data item",
2052 : P_ISLEAF(opaque) ? "leaf" : "internal", targetnext,
2053 : RelationGetRelationName(state->rel))));
2054 0 : return NULL;
2055 : }
2056 :
2057 : /*
2058 : * Return first real item scankey. Note that this relies on right page
2059 : * memory remaining allocated.
2060 : */
2061 13228 : firstitup = (IndexTuple) PageGetItem(rightpage, rightitem);
2062 13228 : return bt_mkscankey_pivotsearch(state->rel, firstitup);
2063 : }
2064 :
2065 : /*
2066 : * Check if two tuples are binary identical except the block number. So,
2067 : * this function is capable to compare pivot keys on different levels.
2068 : */
2069 : static bool
2070 3724 : bt_pivot_tuple_identical(bool heapkeyspace, IndexTuple itup1, IndexTuple itup2)
2071 : {
2072 3724 : if (IndexTupleSize(itup1) != IndexTupleSize(itup2))
2073 0 : return false;
2074 :
2075 3724 : if (heapkeyspace)
2076 : {
2077 : /*
2078 : * Offset number will contain important information in heapkeyspace
2079 : * indexes: the number of attributes left in the pivot tuple following
2080 : * suffix truncation. Don't skip over it (compare it too).
2081 : */
2082 3724 : if (memcmp(&itup1->t_tid.ip_posid, &itup2->t_tid.ip_posid,
2083 3724 : IndexTupleSize(itup1) -
2084 : offsetof(ItemPointerData, ip_posid)) != 0)
2085 0 : return false;
2086 : }
2087 : else
2088 : {
2089 : /*
2090 : * Cannot rely on offset number field having consistent value across
2091 : * levels on pg_upgrade'd !heapkeyspace indexes. Compare contents of
2092 : * tuple starting from just after item pointer (i.e. after block
2093 : * number and offset number).
2094 : */
2095 0 : if (memcmp(&itup1->t_info, &itup2->t_info,
2096 0 : IndexTupleSize(itup1) -
2097 : offsetof(IndexTupleData, t_info)) != 0)
2098 0 : return false;
2099 : }
2100 :
2101 3724 : return true;
2102 : }
2103 :
2104 : /*---
2105 : * Check high keys on the child level. Traverse rightlinks from previous
2106 : * downlink to the current one. Check that there are no intermediate pages
2107 : * with missing downlinks.
2108 : *
2109 : * If 'loaded_child' is given, it's assumed to be the page pointed to by the
2110 : * downlink referenced by 'downlinkoffnum' of the target page.
2111 : *
2112 : * Basically this function is called for each target downlink and checks two
2113 : * invariants:
2114 : *
2115 : * 1) You can reach the next child from previous one via rightlinks;
2116 : * 2) Each child high key have matching pivot key on target level.
2117 : *
2118 : * Consider the sample tree picture.
2119 : *
2120 : * 1
2121 : * / \
2122 : * 2 <-> 3
2123 : * / \ / \
2124 : * 4 <> 5 <> 6 <> 7 <> 8
2125 : *
2126 : * This function will be called for blocks 4, 5, 6 and 8. Consider what is
2127 : * happening for each function call.
2128 : *
2129 : * - The function call for block 4 initializes data structure and matches high
2130 : * key of block 4 to downlink's pivot key of block 2.
2131 : * - The high key of block 5 is matched to the high key of block 2.
2132 : * - The block 6 has an incomplete split flag set, so its high key isn't
2133 : * matched to anything.
2134 : * - The function call for block 8 checks that block 8 can be found while
2135 : * following rightlinks from block 6. The high key of block 7 will be
2136 : * matched to downlink's pivot key in block 3.
2137 : *
2138 : * There is also final call of this function, which checks that there is no
2139 : * missing downlinks for children to the right of the child referenced by
2140 : * rightmost downlink in target level.
2141 : */
2142 : static void
2143 3768 : bt_child_highkey_check(BtreeCheckState *state,
2144 : OffsetNumber target_downlinkoffnum,
2145 : Page loaded_child,
2146 : uint32 target_level)
2147 : {
2148 3768 : BlockNumber blkno = state->prevrightlink;
2149 : Page page;
2150 : BTPageOpaque opaque;
2151 3768 : bool rightsplit = state->previncompletesplit;
2152 3768 : bool first = true;
2153 : ItemId itemid;
2154 : IndexTuple itup;
2155 : BlockNumber downlink;
2156 :
2157 3768 : if (OffsetNumberIsValid(target_downlinkoffnum))
2158 : {
2159 3746 : itemid = PageGetItemIdCareful(state, state->targetblock,
2160 : state->target, target_downlinkoffnum);
2161 3746 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2162 3746 : downlink = BTreeTupleGetDownLink(itup);
2163 : }
2164 : else
2165 : {
2166 22 : downlink = P_NONE;
2167 : }
2168 :
2169 : /*
2170 : * If no previous rightlink is memorized for current level just below
2171 : * target page's level, we are about to start from the leftmost page. We
2172 : * can't follow rightlinks from previous page, because there is no
2173 : * previous page. But we still can match high key.
2174 : *
2175 : * So we initialize variables for the loop above like there is previous
2176 : * page referencing current child. Also we imply previous page to not
2177 : * have incomplete split flag, that would make us require downlink for
2178 : * current child. That's correct, because leftmost page on the level
2179 : * should always have parent downlink.
2180 : */
2181 3768 : if (!BlockNumberIsValid(blkno))
2182 : {
2183 22 : blkno = downlink;
2184 22 : rightsplit = false;
2185 : }
2186 :
2187 : /* Move to the right on the child level */
2188 : while (true)
2189 : {
2190 : /*
2191 : * Did we traverse the whole tree level and this is check for pages to
2192 : * the right of rightmost downlink?
2193 : */
2194 3768 : if (blkno == P_NONE && downlink == P_NONE)
2195 : {
2196 22 : state->prevrightlink = InvalidBlockNumber;
2197 22 : state->previncompletesplit = false;
2198 22 : return;
2199 : }
2200 :
2201 : /* Did we traverse the whole tree level and don't find next downlink? */
2202 3746 : if (blkno == P_NONE)
2203 0 : ereport(ERROR,
2204 : (errcode(ERRCODE_INDEX_CORRUPTED),
2205 : errmsg("can't traverse from downlink %u to downlink %u of index \"%s\"",
2206 : state->prevrightlink, downlink,
2207 : RelationGetRelationName(state->rel))));
2208 :
2209 : /* Load page contents */
2210 3746 : if (blkno == downlink && loaded_child)
2211 3722 : page = loaded_child;
2212 : else
2213 24 : page = palloc_btree_page(state, blkno);
2214 :
2215 3746 : opaque = BTPageGetOpaque(page);
2216 :
2217 : /* The first page we visit at the level should be leftmost */
2218 3746 : if (first && !BlockNumberIsValid(state->prevrightlink) &&
2219 22 : !bt_leftmost_ignoring_half_dead(state, blkno, opaque))
2220 0 : ereport(ERROR,
2221 : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 : errmsg("the first child of leftmost target page is not leftmost of its level in index \"%s\"",
2223 : RelationGetRelationName(state->rel)),
2224 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2225 : state->targetblock, blkno,
2226 : LSN_FORMAT_ARGS(state->targetlsn))));
2227 :
2228 : /* Do level sanity check */
2229 3746 : if ((!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque)) &&
2230 3746 : opaque->btpo_level != target_level - 1)
2231 0 : ereport(ERROR,
2232 : (errcode(ERRCODE_INDEX_CORRUPTED),
2233 : errmsg("block found while following rightlinks from child of index \"%s\" has invalid level",
2234 : RelationGetRelationName(state->rel)),
2235 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
2236 : blkno, target_level - 1, opaque->btpo_level)));
2237 :
2238 : /* Try to detect circular links */
2239 3746 : if ((!first && blkno == state->prevrightlink) || blkno == opaque->btpo_prev)
2240 0 : ereport(ERROR,
2241 : (errcode(ERRCODE_INDEX_CORRUPTED),
2242 : errmsg("circular link chain found in block %u of index \"%s\"",
2243 : blkno, RelationGetRelationName(state->rel))));
2244 :
2245 3746 : if (blkno != downlink && !P_IGNORE(opaque))
2246 : {
2247 : /* blkno probably has missing parent downlink */
2248 0 : bt_downlink_missing_check(state, rightsplit, blkno, page);
2249 : }
2250 :
2251 3746 : rightsplit = P_INCOMPLETE_SPLIT(opaque);
2252 :
2253 : /*
2254 : * If we visit page with high key, check that it is equal to the
2255 : * target key next to corresponding downlink.
