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) 2017-2026, 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 :
46 329 : 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 this field and the one below once
96 : * per index check.
97 : */
98 : IndexInfo *indexinfo;
99 : /* Table scan snapshot for heapallindexed and checkunique */
100 : Snapshot snapshot;
101 :
102 : /*
103 : * Mutable state, for verification of particular page:
104 : */
105 :
106 : /* Current target page */
107 : Page target;
108 : /* Target block number */
109 : BlockNumber targetblock;
110 : /* Target page's LSN */
111 : XLogRecPtr targetlsn;
112 :
113 : /*
114 : * Low key: high key of left sibling of target page. Used only for child
115 : * verification. So, 'lowkey' is kept only when 'readonly' is set.
116 : */
117 : IndexTuple lowkey;
118 :
119 : /*
120 : * The rightlink and incomplete split flag of block one level down to the
121 : * target page, which was visited last time via downlink from target page.
122 : * We use it to check for missing downlinks.
123 : */
124 : BlockNumber prevrightlink;
125 : bool previncompletesplit;
126 :
127 : /*
128 : * Mutable state, for optional heapallindexed verification:
129 : */
130 :
131 : /* Bloom filter fingerprints B-Tree index */
132 : bloom_filter *filter;
133 : /* Debug counter */
134 : int64 heaptuplespresent;
135 : } BtreeCheckState;
136 :
137 : /*
138 : * Starting point for verifying an entire B-Tree index level
139 : */
140 : typedef struct BtreeLevel
141 : {
142 : /* Level number (0 is leaf page level). */
143 : uint32 level;
144 :
145 : /* Left most block on level. Scan of level begins here. */
146 : BlockNumber leftmost;
147 :
148 : /* Is this level reported as "true" root level by meta page? */
149 : bool istruerootlevel;
150 : } BtreeLevel;
151 :
152 : /*
153 : * Information about the last visible entry with current B-tree key. Used
154 : * for validation of the unique constraint.
155 : */
156 : typedef struct BtreeLastVisibleEntry
157 : {
158 : BlockNumber blkno; /* Index block */
159 : OffsetNumber offset; /* Offset on index block */
160 : int postingIndex; /* Number in the posting list (-1 for
161 : * non-deduplicated tuples) */
162 : ItemPointer tid; /* Heap tid */
163 : } BtreeLastVisibleEntry;
164 :
165 : /*
166 : * arguments for the bt_index_check_callback callback
167 : */
168 : typedef struct BTCallbackState
169 : {
170 : bool parentcheck;
171 : bool heapallindexed;
172 : bool rootdescend;
173 : bool checkunique;
174 : } BTCallbackState;
175 :
176 92 : PG_FUNCTION_INFO_V1(bt_index_check);
177 65 : PG_FUNCTION_INFO_V1(bt_index_parent_check);
178 :
179 : static void bt_index_check_callback(Relation indrel, Relation heaprel,
180 : void *state, bool readonly);
181 : static void bt_check_every_level(Relation rel, Relation heaprel,
182 : bool heapkeyspace, bool readonly, bool heapallindexed,
183 : bool rootdescend, bool checkunique);
184 : static BtreeLevel bt_check_level_from_leftmost(BtreeCheckState *state,
185 : BtreeLevel level);
186 : static bool bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
187 : BlockNumber start,
188 : BTPageOpaque start_opaque);
189 : static void bt_recheck_sibling_links(BtreeCheckState *state,
190 : BlockNumber btpo_prev_from_target,
191 : BlockNumber leftcurrent);
192 : static bool heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid);
193 : static void bt_report_duplicate(BtreeCheckState *state,
194 : BtreeLastVisibleEntry *lVis,
195 : ItemPointer nexttid,
196 : BlockNumber nblock, OffsetNumber noffset,
197 : int nposting);
198 : static void bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
199 : BlockNumber targetblock, OffsetNumber offset,
200 : BtreeLastVisibleEntry *lVis);
201 : static void bt_target_page_check(BtreeCheckState *state);
202 : static BTScanInsert bt_right_page_check_scankey(BtreeCheckState *state,
203 : OffsetNumber *rightfirstoffset);
204 : static void bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
205 : OffsetNumber downlinkoffnum);
206 : static void bt_child_highkey_check(BtreeCheckState *state,
207 : OffsetNumber target_downlinkoffnum,
208 : Page loaded_child,
209 : uint32 target_level);
210 : static void bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
211 : BlockNumber blkno, Page page);
212 : static void bt_tuple_present_callback(Relation index, ItemPointer tid,
213 : Datum *values, bool *isnull,
214 : bool tupleIsAlive, void *checkstate);
215 : static IndexTuple bt_normalize_tuple(BtreeCheckState *state,
216 : IndexTuple itup);
217 : static inline IndexTuple bt_posting_plain_tuple(IndexTuple itup, int n);
218 : static bool bt_rootdescend(BtreeCheckState *state, IndexTuple itup);
219 : static inline bool offset_is_negative_infinity(BTPageOpaque opaque,
220 : OffsetNumber offset);
221 : static inline bool invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
222 : OffsetNumber upperbound);
223 : static inline bool invariant_leq_offset(BtreeCheckState *state,
224 : BTScanInsert key,
225 : OffsetNumber upperbound);
226 : static inline bool invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
227 : OffsetNumber lowerbound);
228 : static inline bool invariant_l_nontarget_offset(BtreeCheckState *state,
229 : BTScanInsert key,
230 : BlockNumber nontargetblock,
231 : Page nontarget,
232 : OffsetNumber upperbound);
233 : static Page palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum);
234 : static inline BTScanInsert bt_mkscankey_pivotsearch(Relation rel,
235 : IndexTuple itup);
236 : static ItemId PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block,
237 : Page page, OffsetNumber offset);
238 : static inline ItemPointer BTreeTupleGetHeapTIDCareful(BtreeCheckState *state,
239 : IndexTuple itup, bool nonpivot);
240 : static inline ItemPointer BTreeTupleGetPointsToTID(IndexTuple itup);
241 :
242 : /*
243 : * bt_index_check(index regclass, heapallindexed boolean, checkunique boolean)
244 : *
245 : * Verify integrity of B-Tree index.
246 : *
247 : * Acquires AccessShareLock on heap & index relations. Does not consider
248 : * invariants that exist between parent/child pages. Optionally verifies
249 : * that heap does not contain any unindexed or incorrectly indexed tuples.
250 : */
251 : Datum
252 4208 : bt_index_check(PG_FUNCTION_ARGS)
253 : {
254 4208 : Oid indrelid = PG_GETARG_OID(0);
255 : BTCallbackState args;
256 :
257 4208 : args.heapallindexed = false;
258 4208 : args.rootdescend = false;
259 4208 : args.parentcheck = false;
260 4208 : args.checkunique = false;
261 :
262 4208 : if (PG_NARGS() >= 2)
263 4202 : args.heapallindexed = PG_GETARG_BOOL(1);
264 4208 : if (PG_NARGS() >= 3)
265 734 : args.checkunique = PG_GETARG_BOOL(2);
266 :
267 4208 : amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
268 : bt_index_check_callback,
269 : AccessShareLock, &args);
270 :
271 4182 : PG_RETURN_VOID();
272 : }
273 :
274 : /*
275 : * bt_index_parent_check(index regclass, heapallindexed boolean, rootdescend boolean, checkunique boolean)
276 : *
277 : * Verify integrity of B-Tree index.
278 : *
279 : * Acquires ShareLock on heap & index relations. Verifies that downlinks in
280 : * parent pages are valid lower bounds on child pages. Optionally verifies
281 : * that heap does not contain any unindexed or incorrectly indexed tuples.
282 : */
283 : Datum
284 62 : bt_index_parent_check(PG_FUNCTION_ARGS)
285 : {
286 62 : Oid indrelid = PG_GETARG_OID(0);
287 : BTCallbackState args;
288 :
289 62 : args.heapallindexed = false;
290 62 : args.rootdescend = false;
291 62 : args.parentcheck = true;
292 62 : args.checkunique = false;
293 :
294 62 : if (PG_NARGS() >= 2)
295 56 : args.heapallindexed = PG_GETARG_BOOL(1);
296 62 : if (PG_NARGS() >= 3)
297 52 : args.rootdescend = PG_GETARG_BOOL(2);
298 62 : if (PG_NARGS() >= 4)
299 26 : args.checkunique = PG_GETARG_BOOL(3);
300 :
301 62 : amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
302 : bt_index_check_callback,
303 : ShareLock, &args);
304 :
305 44 : PG_RETURN_VOID();
306 : }
307 :
308 : /*
309 : * Helper for bt_index_[parent_]check, coordinating the bulk of the work.
310 : */
311 : static void
312 4264 : bt_index_check_callback(Relation indrel, Relation heaprel, void *state, bool readonly)
313 : {
314 4264 : BTCallbackState *args = (BTCallbackState *) state;
315 : bool heapkeyspace,
316 : allequalimage;
317 :
318 4264 : if (!smgrexists(RelationGetSmgr(indrel), MAIN_FORKNUM))
319 18 : ereport(ERROR,
320 : (errcode(ERRCODE_INDEX_CORRUPTED),
321 : errmsg("index \"%s\" lacks a main relation fork",
322 : RelationGetRelationName(indrel))));
323 :
324 : /* Extract metadata from metapage, and sanitize it in passing */
325 4246 : _bt_metaversion(indrel, &heapkeyspace, &allequalimage);
326 4246 : if (allequalimage && !heapkeyspace)
327 0 : ereport(ERROR,
328 : (errcode(ERRCODE_INDEX_CORRUPTED),
329 : errmsg("index \"%s\" metapage has equalimage field set on unsupported nbtree version",
330 : RelationGetRelationName(indrel))));
331 4246 : if (allequalimage && !_bt_allequalimage(indrel, false))
332 : {
333 0 : bool has_interval_ops = false;
334 :
335 0 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(indrel); i++)
336 0 : if (indrel->rd_opfamily[i] == INTERVAL_BTREE_FAM_OID)
337 : {
338 0 : has_interval_ops = true;
339 0 : break;
340 : }
341 :
342 0 : ereport(ERROR,
343 : (errcode(ERRCODE_INDEX_CORRUPTED),
344 : errmsg("index \"%s\" metapage incorrectly indicates that deduplication is safe",
345 : RelationGetRelationName(indrel)),
346 : has_interval_ops
347 : ? errhint("This is known of \"interval\" indexes last built on a version predating 2023-11.")
348 : : 0));
349 : }
350 :
351 : /* Check index, possibly against table it is an index on */
352 4246 : bt_check_every_level(indrel, heaprel, heapkeyspace, readonly,
353 4246 : args->heapallindexed, args->rootdescend, args->checkunique);
354 4226 : }
355 :
356 : /*
357 : * Main entry point for B-Tree SQL-callable functions. Walks the B-Tree in
358 : * logical order, verifying invariants as it goes. Optionally, verification
359 : * checks if the heap relation contains any tuples that are not represented in
360 : * the index but should be.
361 : *
362 : * It is the caller's responsibility to acquire appropriate heavyweight lock on
363 : * the index relation, and advise us if extra checks are safe when a ShareLock
364 : * is held. (A lock of the same type must also have been acquired on the heap
365 : * relation.)
366 : *
367 : * A ShareLock is generally assumed to prevent any kind of physical
368 : * modification to the index structure, including modifications that VACUUM may
369 : * make. This does not include setting of the LP_DEAD bit by concurrent index
370 : * scans, although that is just metadata that is not able to directly affect
371 : * any check performed here. Any concurrent process that might act on the
372 : * LP_DEAD bit being set (recycle space) requires a heavyweight lock that
373 : * cannot be held while we hold a ShareLock. (Besides, even if that could
374 : * happen, the ad-hoc recycling when a page might otherwise split is performed
375 : * per-page, and requires an exclusive buffer lock, which wouldn't cause us
376 : * trouble. _bt_delitems_vacuum() may only delete leaf items, and so the extra
377 : * parent/child check cannot be affected.)
378 : */
379 : static void
380 4246 : bt_check_every_level(Relation rel, Relation heaprel, bool heapkeyspace,
381 : bool readonly, bool heapallindexed, bool rootdescend,
382 : bool checkunique)
383 : {
384 : BtreeCheckState *state;
385 : Page metapage;
386 : BTMetaPageData *metad;
387 : uint32 previouslevel;
388 : BtreeLevel current;
389 :
390 4246 : if (!readonly)
391 4196 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\"",
392 : RelationGetRelationName(rel));
393 : else
394 50 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\" with cross-level checks",
395 : RelationGetRelationName(rel));
396 :
397 : /*
398 : * This assertion matches the one in index_getnext_tid(). See page
399 : * recycling/"visible to everyone" notes in nbtree README.
400 : */
401 : Assert(TransactionIdIsValid(RecentXmin));
402 :
403 : /*
404 : * Initialize state for entire verification operation
405 : */
406 4246 : state = palloc0_object(BtreeCheckState);
407 4246 : state->rel = rel;
408 4246 : state->heaprel = heaprel;
409 4246 : state->heapkeyspace = heapkeyspace;
410 4246 : state->readonly = readonly;
411 4246 : state->heapallindexed = heapallindexed;
412 4246 : state->rootdescend = rootdescend;
413 4246 : state->checkunique = checkunique;
414 4246 : state->snapshot = InvalidSnapshot;
415 :
416 4246 : if (state->heapallindexed)
417 : {
418 : int64 total_pages;
419 : int64 total_elems;
420 : uint64 seed;
421 :
422 : /*
423 : * Size Bloom filter based on estimated number of tuples in index,
424 : * while conservatively assuming that each block must contain at least
425 : * MaxTIDsPerBTreePage / 3 "plain" tuples -- see
426 : * bt_posting_plain_tuple() for definition, and details of how posting
427 : * list tuples are handled.
428 : */
429 72 : total_pages = RelationGetNumberOfBlocks(rel);
430 72 : total_elems = Max(total_pages * (MaxTIDsPerBTreePage / 3),
431 : (int64) state->rel->rd_rel->reltuples);
432 : /* Generate a random seed to avoid repetition */
433 72 : seed = pg_prng_uint64(&pg_global_prng_state);
434 : /* Create Bloom filter to fingerprint index */
435 72 : state->filter = bloom_create(total_elems, maintenance_work_mem, seed);
436 72 : state->heaptuplespresent = 0;
437 :
438 : /*
439 : * Register our own snapshot for heapallindexed, rather than asking
440 : * table_index_build_scan() to do this for us later. This needs to
441 : * happen before index fingerprinting begins, so we can later be
442 : * certain that index fingerprinting should have reached all tuples
443 : * returned by table_index_build_scan().
444 : */
445 72 : state->snapshot = RegisterSnapshot(GetTransactionSnapshot());
446 :
447 : /*
448 : * GetTransactionSnapshot() always acquires a new MVCC snapshot in
449 : * READ COMMITTED mode. A new snapshot is guaranteed to have all the
450 : * entries it requires in the index.
451 : *
452 : * We must defend against the possibility that an old xact snapshot
453 : * was returned at higher isolation levels when that snapshot is not
454 : * safe for index scans of the target index. This is possible when
455 : * the snapshot sees tuples that are before the index's indcheckxmin
456 : * horizon. Throwing an error here should be very rare. It doesn't
457 : * seem worth using a secondary snapshot to avoid this.
458 : */
459 72 : if (IsolationUsesXactSnapshot() && rel->rd_index->indcheckxmin &&
460 0 : !TransactionIdPrecedes(HeapTupleHeaderGetXmin(rel->rd_indextuple->t_data),
461 0 : state->snapshot->xmin))
462 0 : ereport(ERROR,
463 : errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
464 : errmsg("index \"%s\" cannot be verified using transaction snapshot",
465 : RelationGetRelationName(rel)));
466 : }
467 :
468 : /*
469 : * We need a snapshot to check the uniqueness of the index. For better
470 : * performance, take it once per index check. If one was already taken
471 : * above, use that.
472 : */
473 4246 : if (state->checkunique)
474 : {
475 756 : state->indexinfo = BuildIndexInfo(state->rel);
476 :
477 756 : if (state->indexinfo->ii_Unique && state->snapshot == InvalidSnapshot)
478 669 : state->snapshot = RegisterSnapshot(GetTransactionSnapshot());
479 : }
480 :
481 : Assert(!state->rootdescend || state->readonly);
482 4246 : if (state->rootdescend && !state->heapkeyspace)
483 0 : ereport(ERROR,
484 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
485 : errmsg("cannot verify that tuples from index \"%s\" can each be found by an independent index search",
486 : RelationGetRelationName(rel)),
487 : errhint("Only B-Tree version 4 indexes support rootdescend verification.")));
488 :
489 : /* Create context for page */
490 4246 : state->targetcontext = AllocSetContextCreate(CurrentMemoryContext,
491 : "amcheck context",
492 : ALLOCSET_DEFAULT_SIZES);
493 4246 : state->checkstrategy = GetAccessStrategy(BAS_BULKREAD);
494 :
495 : /* Get true root block from meta-page */
496 4246 : metapage = palloc_btree_page(state, BTREE_METAPAGE);
497 4246 : metad = BTPageGetMeta(metapage);
498 :
499 : /*
500 : * Certain deletion patterns can result in "skinny" B-Tree indexes, where
501 : * the fast root and true root differ.
502 : *
503 : * Start from the true root, not the fast root, unlike conventional index
504 : * scans. This approach is more thorough, and removes the risk of
505 : * following a stale fast root from the meta page.
506 : */
507 4246 : if (metad->btm_fastroot != metad->btm_root)
508 13 : ereport(DEBUG1,
509 : (errcode(ERRCODE_NO_DATA),
510 : errmsg_internal("harmless fast root mismatch in index \"%s\"",
511 : RelationGetRelationName(rel)),
512 : errdetail_internal("Fast root block %u (level %u) differs from true root block %u (level %u).",
513 : metad->btm_fastroot, metad->btm_fastlevel,
514 : metad->btm_root, metad->btm_level)));
515 :
516 : /*
517 : * Starting at the root, verify every level. Move left to right, top to
518 : * bottom. Note that there may be no pages other than the meta page (meta
519 : * page can indicate that root is P_NONE when the index is totally empty).
520 : */
521 4246 : previouslevel = InvalidBtreeLevel;
522 4246 : current.level = metad->btm_level;
523 4246 : current.leftmost = metad->btm_root;
524 4246 : current.istruerootlevel = true;
525 6736 : while (current.leftmost != P_NONE)
526 : {
527 : /*
528 : * Verify this level, and get left most page for next level down, if
529 : * not at leaf level
530 : */
531 2508 : current = bt_check_level_from_leftmost(state, current);
532 :
533 2490 : if (current.leftmost == InvalidBlockNumber)
534 0 : ereport(ERROR,
535 : (errcode(ERRCODE_INDEX_CORRUPTED),
536 : errmsg("index \"%s\" has no valid pages on level below %u or first level",
537 : RelationGetRelationName(rel), previouslevel)));
538 :
539 2490 : previouslevel = current.level;
540 : }
541 :
542 : /*
543 : * * Check whether heap contains unindexed/malformed tuples *
544 : */
545 4228 : if (state->heapallindexed)
546 : {
547 65 : IndexInfo *indexinfo = BuildIndexInfo(state->rel);
548 : TableScanDesc scan;
549 :
550 : /*
551 : * Create our own scan for table_index_build_scan(), rather than
552 : * getting it to do so for us. This is required so that we can
553 : * actually use the MVCC snapshot registered earlier.
554 : *
555 : * Note that table_index_build_scan() calls heap_endscan() for us.
556 : */
557 65 : scan = table_beginscan_strat(state->heaprel, /* relation */
558 : state->snapshot, /* snapshot */
559 : 0, /* number of keys */
560 : NULL, /* scan key */
561 : true, /* buffer access strategy OK */
562 : true); /* syncscan OK? */
563 :
564 : /*
565 : * Scan will behave as the first scan of a CREATE INDEX CONCURRENTLY
566 : * behaves.
567 : *
568 : * It's okay that we don't actually use the same lock strength for the
569 : * heap relation as any other ii_Concurrent caller would. We have no
570 : * reason to care about a concurrent VACUUM operation, since there
571 : * isn't going to be a second scan of the heap that needs to be sure
572 : * that there was no concurrent recycling of TIDs.
573 : */
574 63 : indexinfo->ii_Concurrent = true;
575 :
576 : /*
577 : * Don't wait for uncommitted tuple xact commit/abort when index is a
578 : * unique index on a catalog (or an index used by an exclusion
579 : * constraint). This could otherwise happen in the readonly case.
580 : */
581 63 : indexinfo->ii_Unique = false;
582 63 : indexinfo->ii_ExclusionOps = NULL;
583 63 : indexinfo->ii_ExclusionProcs = NULL;
584 63 : indexinfo->ii_ExclusionStrats = NULL;
585 :
586 63 : elog(DEBUG1, "verifying that tuples from index \"%s\" are present in \"%s\"",
587 : RelationGetRelationName(state->rel),
588 : RelationGetRelationName(state->heaprel));
589 :
590 63 : table_index_build_scan(state->heaprel, state->rel, indexinfo, true, false,
591 : bt_tuple_present_callback, state, scan);
592 :
593 63 : ereport(DEBUG1,
594 : (errmsg_internal("finished verifying presence of " INT64_FORMAT " tuples from table \"%s\" with bitset %.2f%% set",
595 : state->heaptuplespresent, RelationGetRelationName(heaprel),
596 : 100.0 * bloom_prop_bits_set(state->filter))));
597 :
598 63 : bloom_free(state->filter);
599 : }
600 :
601 : /* Be tidy: */
602 4226 : if (state->snapshot != InvalidSnapshot)
603 732 : UnregisterSnapshot(state->snapshot);
604 4226 : MemoryContextDelete(state->targetcontext);
605 4226 : }
606 :
607 : /*
608 : * Given a left-most block at some level, move right, verifying each page
609 : * individually (with more verification across pages for "readonly"
610 : * callers). Caller should pass the true root page as the leftmost initially,
611 : * working their way down by passing what is returned for the last call here
612 : * until level 0 (leaf page level) was reached.
613 : *
614 : * Returns state for next call, if any. This includes left-most block number
615 : * one level lower that should be passed on next level/call, which is set to
616 : * P_NONE on last call here (when leaf level is verified). Level numbers
617 : * follow the nbtree convention: higher levels have higher numbers, because new
618 : * levels are added only due to a root page split. Note that prior to the
619 : * first root page split, the root is also a leaf page, so there is always a
620 : * level 0 (leaf level), and it's always the last level processed.
621 : *
622 : * Note on memory management: State's per-page context is reset here, between
623 : * each call to bt_target_page_check().
624 : */
625 : static BtreeLevel
626 2508 : bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
627 : {
628 : /* State to establish early, concerning entire level */
629 : BTPageOpaque opaque;
630 : MemoryContext oldcontext;
631 : BtreeLevel nextleveldown;
632 :
633 : /* Variables for iterating across level using right links */
634 2508 : BlockNumber leftcurrent = P_NONE;
635 2508 : BlockNumber current = level.leftmost;
636 :
637 : /* Initialize return state */
638 2508 : nextleveldown.leftmost = InvalidBlockNumber;
639 2508 : nextleveldown.level = InvalidBtreeLevel;
640 2508 : nextleveldown.istruerootlevel = false;
641 :
642 : /* Use page-level context for duration of this call */
643 2508 : oldcontext = MemoryContextSwitchTo(state->targetcontext);
644 :
645 2508 : elog(DEBUG1, "verifying level %u%s", level.level,
646 : level.istruerootlevel ?
647 : " (true root level)" : level.level == 0 ? " (leaf level)" : "");
648 :
649 2508 : state->prevrightlink = InvalidBlockNumber;
650 2508 : state->previncompletesplit = false;
651 :
652 : do
653 : {
654 : /* Don't rely on CHECK_FOR_INTERRUPTS() calls at lower level */
655 9268 : CHECK_FOR_INTERRUPTS();
656 :
657 : /* Initialize state for this iteration */
658 9268 : state->targetblock = current;
659 9268 : state->target = palloc_btree_page(state, state->targetblock);
660 9256 : state->targetlsn = PageGetLSN(state->target);
661 :
662 9256 : opaque = BTPageGetOpaque(state->target);
663 :
664 9256 : if (P_IGNORE(opaque))
665 : {
666 : /*
667 : * Since there cannot be a concurrent VACUUM operation in readonly
668 : * mode, and since a page has no links within other pages
669 : * (siblings and parent) once it is marked fully deleted, it
670 : * should be impossible to land on a fully deleted page in
671 : * readonly mode. See bt_child_check() for further details.
672 : *
673 : * The bt_child_check() P_ISDELETED() check is repeated here so
674 : * that pages that are only reachable through sibling links get
675 : * checked.
676 : */
677 0 : if (state->readonly && P_ISDELETED(opaque))
678 0 : ereport(ERROR,
679 : (errcode(ERRCODE_INDEX_CORRUPTED),
680 : errmsg("downlink or sibling link points to deleted block in index \"%s\"",
681 : RelationGetRelationName(state->rel)),
682 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
683 : current, leftcurrent, opaque->btpo_prev)));
684 :
685 0 : if (P_RIGHTMOST(opaque))
686 0 : ereport(ERROR,
687 : (errcode(ERRCODE_INDEX_CORRUPTED),
688 : errmsg("block %u fell off the end of index \"%s\"",
689 : current, RelationGetRelationName(state->rel))));
690 : else
691 0 : ereport(DEBUG1,
692 : (errcode(ERRCODE_NO_DATA),
693 : errmsg_internal("block %u of index \"%s\" concurrently deleted",
694 : current, RelationGetRelationName(state->rel))));
695 0 : goto nextpage;
696 : }
697 9256 : else if (nextleveldown.leftmost == InvalidBlockNumber)
698 : {
699 : /*
700 : * A concurrent page split could make the caller supplied leftmost
701 : * block no longer contain the leftmost page, or no longer be the
702 : * true root, but where that isn't possible due to heavyweight
703 : * locking, check that the first valid page meets caller's
704 : * expectations.
705 : */
706 2496 : if (state->readonly)
707 : {
708 45 : if (!bt_leftmost_ignoring_half_dead(state, current, opaque))
709 0 : ereport(ERROR,
710 : (errcode(ERRCODE_INDEX_CORRUPTED),
711 : errmsg("block %u is not leftmost in index \"%s\"",
712 : current, RelationGetRelationName(state->rel))));
713 :
714 45 : if (level.istruerootlevel && (!P_ISROOT(opaque) && !P_INCOMPLETE_SPLIT(opaque)))
715 0 : ereport(ERROR,
716 : (errcode(ERRCODE_INDEX_CORRUPTED),
717 : errmsg("block %u is not true root in index \"%s\"",
718 : current, RelationGetRelationName(state->rel))));
719 : }
720 :
721 : /*
722 : * Before beginning any non-trivial examination of level, prepare
723 : * state for next bt_check_level_from_leftmost() invocation for
724 : * the next level for the next level down (if any).
725 : *
726 : * There should be at least one non-ignorable page per level,
727 : * unless this is the leaf level, which is assumed by caller to be
728 : * final level.
729 : */
730 2496 : if (!P_ISLEAF(opaque))
731 : {
732 : IndexTuple itup;
733 : ItemId itemid;
734 :
735 : /* Internal page -- downlink gets leftmost on next level */
736 555 : itemid = PageGetItemIdCareful(state, state->targetblock,
737 : state->target,
738 555 : P_FIRSTDATAKEY(opaque));
739 555 : itup = (IndexTuple) PageGetItem(state->target, itemid);
740 555 : nextleveldown.leftmost = BTreeTupleGetDownLink(itup);
741 555 : nextleveldown.level = opaque->btpo_level - 1;
742 : }
743 : else
744 : {
745 : /*
746 : * Leaf page -- final level caller must process.
747 : *
748 : * Note that this could also be the root page, if there has
749 : * been no root page split yet.
750 : */
751 1941 : nextleveldown.leftmost = P_NONE;
752 1941 : nextleveldown.level = InvalidBtreeLevel;
753 : }
754 :
755 : /*
756 : * Finished setting up state for this call/level. Control will
757 : * never end up back here in any future loop iteration for this
758 : * level.
759 : */
760 : }
761 :
762 : /*
763 : * Sibling links should be in mutual agreement. There arises
764 : * leftcurrent == P_NONE && btpo_prev != P_NONE when the left sibling
765 : * of the parent's low-key downlink is half-dead. (A half-dead page
766 : * has no downlink from its parent.) Under heavyweight locking, the
767 : * last bt_leftmost_ignoring_half_dead() validated this btpo_prev.
768 : * Without heavyweight locking, validation of the P_NONE case remains
769 : * unimplemented.
770 : */
771 9256 : if (opaque->btpo_prev != leftcurrent && leftcurrent != P_NONE)
772 0 : bt_recheck_sibling_links(state, opaque->btpo_prev, leftcurrent);
773 :
774 : /* Check level */
775 9256 : if (level.level != opaque->btpo_level)
776 0 : ereport(ERROR,
777 : (errcode(ERRCODE_INDEX_CORRUPTED),
778 : errmsg("leftmost down link for level points to block in index \"%s\" whose level is not one level down",
779 : RelationGetRelationName(state->rel)),
780 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
781 : current, level.level, opaque->btpo_level)));
782 :
783 : /* Verify invariants for page */
784 9256 : bt_target_page_check(state);
785 :
786 9250 : nextpage:
787 :
788 : /* Try to detect circular links */
789 9250 : if (current == leftcurrent || current == opaque->btpo_prev)
790 0 : ereport(ERROR,
791 : (errcode(ERRCODE_INDEX_CORRUPTED),
792 : errmsg("circular link chain found in block %u of index \"%s\"",
793 : current, RelationGetRelationName(state->rel))));
794 :
795 9250 : leftcurrent = current;
796 9250 : current = opaque->btpo_next;
797 :
798 9250 : if (state->lowkey)
799 : {
800 : Assert(state->readonly);
801 1862 : pfree(state->lowkey);
802 1862 : state->lowkey = NULL;
803 : }
804 :
805 : /*
806 : * Copy current target high key as the low key of right sibling.
807 : * Allocate memory in upper level context, so it would be cleared
808 : * after reset of target context.
809 : *
810 : * We only need the low key in corner cases of checking child high
811 : * keys. We use high key only when incomplete split on the child level
812 : * falls to the boundary of pages on the target level. See
813 : * bt_child_highkey_check() for details. So, typically we won't end
814 : * up doing anything with low key, but it's simpler for general case
815 : * high key verification to always have it available.
816 : *
817 : * The correctness of managing low key in the case of concurrent
818 : * splits wasn't investigated yet. Thankfully we only need low key
819 : * for readonly verification and concurrent splits won't happen.
820 : */
821 9250 : if (state->readonly && !P_RIGHTMOST(opaque))
822 : {
823 : IndexTuple itup;
824 : ItemId itemid;
825 :
826 1862 : itemid = PageGetItemIdCareful(state, state->targetblock,
827 : state->target, P_HIKEY);
828 1862 : itup = (IndexTuple) PageGetItem(state->target, itemid);
829 :
830 1862 : state->lowkey = MemoryContextAlloc(oldcontext, IndexTupleSize(itup));
831 1862 : memcpy(state->lowkey, itup, IndexTupleSize(itup));
832 : }
833 :
834 : /* Free page and associated memory for this iteration */
835 9250 : MemoryContextReset(state->targetcontext);
836 : }
837 9250 : while (current != P_NONE);
838 :
839 2490 : if (state->lowkey)
840 : {
841 : Assert(state->readonly);
842 0 : pfree(state->lowkey);
843 0 : state->lowkey = NULL;
844 : }
845 :
846 : /* Don't change context for caller */
847 2490 : MemoryContextSwitchTo(oldcontext);
848 :
849 2490 : return nextleveldown;
850 : }
851 :
852 : /* Check visibility of the table entry referenced by nbtree index */
853 : static bool
854 306 : heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid)
855 : {
856 : bool tid_visible;
857 :
858 306 : TupleTableSlot *slot = table_slot_create(state->heaprel, NULL);
859 :
860 306 : tid_visible = table_tuple_fetch_row_version(state->heaprel,
861 : tid, state->snapshot, slot);
862 306 : if (slot != NULL)
863 306 : ExecDropSingleTupleTableSlot(slot);
864 :
865 306 : return tid_visible;
866 : }
867 :
868 : /*
869 : * Prepare an error message for unique constraint violation in
870 : * a btree index and report ERROR.