2256 : */
2257 3746 : if (!rightsplit && !P_RIGHTMOST(opaque) && !P_ISHALFDEAD(opaque))
2258 : {
2259 : BTPageOpaque topaque;
2260 : IndexTuple highkey;
2261 : OffsetNumber pivotkey_offset;
2262 :
2263 : /* Get high key */
2264 3724 : itemid = PageGetItemIdCareful(state, blkno, page, P_HIKEY);
2265 3724 : highkey = (IndexTuple) PageGetItem(page, itemid);
2266 :
2267 : /*
2268 : * There might be two situations when we examine high key. If
2269 : * current child page is referenced by given target downlink, we
2270 : * should look to the next offset number for matching key from
2271 : * target page.
2272 : *
2273 : * Alternatively, we're following rightlinks somewhere in the
2274 : * middle between page referenced by previous target's downlink
2275 : * and the page referenced by current target's downlink. If
2276 : * current child page hasn't incomplete split flag set, then its
2277 : * high key should match to the target's key of current offset
2278 : * number. This happens when a previous call here (to
2279 : * bt_child_highkey_check()) found an incomplete split, and we
2280 : * reach a right sibling page without a downlink -- the right
2281 : * sibling page's high key still needs to be matched to a
2282 : * separator key on the parent/target level.
2283 : *
2284 : * Don't apply OffsetNumberNext() to target_downlinkoffnum when we
2285 : * already had to step right on the child level. Our traversal of
2286 : * the child level must try to move in perfect lockstep behind (to
2287 : * the left of) the target/parent level traversal.
2288 : */
2289 3724 : if (blkno == downlink)
2290 3724 : pivotkey_offset = OffsetNumberNext(target_downlinkoffnum);
2291 : else
2292 0 : pivotkey_offset = target_downlinkoffnum;
2293 :
2294 3724 : topaque = BTPageGetOpaque(state->target);
2295 :
2296 3724 : if (!offset_is_negative_infinity(topaque, pivotkey_offset))
2297 : {
2298 : /*
2299 : * If we're looking for the next pivot tuple in target page,
2300 : * but there is no more pivot tuples, then we should match to
2301 : * high key instead.
2302 : */
2303 3724 : if (pivotkey_offset > PageGetMaxOffsetNumber(state->target))
2304 : {
2305 2 : if (P_RIGHTMOST(topaque))
2306 0 : ereport(ERROR,
2307 : (errcode(ERRCODE_INDEX_CORRUPTED),
2308 : errmsg("child high key is greater than rightmost pivot key on target level in index \"%s\"",
2309 : RelationGetRelationName(state->rel)),
2310 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2311 : state->targetblock, blkno,
2312 : LSN_FORMAT_ARGS(state->targetlsn))));
2313 2 : pivotkey_offset = P_HIKEY;
2314 : }
2315 3724 : itemid = PageGetItemIdCareful(state, state->targetblock,
2316 : state->target, pivotkey_offset);
2317 3724 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2318 : }
2319 : else
2320 : {
2321 : /*
2322 : * We cannot try to match child's high key to a negative
2323 : * infinity key in target, since there is nothing to compare.
2324 : * However, it's still possible to match child's high key
2325 : * outside of target page. The reason why we're are is that
2326 : * bt_child_highkey_check() was previously called for the
2327 : * cousin page of 'loaded_child', which is incomplete split.
2328 : * So, now we traverse to the right of that cousin page and
2329 : * current child level page under consideration still belongs
2330 : * to the subtree of target's left sibling. Thus, we need to
2331 : * match child's high key to its left uncle page high key.
2332 : * Thankfully we saved it, it's called a "low key" of target
2333 : * page.
2334 : */
2335 0 : if (!state->lowkey)
2336 0 : ereport(ERROR,
2337 : (errcode(ERRCODE_INDEX_CORRUPTED),
2338 : errmsg("can't find left sibling high key in index \"%s\"",
2339 : RelationGetRelationName(state->rel)),
2340 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2341 : state->targetblock, blkno,
2342 : LSN_FORMAT_ARGS(state->targetlsn))));
2343 0 : itup = state->lowkey;
2344 : }
2345 :
2346 3724 : if (!bt_pivot_tuple_identical(state->heapkeyspace, highkey, itup))
2347 : {
2348 0 : ereport(ERROR,
2349 : (errcode(ERRCODE_INDEX_CORRUPTED),
2350 : errmsg("mismatch between parent key and child high key in index \"%s\"",
2351 : RelationGetRelationName(state->rel)),
2352 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2353 : state->targetblock, blkno,
2354 : LSN_FORMAT_ARGS(state->targetlsn))));
2355 : }
2356 : }
2357 :
2358 : /* Exit if we already found next downlink */
2359 3746 : if (blkno == downlink)
2360 : {
2361 3746 : state->prevrightlink = opaque->btpo_next;
2362 3746 : state->previncompletesplit = rightsplit;
2363 3746 : return;
2364 : }
2365 :
2366 : /* Traverse to the next page using rightlink */
2367 0 : blkno = opaque->btpo_next;
2368 :
2369 : /* Free page contents if it's allocated by us */
2370 0 : if (page != loaded_child)
2371 0 : pfree(page);
2372 0 : first = false;
2373 : }
2374 : }
2375 :
2376 : /*
2377 : * Checks one of target's downlink against its child page.
2378 : *
2379 : * Conceptually, the target page continues to be what is checked here. The
2380 : * target block is still blamed in the event of finding an invariant violation.
2381 : * The downlink insertion into the target is probably where any problem raised
2382 : * here arises, and there is no such thing as a parent link, so doing the
2383 : * verification this way around is much more practical.
2384 : *
2385 : * This function visits child page and it's sequentially called for each
2386 : * downlink of target page. Assuming this we also check downlink connectivity
2387 : * here in order to save child page visits.
2388 : */
2389 : static void
2390 3722 : bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
2391 : OffsetNumber downlinkoffnum)
2392 : {
2393 : ItemId itemid;
2394 : IndexTuple itup;
2395 : BlockNumber childblock;
2396 : OffsetNumber offset;
2397 : OffsetNumber maxoffset;
2398 : Page child;
2399 : BTPageOpaque copaque;
2400 : BTPageOpaque topaque;
2401 :
2402 3722 : itemid = PageGetItemIdCareful(state, state->targetblock,
2403 : state->target, downlinkoffnum);
2404 3722 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2405 3722 : childblock = BTreeTupleGetDownLink(itup);
2406 :
2407 : /*
2408 : * Caller must have ShareLock on target relation, because of
2409 : * considerations around page deletion by VACUUM.
2410 : *
2411 : * NB: In general, page deletion deletes the right sibling's downlink, not
2412 : * the downlink of the page being deleted; the deleted page's downlink is
2413 : * reused for its sibling. The key space is thereby consolidated between
2414 : * the deleted page and its right sibling. (We cannot delete a parent
2415 : * page's rightmost child unless it is the last child page, and we intend
2416 : * to also delete the parent itself.)
2417 : *
2418 : * If this verification happened without a ShareLock, the following race
2419 : * condition could cause false positives:
2420 : *
2421 : * In general, concurrent page deletion might occur, including deletion of
2422 : * the left sibling of the child page that is examined here. If such a
2423 : * page deletion were to occur, closely followed by an insertion into the
2424 : * newly expanded key space of the child, a window for the false positive
2425 : * opens up: the stale parent/target downlink originally followed to get
2426 : * to the child legitimately ceases to be a lower bound on all items in
2427 : * the page, since the key space was concurrently expanded "left".
2428 : * (Insertion followed the "new" downlink for the child, not our now-stale
2429 : * downlink, which was concurrently physically removed in target/parent as
2430 : * part of deletion's first phase.)
2431 : *
2432 : * While we use various techniques elsewhere to perform cross-page
2433 : * verification for !readonly callers, a similar trick seems difficult
2434 : * here. The tricks used by bt_recheck_sibling_links and by
2435 : * bt_right_page_check_scankey both involve verification of a same-level,
2436 : * cross-sibling invariant. Cross-level invariants are far more squishy,
2437 : * though. The nbtree REDO routines do not actually couple buffer locks
2438 : * across levels during page splits, so making any cross-level check work
2439 : * reliably in !readonly mode may be impossible.
2440 : */
2441 : Assert(state->readonly);
2442 :
2443 : /*
2444 : * Verify child page has the downlink key from target page (its parent) as
2445 : * a lower bound; downlink must be strictly less than all keys on the
2446 : * page.
2447 : *
2448 : * Check all items, rather than checking just the first and trusting that
2449 : * the operator class obeys the transitive law.
2450 : */
2451 3722 : topaque = BTPageGetOpaque(state->target);
2452 3722 : child = palloc_btree_page(state, childblock);
2453 3722 : copaque = BTPageGetOpaque(child);
2454 3722 : maxoffset = PageGetMaxOffsetNumber(child);
2455 :
2456 : /*
2457 : * Since we've already loaded the child block, combine this check with
2458 : * check for downlink connectivity.
2459 : */
2460 3722 : bt_child_highkey_check(state, downlinkoffnum,
2461 : child, topaque->btpo_level);
2462 :
2463 : /*
2464 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2465 : * and since a page has no links within other pages (siblings and parent)
2466 : * once it is marked fully deleted, it should be impossible to land on a
2467 : * fully deleted page.
2468 : *
2469 : * It does not quite make sense to enforce that the page cannot even be
2470 : * half-dead, despite the fact the downlink is modified at the same stage
2471 : * that the child leaf page is marked half-dead. That's incorrect because
2472 : * there may occasionally be multiple downlinks from a chain of pages
2473 : * undergoing deletion, where multiple successive calls are made to
2474 : * _bt_unlink_halfdead_page() by VACUUM before it can finally safely mark
2475 : * the leaf page as fully dead. While _bt_mark_page_halfdead() usually
2476 : * removes the downlink to the leaf page that is marked half-dead, that's
2477 : * not guaranteed, so it's possible we'll land on a half-dead page with a
2478 : * downlink due to an interrupted multi-level page deletion.
2479 : *
2480 : * We go ahead with our checks if the child page is half-dead. It's safe
2481 : * to do so because we do not test the child's high key, so it does not
2482 : * matter that the original high key will have been replaced by a dummy
2483 : * truncated high key within _bt_mark_page_halfdead(). All other page
2484 : * items are left intact on a half-dead page, so there is still something
2485 : * to test.
2486 : */
2487 3722 : if (P_ISDELETED(copaque))
2488 0 : ereport(ERROR,
2489 : (errcode(ERRCODE_INDEX_CORRUPTED),
2490 : errmsg("downlink to deleted page found in index \"%s\"",
2491 : RelationGetRelationName(state->rel)),
2492 : errdetail_internal("Parent block=%u child block=%u parent page lsn=%X/%08X.",
2493 : state->targetblock, childblock,
2494 : LSN_FORMAT_ARGS(state->targetlsn))));
2495 :
2496 3722 : for (offset = P_FIRSTDATAKEY(copaque);
2497 1200700 : offset <= maxoffset;
2498 1196978 : offset = OffsetNumberNext(offset))
2499 : {
2500 : /*
2501 : * Skip comparison of target page key against "negative infinity"
2502 : * item, if any. Checking it would indicate that it's not a strict
2503 : * lower bound, but that's only because of the hard-coding for
2504 : * negative infinity items within _bt_compare().
2505 : *
2506 : * If nbtree didn't truncate negative infinity tuples during internal
2507 : * page splits then we'd expect child's negative infinity key to be
2508 : * equal to the scankey/downlink from target/parent (it would be a
2509 : * "low key" in this hypothetical scenario, and so it would still need
2510 : * to be treated as a special case here).
2511 : *
2512 : * Negative infinity items can be thought of as a strict lower bound
2513 : * that works transitively, with the last non-negative-infinity pivot
2514 : * followed during a descent from the root as its "true" strict lower
2515 : * bound. Only a small number of negative infinity items are truly
2516 : * negative infinity; those that are the first items of leftmost
2517 : * internal pages. In more general terms, a negative infinity item is
2518 : * only negative infinity with respect to the subtree that the page is
2519 : * at the root of.
2520 : *
2521 : * See also: bt_rootdescend(), which can even detect transitive
2522 : * inconsistencies on cousin leaf pages.
2523 : */
2524 1196978 : if (offset_is_negative_infinity(copaque, offset))
2525 2 : continue;
2526 :
2527 1196976 : if (!invariant_l_nontarget_offset(state, targetkey, childblock, child,
2528 : offset))
2529 0 : ereport(ERROR,
2530 : (errcode(ERRCODE_INDEX_CORRUPTED),
2531 : errmsg("down-link lower bound invariant violated for index \"%s\"",
2532 : RelationGetRelationName(state->rel)),
2533 : errdetail_internal("Parent block=%u child index tid=(%u,%u) parent page lsn=%X/%08X.",
2534 : state->targetblock, childblock, offset,
2535 : LSN_FORMAT_ARGS(state->targetlsn))));
2536 : }
2537 :
2538 3722 : pfree(child);
2539 3722 : }
2540 :
2541 : /*
2542 : * Checks if page is missing a downlink that it should have.
2543 : *
2544 : * A page that lacks a downlink/parent may indicate corruption. However, we
2545 : * must account for the fact that a missing downlink can occasionally be
2546 : * encountered in a non-corrupt index. This can be due to an interrupted page
2547 : * split, or an interrupted multi-level page deletion (i.e. there was a hard
2548 : * crash or an error during a page split, or while VACUUM was deleting a
2549 : * multi-level chain of pages).
2550 : *
2551 : * Note that this can only be called in readonly mode, so there is no need to
2552 : * be concerned about concurrent page splits or page deletions.
2553 : */
2554 : static void
2555 0 : bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
2556 : BlockNumber blkno, Page page)
2557 : {
2558 0 : BTPageOpaque opaque = BTPageGetOpaque(page);
2559 : ItemId itemid;
2560 : IndexTuple itup;
2561 : Page child;
2562 : BTPageOpaque copaque;
2563 : uint32 level;
2564 : BlockNumber childblk;
2565 : XLogRecPtr pagelsn;
2566 :
2567 : Assert(state->readonly);
2568 : Assert(!P_IGNORE(opaque));
2569 :
2570 : /* No next level up with downlinks to fingerprint from the true root */
2571 0 : if (P_ISROOT(opaque))
2572 0 : return;
2573 :
2574 0 : pagelsn = PageGetLSN(page);
2575 :
2576 : /*
2577 : * Incomplete (interrupted) page splits can account for the lack of a
2578 : * downlink. Some inserting transaction should eventually complete the
2579 : * page split in passing, when it notices that the left sibling page is
2580 : * P_INCOMPLETE_SPLIT().
2581 : *
2582 : * In general, VACUUM is not prepared for there to be no downlink to a
2583 : * page that it deletes. This is the main reason why the lack of a
2584 : * downlink can be reported as corruption here. It's not obvious that an
2585 : * invalid missing downlink can result in wrong answers to queries,
2586 : * though, since index scans that land on the child may end up
2587 : * consistently moving right. The handling of concurrent page splits (and
2588 : * page deletions) within _bt_moveright() cannot distinguish
2589 : * inconsistencies that last for a moment from inconsistencies that are
2590 : * permanent and irrecoverable.
2591 : *
2592 : * VACUUM isn't even prepared to delete pages that have no downlink due to
2593 : * an incomplete page split, but it can detect and reason about that case
2594 : * by design, so it shouldn't be taken to indicate corruption. See
2595 : * _bt_pagedel() for full details.
2596 : */
2597 0 : if (rightsplit)
2598 : {
2599 0 : ereport(DEBUG1,
2600 : (errcode(ERRCODE_NO_DATA),
2601 : errmsg_internal("harmless interrupted page split detected in index \"%s\"",
2602 : RelationGetRelationName(state->rel)),
2603 : errdetail_internal("Block=%u level=%u left sibling=%u page lsn=%X/%08X.",
2604 : blkno, opaque->btpo_level,
2605 : opaque->btpo_prev,
2606 : LSN_FORMAT_ARGS(pagelsn))));
2607 0 : return;
2608 : }
2609 :
2610 : /*
2611 : * Page under check is probably the "top parent" of a multi-level page
2612 : * deletion. We'll need to descend the subtree to make sure that
2613 : * descendant pages are consistent with that, though.
2614 : *
2615 : * If the page (which must be non-ignorable) is a leaf page, then clearly
2616 : * it can't be the top parent. The lack of a downlink is probably a
2617 : * symptom of a broad problem that could just as easily cause
2618 : * inconsistencies anywhere else.
2619 : */
2620 0 : if (P_ISLEAF(opaque))
2621 0 : ereport(ERROR,
2622 : (errcode(ERRCODE_INDEX_CORRUPTED),
2623 : errmsg("leaf index block lacks downlink in index \"%s\"",
2624 : RelationGetRelationName(state->rel)),
2625 : errdetail_internal("Block=%u page lsn=%X/%08X.",
2626 : blkno,
2627 : LSN_FORMAT_ARGS(pagelsn))));
2628 :
2629 : /* Descend from the given page, which is an internal page */
2630 0 : elog(DEBUG1, "checking for interrupted multi-level deletion due to missing downlink in index \"%s\"",
2631 : RelationGetRelationName(state->rel));
2632 :
2633 0 : level = opaque->btpo_level;
2634 0 : itemid = PageGetItemIdCareful(state, blkno, page, P_FIRSTDATAKEY(opaque));
2635 0 : itup = (IndexTuple) PageGetItem(page, itemid);
2636 0 : childblk = BTreeTupleGetDownLink(itup);
2637 : for (;;)
2638 : {
2639 0 : CHECK_FOR_INTERRUPTS();
2640 :
2641 0 : child = palloc_btree_page(state, childblk);
2642 0 : copaque = BTPageGetOpaque(child);
2643 :
2644 0 : if (P_ISLEAF(copaque))
2645 0 : break;
2646 :
2647 : /* Do an extra sanity check in passing on internal pages */
2648 0 : if (copaque->btpo_level != level - 1)
2649 0 : ereport(ERROR,
2650 : (errcode(ERRCODE_INDEX_CORRUPTED),
2651 : errmsg_internal("downlink points to block in index \"%s\" whose level is not one level down",
2652 : RelationGetRelationName(state->rel)),
2653 : errdetail_internal("Top parent/under check block=%u block pointed to=%u expected level=%u level in pointed to block=%u.",
2654 : blkno, childblk,
2655 : level - 1, copaque->btpo_level)));
2656 :
2657 0 : level = copaque->btpo_level;
2658 0 : itemid = PageGetItemIdCareful(state, childblk, child,
2659 0 : P_FIRSTDATAKEY(copaque));
2660 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2661 0 : childblk = BTreeTupleGetDownLink(itup);
2662 : /* Be slightly more pro-active in freeing this memory, just in case */
2663 0 : pfree(child);
2664 : }
2665 :
2666 : /*
2667 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2668 : * and since a page has no links within other pages (siblings and parent)
2669 : * once it is marked fully deleted, it should be impossible to land on a
2670 : * fully deleted page. See bt_child_check() for further details.