871 : */
872 : static void
873 3 : bt_report_duplicate(BtreeCheckState *state,
874 : BtreeLastVisibleEntry *lVis,
875 : ItemPointer nexttid, BlockNumber nblock, OffsetNumber noffset,
876 : int nposting)
877 : {
878 : char *htid,
879 : *nhtid,
880 : *itid,
881 3 : *nitid = "",
882 3 : *pposting = "",
883 3 : *pnposting = "";
884 :
885 3 : htid = psprintf("tid=(%u,%u)",
886 3 : ItemPointerGetBlockNumberNoCheck(lVis->tid),
887 3 : ItemPointerGetOffsetNumberNoCheck(lVis->tid));
888 3 : nhtid = psprintf("tid=(%u,%u)",
889 : ItemPointerGetBlockNumberNoCheck(nexttid),
890 3 : ItemPointerGetOffsetNumberNoCheck(nexttid));
891 3 : itid = psprintf("tid=(%u,%u)", lVis->blkno, lVis->offset);
892 :
893 3 : if (nblock != lVis->blkno || noffset != lVis->offset)
894 3 : nitid = psprintf(" tid=(%u,%u)", nblock, noffset);
895 :
896 3 : if (lVis->postingIndex >= 0)
897 0 : pposting = psprintf(" posting %u", lVis->postingIndex);
898 :
899 3 : if (nposting >= 0)
900 0 : pnposting = psprintf(" posting %u", nposting);
901 :
902 3 : ereport(ERROR,
903 : (errcode(ERRCODE_INDEX_CORRUPTED),
904 : errmsg("index uniqueness is violated for index \"%s\"",
905 : RelationGetRelationName(state->rel)),
906 : errdetail("Index %s%s and%s%s (point to heap %s and %s) page lsn=%X/%08X.",
907 : itid, pposting, nitid, pnposting, htid, nhtid,
908 : LSN_FORMAT_ARGS(state->targetlsn))));
909 : }
910 :
911 : /* Check if current nbtree leaf entry complies with UNIQUE constraint */
912 : static void
913 282 : bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
914 : BlockNumber targetblock, OffsetNumber offset,
915 : BtreeLastVisibleEntry *lVis)
916 : {
917 : ItemPointer tid;
918 282 : bool has_visible_entry = false;
919 :
920 : Assert(targetblock != P_NONE);
921 :
922 : /*
923 : * Current tuple has posting list. Report duplicate if TID of any posting
924 : * list entry is visible and lVis->tid is valid.
925 : */
926 282 : if (BTreeTupleIsPosting(itup))
927 : {
928 66 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
929 : {
930 45 : tid = BTreeTupleGetPostingN(itup, i);
931 45 : if (heap_entry_is_visible(state, tid))
932 : {
933 9 : has_visible_entry = true;
934 9 : if (ItemPointerIsValid(lVis->tid))
935 : {
936 0 : bt_report_duplicate(state,
937 : lVis,
938 : tid, targetblock,
939 : offset, i);
940 : }
941 :
942 : /*
943 : * Prevent double reporting unique constraint violation
944 : * between the posting list entries of the first tuple on the
945 : * page after cross-page check.
946 : */
947 9 : if (lVis->blkno != targetblock && ItemPointerIsValid(lVis->tid))
948 0 : return;
949 :
950 9 : lVis->blkno = targetblock;
951 9 : lVis->offset = offset;
952 9 : lVis->postingIndex = i;
953 9 : lVis->tid = tid;
954 : }
955 : }
956 : }
957 :
958 : /*
959 : * Current tuple has no posting list. If TID is visible save info about it
960 : * for the next comparisons in the loop in bt_target_page_check(). Report
961 : * duplicate if lVis->tid is already valid.
962 : */
963 : else
964 : {
965 261 : tid = BTreeTupleGetHeapTID(itup);
966 261 : if (heap_entry_is_visible(state, tid))
967 : {
968 18 : has_visible_entry = true;
969 18 : if (ItemPointerIsValid(lVis->tid))
970 : {
971 3 : bt_report_duplicate(state,
972 : lVis,
973 : tid, targetblock,
974 : offset, -1);
975 : }
976 :
977 15 : lVis->blkno = targetblock;
978 15 : lVis->offset = offset;
979 15 : lVis->tid = tid;
980 15 : lVis->postingIndex = -1;
981 : }
982 : }
983 :
984 279 : if (!has_visible_entry &&
985 255 : lVis->blkno != InvalidBlockNumber &&
986 15 : lVis->blkno != targetblock)
987 : {
988 0 : char *posting = "";
989 :
990 0 : if (lVis->postingIndex >= 0)
991 0 : posting = psprintf(" posting %u", lVis->postingIndex);
992 0 : ereport(DEBUG1,
993 : (errcode(ERRCODE_NO_DATA),
994 : errmsg("index uniqueness can not be checked for index tid=(%u,%u) in index \"%s\"",
995 : targetblock, offset,
996 : RelationGetRelationName(state->rel)),
997 : errdetail("It doesn't have visible heap tids and key is equal to the tid=(%u,%u)%s (points to heap tid=(%u,%u)).",
998 : lVis->blkno, lVis->offset, posting,
999 : ItemPointerGetBlockNumberNoCheck(lVis->tid),
1000 : ItemPointerGetOffsetNumberNoCheck(lVis->tid)),
1001 : errhint("VACUUM the table and repeat the check.")));
1002 : }
1003 : }
1004 :
1005 : /*
1006 : * Like P_LEFTMOST(start_opaque), but accept an arbitrarily-long chain of
1007 : * half-dead, sibling-linked pages to the left. If a half-dead page appears
1008 : * under state->readonly, the database exited recovery between the first-stage
1009 : * and second-stage WAL records of a deletion.
1010 : */
1011 : static bool
1012 56 : bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
1013 : BlockNumber start,
1014 : BTPageOpaque start_opaque)
1015 : {
1016 56 : BlockNumber reached = start_opaque->btpo_prev,
1017 56 : reached_from = start;
1018 56 : bool all_half_dead = true;
1019 :
1020 : /*
1021 : * To handle the !readonly case, we'd need to accept BTP_DELETED pages and
1022 : * potentially observe nbtree/README "Page deletion and backwards scans".
1023 : */
1024 : Assert(state->readonly);
1025 :
1026 58 : while (reached != P_NONE && all_half_dead)
1027 : {
1028 2 : Page page = palloc_btree_page(state, reached);
1029 2 : BTPageOpaque reached_opaque = BTPageGetOpaque(page);
1030 :
1031 2 : CHECK_FOR_INTERRUPTS();
1032 :
1033 : /*
1034 : * Try to detect btpo_prev circular links. _bt_unlink_halfdead_page()
1035 : * writes that side-links will continue to point to the siblings.
1036 : * Check btpo_next for that property.
1037 : */
1038 2 : all_half_dead = P_ISHALFDEAD(reached_opaque) &&
1039 2 : reached != start &&
1040 4 : reached != reached_from &&
1041 2 : reached_opaque->btpo_next == reached_from;
1042 2 : if (all_half_dead)
1043 : {
1044 2 : XLogRecPtr pagelsn = PageGetLSN(page);
1045 :
1046 : /* pagelsn should point to an XLOG_BTREE_MARK_PAGE_HALFDEAD */
1047 2 : ereport(DEBUG1,
1048 : (errcode(ERRCODE_NO_DATA),
1049 : errmsg_internal("harmless interrupted page deletion detected in index \"%s\"",
1050 : RelationGetRelationName(state->rel)),
1051 : errdetail_internal("Block=%u right block=%u page lsn=%X/%08X.",
1052 : reached, reached_from,
1053 : LSN_FORMAT_ARGS(pagelsn))));
1054 :
1055 2 : reached_from = reached;
1056 2 : reached = reached_opaque->btpo_prev;
1057 : }
1058 :
1059 2 : pfree(page);
1060 : }
1061 :
1062 56 : return all_half_dead;
1063 : }
1064 :
1065 : /*
1066 : * Raise an error when target page's left link does not point back to the
1067 : * previous target page, called leftcurrent here. The leftcurrent page's
1068 : * right link was followed to get to the current target page, and we expect
1069 : * mutual agreement among leftcurrent and the current target page. Make sure
1070 : * that this condition has definitely been violated in the !readonly case,
1071 : * where concurrent page splits are something that we need to deal with.
1072 : *
1073 : * Cross-page inconsistencies involving pages that don't agree about being
1074 : * siblings are known to be a particularly good indicator of corruption
1075 : * involving partial writes/lost updates. The bt_right_page_check_scankey
1076 : * check also provides a way of detecting cross-page inconsistencies for
1077 : * !readonly callers, but it can only detect sibling pages that have an
1078 : * out-of-order keyspace, which can't catch many of the problems that we
1079 : * expect to catch here.
1080 : *
1081 : * The classic example of the kind of inconsistency that we can only catch
1082 : * with this check (when in !readonly mode) involves three sibling pages that
1083 : * were affected by a faulty page split at some point in the past. The
1084 : * effects of the split are reflected in the original page and its new right
1085 : * sibling page, with a lack of any accompanying changes for the _original_
1086 : * right sibling page. The original right sibling page's left link fails to
1087 : * point to the new right sibling page (its left link still points to the
1088 : * original page), even though the first phase of a page split is supposed to
1089 : * work as a single atomic action. This subtle inconsistency will probably
1090 : * only break backwards scans in practice.
1091 : *
1092 : * Note that this is the only place where amcheck will "couple" buffer locks
1093 : * (and only for !readonly callers). In general we prefer to avoid more
1094 : * thorough cross-page checks in !readonly mode, but it seems worth the
1095 : * complexity here. Also, the performance overhead of performing lock
1096 : * coupling here is negligible in practice. Control only reaches here with a
1097 : * non-corrupt index when there is a concurrent page split at the instant
1098 : * caller crossed over to target page from leftcurrent page.
1099 : */
1100 : static void
1101 0 : bt_recheck_sibling_links(BtreeCheckState *state,
1102 : BlockNumber btpo_prev_from_target,
1103 : BlockNumber leftcurrent)
1104 : {
1105 : /* passing metapage to BTPageGetOpaque() would give irrelevant findings */
1106 : Assert(leftcurrent != P_NONE);
1107 :
1108 0 : if (!state->readonly)
1109 : {
1110 : Buffer lbuf;
1111 : Buffer newtargetbuf;
1112 : Page page;
1113 : BTPageOpaque opaque;
1114 : BlockNumber newtargetblock;
1115 :
1116 : /* Couple locks in the usual order for nbtree: Left to right */
1117 0 : lbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM, leftcurrent,
1118 : RBM_NORMAL, state->checkstrategy);
1119 0 : LockBuffer(lbuf, BT_READ);
1120 0 : _bt_checkpage(state->rel, lbuf);
1121 0 : page = BufferGetPage(lbuf);
1122 0 : opaque = BTPageGetOpaque(page);
1123 0 : if (P_ISDELETED(opaque))
1124 : {
1125 : /*
1126 : * Cannot reason about concurrently deleted page -- the left link
1127 : * in the page to the right is expected to point to some other
1128 : * page to the left (not leftcurrent page).
1129 : *
1130 : * Note that we deliberately don't give up with a half-dead page.
1131 : */
1132 0 : UnlockReleaseBuffer(lbuf);
1133 0 : return;
1134 : }
1135 :
1136 0 : newtargetblock = opaque->btpo_next;
1137 : /* Avoid self-deadlock when newtargetblock == leftcurrent */
1138 0 : if (newtargetblock != leftcurrent)
1139 : {
1140 0 : newtargetbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM,
1141 : newtargetblock, RBM_NORMAL,
1142 : state->checkstrategy);
1143 0 : LockBuffer(newtargetbuf, BT_READ);
1144 0 : _bt_checkpage(state->rel, newtargetbuf);
1145 0 : page = BufferGetPage(newtargetbuf);
1146 0 : opaque = BTPageGetOpaque(page);
1147 : /* btpo_prev_from_target may have changed; update it */
1148 0 : btpo_prev_from_target = opaque->btpo_prev;
1149 : }
1150 : else
1151 : {
1152 : /*
1153 : * leftcurrent right sibling points back to leftcurrent block.
1154 : * Index is corrupt. Easiest way to handle this is to pretend
1155 : * that we actually read from a distinct page that has an invalid
1156 : * block number in its btpo_prev.
1157 : */
1158 0 : newtargetbuf = InvalidBuffer;
1159 0 : btpo_prev_from_target = InvalidBlockNumber;
1160 : }
1161 :
1162 : /*
1163 : * No need to check P_ISDELETED here, since new target block cannot be
1164 : * marked deleted as long as we hold a lock on lbuf
1165 : */
1166 0 : if (BufferIsValid(newtargetbuf))
1167 0 : UnlockReleaseBuffer(newtargetbuf);
1168 0 : UnlockReleaseBuffer(lbuf);
1169 :
1170 0 : if (btpo_prev_from_target == leftcurrent)
1171 : {
1172 : /* Report split in left sibling, not target (or new target) */
1173 0 : ereport(DEBUG1,
1174 : (errcode(ERRCODE_INTERNAL_ERROR),
1175 : errmsg_internal("harmless concurrent page split detected in index \"%s\"",
1176 : RelationGetRelationName(state->rel)),
1177 : errdetail_internal("Block=%u new right sibling=%u original right sibling=%u.",
1178 : leftcurrent, newtargetblock,
1179 : state->targetblock)));
1180 0 : return;
1181 : }
1182 :
1183 : /*
1184 : * Index is corrupt. Make sure that we report correct target page.
1185 : *
1186 : * This could have changed in cases where there was a concurrent page
1187 : * split, as well as index corruption (at least in theory). Note that
1188 : * btpo_prev_from_target was already updated above.
1189 : */
1190 0 : state->targetblock = newtargetblock;
1191 : }
1192 :
1193 0 : ereport(ERROR,
1194 : (errcode(ERRCODE_INDEX_CORRUPTED),
1195 : errmsg("left link/right link pair in index \"%s\" not in agreement",
1196 : RelationGetRelationName(state->rel)),
1197 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
1198 : state->targetblock, leftcurrent,
1199 : btpo_prev_from_target)));
1200 : }
1201 :
1202 : /*
1203 : * Function performs the following checks on target page, or pages ancillary to
1204 : * target page:
1205 : *
1206 : * - That every "real" data item is less than or equal to the high key, which
1207 : * is an upper bound on the items on the page. Data items should be
1208 : * strictly less than the high key when the page is an internal page.
1209 : *
1210 : * - That within the page, every data item is strictly less than the item
1211 : * immediately to its right, if any (i.e., that the items are in order
1212 : * within the page, so that the binary searches performed by index scans are
1213 : * sane).
1214 : *
1215 : * - That the last data item stored on the page is strictly less than the
1216 : * first data item on the page to the right (when such a first item is
1217 : * available).
1218 : *
1219 : * - Various checks on the structure of tuples themselves. For example, check
1220 : * that non-pivot tuples have no truncated attributes.
1221 : *
1222 : * - For index with unique constraint make sure that only one of table entries
1223 : * for equal keys is visible.
1224 : *
1225 : * Furthermore, when state passed shows ShareLock held, function also checks:
1226 : *
1227 : * - That all child pages respect strict lower bound from parent's pivot
1228 : * tuple.
1229 : *
1230 : * - That downlink to block was encountered in parent where that's expected.
1231 : *
1232 : * - That high keys of child pages matches corresponding pivot keys in parent.
1233 : *
1234 : * This is also where heapallindexed callers use their Bloom filter to
1235 : * fingerprint IndexTuples for later table_index_build_scan() verification.
1236 : *
1237 : * Note: Memory allocated in this routine is expected to be released by caller
1238 : * resetting state->targetcontext.
1239 : */
1240 : static void
1241 9256 : bt_target_page_check(BtreeCheckState *state)
1242 : {
1243 : OffsetNumber offset;
1244 : OffsetNumber max;
1245 : BTPageOpaque topaque;
1246 :
1247 : /* Last visible entry info for checking indexes with unique constraint */
1248 9256 : BtreeLastVisibleEntry lVis = {InvalidBlockNumber, InvalidOffsetNumber, -1, NULL};
1249 :
1250 9256 : topaque = BTPageGetOpaque(state->target);
1251 9256 : max = PageGetMaxOffsetNumber(state->target);
1252 :
1253 9256 : elog(DEBUG2, "verifying %u items on %s block %u", max,
1254 : P_ISLEAF(topaque) ? "leaf" : "internal", state->targetblock);
1255 :
1256 : /*
1257 : * Check the number of attributes in high key. Note, rightmost page
1258 : * doesn't contain a high key, so nothing to check
1259 : */
1260 9256 : if (!P_RIGHTMOST(topaque))
1261 : {
1262 : ItemId itemid;
1263 : IndexTuple itup;
1264 :
1265 : /* Verify line pointer before checking tuple */
1266 6764 : itemid = PageGetItemIdCareful(state, state->targetblock,
1267 : state->target, P_HIKEY);
1268 6764 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1269 : P_HIKEY))
1270 : {
1271 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1272 0 : ereport(ERROR,
1273 : (errcode(ERRCODE_INDEX_CORRUPTED),
1274 : errmsg("wrong number of high key index tuple attributes in index \"%s\"",
1275 : RelationGetRelationName(state->rel)),
1276 : errdetail_internal("Index block=%u natts=%u block type=%s page lsn=%X/%08X.",
1277 : state->targetblock,
1278 : BTreeTupleGetNAtts(itup, state->rel),
1279 : P_ISLEAF(topaque) ? "heap" : "index",
1280 : LSN_FORMAT_ARGS(state->targetlsn))));
1281 : }
1282 : }
1283 :
1284 : /*
1285 : * Loop over page items, starting from first non-highkey item, not high
1286 : * key (if any). Most tests are not performed for the "negative infinity"
1287 : * real item (if any).