2671 : *
2672 : * The bt_child_check() P_ISDELETED() check is repeated here because
2673 : * bt_child_check() does not visit pages reachable through negative
2674 : * infinity items. Besides, bt_child_check() is unwilling to descend
2675 : * multiple levels. (The similar bt_child_check() P_ISDELETED() check
2676 : * within bt_check_level_from_leftmost() won't reach the page either,
2677 : * since the leaf's live siblings should have their sibling links updated
2678 : * to bypass the deletion target page when it is marked fully dead.)
2679 : *
2680 : * If this error is raised, it might be due to a previous multi-level page
2681 : * deletion that failed to realize that it wasn't yet safe to mark the
2682 : * leaf page as fully dead. A "dangling downlink" will still remain when
2683 : * this happens. The fact that the dangling downlink's page (the leaf's
2684 : * parent/ancestor page) lacked a downlink is incidental.
2685 : */
2686 0 : if (P_ISDELETED(copaque))
2687 0 : ereport(ERROR,
2688 : (errcode(ERRCODE_INDEX_CORRUPTED),
2689 : errmsg_internal("downlink to deleted leaf page found in index \"%s\"",
2690 : RelationGetRelationName(state->rel)),
2691 : errdetail_internal("Top parent/target block=%u leaf block=%u top parent/under check lsn=%X/%08X.",
2692 : blkno, childblk,
2693 : LSN_FORMAT_ARGS(pagelsn))));
2694 :
2695 : /*
2696 : * Iff leaf page is half-dead, its high key top parent link should point
2697 : * to what VACUUM considered to be the top parent page at the instant it
2698 : * was interrupted. Provided the high key link actually points to the
2699 : * page under check, the missing downlink we detected is consistent with
2700 : * there having been an interrupted multi-level page deletion. This means
2701 : * that the subtree with the page under check at its root (a page deletion
2702 : * chain) is in a consistent state, enabling VACUUM to resume deleting the
2703 : * entire chain the next time it encounters the half-dead leaf page.
2704 : */
2705 0 : if (P_ISHALFDEAD(copaque) && !P_RIGHTMOST(copaque))
2706 : {
2707 0 : itemid = PageGetItemIdCareful(state, childblk, child, P_HIKEY);
2708 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2709 0 : if (BTreeTupleGetTopParent(itup) == blkno)
2710 0 : return;
2711 : }
2712 :
2713 0 : ereport(ERROR,
2714 : (errcode(ERRCODE_INDEX_CORRUPTED),
2715 : errmsg("internal index block lacks downlink in index \"%s\"",
2716 : RelationGetRelationName(state->rel)),
2717 : errdetail_internal("Block=%u level=%u page lsn=%X/%08X.",
2718 : blkno, opaque->btpo_level,
2719 : LSN_FORMAT_ARGS(pagelsn))));
2720 : }
2721 :
2722 : /*
2723 : * Per-tuple callback from table_index_build_scan, used to determine if index has
2724 : * all the entries that definitely should have been observed in leaf pages of
2725 : * the target index (that is, all IndexTuples that were fingerprinted by our
2726 : * Bloom filter). All heapallindexed checks occur here.
2727 : *
2728 : * The redundancy between an index and the table it indexes provides a good
2729 : * opportunity to detect corruption, especially corruption within the table.
2730 : * The high level principle behind the verification performed here is that any
2731 : * IndexTuple that should be in an index following a fresh CREATE INDEX (based
2732 : * on the same index definition) should also have been in the original,
2733 : * existing index, which should have used exactly the same representation
2734 : *
2735 : * Since the overall structure of the index has already been verified, the most
2736 : * likely explanation for error here is a corrupt heap page (could be logical
2737 : * or physical corruption). Index corruption may still be detected here,
2738 : * though. Only readonly callers will have verified that left links and right
2739 : * links are in agreement, and so it's possible that a leaf page transposition
2740 : * within index is actually the source of corruption detected here (for
2741 : * !readonly callers). The checks performed only for readonly callers might
2742 : * more accurately frame the problem as a cross-page invariant issue (this
2743 : * could even be due to recovery not replaying all WAL records). The !readonly
2744 : * ERROR message raised here includes a HINT about retrying with readonly
2745 : * verification, just in case it's a cross-page invariant issue, though that
2746 : * isn't particularly likely.
2747 : *
2748 : * table_index_build_scan() expects to be able to find the root tuple when a
2749 : * heap-only tuple (the live tuple at the end of some HOT chain) needs to be
2750 : * indexed, in order to replace the actual tuple's TID with the root tuple's
2751 : * TID (which is what we're actually passed back here). The index build heap
2752 : * scan code will raise an error when a tuple that claims to be the root of the
2753 : * heap-only tuple's HOT chain cannot be located. This catches cases where the
2754 : * original root item offset/root tuple for a HOT chain indicates (for whatever
2755 : * reason) that the entire HOT chain is dead, despite the fact that the latest
2756 : * heap-only tuple should be indexed. When this happens, sequential scans may
2757 : * always give correct answers, and all indexes may be considered structurally
2758 : * consistent (i.e. the nbtree structural checks would not detect corruption).
2759 : * It may be the case that only index scans give wrong answers, and yet heap or
2760 : * SLRU corruption is the real culprit. (While it's true that LP_DEAD bit
2761 : * setting will probably also leave the index in a corrupt state before too
2762 : * long, the problem is nonetheless that there is heap corruption.)
2763 : *
2764 : * Heap-only tuple handling within table_index_build_scan() works in a way that
2765 : * helps us to detect index tuples that contain the wrong values (values that
2766 : * don't match the latest tuple in the HOT chain). This can happen when there
2767 : * is no superseding index tuple due to a faulty assessment of HOT safety,
2768 : * perhaps during the original CREATE INDEX. Because the latest tuple's
2769 : * contents are used with the root TID, an error will be raised when a tuple
2770 : * with the same TID but non-matching attribute values is passed back to us.
2771 : * Faulty assessment of HOT-safety was behind at least two distinct CREATE
2772 : * INDEX CONCURRENTLY bugs that made it into stable releases, one of which was
2773 : * undetected for many years. In short, the same principle that allows a
2774 : * REINDEX to repair corruption when there was an (undetected) broken HOT chain
2775 : * also allows us to detect the corruption in many cases.
2776 : */
2777 : static void
2778 1066710 : bt_tuple_present_callback(Relation index, ItemPointer tid, Datum *values,
2779 : bool *isnull, bool tupleIsAlive, void *checkstate)
2780 : {
2781 1066710 : BtreeCheckState *state = (BtreeCheckState *) checkstate;
2782 : IndexTuple itup,
2783 : norm;
2784 :
2785 : Assert(state->heapallindexed);
2786 :
2787 : /* Generate a normalized index tuple for fingerprinting */
2788 1066710 : itup = index_form_tuple(RelationGetDescr(index), values, isnull);
2789 1066710 : itup->t_tid = *tid;
2790 1066710 : norm = bt_normalize_tuple(state, itup);
2791 :
2792 : /* Probe Bloom filter -- tuple should be present */
2793 1066710 : if (bloom_lacks_element(state->filter, (unsigned char *) norm,
2794 : IndexTupleSize(norm)))
2795 0 : ereport(ERROR,
2796 : (errcode(ERRCODE_DATA_CORRUPTED),
2797 : errmsg("heap tuple (%u,%u) from table \"%s\" lacks matching index tuple within index \"%s\"",
2798 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2799 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2800 : RelationGetRelationName(state->heaprel),
2801 : RelationGetRelationName(state->rel)),
2802 : !state->readonly
2803 : ? errhint("Retrying verification using the function bt_index_parent_check() might provide a more specific error.")
2804 : : 0));
2805 :
2806 1066710 : state->heaptuplespresent++;
2807 1066710 : pfree(itup);
2808 : /* Cannot leak memory here */
2809 1066710 : if (norm != itup)
2810 10 : pfree(norm);
2811 1066710 : }
2812 :
2813 : /*
2814 : * Normalize an index tuple for fingerprinting.
2815 : *
2816 : * In general, index tuple formation is assumed to be deterministic by
2817 : * heapallindexed verification, and IndexTuples are assumed immutable. While
2818 : * the LP_DEAD bit is mutable in leaf pages, that's ItemId metadata, which is
2819 : * not fingerprinted. Normalization is required to compensate for corner
2820 : * cases where the determinism assumption doesn't quite work.
2821 : *
2822 : * There is currently one such case: index_form_tuple() does not try to hide
2823 : * the source TOAST state of input datums. The executor applies TOAST
2824 : * compression for heap tuples based on different criteria to the compression
2825 : * applied within btinsert()'s call to index_form_tuple(): it sometimes
2826 : * compresses more aggressively, resulting in compressed heap tuple datums but
2827 : * uncompressed corresponding index tuple datums. A subsequent heapallindexed
2828 : * verification will get a logically equivalent though bitwise unequal tuple
2829 : * from index_form_tuple(). False positive heapallindexed corruption reports
2830 : * could occur without normalizing away the inconsistency.
2831 : *
2832 : * Returned tuple is often caller's own original tuple. Otherwise, it is a
2833 : * new representation of caller's original index tuple, palloc()'d in caller's
2834 : * memory context.
2835 : *
2836 : * Note: This routine is not concerned with distinctions about the
2837 : * representation of tuples beyond those that might break heapallindexed
2838 : * verification. In particular, it won't try to normalize opclass-equal
2839 : * datums with potentially distinct representations (e.g., btree/numeric_ops
2840 : * index datums will not get their display scale normalized-away here).
2841 : * Caller does normalization for non-pivot tuples that have a posting list,
2842 : * since dummy CREATE INDEX callback code generates new tuples with the same
2843 : * normalized representation.
2844 : */
2845 : static IndexTuple
2846 2138204 : bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
2847 : {
2848 2138204 : TupleDesc tupleDescriptor = RelationGetDescr(state->rel);
2849 : Datum normalized[INDEX_MAX_KEYS];
2850 : bool isnull[INDEX_MAX_KEYS];
2851 : bool need_free[INDEX_MAX_KEYS];
2852 2138204 : bool formnewtup = false;
2853 : IndexTuple reformed;
2854 : int i;
2855 :
2856 : /* Caller should only pass "logical" non-pivot tuples here */
2857 : Assert(!BTreeTupleIsPosting(itup) && !BTreeTupleIsPivot(itup));
2858 :
2859 : /* Easy case: It's immediately clear that tuple has no varlena datums */
2860 2138204 : if (!IndexTupleHasVarwidths(itup))
2861 2138156 : return itup;
2862 :
2863 96 : for (i = 0; i < tupleDescriptor->natts; i++)
2864 : {
2865 : Form_pg_attribute att;
2866 :
2867 48 : att = TupleDescAttr(tupleDescriptor, i);
2868 :
2869 : /* Assume untoasted/already normalized datum initially */
2870 48 : need_free[i] = false;
2871 48 : normalized[i] = index_getattr(itup, att->attnum,
2872 : tupleDescriptor,
2873 : &isnull[i]);
2874 48 : if (att->attbyval || att->attlen != -1 || isnull[i])
2875 0 : continue;
2876 :
2877 : /*
2878 : * Callers always pass a tuple that could safely be inserted into the
2879 : * index without further processing, so an external varlena header
2880 : * should never be encountered here
2881 : */
2882 48 : if (VARATT_IS_EXTERNAL(DatumGetPointer(normalized[i])))
2883 0 : ereport(ERROR,
2884 : (errcode(ERRCODE_INDEX_CORRUPTED),
2885 : errmsg("external varlena datum in tuple that references heap row (%u,%u) in index \"%s\"",
2886 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2887 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2888 : RelationGetRelationName(state->rel))));
2889 92 : else if (!VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])) &&
2890 44 : VARSIZE(DatumGetPointer(normalized[i])) > TOAST_INDEX_TARGET &&
2891 42 : (att->attstorage == TYPSTORAGE_EXTENDED ||
2892 32 : att->attstorage == TYPSTORAGE_MAIN))
2893 : {
2894 : /*
2895 : * This value will be compressed by index_form_tuple() with the
2896 : * current storage settings. We may be here because this tuple
2897 : * was formed with different storage settings. So, force forming.
2898 : */
2899 10 : formnewtup = true;
2900 : }
2901 38 : else if (VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])))
2902 : {
2903 4 : formnewtup = true;
2904 4 : normalized[i] = PointerGetDatum(PG_DETOAST_DATUM(normalized[i]));
2905 4 : need_free[i] = true;
2906 : }
2907 :
2908 : /*
2909 : * Short tuples may have 1B or 4B header. Convert 4B header of short
2910 : * tuples to 1B
2911 : */
2912 34 : else if (VARATT_CAN_MAKE_SHORT(DatumGetPointer(normalized[i])))
2913 : {
2914 : /* convert to short varlena */
2915 2 : Size len = VARATT_CONVERTED_SHORT_SIZE(DatumGetPointer(normalized[i]));
2916 2 : char *data = palloc(len);
2917 :
2918 2 : SET_VARSIZE_SHORT(data, len);
2919 2 : memcpy(data + 1, VARDATA(DatumGetPointer(normalized[i])), len - 1);
2920 :
2921 2 : formnewtup = true;
2922 2 : normalized[i] = PointerGetDatum(data);
2923 2 : need_free[i] = true;
2924 : }
2925 : }
2926 :
2927 : /*
2928 : * Easier case: Tuple has varlena datums, none of which are compressed or
2929 : * short with 4B header
2930 : */
2931 48 : if (!formnewtup)
2932 32 : return itup;
2933 :
2934 : /*
2935 : * Hard case: Tuple had compressed varlena datums that necessitate
2936 : * creating normalized version of the tuple from uncompressed input datums
2937 : * (normalized input datums). This is rather naive, but shouldn't be
2938 : * necessary too often.
2939 : *
2940 : * In the heap, tuples may contain short varlena datums with both 1B
2941 : * header and 4B headers. But the corresponding index tuple should always
2942 : * have such varlena's with 1B headers. So, if there is a short varlena
2943 : * with 4B header, we need to convert it for fingerprinting.
2944 : *
2945 : * Note that we rely on deterministic index_form_tuple() TOAST compression
2946 : * of normalized input.
2947 : */
2948 16 : reformed = index_form_tuple(tupleDescriptor, normalized, isnull);
2949 16 : reformed->t_tid = itup->t_tid;
2950 :
2951 : /* Cannot leak memory here */
2952 32 : for (i = 0; i < tupleDescriptor->natts; i++)
2953 16 : if (need_free[i])
2954 6 : pfree(DatumGetPointer(normalized[i]));
2955 :
2956 16 : return reformed;
2957 : }
2958 :
2959 : /*
2960 : * Produce palloc()'d "plain" tuple for nth posting list entry/TID.
2961 : *
2962 : * In general, deduplication is not supposed to change the logical contents of
2963 : * an index. Multiple index tuples are merged together into one equivalent
2964 : * posting list index tuple when convenient.
2965 : *
2966 : * heapallindexed verification must normalize-away this variation in
2967 : * representation by converting posting list tuples into two or more "plain"
2968 : * tuples. Each tuple must be fingerprinted separately -- there must be one
2969 : * tuple for each corresponding Bloom filter probe during the heap scan.
2970 : *
2971 : * Note: Caller still needs to call bt_normalize_tuple() with returned tuple.
2972 : */
2973 : static inline IndexTuple
2974 57804 : bt_posting_plain_tuple(IndexTuple itup, int n)
2975 : {
2976 : Assert(BTreeTupleIsPosting(itup));
2977 :
2978 : /* Returns non-posting-list tuple */
2979 57804 : return _bt_form_posting(itup, BTreeTupleGetPostingN(itup, n), 1);
2980 : }
2981 :
2982 : /*
2983 : * Search for itup in index, starting from fast root page. itup must be a
2984 : * non-pivot tuple. This is only supported with heapkeyspace indexes, since
2985 : * we rely on having fully unique keys to find a match with only a single
2986 : * visit to a leaf page, barring an interrupted page split, where we may have
2987 : * to move right. (A concurrent page split is impossible because caller must
2988 : * be readonly caller.)
2989 : *
2990 : * This routine can detect very subtle transitive consistency issues across
2991 : * more than one level of the tree. Leaf pages all have a high key (even the
2992 : * rightmost page has a conceptual positive infinity high key), but not a low
2993 : * key. Their downlink in parent is a lower bound, which along with the high
2994 : * key is almost enough to detect every possible inconsistency. A downlink
2995 : * separator key value won't always be available from parent, though, because
2996 : * the first items of internal pages are negative infinity items, truncated
2997 : * down to zero attributes during internal page splits. While it's true that
2998 : * bt_child_check() and the high key check can detect most imaginable key
2999 : * space problems, there are remaining problems it won't detect with non-pivot
3000 : * tuples in cousin leaf pages. Starting a search from the root for every
3001 : * existing leaf tuple detects small inconsistencies in upper levels of the
3002 : * tree that cannot be detected any other way. (Besides all this, this is
3003 : * probably also useful as a direct test of the code used by index scans
3004 : * themselves.)