1288 : */
1289 9256 : for (offset = P_FIRSTDATAKEY(topaque);
1290 2058890 : offset <= max;
1291 2049634 : offset = OffsetNumberNext(offset))
1292 : {
1293 : ItemId itemid;
1294 : IndexTuple itup;
1295 : size_t tupsize;
1296 : BTScanInsert skey;
1297 : bool lowersizelimit;
1298 : ItemPointer scantid;
1299 :
1300 : /*
1301 : * True if we already called bt_entry_unique_check() for the current
1302 : * item. This helps to avoid visiting the heap for keys, which are
1303 : * anyway presented only once and can't comprise a unique violation.
1304 : */
1305 2049640 : bool unique_checked = false;
1306 :
1307 2049640 : CHECK_FOR_INTERRUPTS();
1308 :
1309 2049640 : itemid = PageGetItemIdCareful(state, state->targetblock,
1310 : state->target, offset);
1311 2049640 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1312 2049640 : tupsize = IndexTupleSize(itup);
1313 :
1314 : /*
1315 : * lp_len should match the IndexTuple reported length exactly, since
1316 : * lp_len is completely redundant in indexes, and both sources of
1317 : * tuple length are MAXALIGN()'d. nbtree does not use lp_len all that
1318 : * frequently, and is surprisingly tolerant of corrupt lp_len fields.
1319 : */
1320 2049640 : if (tupsize != ItemIdGetLength(itemid))
1321 0 : ereport(ERROR,
1322 : (errcode(ERRCODE_INDEX_CORRUPTED),
1323 : errmsg("index tuple size does not equal lp_len in index \"%s\"",
1324 : RelationGetRelationName(state->rel)),
1325 : errdetail_internal("Index tid=(%u,%u) tuple size=%zu lp_len=%u page lsn=%X/%08X.",
1326 : state->targetblock, offset,
1327 : tupsize, ItemIdGetLength(itemid),
1328 : LSN_FORMAT_ARGS(state->targetlsn)),
1329 : errhint("This could be a torn page problem.")));
1330 :
1331 : /* Check the number of index tuple attributes */
1332 2049640 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1333 : offset))
1334 : {
1335 : ItemPointer tid;
1336 : char *itid,
1337 : *htid;
1338 :
1339 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1340 0 : tid = BTreeTupleGetPointsToTID(itup);
1341 0 : htid = psprintf("(%u,%u)",
1342 : ItemPointerGetBlockNumberNoCheck(tid),
1343 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1344 :
1345 0 : ereport(ERROR,
1346 : (errcode(ERRCODE_INDEX_CORRUPTED),
1347 : errmsg("wrong number of index tuple attributes in index \"%s\"",
1348 : RelationGetRelationName(state->rel)),
1349 : errdetail_internal("Index tid=%s natts=%u points to %s tid=%s page lsn=%X/%08X.",
1350 : itid,
1351 : BTreeTupleGetNAtts(itup, state->rel),
1352 : P_ISLEAF(topaque) ? "heap" : "index",
1353 : htid,
1354 : LSN_FORMAT_ARGS(state->targetlsn))));
1355 : }
1356 :
1357 : /*
1358 : * Don't try to generate scankey using "negative infinity" item on
1359 : * internal pages. They are always truncated to zero attributes.
1360 : */
1361 2049640 : if (offset_is_negative_infinity(topaque, offset))
1362 : {
1363 : /*
1364 : * We don't call bt_child_check() for "negative infinity" items.
1365 : * But if we're performing downlink connectivity check, we do it
1366 : * for every item including "negative infinity" one.
1367 : */
1368 557 : if (!P_ISLEAF(topaque) && state->readonly)
1369 : {
1370 12 : bt_child_highkey_check(state,
1371 : offset,
1372 : NULL,
1373 : topaque->btpo_level);
1374 : }
1375 557 : continue;
1376 : }
1377 :
1378 : /*
1379 : * Readonly callers may optionally verify that non-pivot tuples can
1380 : * each be found by an independent search that starts from the root.
1381 : * Note that we deliberately don't do individual searches for each
1382 : * TID, since the posting list itself is validated by other checks.
1383 : */
1384 2049083 : if (state->rootdescend && P_ISLEAF(topaque) &&
1385 201098 : !bt_rootdescend(state, itup))
1386 : {
1387 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1388 : char *itid,
1389 : *htid;
1390 :
1391 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1392 0 : htid = psprintf("(%u,%u)", ItemPointerGetBlockNumber(tid),
1393 0 : ItemPointerGetOffsetNumber(tid));
1394 :
1395 0 : ereport(ERROR,
1396 : (errcode(ERRCODE_INDEX_CORRUPTED),
1397 : errmsg("could not find tuple using search from root page in index \"%s\"",
1398 : RelationGetRelationName(state->rel)),
1399 : errdetail_internal("Index tid=%s points to heap tid=%s page lsn=%X/%08X.",
1400 : itid, htid,
1401 : LSN_FORMAT_ARGS(state->targetlsn))));
1402 : }
1403 :
1404 : /*
1405 : * If tuple is a posting list tuple, make sure posting list TIDs are
1406 : * in order
1407 : */
1408 2049083 : if (BTreeTupleIsPosting(itup))
1409 : {
1410 : ItemPointerData last;
1411 : ItemPointer current;
1412 :
1413 11800 : ItemPointerCopy(BTreeTupleGetHeapTID(itup), &last);
1414 :
1415 86295 : for (int i = 1; i < BTreeTupleGetNPosting(itup); i++)
1416 : {
1417 :
1418 74495 : current = BTreeTupleGetPostingN(itup, i);
1419 :
1420 74495 : if (ItemPointerCompare(current, &last) <= 0)
1421 : {
1422 0 : char *itid = psprintf("(%u,%u)", state->targetblock, offset);
1423 :
1424 0 : ereport(ERROR,
1425 : (errcode(ERRCODE_INDEX_CORRUPTED),
1426 : errmsg_internal("posting list contains misplaced TID in index \"%s\"",
1427 : RelationGetRelationName(state->rel)),
1428 : errdetail_internal("Index tid=%s posting list offset=%d page lsn=%X/%08X.",
1429 : itid, i,
1430 : LSN_FORMAT_ARGS(state->targetlsn))));
1431 : }
1432 :
1433 74495 : ItemPointerCopy(current, &last);
1434 : }
1435 : }
1436 :
1437 : /* Build insertion scankey for current page offset */
1438 2049083 : skey = bt_mkscankey_pivotsearch(state->rel, itup);
1439 :
1440 : /*
1441 : * Make sure tuple size does not exceed the relevant BTREE_VERSION
1442 : * specific limit.
1443 : *
1444 : * BTREE_VERSION 4 (which introduced heapkeyspace rules) requisitioned
1445 : * a small amount of space from BTMaxItemSize() in order to ensure
1446 : * that suffix truncation always has enough space to add an explicit
1447 : * heap TID back to a tuple -- we pessimistically assume that every
1448 : * newly inserted tuple will eventually need to have a heap TID
1449 : * appended during a future leaf page split, when the tuple becomes
1450 : * the basis of the new high key (pivot tuple) for the leaf page.
1451 : *
1452 : * Since the reclaimed space is reserved for that purpose, we must not
1453 : * enforce the slightly lower limit when the extra space has been used
1454 : * as intended. In other words, there is only a cross-version
1455 : * difference in the limit on tuple size within leaf pages.
1456 : *
1457 : * Still, we're particular about the details within BTREE_VERSION 4
1458 : * internal pages. Pivot tuples may only use the extra space for its
1459 : * designated purpose. Enforce the lower limit for pivot tuples when
1460 : * an explicit heap TID isn't actually present. (In all other cases
1461 : * suffix truncation is guaranteed to generate a pivot tuple that's no
1462 : * larger than the firstright tuple provided to it by its caller.)
1463 : */
1464 4098166 : lowersizelimit = skey->heapkeyspace &&
1465 2049083 : (P_ISLEAF(topaque) || BTreeTupleGetHeapTID(itup) == NULL);
1466 2049083 : if (tupsize > (lowersizelimit ? BTMaxItemSize : BTMaxItemSizeNoHeapTid))
1467 : {
1468 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1469 : char *itid,
1470 : *htid;
1471 :
1472 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1473 0 : htid = psprintf("(%u,%u)",
1474 : ItemPointerGetBlockNumberNoCheck(tid),
1475 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1476 :
1477 0 : ereport(ERROR,
1478 : (errcode(ERRCODE_INDEX_CORRUPTED),
1479 : errmsg("index row size %zu exceeds maximum for index \"%s\"",
1480 : tupsize, RelationGetRelationName(state->rel)),
1481 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%08X.",
1482 : itid,
1483 : P_ISLEAF(topaque) ? "heap" : "index",
1484 : htid,
1485 : LSN_FORMAT_ARGS(state->targetlsn))));
1486 : }
1487 :
1488 : /* Fingerprint leaf page tuples (those that point to the heap) */
1489 2049083 : if (state->heapallindexed && P_ISLEAF(topaque) && !ItemIdIsDead(itemid))
1490 : {
1491 : IndexTuple norm;
1492 :
1493 506986 : if (BTreeTupleIsPosting(itup))
1494 : {
1495 : /* Fingerprint all elements as distinct "plain" tuples */
1496 27584 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
1497 : {
1498 : IndexTuple logtuple;
1499 :
1500 27390 : logtuple = bt_posting_plain_tuple(itup, i);
1501 27390 : norm = bt_normalize_tuple(state, logtuple);
1502 27390 : bloom_add_element(state->filter, (unsigned char *) norm,
1503 : IndexTupleSize(norm));
1504 : /* Be tidy */
1505 27390 : if (norm != logtuple)
1506 0 : pfree(norm);
1507 27390 : pfree(logtuple);
1508 : }
1509 : }
1510 : else
1511 : {
1512 506792 : norm = bt_normalize_tuple(state, itup);
1513 506792 : bloom_add_element(state->filter, (unsigned char *) norm,
1514 : IndexTupleSize(norm));
1515 : /* Be tidy */
1516 506792 : if (norm != itup)
1517 1 : pfree(norm);
1518 : }
1519 : }
1520 :
1521 : /*
1522 : * * High key check *
1523 : *
1524 : * If there is a high key (if this is not the rightmost page on its
1525 : * entire level), check that high key actually is upper bound on all
1526 : * page items. If this is a posting list tuple, we'll need to set
1527 : * scantid to be highest TID in posting list.
1528 : *
1529 : * We prefer to check all items against high key rather than checking
1530 : * just the last and trusting that the operator class obeys the
1531 : * transitive law (which implies that all previous items also
1532 : * respected the high key invariant if they pass the item order
1533 : * check).
1534 : *
1535 : * Ideally, we'd compare every item in the index against every other
1536 : * item in the index, and not trust opclass obedience of the
1537 : * transitive law to bridge the gap between children and their
1538 : * grandparents (as well as great-grandparents, and so on). We don't
1539 : * go to those lengths because that would be prohibitively expensive,
1540 : * and probably not markedly more effective in practice.
1541 : *
1542 : * On the leaf level, we check that the key is <= the highkey.
1543 : * However, on non-leaf levels we check that the key is < the highkey,
1544 : * because the high key is "just another separator" rather than a copy
1545 : * of some existing key item; we expect it to be unique among all keys
1546 : * on the same level. (Suffix truncation will sometimes produce a
1547 : * leaf highkey that is an untruncated copy of the lastleft item, but
1548 : * never any other item, which necessitates weakening the leaf level
1549 : * check to <=.)
1550 : *
1551 : * Full explanation for why a highkey is never truly a copy of another
1552 : * item from the same level on internal levels:
1553 : *
1554 : * While the new left page's high key is copied from the first offset
1555 : * on the right page during an internal page split, that's not the
1556 : * full story. In effect, internal pages are split in the middle of
1557 : * the firstright tuple, not between the would-be lastleft and
1558 : * firstright tuples: the firstright key ends up on the left side as
1559 : * left's new highkey, and the firstright downlink ends up on the
1560 : * right side as right's new "negative infinity" item. The negative
1561 : * infinity tuple is truncated to zero attributes, so we're only left
1562 : * with the downlink. In other words, the copying is just an
1563 : * implementation detail of splitting in the middle of a (pivot)
1564 : * tuple. (See also: "Notes About Data Representation" in the nbtree
1565 : * README.)
1566 : */
1567 2049083 : scantid = skey->scantid;
1568 2049083 : if (state->heapkeyspace && BTreeTupleIsPosting(itup))
1569 11800 : skey->scantid = BTreeTupleGetMaxHeapTID(itup);
1570 :
1571 3932792 : if (!P_RIGHTMOST(topaque) &&
1572 1883709 : !(P_ISLEAF(topaque) ? invariant_leq_offset(state, skey, P_HIKEY) :
1573 566 : invariant_l_offset(state, skey, P_HIKEY)))
1574 : {
1575 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1576 : char *itid,
1577 : *htid;
1578 :
1579 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1580 0 : htid = psprintf("(%u,%u)",
1581 : ItemPointerGetBlockNumberNoCheck(tid),
1582 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1583 :
1584 0 : ereport(ERROR,
1585 : (errcode(ERRCODE_INDEX_CORRUPTED),
1586 : errmsg("high key invariant violated for index \"%s\"",
1587 : RelationGetRelationName(state->rel)),
1588 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%08X.",
1589 : itid,
1590 : P_ISLEAF(topaque) ? "heap" : "index",
1591 : htid,
1592 : LSN_FORMAT_ARGS(state->targetlsn))));
1593 : }
1594 : /* Reset, in case scantid was set to (itup) posting tuple's max TID */
1595 2049083 : skey->scantid = scantid;
1596 :
1597 : /*
1598 : * * Item order check *
1599 : *
1600 : * Check that items are stored on page in logical order, by checking
1601 : * current item is strictly less than next item (if any).
1602 : */
1603 2049083 : if (OffsetNumberNext(offset) <= max &&
1604 2039835 : !invariant_l_offset(state, skey, OffsetNumberNext(offset)))
1605 : {
1606 : ItemPointer tid;
1607 : char *itid,
1608 : *htid,
1609 : *nitid,
1610 : *nhtid;
1611 :
1612 3 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1613 3 : tid = BTreeTupleGetPointsToTID(itup);
1614 3 : htid = psprintf("(%u,%u)",
1615 : ItemPointerGetBlockNumberNoCheck(tid),
1616 3 : ItemPointerGetOffsetNumberNoCheck(tid));
1617 3 : nitid = psprintf("(%u,%u)", state->targetblock,
1618 3 : OffsetNumberNext(offset));
1619 :
1620 : /* Reuse itup to get pointed-to heap location of second item */
1621 3 : itemid = PageGetItemIdCareful(state, state->targetblock,
1622 : state->target,
1623 3 : OffsetNumberNext(offset));
1624 3 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1625 3 : tid = BTreeTupleGetPointsToTID(itup);
1626 3 : nhtid = psprintf("(%u,%u)",
1627 : ItemPointerGetBlockNumberNoCheck(tid),
1628 3 : ItemPointerGetOffsetNumberNoCheck(tid));
1629 :
1630 3 : ereport(ERROR,
1631 : (errcode(ERRCODE_INDEX_CORRUPTED),
1632 : errmsg("item order invariant violated for index \"%s\"",
1633 : RelationGetRelationName(state->rel)),
1634 : errdetail_internal("Lower index tid=%s (points to %s tid=%s) higher index tid=%s (points to %s tid=%s) page lsn=%X/%08X.",
1635 : itid,
1636 : P_ISLEAF(topaque) ? "heap" : "index",
1637 : htid,
1638 : nitid,
1639 : P_ISLEAF(topaque) ? "heap" : "index",
1640 : nhtid,
1641 : LSN_FORMAT_ARGS(state->targetlsn))));
1642 : }
1643 :
1644 : /*
1645 : * If the index is unique verify entries uniqueness by checking the
1646 : * heap tuples visibility. Immediately check posting tuples and
1647 : * tuples with repeated keys. Postpone check for keys, which have the
1648 : * first appearance.
1649 : */
1650 2049080 : if (state->checkunique && state->indexinfo->ii_Unique &&
1651 297461 : P_ISLEAF(topaque) && !skey->anynullkeys &&
1652 296825 : (BTreeTupleIsPosting(itup) || ItemPointerIsValid(lVis.tid)))
1653 : {
1654 39 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1655 : &lVis);
1656 36 : unique_checked = true;
1657 : }
1658 :
1659 2049077 : if (state->checkunique && state->indexinfo->ii_Unique &&
1660 149035 : P_ISLEAF(topaque) && OffsetNumberNext(offset) <= max)
1661 : {
1662 : /* Save current scankey tid */
1663 147679 : scantid = skey->scantid;
1664 :
1665 : /*
1666 : * Invalidate scankey tid to make _bt_compare compare only keys in
1667 : * the item to report equality even if heap TIDs are different
1668 : */
1669 147679 : skey->scantid = NULL;
1670 :
1671 : /*
1672 : * If next key tuple is different, invalidate last visible entry
1673 : * data (whole index tuple or last posting in index tuple). Key
1674 : * containing null value does not violate unique constraint and
1675 : * treated as different to any other key.
1676 : *
1677 : * If the next key is the same as the previous one, do the
1678 : * bt_entry_unique_check() call if it was postponed.
1679 : */
1680 147679 : if (_bt_compare(state->rel, skey, state->target,
1681 147995 : OffsetNumberNext(offset)) != 0 || skey->anynullkeys)
1682 : {
1683 147433 : lVis.blkno = InvalidBlockNumber;
1684 147433 : lVis.offset = InvalidOffsetNumber;
1685 147433 : lVis.postingIndex = -1;
1686 147433 : lVis.tid = NULL;
1687 : }
1688 246 : else if (!unique_checked)
1689 : {
1690 243 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1691 : &lVis);
1692 : }
1693 147679 : skey->scantid = scantid; /* Restore saved scan key state */
1694 : }
1695 :
1696 : /*
1697 : * * Last item check *
1698 : *
1699 : * Check last item against next/right page's first data item's when
1700 : * last item on page is reached. This additional check will detect
1701 : * transposed pages iff the supposed right sibling page happens to
1702 : * belong before target in the key space. (Otherwise, a subsequent
1703 : * heap verification will probably detect the problem.)
1704 : *
1705 : * This check is similar to the item order check that will have
1706 : * already been performed for every other "real" item on target page
1707 : * when last item is checked. The difference is that the next item
1708 : * (the item that is compared to target's last item) needs to come
1709 : * from the next/sibling page. There may not be such an item
1710 : * available from sibling for various reasons, though (e.g., target is
1711 : * the rightmost page on level).
1712 : */
1713 2049077 : if (offset == max)
1714 : {
1715 : BTScanInsert rightkey;
1716 :
1717 : /* first offset on a right index page (log only) */
1718 9248 : OffsetNumber rightfirstoffset = InvalidOffsetNumber;
1719 :
1720 : /* Get item in next/right page */
1721 9248 : rightkey = bt_right_page_check_scankey(state, &rightfirstoffset);
1722 :
1723 9248 : if (rightkey &&
1724 6760 : !invariant_g_offset(state, rightkey, max))
1725 : {
1726 : /*
1727 : * As explained at length in bt_right_page_check_scankey(),
1728 : * there is a known !readonly race that could account for
1729 : * apparent violation of invariant, which we must check for
1730 : * before actually proceeding with raising error. Our canary
1731 : * condition is that target page was deleted.
1732 : */
1733 0 : if (!state->readonly)
1734 : {
1735 : /* Get fresh copy of target page */
1736 0 : state->target = palloc_btree_page(state, state->targetblock);
1737 : /* Note that we deliberately do not update target LSN */
1738 0 : topaque = BTPageGetOpaque(state->target);
1739 :
1740 : /*
1741 : * All !readonly checks now performed; just return
1742 : */
1743 0 : if (P_IGNORE(topaque))
1744 0 : return;
1745 : }
1746 :
1747 0 : ereport(ERROR,
1748 : (errcode(ERRCODE_INDEX_CORRUPTED),
1749 : errmsg("cross page item order invariant violated for index \"%s\"",
1750 : RelationGetRelationName(state->rel)),
1751 : errdetail_internal("Last item on page tid=(%u,%u) page lsn=%X/%08X.",
1752 : state->targetblock, offset,
1753 : LSN_FORMAT_ARGS(state->targetlsn))));
1754 : }
1755 :
1756 : /*
1757 : * If index has unique constraint make sure that no more than one
1758 : * found equal items is visible.
1759 : */
1760 9248 : if (state->checkunique && state->indexinfo->ii_Unique &&
1761 533 : rightkey && P_ISLEAF(topaque) && !P_RIGHTMOST(topaque))
1762 : {
1763 533 : BlockNumber rightblock_number = topaque->btpo_next;
1764 :
1765 533 : elog(DEBUG2, "check cross page unique condition");
1766 :
1767 : /*
1768 : * Make _bt_compare compare only index keys without heap TIDs.
1769 : * rightkey->scantid is modified destructively but it is ok
1770 : * for it is not used later.
1771 : */
1772 533 : rightkey->scantid = NULL;
1773 :
1774 : /* The first key on the next page is the same */
1775 533 : if (_bt_compare(state->rel, rightkey, state->target, max) == 0 &&
1776 7 : !rightkey->anynullkeys)
1777 : {
1778 : Page rightpage;
1779 :
1780 : /*
1781 : * Do the bt_entry_unique_check() call if it was
1782 : * postponed.
1783 : */
1784 0 : if (!unique_checked)
1785 0 : bt_entry_unique_check(state, itup, state->targetblock,
1786 : offset, &lVis);
1787 :
1788 0 : elog(DEBUG2, "cross page equal keys");
1789 0 : rightpage = palloc_btree_page(state,
1790 : rightblock_number);
1791 0 : topaque = BTPageGetOpaque(rightpage);
1792 :
1793 0 : if (P_IGNORE(topaque))
1794 : {
1795 0 : pfree(rightpage);
1796 0 : break;
1797 : }
1798 :
1799 0 : if (unlikely(!P_ISLEAF(topaque)))
1800 0 : ereport(ERROR,
1801 : (errcode(ERRCODE_INDEX_CORRUPTED),
1802 : errmsg("right block of leaf block is non-leaf for index \"%s\"",
1803 : RelationGetRelationName(state->rel)),
1804 : errdetail_internal("Block=%u page lsn=%X/%08X.",
1805 : state->targetblock,
1806 : LSN_FORMAT_ARGS(state->targetlsn))));
1807 :
1808 0 : itemid = PageGetItemIdCareful(state, rightblock_number,
1809 : rightpage,
1810 : rightfirstoffset);
1811 0 : itup = (IndexTuple) PageGetItem(rightpage, itemid);
1812 :
1813 0 : bt_entry_unique_check(state, itup, rightblock_number, rightfirstoffset, &lVis);
1814 :
1815 0 : pfree(rightpage);
1816 : }
1817 : }
1818 : }
1819 :
1820 : /*
1821 : * * Downlink check *
1822 : *
1823 : * Additional check of child items iff this is an internal page and
1824 : * caller holds a ShareLock. This happens for every downlink (item)
1825 : * in target excluding the negative-infinity downlink (again, this is
1826 : * because it has no useful value to compare).
1827 : */
1828 2049077 : if (!P_ISLEAF(topaque) && state->readonly)
1829 1861 : bt_child_check(state, skey, offset);
1830 : }
1831 :
1832 : /*
1833 : * Special case bt_child_highkey_check() call
1834 : *
1835 : * We don't pass a real downlink, but we've to finish the level
1836 : * processing. If condition is satisfied, we've already processed all the
1837 : * downlinks from the target level. But there still might be pages to the
1838 : * right of the child page pointer to by our rightmost downlink. And they
1839 : * might have missing downlinks. This final call checks for them.
1840 : */
1841 9250 : if (!P_ISLEAF(topaque) && P_RIGHTMOST(topaque) && state->readonly)
1842 : {
1843 11 : bt_child_highkey_check(state, InvalidOffsetNumber,
1844 : NULL, topaque->btpo_level);
1845 : }
1846 : }
1847 :
1848 : /*
1849 : * Return a scankey for an item on page to right of current target (or the
1850 : * first non-ignorable page), sufficient to check ordering invariant on last
1851 : * item in current target page. Returned scankey relies on local memory
1852 : * allocated for the child page, which caller cannot pfree(). Caller's memory
1853 : * context should be reset between calls here.
1854 : *
1855 : * This is the first data item, and so all adjacent items are checked against
1856 : * their immediate sibling item (which may be on a sibling page, or even a
1857 : * "cousin" page at parent boundaries where target's rightlink points to page
1858 : * with different parent page). If no such valid item is available, return
1859 : * NULL instead.
1860 : *
1861 : * Note that !readonly callers must reverify that target page has not
1862 : * been concurrently deleted.
1863 : *
1864 : * Save rightfirstoffset for detailed error message.
1865 : */
1866 : static BTScanInsert
1867 9248 : bt_right_page_check_scankey(BtreeCheckState *state, OffsetNumber *rightfirstoffset)
1868 : {
1869 : BTPageOpaque opaque;
1870 : ItemId rightitem;
1871 : IndexTuple firstitup;
1872 : BlockNumber targetnext;
1873 : Page rightpage;
1874 : OffsetNumber nline;
1875 :
1876 : /* Determine target's next block number */
1877 9248 : opaque = BTPageGetOpaque(state->target);
1878 :
1879 : /* If target is already rightmost, no right sibling; nothing to do here */
1880 9248 : if (P_RIGHTMOST(opaque))
1881 2488 : return NULL;
1882 :
1883 : /*
1884 : * General notes on concurrent page splits and page deletion:
1885 : *
1886 : * Routines like _bt_search() don't require *any* page split interlock
1887 : * when descending the tree, including something very light like a buffer
1888 : * pin. That's why it's okay that we don't either. This avoidance of any
1889 : * need to "couple" buffer locks is the raison d' etre of the Lehman & Yao
1890 : * algorithm, in fact.
1891 : *
1892 : * That leaves deletion. A deleted page won't actually be recycled by
1893 : * VACUUM early enough for us to fail to at least follow its right link
1894 : * (or left link, or downlink) and find its sibling, because recycling
1895 : * does not occur until no possible index scan could land on the page.
1896 : * Index scans can follow links with nothing more than their snapshot as
1897 : * an interlock and be sure of at least that much. (See page
1898 : * recycling/"visible to everyone" notes in nbtree README.)
1899 : *
1900 : * Furthermore, it's okay if we follow a rightlink and find a half-dead or
1901 : * dead (ignorable) page one or more times. There will either be a
1902 : * further right link to follow that leads to a live page before too long
1903 : * (before passing by parent's rightmost child), or we will find the end
1904 : * of the entire level instead (possible when parent page is itself the
1905 : * rightmost on its level).
1906 : */
1907 6760 : targetnext = opaque->btpo_next;
1908 : for (;;)
1909 : {
1910 6760 : CHECK_FOR_INTERRUPTS();
1911 :
1912 6760 : rightpage = palloc_btree_page(state, targetnext);
1913 6760 : opaque = BTPageGetOpaque(rightpage);
1914 :
1915 6760 : if (!P_IGNORE(opaque) || P_RIGHTMOST(opaque))
1916 : break;
1917 :
1918 : /*
1919 : * We landed on a deleted or half-dead sibling page. Step right until
1920 : * we locate a live sibling page.
1921 : */
1922 0 : ereport(DEBUG2,
1923 : (errcode(ERRCODE_NO_DATA),
1924 : errmsg_internal("level %u sibling page in block %u of index \"%s\" was found deleted or half dead",
1925 : opaque->btpo_level, targetnext, RelationGetRelationName(state->rel)),
1926 : errdetail_internal("Deleted page found when building scankey from right sibling.")));
1927 :
1928 0 : targetnext = opaque->btpo_next;
1929 :
1930 : /* Be slightly more pro-active in freeing this memory, just in case */
1931 0 : pfree(rightpage);
1932 : }
1933 :
1934 : /*
1935 : * No ShareLock held case -- why it's safe to proceed.
1936 : *
1937 : * Problem:
1938 : *
1939 : * We must avoid false positive reports of corruption when caller treats
1940 : * item returned here as an upper bound on target's last item. In
1941 : * general, false positives are disallowed. Avoiding them here when
1942 : * caller is !readonly is subtle.
1943 : *
1944 : * A concurrent page deletion by VACUUM of the target page can result in
1945 : * the insertion of items on to this right sibling page that would
1946 : * previously have been inserted on our target page. There might have
1947 : * been insertions that followed the target's downlink after it was made
1948 : * to point to right sibling instead of target by page deletion's first
1949 : * phase. The inserters insert items that would belong on target page.
1950 : * This race is very tight, but it's possible. This is our only problem.
1951 : *
1952 : * Non-problems:
1953 : *
1954 : * We are not hindered by a concurrent page split of the target; we'll
1955 : * never land on the second half of the page anyway. A concurrent split
1956 : * of the right page will also not matter, because the first data item
1957 : * remains the same within the left half, which we'll reliably land on. If
1958 : * we had to skip over ignorable/deleted pages, it cannot matter because
1959 : * their key space has already been atomically merged with the first
1960 : * non-ignorable page we eventually find (doesn't matter whether the page
1961 : * we eventually find is a true sibling or a cousin of target, which we go
1962 : * into below).
1963 : *
1964 : * Solution:
1965 : *
1966 : * Caller knows that it should reverify that target is not ignorable
1967 : * (half-dead or deleted) when cross-page sibling item comparison appears
1968 : * to indicate corruption (invariant fails). This detects the single race
1969 : * condition that exists for caller. This is correct because the
1970 : * continued existence of target block as non-ignorable (not half-dead or
1971 : * deleted) implies that target page was not merged into from the right by
1972 : * deletion; the key space at or after target never moved left. Target's
1973 : * parent either has the same downlink to target as before, or a <
1974 : * downlink due to deletion at the left of target. Target either has the
1975 : * same highkey as before, or a highkey < before when there is a page
1976 : * split. (The rightmost concurrently-split-from-target-page page will
1977 : * still have the same highkey as target was originally found to have,
1978 : * which for our purposes is equivalent to target's highkey itself never
1979 : * changing, since we reliably skip over
1980 : * concurrently-split-from-target-page pages.)
1981 : *
1982 : * In simpler terms, we allow that the key space of the target may expand
1983 : * left (the key space can move left on the left side of target only), but
1984 : * the target key space cannot expand right and get ahead of us without
1985 : * our detecting it. The key space of the target cannot shrink, unless it
1986 : * shrinks to zero due to the deletion of the original page, our canary
1987 : * condition. (To be very precise, we're a bit stricter than that because
1988 : * it might just have been that the target page split and only the
1989 : * original target page was deleted. We can be more strict, just not more
1990 : * lax.)
1991 : *
1992 : * Top level tree walk caller moves on to next page (makes it the new
1993 : * target) following recovery from this race. (cf. The rationale for
1994 : * child/downlink verification needing a ShareLock within
1995 : * bt_child_check(), where page deletion is also the main source of
1996 : * trouble.)
1997 : *
1998 : * Note that it doesn't matter if right sibling page here is actually a
1999 : * cousin page, because in order for the key space to be readjusted in a
2000 : * way that causes us issues in next level up (guiding problematic
2001 : * concurrent insertions to the cousin from the grandparent rather than to
2002 : * the sibling from the parent), there'd have to be page deletion of
2003 : * target's parent page (affecting target's parent's downlink in target's
2004 : * grandparent page). Internal page deletion only occurs when there are
2005 : * no child pages (they were all fully deleted), and caller is checking
2006 : * that the target's parent has at least one non-deleted (so
2007 : * non-ignorable) child: the target page. (Note that the first phase of
2008 : * deletion atomically marks the page to be deleted half-dead/ignorable at
2009 : * the same time downlink in its parent is removed, so caller will
2010 : * definitely not fail to detect that this happened.)
2011 : *
2012 : * This trick is inspired by the method backward scans use for dealing
2013 : * with concurrent page splits; concurrent page deletion is a problem that
2014 : * similarly receives special consideration sometimes (it's possible that
2015 : * the backwards scan will re-read its "original" block after failing to
2016 : * find a right-link to it, having already moved in the opposite direction
2017 : * (right/"forwards") a few times to try to locate one). Just like us,
2018 : * that happens only to determine if there was a concurrent page deletion
2019 : * of a reference page, and just like us if there was a page deletion of
2020 : * that reference page it means we can move on from caring about the
2021 : * reference page. See the nbtree README for a full description of how
2022 : * that works.
2023 : */
2024 6760 : nline = PageGetMaxOffsetNumber(rightpage);
2025 :
2026 : /*
2027 : * Get first data item, if any
2028 : */
2029 6760 : if (P_ISLEAF(opaque) && nline >= P_FIRSTDATAKEY(opaque))
2030 : {
2031 : /* Return first data item (if any) */
2032 6758 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2033 6758 : P_FIRSTDATAKEY(opaque));
2034 6758 : *rightfirstoffset = P_FIRSTDATAKEY(opaque);
2035 : }
2036 4 : else if (!P_ISLEAF(opaque) &&
2037 2 : nline >= OffsetNumberNext(P_FIRSTDATAKEY(opaque)))
2038 : {
2039 : /*
2040 : * Return first item after the internal page's "negative infinity"
2041 : * item
2042 : */
2043 2 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2044 2 : OffsetNumberNext(P_FIRSTDATAKEY(opaque)));
2045 : }
2046 : else
2047 : {
2048 : /*
2049 : * No first item. Page is probably empty leaf page, but it's also
2050 : * possible that it's an internal page with only a negative infinity
2051 : * item.