3005 : */
3006 : static bool
3007 402196 : bt_rootdescend(BtreeCheckState *state, IndexTuple itup)
3008 : {
3009 : BTScanInsert key;
3010 : BTStack stack;
3011 : Buffer lbuf;
3012 : bool exists;
3013 :
3014 402196 : key = _bt_mkscankey(state->rel, itup);
3015 : Assert(key->heapkeyspace && key->scantid != NULL);
3016 :
3017 : /*
3018 : * Search from root.
3019 : *
3020 : * Ideally, we would arrange to only move right within _bt_search() when
3021 : * an interrupted page split is detected (i.e. when the incomplete split
3022 : * bit is found to be set), but for now we accept the possibility that
3023 : * that could conceal an inconsistency.
3024 : */
3025 : Assert(state->readonly && state->rootdescend);
3026 402196 : exists = false;
3027 402196 : stack = _bt_search(state->rel, NULL, key, &lbuf, BT_READ);
3028 :
3029 402196 : if (BufferIsValid(lbuf))
3030 : {
3031 : BTInsertStateData insertstate;
3032 : OffsetNumber offnum;
3033 : Page page;
3034 :
3035 402196 : insertstate.itup = itup;
3036 402196 : insertstate.itemsz = MAXALIGN(IndexTupleSize(itup));
3037 402196 : insertstate.itup_key = key;
3038 402196 : insertstate.postingoff = 0;
3039 402196 : insertstate.bounds_valid = false;
3040 402196 : insertstate.buf = lbuf;
3041 :
3042 : /* Get matching tuple on leaf page */
3043 402196 : offnum = _bt_binsrch_insert(state->rel, &insertstate);
3044 : /* Compare first >= matching item on leaf page, if any */
3045 402196 : page = BufferGetPage(lbuf);
3046 : /* Should match on first heap TID when tuple has a posting list */
3047 402196 : if (offnum <= PageGetMaxOffsetNumber(page) &&
3048 804392 : insertstate.postingoff <= 0 &&
3049 402196 : _bt_compare(state->rel, key, page, offnum) == 0)
3050 402196 : exists = true;
3051 402196 : _bt_relbuf(state->rel, lbuf);
3052 : }
3053 :
3054 402196 : _bt_freestack(stack);
3055 402196 : pfree(key);
3056 :
3057 402196 : return exists;
3058 : }
3059 :
3060 : /*
3061 : * Is particular offset within page (whose special state is passed by caller)
3062 : * the page negative-infinity item?
3063 : *
3064 : * As noted in comments above _bt_compare(), there is special handling of the
3065 : * first data item as a "negative infinity" item. The hard-coding within
3066 : * _bt_compare() makes comparing this item for the purposes of verification
3067 : * pointless at best, since the IndexTuple only contains a valid TID (a
3068 : * reference TID to child page).
3069 : */
3070 : static inline bool
3071 5250110 : offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
3072 : {
3073 : /*
3074 : * For internal pages only, the first item after high key, if any, is
3075 : * negative infinity item. Internal pages always have a negative infinity
3076 : * item, whereas leaf pages never have one. This implies that negative
3077 : * infinity item is either first or second line item, or there is none
3078 : * within page.
3079 : *
3080 : * Negative infinity items are a special case among pivot tuples. They
3081 : * always have zero attributes, while all other pivot tuples always have
3082 : * nkeyatts attributes.
3083 : *
3084 : * Right-most pages don't have a high key, but could be said to
3085 : * conceptually have a "positive infinity" high key. Thus, there is a
3086 : * symmetry between down link items in parent pages, and high keys in
3087 : * children. Together, they represent the part of the key space that
3088 : * belongs to each page in the index. For example, all children of the
3089 : * root page will have negative infinity as a lower bound from root
3090 : * negative infinity downlink, and positive infinity as an upper bound
3091 : * (implicitly, from "imaginary" positive infinity high key in root).
3092 : */
3093 5250110 : return !P_ISLEAF(opaque) && offset == P_FIRSTDATAKEY(opaque);
3094 : }
3095 :
3096 : /*
3097 : * Does the invariant hold that the key is strictly less than a given upper
3098 : * bound offset item?
3099 : *
3100 : * Verifies line pointer on behalf of caller.
3101 : *
3102 : * If this function returns false, convention is that caller throws error due
3103 : * to corruption.
3104 : */
3105 : static inline bool
3106 4031214 : invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
3107 : OffsetNumber upperbound)
3108 : {
3109 : ItemId itemid;
3110 : int32 cmp;
3111 :
3112 : Assert(!key->nextkey && key->backward);
3113 :
3114 : /* Verify line pointer before checking tuple */
3115 4031214 : itemid = PageGetItemIdCareful(state, state->targetblock, state->target,
3116 : upperbound);
3117 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3118 4031214 : if (!key->heapkeyspace)
3119 0 : return invariant_leq_offset(state, key, upperbound);
3120 :
3121 4031214 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3122 :
3123 : /*
3124 : * _bt_compare() is capable of determining that a scankey with a
3125 : * filled-out attribute is greater than pivot tuples where the comparison
3126 : * is resolved at a truncated attribute (value of attribute in pivot is
3127 : * minus infinity). However, it is not capable of determining that a
3128 : * scankey is _less than_ a tuple on the basis of a comparison resolved at
3129 : * _scankey_ minus infinity attribute. Complete an extra step to simulate
3130 : * having minus infinity values for omitted scankey attribute(s).
3131 : */
3132 4031214 : if (cmp == 0)
3133 : {
3134 : BTPageOpaque topaque;
3135 : IndexTuple ritup;
3136 : int uppnkeyatts;
3137 : ItemPointer rheaptid;
3138 : bool nonpivot;
3139 :
3140 0 : ritup = (IndexTuple) PageGetItem(state->target, itemid);
3141 0 : topaque = BTPageGetOpaque(state->target);
3142 0 : nonpivot = P_ISLEAF(topaque) && upperbound >= P_FIRSTDATAKEY(topaque);
3143 :
3144 : /* Get number of keys + heap TID for item to the right */
3145 0 : uppnkeyatts = BTreeTupleGetNKeyAtts(ritup, state->rel);
3146 0 : rheaptid = BTreeTupleGetHeapTIDCareful(state, ritup, nonpivot);
3147 :
3148 : /* Heap TID is tiebreaker key attribute */
3149 0 : if (key->keysz == uppnkeyatts)
3150 0 : return key->scantid == NULL && rheaptid != NULL;
3151 :
3152 0 : return key->keysz < uppnkeyatts;
3153 : }
3154 :
3155 4031214 : return cmp < 0;
3156 : }
3157 :
3158 : /*
3159 : * Does the invariant hold that the key is less than or equal to a given upper
3160 : * bound offset item?
3161 : *
3162 : * Caller should have verified that upperbound's line pointer is consistent
3163 : * using PageGetItemIdCareful() call.
3164 : *
3165 : * If this function returns false, convention is that caller throws error due
3166 : * to corruption.
3167 : */
3168 : static inline bool
3169 3700602 : invariant_leq_offset(BtreeCheckState *state, BTScanInsert key,
3170 : OffsetNumber upperbound)
3171 : {
3172 : int32 cmp;
3173 :
3174 : Assert(!key->nextkey && key->backward);
3175 :
3176 3700602 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3177 :
3178 3700602 : return cmp <= 0;
3179 : }
3180 :
3181 : /*
3182 : * Does the invariant hold that the key is strictly greater than a given lower
3183 : * bound offset item?
3184 : *
3185 : * Caller should have verified that lowerbound's line pointer is consistent
3186 : * using PageGetItemIdCareful() call.
3187 : *
3188 : * If this function returns false, convention is that caller throws error due
3189 : * to corruption.
3190 : */
3191 : static inline bool
3192 13228 : invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
3193 : OffsetNumber lowerbound)
3194 : {
3195 : int32 cmp;
3196 :
3197 : Assert(!key->nextkey && key->backward);
3198 :
3199 13228 : cmp = _bt_compare(state->rel, key, state->target, lowerbound);
3200 :
3201 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3202 13228 : if (!key->heapkeyspace)
3203 0 : return cmp >= 0;
3204 :
3205 : /*
3206 : * No need to consider the possibility that scankey has attributes that we
3207 : * need to force to be interpreted as negative infinity. _bt_compare() is
3208 : * able to determine that scankey is greater than negative infinity. The
3209 : * distinction between "==" and "<" isn't interesting here, since
3210 : * corruption is indicated either way.
3211 : */
3212 13228 : return cmp > 0;
3213 : }
3214 :
3215 : /*
3216 : * Does the invariant hold that the key is strictly less than a given upper
3217 : * bound offset item, with the offset relating to a caller-supplied page that
3218 : * is not the current target page?