2052 : */
2053 0 : ereport(DEBUG2,
2054 : (errcode(ERRCODE_NO_DATA),
2055 : errmsg_internal("%s block %u of index \"%s\" has no first data item",
2056 : P_ISLEAF(opaque) ? "leaf" : "internal", targetnext,
2057 : RelationGetRelationName(state->rel))));
2058 0 : return NULL;
2059 : }
2060 :
2061 : /*
2062 : * Return first real item scankey. Note that this relies on right page
2063 : * memory remaining allocated.
2064 : */
2065 6760 : firstitup = (IndexTuple) PageGetItem(rightpage, rightitem);
2066 6760 : return bt_mkscankey_pivotsearch(state->rel, firstitup);
2067 : }
2068 :
2069 : /*
2070 : * Check if two tuples are binary identical except the block number. So,
2071 : * this function is capable to compare pivot keys on different levels.
2072 : */
2073 : static bool
2074 1862 : bt_pivot_tuple_identical(bool heapkeyspace, IndexTuple itup1, IndexTuple itup2)
2075 : {
2076 1862 : if (IndexTupleSize(itup1) != IndexTupleSize(itup2))
2077 0 : return false;
2078 :
2079 1862 : if (heapkeyspace)
2080 : {
2081 : /*
2082 : * Offset number will contain important information in heapkeyspace
2083 : * indexes: the number of attributes left in the pivot tuple following
2084 : * suffix truncation. Don't skip over it (compare it too).
2085 : */
2086 1862 : if (memcmp(&itup1->t_tid.ip_posid, &itup2->t_tid.ip_posid,
2087 1862 : IndexTupleSize(itup1) -
2088 : offsetof(ItemPointerData, ip_posid)) != 0)
2089 0 : return false;
2090 : }
2091 : else
2092 : {
2093 : /*
2094 : * Cannot rely on offset number field having consistent value across
2095 : * levels on pg_upgrade'd !heapkeyspace indexes. Compare contents of
2096 : * tuple starting from just after item pointer (i.e. after block
2097 : * number and offset number).
2098 : */
2099 0 : if (memcmp(&itup1->t_info, &itup2->t_info,
2100 0 : IndexTupleSize(itup1) -
2101 : offsetof(IndexTupleData, t_info)) != 0)
2102 0 : return false;
2103 : }
2104 :
2105 1862 : return true;
2106 : }
2107 :
2108 : /*---
2109 : * Check high keys on the child level. Traverse rightlinks from previous
2110 : * downlink to the current one. Check that there are no intermediate pages
2111 : * with missing downlinks.
2112 : *
2113 : * If 'loaded_child' is given, it's assumed to be the page pointed to by the
2114 : * downlink referenced by 'downlinkoffnum' of the target page.
2115 : *
2116 : * Basically this function is called for each target downlink and checks two
2117 : * invariants:
2118 : *
2119 : * 1) You can reach the next child from previous one via rightlinks;
2120 : * 2) Each child high key have matching pivot key on target level.
2121 : *
2122 : * Consider the sample tree picture.
2123 : *
2124 : * 1
2125 : * / \
2126 : * 2 <-> 3
2127 : * / \ / \
2128 : * 4 <> 5 <> 6 <> 7 <> 8
2129 : *
2130 : * This function will be called for blocks 4, 5, 6 and 8. Consider what is
2131 : * happening for each function call.
2132 : *
2133 : * - The function call for block 4 initializes data structure and matches high
2134 : * key of block 4 to downlink's pivot key of block 2.
2135 : * - The high key of block 5 is matched to the high key of block 2.
2136 : * - The block 6 has an incomplete split flag set, so its high key isn't
2137 : * matched to anything.
2138 : * - The function call for block 8 checks that block 8 can be found while
2139 : * following rightlinks from block 6. The high key of block 7 will be
2140 : * matched to downlink's pivot key in block 3.
2141 : *
2142 : * There is also final call of this function, which checks that there is no
2143 : * missing downlinks for children to the right of the child referenced by
2144 : * rightmost downlink in target level.
2145 : */
2146 : static void
2147 1884 : bt_child_highkey_check(BtreeCheckState *state,
2148 : OffsetNumber target_downlinkoffnum,
2149 : Page loaded_child,
2150 : uint32 target_level)
2151 : {
2152 1884 : BlockNumber blkno = state->prevrightlink;
2153 : Page page;
2154 : BTPageOpaque opaque;
2155 1884 : bool rightsplit = state->previncompletesplit;
2156 1884 : bool first = true;
2157 : ItemId itemid;
2158 : IndexTuple itup;
2159 : BlockNumber downlink;
2160 :
2161 1884 : if (OffsetNumberIsValid(target_downlinkoffnum))
2162 : {
2163 1873 : itemid = PageGetItemIdCareful(state, state->targetblock,
2164 : state->target, target_downlinkoffnum);
2165 1873 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2166 1873 : downlink = BTreeTupleGetDownLink(itup);
2167 : }
2168 : else
2169 : {
2170 11 : downlink = P_NONE;
2171 : }
2172 :
2173 : /*
2174 : * If no previous rightlink is memorized for current level just below
2175 : * target page's level, we are about to start from the leftmost page. We
2176 : * can't follow rightlinks from previous page, because there is no
2177 : * previous page. But we still can match high key.
2178 : *
2179 : * So we initialize variables for the loop above like there is previous
2180 : * page referencing current child. Also we imply previous page to not
2181 : * have incomplete split flag, that would make us require downlink for
2182 : * current child. That's correct, because leftmost page on the level
2183 : * should always have parent downlink.
2184 : */
2185 1884 : if (!BlockNumberIsValid(blkno))
2186 : {
2187 11 : blkno = downlink;
2188 11 : rightsplit = false;
2189 : }
2190 :
2191 : /* Move to the right on the child level */
2192 : while (true)
2193 : {
2194 : /*
2195 : * Did we traverse the whole tree level and this is check for pages to
2196 : * the right of rightmost downlink?
2197 : */
2198 1884 : if (blkno == P_NONE && downlink == P_NONE)
2199 : {
2200 11 : state->prevrightlink = InvalidBlockNumber;
2201 11 : state->previncompletesplit = false;
2202 11 : return;
2203 : }
2204 :
2205 : /* Did we traverse the whole tree level and don't find next downlink? */
2206 1873 : if (blkno == P_NONE)
2207 0 : ereport(ERROR,
2208 : (errcode(ERRCODE_INDEX_CORRUPTED),
2209 : errmsg("can't traverse from downlink %u to downlink %u of index \"%s\"",
2210 : state->prevrightlink, downlink,
2211 : RelationGetRelationName(state->rel))));
2212 :
2213 : /* Load page contents */
2214 1873 : if (blkno == downlink && loaded_child)
2215 1861 : page = loaded_child;
2216 : else
2217 12 : page = palloc_btree_page(state, blkno);
2218 :
2219 1873 : opaque = BTPageGetOpaque(page);
2220 :
2221 : /* The first page we visit at the level should be leftmost */
2222 1873 : if (first && !BlockNumberIsValid(state->prevrightlink) &&
2223 11 : !bt_leftmost_ignoring_half_dead(state, blkno, opaque))
2224 0 : ereport(ERROR,
2225 : (errcode(ERRCODE_INDEX_CORRUPTED),
2226 : errmsg("the first child of leftmost target page is not leftmost of its level in index \"%s\"",
2227 : RelationGetRelationName(state->rel)),
2228 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2229 : state->targetblock, blkno,
2230 : LSN_FORMAT_ARGS(state->targetlsn))));
2231 :
2232 : /* Do level sanity check */
2233 1873 : if ((!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque)) &&
2234 1873 : opaque->btpo_level != target_level - 1)
2235 0 : ereport(ERROR,
2236 : (errcode(ERRCODE_INDEX_CORRUPTED),
2237 : errmsg("block found while following rightlinks from child of index \"%s\" has invalid level",
2238 : RelationGetRelationName(state->rel)),
2239 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
2240 : blkno, target_level - 1, opaque->btpo_level)));
2241 :
2242 : /* Try to detect circular links */
2243 1873 : if ((!first && blkno == state->prevrightlink) || blkno == opaque->btpo_prev)
2244 0 : ereport(ERROR,
2245 : (errcode(ERRCODE_INDEX_CORRUPTED),
2246 : errmsg("circular link chain found in block %u of index \"%s\"",
2247 : blkno, RelationGetRelationName(state->rel))));
2248 :
2249 1873 : if (blkno != downlink && !P_IGNORE(opaque))
2250 : {
2251 : /* blkno probably has missing parent downlink */
2252 0 : bt_downlink_missing_check(state, rightsplit, blkno, page);
2253 : }
2254 :
2255 1873 : rightsplit = P_INCOMPLETE_SPLIT(opaque);
2256 :
2257 : /*
2258 : * If we visit page with high key, check that it is equal to the
2259 : * target key next to corresponding downlink.
2260 : */
2261 1873 : if (!rightsplit && !P_RIGHTMOST(opaque) && !P_ISHALFDEAD(opaque))
2262 : {
2263 : BTPageOpaque topaque;
2264 : IndexTuple highkey;
2265 : OffsetNumber pivotkey_offset;
2266 :
2267 : /* Get high key */
2268 1862 : itemid = PageGetItemIdCareful(state, blkno, page, P_HIKEY);
2269 1862 : highkey = (IndexTuple) PageGetItem(page, itemid);
2270 :
2271 : /*
2272 : * There might be two situations when we examine high key. If
2273 : * current child page is referenced by given target downlink, we
2274 : * should look to the next offset number for matching key from
2275 : * target page.
2276 : *
2277 : * Alternatively, we're following rightlinks somewhere in the
2278 : * middle between page referenced by previous target's downlink
2279 : * and the page referenced by current target's downlink. If
2280 : * current child page hasn't incomplete split flag set, then its
2281 : * high key should match to the target's key of current offset
2282 : * number. This happens when a previous call here (to
2283 : * bt_child_highkey_check()) found an incomplete split, and we
2284 : * reach a right sibling page without a downlink -- the right
2285 : * sibling page's high key still needs to be matched to a
2286 : * separator key on the parent/target level.
2287 : *
2288 : * Don't apply OffsetNumberNext() to target_downlinkoffnum when we
2289 : * already had to step right on the child level. Our traversal of
2290 : * the child level must try to move in perfect lockstep behind (to
2291 : * the left of) the target/parent level traversal.
2292 : */
2293 1862 : if (blkno == downlink)
2294 1862 : pivotkey_offset = OffsetNumberNext(target_downlinkoffnum);
2295 : else
2296 0 : pivotkey_offset = target_downlinkoffnum;
2297 :
2298 1862 : topaque = BTPageGetOpaque(state->target);
2299 :
2300 1862 : if (!offset_is_negative_infinity(topaque, pivotkey_offset))
2301 : {
2302 : /*
2303 : * If we're looking for the next pivot tuple in target page,
2304 : * but there is no more pivot tuples, then we should match to
2305 : * high key instead.
2306 : */
2307 1862 : if (pivotkey_offset > PageGetMaxOffsetNumber(state->target))
2308 : {
2309 1 : if (P_RIGHTMOST(topaque))
2310 0 : ereport(ERROR,
2311 : (errcode(ERRCODE_INDEX_CORRUPTED),
2312 : errmsg("child high key is greater than rightmost pivot key on target level in index \"%s\"",
2313 : RelationGetRelationName(state->rel)),
2314 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2315 : state->targetblock, blkno,
2316 : LSN_FORMAT_ARGS(state->targetlsn))));
2317 1 : pivotkey_offset = P_HIKEY;
2318 : }
2319 1862 : itemid = PageGetItemIdCareful(state, state->targetblock,
2320 : state->target, pivotkey_offset);
2321 1862 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2322 : }
2323 : else
2324 : {
2325 : /*
2326 : * We cannot try to match child's high key to a negative
2327 : * infinity key in target, since there is nothing to compare.
2328 : * However, it's still possible to match child's high key
2329 : * outside of target page. The reason why we're are is that
2330 : * bt_child_highkey_check() was previously called for the
2331 : * cousin page of 'loaded_child', which is incomplete split.
2332 : * So, now we traverse to the right of that cousin page and
2333 : * current child level page under consideration still belongs
2334 : * to the subtree of target's left sibling. Thus, we need to
2335 : * match child's high key to its left uncle page high key.
2336 : * Thankfully we saved it, it's called a "low key" of target
2337 : * page.
2338 : */
2339 0 : if (!state->lowkey)
2340 0 : ereport(ERROR,
2341 : (errcode(ERRCODE_INDEX_CORRUPTED),
2342 : errmsg("can't find left sibling high key in index \"%s\"",
2343 : RelationGetRelationName(state->rel)),
2344 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2345 : state->targetblock, blkno,
2346 : LSN_FORMAT_ARGS(state->targetlsn))));
2347 0 : itup = state->lowkey;
2348 : }
2349 :
2350 1862 : if (!bt_pivot_tuple_identical(state->heapkeyspace, highkey, itup))
2351 : {
2352 0 : ereport(ERROR,
2353 : (errcode(ERRCODE_INDEX_CORRUPTED),
2354 : errmsg("mismatch between parent key and child high key in index \"%s\"",
2355 : RelationGetRelationName(state->rel)),
2356 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%08X.",
2357 : state->targetblock, blkno,
2358 : LSN_FORMAT_ARGS(state->targetlsn))));
2359 : }
2360 : }
2361 :
2362 : /* Exit if we already found next downlink */
2363 1873 : if (blkno == downlink)
2364 : {
2365 1873 : state->prevrightlink = opaque->btpo_next;
2366 1873 : state->previncompletesplit = rightsplit;
2367 1873 : return;
2368 : }
2369 :
2370 : /* Traverse to the next page using rightlink */
2371 0 : blkno = opaque->btpo_next;
2372 :
2373 : /* Free page contents if it's allocated by us */
2374 0 : if (page != loaded_child)
2375 0 : pfree(page);
2376 0 : first = false;
2377 : }
2378 : }
2379 :
2380 : /*
2381 : * Checks one of target's downlink against its child page.
2382 : *
2383 : * Conceptually, the target page continues to be what is checked here. The
2384 : * target block is still blamed in the event of finding an invariant violation.
2385 : * The downlink insertion into the target is probably where any problem raised
2386 : * here arises, and there is no such thing as a parent link, so doing the
2387 : * verification this way around is much more practical.
2388 : *
2389 : * This function visits child page and it's sequentially called for each
2390 : * downlink of target page. Assuming this we also check downlink connectivity
2391 : * here in order to save child page visits.
2392 : */
2393 : static void
2394 1861 : bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
2395 : OffsetNumber downlinkoffnum)
2396 : {
2397 : ItemId itemid;
2398 : IndexTuple itup;
2399 : BlockNumber childblock;
2400 : OffsetNumber offset;
2401 : OffsetNumber maxoffset;
2402 : Page child;
2403 : BTPageOpaque copaque;
2404 : BTPageOpaque topaque;
2405 :
2406 1861 : itemid = PageGetItemIdCareful(state, state->targetblock,
2407 : state->target, downlinkoffnum);
2408 1861 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2409 1861 : childblock = BTreeTupleGetDownLink(itup);
2410 :
2411 : /*
2412 : * Caller must have ShareLock on target relation, because of
2413 : * considerations around page deletion by VACUUM.
2414 : *
2415 : * NB: In general, page deletion deletes the right sibling's downlink, not
2416 : * the downlink of the page being deleted; the deleted page's downlink is
2417 : * reused for its sibling. The key space is thereby consolidated between
2418 : * the deleted page and its right sibling. (We cannot delete a parent
2419 : * page's rightmost child unless it is the last child page, and we intend
2420 : * to also delete the parent itself.)
2421 : *
2422 : * If this verification happened without a ShareLock, the following race
2423 : * condition could cause false positives:
2424 : *
2425 : * In general, concurrent page deletion might occur, including deletion of
2426 : * the left sibling of the child page that is examined here. If such a
2427 : * page deletion were to occur, closely followed by an insertion into the
2428 : * newly expanded key space of the child, a window for the false positive
2429 : * opens up: the stale parent/target downlink originally followed to get
2430 : * to the child legitimately ceases to be a lower bound on all items in
2431 : * the page, since the key space was concurrently expanded "left".
2432 : * (Insertion followed the "new" downlink for the child, not our now-stale
2433 : * downlink, which was concurrently physically removed in target/parent as
2434 : * part of deletion's first phase.)
2435 : *
2436 : * While we use various techniques elsewhere to perform cross-page
2437 : * verification for !readonly callers, a similar trick seems difficult
2438 : * here. The tricks used by bt_recheck_sibling_links and by
2439 : * bt_right_page_check_scankey both involve verification of a same-level,
2440 : * cross-sibling invariant. Cross-level invariants are far more squishy,
2441 : * though. The nbtree REDO routines do not actually couple buffer locks
2442 : * across levels during page splits, so making any cross-level check work
2443 : * reliably in !readonly mode may be impossible.
2444 : */
2445 : Assert(state->readonly);
2446 :
2447 : /*
2448 : * Verify child page has the downlink key from target page (its parent) as
2449 : * a lower bound; downlink must be strictly less than all keys on the
2450 : * page.
2451 : *
2452 : * Check all items, rather than checking just the first and trusting that
2453 : * the operator class obeys the transitive law.
2454 : */
2455 1861 : topaque = BTPageGetOpaque(state->target);
2456 1861 : child = palloc_btree_page(state, childblock);
2457 1861 : copaque = BTPageGetOpaque(child);
2458 1861 : maxoffset = PageGetMaxOffsetNumber(child);
2459 :
2460 : /*
2461 : * Since we've already loaded the child block, combine this check with
2462 : * check for downlink connectivity.
2463 : */
2464 1861 : bt_child_highkey_check(state, downlinkoffnum,
2465 : child, topaque->btpo_level);
2466 :
2467 : /*
2468 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2469 : * and since a page has no links within other pages (siblings and parent)
2470 : * once it is marked fully deleted, it should be impossible to land on a
2471 : * fully deleted page.
2472 : *
2473 : * It does not quite make sense to enforce that the page cannot even be
2474 : * half-dead, despite the fact the downlink is modified at the same stage
2475 : * that the child leaf page is marked half-dead. That's incorrect because
2476 : * there may occasionally be multiple downlinks from a chain of pages
2477 : * undergoing deletion, where multiple successive calls are made to
2478 : * _bt_unlink_halfdead_page() by VACUUM before it can finally safely mark
2479 : * the leaf page as fully dead. While _bt_mark_page_halfdead() usually
2480 : * removes the downlink to the leaf page that is marked half-dead, that's
2481 : * not guaranteed, so it's possible we'll land on a half-dead page with a
2482 : * downlink due to an interrupted multi-level page deletion.
2483 : *
2484 : * We go ahead with our checks if the child page is half-dead. It's safe
2485 : * to do so because we do not test the child's high key, so it does not
2486 : * matter that the original high key will have been replaced by a dummy
2487 : * truncated high key within _bt_mark_page_halfdead(). All other page
2488 : * items are left intact on a half-dead page, so there is still something
2489 : * to test.
2490 : */
2491 1861 : if (P_ISDELETED(copaque))
2492 0 : ereport(ERROR,
2493 : (errcode(ERRCODE_INDEX_CORRUPTED),
2494 : errmsg("downlink to deleted page found in index \"%s\"",
2495 : RelationGetRelationName(state->rel)),
2496 : errdetail_internal("Parent block=%u child block=%u parent page lsn=%X/%08X.",
2497 : state->targetblock, childblock,
2498 : LSN_FORMAT_ARGS(state->targetlsn))));
2499 :
2500 1861 : for (offset = P_FIRSTDATAKEY(copaque);
2501 600350 : offset <= maxoffset;
2502 598489 : offset = OffsetNumberNext(offset))
2503 : {
2504 : /*
2505 : * Skip comparison of target page key against "negative infinity"
2506 : * item, if any. Checking it would indicate that it's not a strict
2507 : * lower bound, but that's only because of the hard-coding for
2508 : * negative infinity items within _bt_compare().
2509 : *
2510 : * If nbtree didn't truncate negative infinity tuples during internal
2511 : * page splits then we'd expect child's negative infinity key to be
2512 : * equal to the scankey/downlink from target/parent (it would be a
2513 : * "low key" in this hypothetical scenario, and so it would still need
2514 : * to be treated as a special case here).
2515 : *
2516 : * Negative infinity items can be thought of as a strict lower bound
2517 : * that works transitively, with the last non-negative-infinity pivot
2518 : * followed during a descent from the root as its "true" strict lower
2519 : * bound. Only a small number of negative infinity items are truly
2520 : * negative infinity; those that are the first items of leftmost
2521 : * internal pages. In more general terms, a negative infinity item is
2522 : * only negative infinity with respect to the subtree that the page is
2523 : * at the root of.
2524 : *
2525 : * See also: bt_rootdescend(), which can even detect transitive
2526 : * inconsistencies on cousin leaf pages.
2527 : */
2528 598489 : if (offset_is_negative_infinity(copaque, offset))
2529 1 : continue;
2530 :
2531 598488 : if (!invariant_l_nontarget_offset(state, targetkey, childblock, child,
2532 : offset))
2533 0 : ereport(ERROR,
2534 : (errcode(ERRCODE_INDEX_CORRUPTED),
2535 : errmsg("down-link lower bound invariant violated for index \"%s\"",
2536 : RelationGetRelationName(state->rel)),
2537 : errdetail_internal("Parent block=%u child index tid=(%u,%u) parent page lsn=%X/%08X.",
2538 : state->targetblock, childblock, offset,
2539 : LSN_FORMAT_ARGS(state->targetlsn))));
2540 : }
2541 :
2542 1861 : pfree(child);
2543 1861 : }
2544 :
2545 : /*
2546 : * Checks if page is missing a downlink that it should have.
2547 : *
2548 : * A page that lacks a downlink/parent may indicate corruption. However, we
2549 : * must account for the fact that a missing downlink can occasionally be
2550 : * encountered in a non-corrupt index. This can be due to an interrupted page
2551 : * split, or an interrupted multi-level page deletion (i.e. there was a hard
2552 : * crash or an error during a page split, or while VACUUM was deleting a
2553 : * multi-level chain of pages).
2554 : *
2555 : * Note that this can only be called in readonly mode, so there is no need to
2556 : * be concerned about concurrent page splits or page deletions.
2557 : */
2558 : static void
2559 0 : bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
2560 : BlockNumber blkno, Page page)
2561 : {
2562 0 : BTPageOpaque opaque = BTPageGetOpaque(page);
2563 : ItemId itemid;
2564 : IndexTuple itup;
2565 : Page child;
2566 : BTPageOpaque copaque;
2567 : uint32 level;
2568 : BlockNumber childblk;
2569 : XLogRecPtr pagelsn;
2570 :
2571 : Assert(state->readonly);
2572 : Assert(!P_IGNORE(opaque));
2573 :
2574 : /* No next level up with downlinks to fingerprint from the true root */
2575 0 : if (P_ISROOT(opaque))
2576 0 : return;
2577 :
2578 0 : pagelsn = PageGetLSN(page);
2579 :
2580 : /*
2581 : * Incomplete (interrupted) page splits can account for the lack of a
2582 : * downlink. Some inserting transaction should eventually complete the
2583 : * page split in passing, when it notices that the left sibling page is
2584 : * P_INCOMPLETE_SPLIT().
2585 : *
2586 : * In general, VACUUM is not prepared for there to be no downlink to a
2587 : * page that it deletes. This is the main reason why the lack of a
2588 : * downlink can be reported as corruption here. It's not obvious that an
2589 : * invalid missing downlink can result in wrong answers to queries,
2590 : * though, since index scans that land on the child may end up
2591 : * consistently moving right. The handling of concurrent page splits (and
2592 : * page deletions) within _bt_moveright() cannot distinguish
2593 : * inconsistencies that last for a moment from inconsistencies that are
2594 : * permanent and irrecoverable.
2595 : *
2596 : * VACUUM isn't even prepared to delete pages that have no downlink due to
2597 : * an incomplete page split, but it can detect and reason about that case
2598 : * by design, so it shouldn't be taken to indicate corruption. See
2599 : * _bt_pagedel() for full details.
2600 : */
2601 0 : if (rightsplit)
2602 : {
2603 0 : ereport(DEBUG1,
2604 : (errcode(ERRCODE_NO_DATA),
2605 : errmsg_internal("harmless interrupted page split detected in index \"%s\"",
2606 : RelationGetRelationName(state->rel)),
2607 : errdetail_internal("Block=%u level=%u left sibling=%u page lsn=%X/%08X.",
2608 : blkno, opaque->btpo_level,
2609 : opaque->btpo_prev,
2610 : LSN_FORMAT_ARGS(pagelsn))));
2611 0 : return;
2612 : }
2613 :
2614 : /*
2615 : * Page under check is probably the "top parent" of a multi-level page
2616 : * deletion. We'll need to descend the subtree to make sure that
2617 : * descendant pages are consistent with that, though.
2618 : *
2619 : * If the page (which must be non-ignorable) is a leaf page, then clearly
2620 : * it can't be the top parent. The lack of a downlink is probably a
2621 : * symptom of a broad problem that could just as easily cause
2622 : * inconsistencies anywhere else.
2623 : */
2624 0 : if (P_ISLEAF(opaque))
2625 0 : ereport(ERROR,
2626 : (errcode(ERRCODE_INDEX_CORRUPTED),
2627 : errmsg("leaf index block lacks downlink in index \"%s\"",
2628 : RelationGetRelationName(state->rel)),
2629 : errdetail_internal("Block=%u page lsn=%X/%08X.",
2630 : blkno,
2631 : LSN_FORMAT_ARGS(pagelsn))));
2632 :
2633 : /* Descend from the given page, which is an internal page */
2634 0 : elog(DEBUG1, "checking for interrupted multi-level deletion due to missing downlink in index \"%s\"",
2635 : RelationGetRelationName(state->rel));
2636 :
2637 0 : level = opaque->btpo_level;
2638 0 : itemid = PageGetItemIdCareful(state, blkno, page, P_FIRSTDATAKEY(opaque));
2639 0 : itup = (IndexTuple) PageGetItem(page, itemid);
2640 0 : childblk = BTreeTupleGetDownLink(itup);
2641 : for (;;)
2642 : {
2643 0 : CHECK_FOR_INTERRUPTS();
2644 :
2645 0 : child = palloc_btree_page(state, childblk);
2646 0 : copaque = BTPageGetOpaque(child);
2647 :
2648 0 : if (P_ISLEAF(copaque))
2649 0 : break;
2650 :
2651 : /* Do an extra sanity check in passing on internal pages */
2652 0 : if (copaque->btpo_level != level - 1)
2653 0 : ereport(ERROR,
2654 : (errcode(ERRCODE_INDEX_CORRUPTED),
2655 : errmsg_internal("downlink points to block in index \"%s\" whose level is not one level down",
2656 : RelationGetRelationName(state->rel)),
2657 : errdetail_internal("Top parent/under check block=%u block pointed to=%u expected level=%u level in pointed to block=%u.",
2658 : blkno, childblk,
2659 : level - 1, copaque->btpo_level)));
2660 :
2661 0 : level = copaque->btpo_level;
2662 0 : itemid = PageGetItemIdCareful(state, childblk, child,
2663 0 : P_FIRSTDATAKEY(copaque));
2664 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2665 0 : childblk = BTreeTupleGetDownLink(itup);
2666 : /* Be slightly more pro-active in freeing this memory, just in case */
2667 0 : pfree(child);
2668 : }
2669 :
2670 : /*
2671 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2672 : * and since a page has no links within other pages (siblings and parent)
2673 : * once it is marked fully deleted, it should be impossible to land on a
2674 : * fully deleted page. See bt_child_check() for further details.
2675 : *
2676 : * The bt_child_check() P_ISDELETED() check is repeated here because
2677 : * bt_child_check() does not visit pages reachable through negative
2678 : * infinity items. Besides, bt_child_check() is unwilling to descend
2679 : * multiple levels. (The similar bt_child_check() P_ISDELETED() check
2680 : * within bt_check_level_from_leftmost() won't reach the page either,
2681 : * since the leaf's live siblings should have their sibling links updated
2682 : * to bypass the deletion target page when it is marked fully dead.)
2683 : *
2684 : * If this error is raised, it might be due to a previous multi-level page
2685 : * deletion that failed to realize that it wasn't yet safe to mark the
2686 : * leaf page as fully dead. A "dangling downlink" will still remain when
2687 : * this happens. The fact that the dangling downlink's page (the leaf's
2688 : * parent/ancestor page) lacked a downlink is incidental.
2689 : */
2690 0 : if (P_ISDELETED(copaque))
2691 0 : ereport(ERROR,
2692 : (errcode(ERRCODE_INDEX_CORRUPTED),
2693 : errmsg_internal("downlink to deleted leaf page found in index \"%s\"",
2694 : RelationGetRelationName(state->rel)),
2695 : errdetail_internal("Top parent/target block=%u leaf block=%u top parent/under check lsn=%X/%08X.",
2696 : blkno, childblk,
2697 : LSN_FORMAT_ARGS(pagelsn))));
2698 :
2699 : /*
2700 : * Iff leaf page is half-dead, its high key top parent link should point
2701 : * to what VACUUM considered to be the top parent page at the instant it
2702 : * was interrupted. Provided the high key link actually points to the
2703 : * page under check, the missing downlink we detected is consistent with
2704 : * there having been an interrupted multi-level page deletion. This means
2705 : * that the subtree with the page under check at its root (a page deletion
2706 : * chain) is in a consistent state, enabling VACUUM to resume deleting the
2707 : * entire chain the next time it encounters the half-dead leaf page.
2708 : */
2709 0 : if (P_ISHALFDEAD(copaque) && !P_RIGHTMOST(copaque))
2710 : {
2711 0 : itemid = PageGetItemIdCareful(state, childblk, child, P_HIKEY);
2712 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2713 0 : if (BTreeTupleGetTopParent(itup) == blkno)
2714 0 : return;
2715 : }
2716 :
2717 0 : ereport(ERROR,
2718 : (errcode(ERRCODE_INDEX_CORRUPTED),
2719 : errmsg("internal index block lacks downlink in index \"%s\"",
2720 : RelationGetRelationName(state->rel)),
2721 : errdetail_internal("Block=%u level=%u page lsn=%X/%08X.",
2722 : blkno, opaque->btpo_level,
2723 : LSN_FORMAT_ARGS(pagelsn))));
2724 : }
2725 :
2726 : /*
2727 : * Per-tuple callback from table_index_build_scan, used to determine if index has
2728 : * all the entries that definitely should have been observed in leaf pages of
2729 : * the target index (that is, all IndexTuples that were fingerprinted by our
2730 : * Bloom filter). All heapallindexed checks occur here.
2731 : *
2732 : * The redundancy between an index and the table it indexes provides a good
2733 : * opportunity to detect corruption, especially corruption within the table.
2734 : * The high level principle behind the verification performed here is that any
2735 : * IndexTuple that should be in an index following a fresh CREATE INDEX (based
2736 : * on the same index definition) should also have been in the original,
2737 : * existing index, which should have used exactly the same representation
2738 : *
2739 : * Since the overall structure of the index has already been verified, the most
2740 : * likely explanation for error here is a corrupt heap page (could be logical
2741 : * or physical corruption). Index corruption may still be detected here,
2742 : * though. Only readonly callers will have verified that left links and right
2743 : * links are in agreement, and so it's possible that a leaf page transposition
2744 : * within index is actually the source of corruption detected here (for
2745 : * !readonly callers). The checks performed only for readonly callers might
2746 : * more accurately frame the problem as a cross-page invariant issue (this
2747 : * could even be due to recovery not replaying all WAL records). The !readonly
2748 : * ERROR message raised here includes a HINT about retrying with readonly
2749 : * verification, just in case it's a cross-page invariant issue, though that
2750 : * isn't particularly likely.
2751 : *
2752 : * table_index_build_scan() expects to be able to find the root tuple when a
2753 : * heap-only tuple (the live tuple at the end of some HOT chain) needs to be
2754 : * indexed, in order to replace the actual tuple's TID with the root tuple's
2755 : * TID (which is what we're actually passed back here). The index build heap
2756 : * scan code will raise an error when a tuple that claims to be the root of the
2757 : * heap-only tuple's HOT chain cannot be located. This catches cases where the
2758 : * original root item offset/root tuple for a HOT chain indicates (for whatever
2759 : * reason) that the entire HOT chain is dead, despite the fact that the latest
2760 : * heap-only tuple should be indexed. When this happens, sequential scans may
2761 : * always give correct answers, and all indexes may be considered structurally
2762 : * consistent (i.e. the nbtree structural checks would not detect corruption).
2763 : * It may be the case that only index scans give wrong answers, and yet heap or
2764 : * SLRU corruption is the real culprit. (While it's true that LP_DEAD bit
2765 : * setting will probably also leave the index in a corrupt state before too
2766 : * long, the problem is nonetheless that there is heap corruption.)