3219 : *
3220 : * Caller's non-target page is a child page of the target, checked as part of
3221 : * checking a property of the target page (i.e. the key comes from the
3222 : * target). Verifies line pointer on behalf of caller.
3223 : *
3224 : * If this function returns false, convention is that caller throws error due
3225 : * to corruption.
3226 : */
3227 : static inline bool
3228 1196976 : invariant_l_nontarget_offset(BtreeCheckState *state, BTScanInsert key,
3229 : BlockNumber nontargetblock, Page nontarget,
3230 : OffsetNumber upperbound)
3231 : {
3232 : ItemId itemid;
3233 : int32 cmp;
3234 :
3235 : Assert(!key->nextkey && key->backward);
3236 :
3237 : /* Verify line pointer before checking tuple */
3238 1196976 : itemid = PageGetItemIdCareful(state, nontargetblock, nontarget,
3239 : upperbound);
3240 1196976 : cmp = _bt_compare(state->rel, key, nontarget, upperbound);
3241 :
3242 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3243 1196976 : if (!key->heapkeyspace)
3244 0 : return cmp <= 0;
3245 :
3246 : /* See invariant_l_offset() for an explanation of this extra step */
3247 1196976 : if (cmp == 0)
3248 : {
3249 : IndexTuple child;
3250 : int uppnkeyatts;
3251 : ItemPointer childheaptid;
3252 : BTPageOpaque copaque;
3253 : bool nonpivot;
3254 :
3255 3720 : child = (IndexTuple) PageGetItem(nontarget, itemid);
3256 3720 : copaque = BTPageGetOpaque(nontarget);
3257 3720 : nonpivot = P_ISLEAF(copaque) && upperbound >= P_FIRSTDATAKEY(copaque);
3258 :
3259 : /* Get number of keys + heap TID for child/non-target item */
3260 3720 : uppnkeyatts = BTreeTupleGetNKeyAtts(child, state->rel);
3261 3720 : childheaptid = BTreeTupleGetHeapTIDCareful(state, child, nonpivot);
3262 :
3263 : /* Heap TID is tiebreaker key attribute */
3264 3720 : if (key->keysz == uppnkeyatts)
3265 3720 : return key->scantid == NULL && childheaptid != NULL;
3266 :
3267 0 : return key->keysz < uppnkeyatts;
3268 : }
3269 :
3270 1193256 : return cmp < 0;
3271 : }
3272 :
3273 : /*
3274 : * Given a block number of a B-Tree page, return page in palloc()'d memory.
3275 : * While at it, perform some basic checks of the page.
3276 : *
3277 : * There is never an attempt to get a consistent view of multiple pages using
3278 : * multiple concurrent buffer locks; in general, we only acquire a single pin
3279 : * and buffer lock at a time, which is often all that the nbtree code requires.
3280 : * (Actually, bt_recheck_sibling_links couples buffer locks, which is the only
3281 : * exception to this general rule.)
3282 : *
3283 : * Operating on a copy of the page is useful because it prevents control
3284 : * getting stuck in an uninterruptible state when an underlying operator class
3285 : * misbehaves.
3286 : */
3287 : static Page
3288 43184 : palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum)
3289 : {
3290 : Buffer buffer;
3291 : Page page;
3292 : BTPageOpaque opaque;
3293 : OffsetNumber maxoffset;
3294 :
3295 43184 : page = palloc(BLCKSZ);
3296 :
3297 : /*
3298 : * We copy the page into local storage to avoid holding pin on the buffer
3299 : * longer than we must.
3300 : */
3301 43184 : buffer = ReadBufferExtended(state->rel, MAIN_FORKNUM, blocknum, RBM_NORMAL,
3302 : state->checkstrategy);
3303 43160 : LockBuffer(buffer, BT_READ);
3304 :
3305 : /*
3306 : * Perform the same basic sanity checking that nbtree itself performs for
3307 : * every page:
3308 : */
3309 43160 : _bt_checkpage(state->rel, buffer);
3310 :
3311 : /* Only use copy of page in palloc()'d memory */
3312 43160 : memcpy(page, BufferGetPage(buffer), BLCKSZ);
3313 43160 : UnlockReleaseBuffer(buffer);
3314 :
3315 43160 : opaque = BTPageGetOpaque(page);
3316 :
3317 43160 : if (P_ISMETA(opaque) && blocknum != BTREE_METAPAGE)
3318 0 : ereport(ERROR,
3319 : (errcode(ERRCODE_INDEX_CORRUPTED),
3320 : errmsg("invalid meta page found at block %u in index \"%s\"",
3321 : blocknum, RelationGetRelationName(state->rel))));
3322 :
3323 : /* Check page from block that ought to be meta page */
3324 43160 : if (blocknum == BTREE_METAPAGE)
3325 : {
3326 7954 : BTMetaPageData *metad = BTPageGetMeta(page);
3327 :
3328 7954 : if (!P_ISMETA(opaque) ||
3329 7954 : metad->btm_magic != BTREE_MAGIC)
3330 0 : ereport(ERROR,
3331 : (errcode(ERRCODE_INDEX_CORRUPTED),
3332 : errmsg("index \"%s\" meta page is corrupt",
3333 : RelationGetRelationName(state->rel))));
3334 :
3335 7954 : if (metad->btm_version < BTREE_MIN_VERSION ||
3336 7954 : metad->btm_version > BTREE_VERSION)
3337 0 : ereport(ERROR,
3338 : (errcode(ERRCODE_INDEX_CORRUPTED),
3339 : errmsg("version mismatch in index \"%s\": file version %d, "
3340 : "current version %d, minimum supported version %d",
3341 : RelationGetRelationName(state->rel),
3342 : metad->btm_version, BTREE_VERSION,
3343 : BTREE_MIN_VERSION)));
3344 :
3345 : /* Finished with metapage checks */
3346 7954 : return page;
3347 : }
3348 :
3349 : /*
3350 : * Deleted pages that still use the old 32-bit XID representation have no
3351 : * sane "level" field because they type pun the field, but all other pages
3352 : * (including pages deleted on Postgres 14+) have a valid value.
3353 : */
3354 35206 : if (!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque))
3355 : {
3356 : /* Okay, no reason not to trust btpo_level field from page */
3357 :
3358 35206 : if (P_ISLEAF(opaque) && opaque->btpo_level != 0)
3359 0 : ereport(ERROR,
3360 : (errcode(ERRCODE_INDEX_CORRUPTED),
3361 : errmsg_internal("invalid leaf page level %u for block %u in index \"%s\"",
3362 : opaque->btpo_level, blocknum,
3363 : RelationGetRelationName(state->rel))));
3364 :
3365 35206 : if (!P_ISLEAF(opaque) && opaque->btpo_level == 0)
3366 0 : ereport(ERROR,
3367 : (errcode(ERRCODE_INDEX_CORRUPTED),
3368 : errmsg_internal("invalid internal page level 0 for block %u in index \"%s\"",
3369 : blocknum,
3370 : RelationGetRelationName(state->rel))));
3371 : }
3372 :
3373 : /*
3374 : * Sanity checks for number of items on page.
3375 : *
3376 : * As noted at the beginning of _bt_binsrch(), an internal page must have
3377 : * children, since there must always be a negative infinity downlink
3378 : * (there may also be a highkey). In the case of non-rightmost leaf
3379 : * pages, there must be at least a highkey. The exceptions are deleted
3380 : * pages, which contain no items.
3381 : *
3382 : * This is correct when pages are half-dead, since internal pages are
3383 : * never half-dead, and leaf pages must have a high key when half-dead
3384 : * (the rightmost page can never be deleted). It's also correct with
3385 : * fully deleted pages: _bt_unlink_halfdead_page() doesn't change anything
3386 : * about the target page other than setting the page as fully dead, and
3387 : * setting its xact field. In particular, it doesn't change the sibling
3388 : * links in the deletion target itself, since they're required when index
3389 : * scans land on the deletion target, and then need to move right (or need
3390 : * to move left, in the case of backward index scans).
3391 : */
3392 35206 : maxoffset = PageGetMaxOffsetNumber(page);
3393 35206 : if (maxoffset > MaxIndexTuplesPerPage)
3394 0 : ereport(ERROR,
3395 : (errcode(ERRCODE_INDEX_CORRUPTED),
3396 : errmsg("Number of items on block %u of index \"%s\" exceeds MaxIndexTuplesPerPage (%u)",
3397 : blocknum, RelationGetRelationName(state->rel),
3398 : MaxIndexTuplesPerPage)));
3399 :
3400 35206 : if (!P_ISLEAF(opaque) && !P_ISDELETED(opaque) && maxoffset < P_FIRSTDATAKEY(opaque))
3401 0 : ereport(ERROR,
3402 : (errcode(ERRCODE_INDEX_CORRUPTED),
3403 : errmsg("internal block %u in index \"%s\" lacks high key and/or at least one downlink",
3404 : blocknum, RelationGetRelationName(state->rel))));
3405 :
3406 35206 : if (P_ISLEAF(opaque) && !P_ISDELETED(opaque) && !P_RIGHTMOST(opaque) && maxoffset < P_HIKEY)
3407 0 : ereport(ERROR,
3408 : (errcode(ERRCODE_INDEX_CORRUPTED),
3409 : errmsg("non-rightmost leaf block %u in index \"%s\" lacks high key item",
3410 : blocknum, RelationGetRelationName(state->rel))));
3411 :
3412 : /*
3413 : * In general, internal pages are never marked half-dead, except on
3414 : * versions of Postgres prior to 9.4, where it can be valid transient
3415 : * state. This state is nonetheless treated as corruption by VACUUM on
3416 : * from version 9.4 on, so do the same here. See _bt_pagedel() for full
3417 : * details.
3418 : */
3419 35206 : if (!P_ISLEAF(opaque) && P_ISHALFDEAD(opaque))
3420 0 : ereport(ERROR,
3421 : (errcode(ERRCODE_INDEX_CORRUPTED),
3422 : errmsg("internal page block %u in index \"%s\" is half-dead",
3423 : blocknum, RelationGetRelationName(state->rel)),
3424 : errhint("This can be caused by an interrupted VACUUM in version 9.3 or older, before upgrade. Please REINDEX it.")));
3425 :
3426 : /*
3427 : * Check that internal pages have no garbage items, and that no page has
3428 : * an invalid combination of deletion-related page level flags
3429 : */
3430 35206 : if (!P_ISLEAF(opaque) && P_HAS_GARBAGE(opaque))
3431 0 : ereport(ERROR,
3432 : (errcode(ERRCODE_INDEX_CORRUPTED),
3433 : errmsg_internal("internal page block %u in index \"%s\" has garbage items",
3434 : blocknum, RelationGetRelationName(state->rel))));
3435 :
3436 35206 : if (P_HAS_FULLXID(opaque) && !P_ISDELETED(opaque))
3437 0 : ereport(ERROR,
3438 : (errcode(ERRCODE_INDEX_CORRUPTED),
3439 : errmsg_internal("full transaction id page flag appears in non-deleted block %u in index \"%s\"",
3440 : blocknum, RelationGetRelationName(state->rel))));
3441 :
3442 35206 : if (P_ISDELETED(opaque) && P_ISHALFDEAD(opaque))
3443 0 : ereport(ERROR,
3444 : (errcode(ERRCODE_INDEX_CORRUPTED),
3445 : errmsg_internal("deleted page block %u in index \"%s\" is half-dead",
3446 : blocknum, RelationGetRelationName(state->rel))));
3447 :
3448 35206 : return page;
3449 : }
3450 :
3451 : /*
3452 : * _bt_mkscankey() wrapper that automatically prevents insertion scankey from
3453 : * being considered greater than the pivot tuple that its values originated
3454 : * from (or some other identical pivot tuple) in the common case where there
3455 : * are truncated/minus infinity attributes. Without this extra step, there
3456 : * are forms of corruption that amcheck could theoretically fail to report.
3457 : *
3458 : * For example, invariant_g_offset() might miss a cross-page invariant failure
3459 : * on an internal level if the scankey built from the first item on the
3460 : * target's right sibling page happened to be equal to (not greater than) the
3461 : * last item on target page. The !backward tiebreaker in _bt_compare() might
3462 : * otherwise cause amcheck to assume (rather than actually verify) that the
3463 : * scankey is greater.
3464 : */
3465 : static inline BTScanInsert
3466 4061522 : bt_mkscankey_pivotsearch(Relation rel, IndexTuple itup)
3467 : {
3468 : BTScanInsert skey;
3469 :
3470 4061522 : skey = _bt_mkscankey(rel, itup);
3471 4061522 : skey->backward = true;
3472 :
3473 4061522 : return skey;
3474 : }
3475 :
3476 : /*
3477 : * PageGetItemId() wrapper that validates returned line pointer.
3478 : *
3479 : * Buffer page/page item access macros generally trust that line pointers are
3480 : * not corrupt, which might cause problems for verification itself. For
3481 : * example, there is no bounds checking in PageGetItem(). Passing it a
3482 : * corrupt line pointer can cause it to return a tuple/pointer that is unsafe
3483 : * to dereference.
3484 : *
3485 : * Validating line pointers before tuples avoids undefined behavior and
3486 : * assertion failures with corrupt indexes, making the verification process
3487 : * more robust and predictable.
3488 : */
3489 : static ItemId
3490 9323818 : PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block, Page page,
3491 : OffsetNumber offset)
3492 : {
3493 9323818 : ItemId itemid = PageGetItemId(page, offset);
3494 :
3495 9323818 : if (ItemIdGetOffset(itemid) + ItemIdGetLength(itemid) >
3496 : BLCKSZ - MAXALIGN(sizeof(BTPageOpaqueData)))
3497 0 : ereport(ERROR,
3498 : (errcode(ERRCODE_INDEX_CORRUPTED),
3499 : errmsg("line pointer points past end of tuple space in index \"%s\"",
3500 : RelationGetRelationName(state->rel)),
3501 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3502 : block, offset, ItemIdGetOffset(itemid),
3503 : ItemIdGetLength(itemid),
3504 : ItemIdGetFlags(itemid))));
3505 :
3506 : /*
3507 : * Verify that line pointer isn't LP_REDIRECT or LP_UNUSED, since nbtree
3508 : * never uses either. Verify that line pointer has storage, too, since
3509 : * even LP_DEAD items should within nbtree.
3510 : */
3511 9323818 : if (ItemIdIsRedirected(itemid) || !ItemIdIsUsed(itemid) ||
3512 9323818 : ItemIdGetLength(itemid) == 0)
3513 0 : ereport(ERROR,
3514 : (errcode(ERRCODE_INDEX_CORRUPTED),
3515 : errmsg("invalid line pointer storage 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 9323818 : return itemid;
3523 : }
3524 :
3525 : /*
3526 : * BTreeTupleGetHeapTID() wrapper that enforces that a heap TID is present in
3527 : * cases where that is mandatory (i.e. for non-pivot tuples)
3528 : */
3529 : static inline ItemPointer
3530 3720 : BTreeTupleGetHeapTIDCareful(BtreeCheckState *state, IndexTuple itup,
3531 : bool nonpivot)
3532 : {
3533 : ItemPointer htid;
3534 :
3535 : /*
3536 : * Caller determines whether this is supposed to be a pivot or non-pivot
3537 : * tuple using page type and item offset number. Verify that tuple
3538 : * metadata agrees with this.
3539 : */
3540 : Assert(state->heapkeyspace);
3541 3720 : if (BTreeTupleIsPivot(itup) && nonpivot)
3542 0 : ereport(ERROR,
3543 : (errcode(ERRCODE_INDEX_CORRUPTED),
3544 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected pivot tuple",
3545 : state->targetblock,
3546 : RelationGetRelationName(state->rel))));
3547 :
3548 3720 : if (!BTreeTupleIsPivot(itup) && !nonpivot)
3549 0 : ereport(ERROR,
3550 : (errcode(ERRCODE_INDEX_CORRUPTED),
3551 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected non-pivot tuple",
3552 : state->targetblock,
3553 : RelationGetRelationName(state->rel))));
3554 :
3555 3720 : htid = BTreeTupleGetHeapTID(itup);
3556 3720 : if (!ItemPointerIsValid(htid) && nonpivot)
3557 0 : ereport(ERROR,
3558 : (errcode(ERRCODE_INDEX_CORRUPTED),
3559 : errmsg("block %u or its right sibling block or child block in index \"%s\" contains non-pivot tuple that lacks a heap TID",
3560 : state->targetblock,
3561 : RelationGetRelationName(state->rel))));
3562 :
3563 3720 : return htid;
3564 : }
3565 :
3566 : /*
3567 : * Return the "pointed to" TID for itup, which is used to generate a
3568 : * descriptive error message. itup must be a "data item" tuple (it wouldn't
3569 : * make much sense to call here with a high key tuple, since there won't be a
3570 : * valid downlink/block number to display).
3571 : *
3572 : * Returns either a heap TID (which will be the first heap TID in posting list
3573 : * if itup is posting list tuple), or a TID that contains downlink block
3574 : * number, plus some encoded metadata (e.g., the number of attributes present
3575 : * in itup).
3576 : */
3577 : static inline ItemPointer
3578 12 : BTreeTupleGetPointsToTID(IndexTuple itup)
3579 : {
3580 : /*
3581 : * Rely on the assumption that !heapkeyspace internal page data items will
3582 : * correctly return TID with downlink here -- BTreeTupleGetHeapTID() won't
3583 : * recognize it as a pivot tuple, but everything still works out because
3584 : * the t_tid field is still returned
3585 : */
3586 12 : if (!BTreeTupleIsPivot(itup))
3587 8 : return BTreeTupleGetHeapTID(itup);
3588 :
3589 : /* Pivot tuple returns TID with downlink block (heapkeyspace variant) */
3590 4 : return &itup->t_tid;
3591 : }
|