2767 : *
2768 : * Heap-only tuple handling within table_index_build_scan() works in a way that
2769 : * helps us to detect index tuples that contain the wrong values (values that
2770 : * don't match the latest tuple in the HOT chain). This can happen when there
2771 : * is no superseding index tuple due to a faulty assessment of HOT safety,
2772 : * perhaps during the original CREATE INDEX. Because the latest tuple's
2773 : * contents are used with the root TID, an error will be raised when a tuple
2774 : * with the same TID but non-matching attribute values is passed back to us.
2775 : * Faulty assessment of HOT-safety was behind at least two distinct CREATE
2776 : * INDEX CONCURRENTLY bugs that made it into stable releases, one of which was
2777 : * undetected for many years. In short, the same principle that allows a
2778 : * REINDEX to repair corruption when there was an (undetected) broken HOT chain
2779 : * also allows us to detect the corruption in many cases.
2780 : */
2781 : static void
2782 531788 : bt_tuple_present_callback(Relation index, ItemPointer tid, Datum *values,
2783 : bool *isnull, bool tupleIsAlive, void *checkstate)
2784 : {
2785 531788 : BtreeCheckState *state = (BtreeCheckState *) checkstate;
2786 : IndexTuple itup,
2787 : norm;
2788 :
2789 : Assert(state->heapallindexed);
2790 :
2791 : /* Generate a normalized index tuple for fingerprinting */
2792 531788 : itup = index_form_tuple(RelationGetDescr(index), values, isnull);
2793 531788 : itup->t_tid = *tid;
2794 531788 : norm = bt_normalize_tuple(state, itup);
2795 :
2796 : /* Probe Bloom filter -- tuple should be present */
2797 531788 : if (bloom_lacks_element(state->filter, (unsigned char *) norm,
2798 : IndexTupleSize(norm)))
2799 0 : ereport(ERROR,
2800 : (errcode(ERRCODE_DATA_CORRUPTED),
2801 : errmsg("heap tuple (%u,%u) from table \"%s\" lacks matching index tuple within index \"%s\"",
2802 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2803 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2804 : RelationGetRelationName(state->heaprel),
2805 : RelationGetRelationName(state->rel)),
2806 : !state->readonly
2807 : ? errhint("Retrying verification using the function bt_index_parent_check() might provide a more specific error.")
2808 : : 0));
2809 :
2810 531788 : state->heaptuplespresent++;
2811 531788 : pfree(itup);
2812 : /* Cannot leak memory here */
2813 531788 : if (norm != itup)
2814 3 : pfree(norm);
2815 531788 : }
2816 :
2817 : /*
2818 : * Normalize an index tuple for fingerprinting.
2819 : *
2820 : * In general, index tuple formation is assumed to be deterministic by
2821 : * heapallindexed verification, and IndexTuples are assumed immutable. While
2822 : * the LP_DEAD bit is mutable in leaf pages, that's ItemId metadata, which is
2823 : * not fingerprinted. Normalization is required to compensate for corner
2824 : * cases where the determinism assumption doesn't quite work.
2825 : *
2826 : * There is currently one such case: index_form_tuple() does not try to hide
2827 : * the source TOAST state of input datums. The executor applies TOAST
2828 : * compression for heap tuples based on different criteria to the compression
2829 : * applied within btinsert()'s call to index_form_tuple(): it sometimes
2830 : * compresses more aggressively, resulting in compressed heap tuple datums but
2831 : * uncompressed corresponding index tuple datums. A subsequent heapallindexed
2832 : * verification will get a logically equivalent though bitwise unequal tuple
2833 : * from index_form_tuple(). False positive heapallindexed corruption reports
2834 : * could occur without normalizing away the inconsistency.
2835 : *
2836 : * Returned tuple is often caller's own original tuple. Otherwise, it is a
2837 : * new representation of caller's original index tuple, palloc()'d in caller's
2838 : * memory context.
2839 : *
2840 : * Note: This routine is not concerned with distinctions about the
2841 : * representation of tuples beyond those that might break heapallindexed
2842 : * verification. In particular, it won't try to normalize opclass-equal
2843 : * datums with potentially distinct representations (e.g., btree/numeric_ops
2844 : * index datums will not get their display scale normalized-away here).
2845 : * Caller does normalization for non-pivot tuples that have a posting list,
2846 : * since dummy CREATE INDEX callback code generates new tuples with the same
2847 : * normalized representation.
2848 : */
2849 : static IndexTuple
2850 1065970 : bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
2851 : {
2852 1065970 : TupleDesc tupleDescriptor = RelationGetDescr(state->rel);
2853 : Datum normalized[INDEX_MAX_KEYS];
2854 : bool isnull[INDEX_MAX_KEYS];
2855 : bool need_free[INDEX_MAX_KEYS];
2856 1065970 : bool formnewtup = false;
2857 : IndexTuple reformed;
2858 : int i;
2859 :
2860 : /* Caller should only pass "logical" non-pivot tuples here */
2861 : Assert(!BTreeTupleIsPosting(itup) && !BTreeTupleIsPivot(itup));
2862 :
2863 : /* Easy case: It's immediately clear that tuple has no varlena datums */
2864 1065970 : if (!IndexTupleHasVarwidths(itup))
2865 1065946 : return itup;
2866 :
2867 48 : for (i = 0; i < tupleDescriptor->natts; i++)
2868 : {
2869 : Form_pg_attribute att;
2870 :
2871 24 : att = TupleDescAttr(tupleDescriptor, i);
2872 :
2873 : /* Assume untoasted/already normalized datum initially */
2874 24 : need_free[i] = false;
2875 24 : normalized[i] = index_getattr(itup, att->attnum,
2876 : tupleDescriptor,
2877 : &isnull[i]);
2878 24 : if (att->attbyval || att->attlen != -1 || isnull[i])
2879 0 : continue;
2880 :
2881 : /*
2882 : * Callers always pass a tuple that could safely be inserted into the
2883 : * index without further processing, so an external varlena header
2884 : * should never be encountered here
2885 : */
2886 24 : if (VARATT_IS_EXTERNAL(DatumGetPointer(normalized[i])))
2887 0 : ereport(ERROR,
2888 : (errcode(ERRCODE_INDEX_CORRUPTED),
2889 : errmsg("external varlena datum in tuple that references heap row (%u,%u) in index \"%s\"",
2890 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2891 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2892 : RelationGetRelationName(state->rel))));
2893 46 : else if (!VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])) &&
2894 22 : VARSIZE_ANY(DatumGetPointer(normalized[i])) > TOAST_INDEX_TARGET &&
2895 14 : (att->attstorage == TYPSTORAGE_EXTENDED ||
2896 13 : att->attstorage == TYPSTORAGE_MAIN))
2897 : {
2898 : /*
2899 : * This value will be compressed by index_form_tuple() with the
2900 : * current storage settings. We may be here because this tuple
2901 : * was formed with different storage settings. So, force forming.
2902 : */
2903 1 : formnewtup = true;
2904 : }
2905 23 : else if (VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])))
2906 : {
2907 2 : formnewtup = true;
2908 2 : normalized[i] = PointerGetDatum(PG_DETOAST_DATUM(normalized[i]));
2909 2 : need_free[i] = true;
2910 : }
2911 :
2912 : /*
2913 : * Short tuples may have 1B or 4B header. Convert 4B header of short
2914 : * tuples to 1B
2915 : */
2916 21 : else if (VARATT_CAN_MAKE_SHORT(DatumGetPointer(normalized[i])))
2917 : {
2918 : /* convert to short varlena */
2919 1 : Size len = VARATT_CONVERTED_SHORT_SIZE(DatumGetPointer(normalized[i]));
2920 1 : char *data = palloc(len);
2921 :
2922 1 : SET_VARSIZE_SHORT(data, len);
2923 1 : memcpy(data + 1, VARDATA(DatumGetPointer(normalized[i])), len - 1);
2924 :
2925 1 : formnewtup = true;
2926 1 : normalized[i] = PointerGetDatum(data);
2927 1 : need_free[i] = true;
2928 : }
2929 : }
2930 :
2931 : /*
2932 : * Easier case: Tuple has varlena datums, none of which are compressed or
2933 : * short with 4B header
2934 : */
2935 24 : if (!formnewtup)
2936 20 : return itup;
2937 :
2938 : /*
2939 : * Hard case: Tuple had compressed varlena datums that necessitate
2940 : * creating normalized version of the tuple from uncompressed input datums
2941 : * (normalized input datums). This is rather naive, but shouldn't be
2942 : * necessary too often.
2943 : *
2944 : * In the heap, tuples may contain short varlena datums with both 1B
2945 : * header and 4B headers. But the corresponding index tuple should always
2946 : * have such varlena's with 1B headers. So, if there is a short varlena
2947 : * with 4B header, we need to convert it for fingerprinting.
2948 : *
2949 : * Note that we rely on deterministic index_form_tuple() TOAST compression
2950 : * of normalized input.
2951 : */
2952 4 : reformed = index_form_tuple(tupleDescriptor, normalized, isnull);
2953 4 : reformed->t_tid = itup->t_tid;
2954 :
2955 : /* Cannot leak memory here */
2956 8 : for (i = 0; i < tupleDescriptor->natts; i++)
2957 4 : if (need_free[i])
2958 3 : pfree(DatumGetPointer(normalized[i]));
2959 :
2960 4 : return reformed;
2961 : }
2962 :
2963 : /*
2964 : * Produce palloc()'d "plain" tuple for nth posting list entry/TID.
2965 : *
2966 : * In general, deduplication is not supposed to change the logical contents of
2967 : * an index. Multiple index tuples are merged together into one equivalent
2968 : * posting list index tuple when convenient.
2969 : *
2970 : * heapallindexed verification must normalize-away this variation in
2971 : * representation by converting posting list tuples into two or more "plain"
2972 : * tuples. Each tuple must be fingerprinted separately -- there must be one
2973 : * tuple for each corresponding Bloom filter probe during the heap scan.
2974 : *
2975 : * Note: Caller still needs to call bt_normalize_tuple() with returned tuple.
2976 : */
2977 : static inline IndexTuple
2978 27390 : bt_posting_plain_tuple(IndexTuple itup, int n)
2979 : {
2980 : Assert(BTreeTupleIsPosting(itup));
2981 :
2982 : /* Returns non-posting-list tuple */
2983 27390 : return _bt_form_posting(itup, BTreeTupleGetPostingN(itup, n), 1);
2984 : }
2985 :
2986 : /*
2987 : * Search for itup in index, starting from fast root page. itup must be a
2988 : * non-pivot tuple. This is only supported with heapkeyspace indexes, since
2989 : * we rely on having fully unique keys to find a match with only a single
2990 : * visit to a leaf page, barring an interrupted page split, where we may have
2991 : * to move right. (A concurrent page split is impossible because caller must
2992 : * be readonly caller.)
2993 : *
2994 : * This routine can detect very subtle transitive consistency issues across
2995 : * more than one level of the tree. Leaf pages all have a high key (even the
2996 : * rightmost page has a conceptual positive infinity high key), but not a low
2997 : * key. Their downlink in parent is a lower bound, which along with the high
2998 : * key is almost enough to detect every possible inconsistency. A downlink
2999 : * separator key value won't always be available from parent, though, because
3000 : * the first items of internal pages are negative infinity items, truncated
3001 : * down to zero attributes during internal page splits. While it's true that
3002 : * bt_child_check() and the high key check can detect most imaginable key
3003 : * space problems, there are remaining problems it won't detect with non-pivot
3004 : * tuples in cousin leaf pages. Starting a search from the root for every
3005 : * existing leaf tuple detects small inconsistencies in upper levels of the
3006 : * tree that cannot be detected any other way. (Besides all this, this is
3007 : * probably also useful as a direct test of the code used by index scans
3008 : * themselves.)
3009 : */
3010 : static bool
3011 201098 : bt_rootdescend(BtreeCheckState *state, IndexTuple itup)
3012 : {
3013 : BTScanInsert key;
3014 : Buffer lbuf;
3015 : bool exists;
3016 :
3017 201098 : key = _bt_mkscankey(state->rel, itup);
3018 : Assert(key->heapkeyspace && key->scantid != NULL);
3019 :
3020 : /*
3021 : * Search from root.
3022 : *
3023 : * Ideally, we would arrange to only move right within _bt_search() when
3024 : * an interrupted page split is detected (i.e. when the incomplete split
3025 : * bit is found to be set), but for now we accept the possibility that
3026 : * that could conceal an inconsistency.
3027 : */
3028 : Assert(state->readonly && state->rootdescend);
3029 201098 : exists = false;
3030 201098 : _bt_search(state->rel, NULL, key, &lbuf, BT_READ, false);
3031 :
3032 201098 : if (BufferIsValid(lbuf))
3033 : {
3034 : BTInsertStateData insertstate;
3035 : OffsetNumber offnum;
3036 : Page page;
3037 :
3038 201098 : insertstate.itup = itup;
3039 201098 : insertstate.itemsz = MAXALIGN(IndexTupleSize(itup));
3040 201098 : insertstate.itup_key = key;
3041 201098 : insertstate.postingoff = 0;
3042 201098 : insertstate.bounds_valid = false;
3043 201098 : insertstate.buf = lbuf;
3044 :
3045 : /* Get matching tuple on leaf page */
3046 201098 : offnum = _bt_binsrch_insert(state->rel, &insertstate);
3047 : /* Compare first >= matching item on leaf page, if any */
3048 201098 : page = BufferGetPage(lbuf);
3049 : /* Should match on first heap TID when tuple has a posting list */
3050 201098 : if (offnum <= PageGetMaxOffsetNumber(page) &&
3051 402196 : insertstate.postingoff <= 0 &&
3052 201098 : _bt_compare(state->rel, key, page, offnum) == 0)
3053 201098 : exists = true;
3054 201098 : _bt_relbuf(state->rel, lbuf);
3055 : }
3056 :
3057 201098 : pfree(key);
3058 :
3059 201098 : return exists;
3060 : }
3061 :
3062 : /*
3063 : * Is particular offset within page (whose special state is passed by caller)
3064 : * the page negative-infinity item?
3065 : *
3066 : * As noted in comments above _bt_compare(), there is special handling of the
3067 : * first data item as a "negative infinity" item. The hard-coding within
3068 : * _bt_compare() makes comparing this item for the purposes of verification
3069 : * pointless at best, since the IndexTuple only contains a valid TID (a
3070 : * reference TID to child page).
3071 : */
3072 : static inline bool
3073 2649991 : offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
3074 : {
3075 : /*
3076 : * For internal pages only, the first item after high key, if any, is
3077 : * negative infinity item. Internal pages always have a negative infinity
3078 : * item, whereas leaf pages never have one. This implies that negative
3079 : * infinity item is either first or second line item, or there is none
3080 : * within page.
3081 : *
3082 : * Negative infinity items are a special case among pivot tuples. They
3083 : * always have zero attributes, while all other pivot tuples always have
3084 : * nkeyatts attributes.
3085 : *
3086 : * Right-most pages don't have a high key, but could be said to
3087 : * conceptually have a "positive infinity" high key. Thus, there is a
3088 : * symmetry between down link items in parent pages, and high keys in
3089 : * children. Together, they represent the part of the key space that
3090 : * belongs to each page in the index. For example, all children of the
3091 : * root page will have negative infinity as a lower bound from root
3092 : * negative infinity downlink, and positive infinity as an upper bound
3093 : * (implicitly, from "imaginary" positive infinity high key in root).
3094 : */
3095 2649991 : return !P_ISLEAF(opaque) && offset == P_FIRSTDATAKEY(opaque);
3096 : }
3097 :
3098 : /*
3099 : * Does the invariant hold that the key is strictly less than a given upper
3100 : * bound offset item?
3101 : *
3102 : * Verifies line pointer on behalf of caller.
3103 : *
3104 : * If this function returns false, convention is that caller throws error due
3105 : * to corruption.
3106 : */
3107 : static inline bool
3108 2040401 : invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
3109 : OffsetNumber upperbound)
3110 : {
3111 : ItemId itemid;
3112 : int32 cmp;
3113 :
3114 : Assert(!key->nextkey && key->backward);
3115 :
3116 : /* Verify line pointer before checking tuple */
3117 2040401 : itemid = PageGetItemIdCareful(state, state->targetblock, state->target,
3118 : upperbound);
3119 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3120 2040401 : if (!key->heapkeyspace)
3121 0 : return invariant_leq_offset(state, key, upperbound);
3122 :
3123 2040401 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3124 :
3125 : /*
3126 : * _bt_compare() is capable of determining that a scankey with a
3127 : * filled-out attribute is greater than pivot tuples where the comparison
3128 : * is resolved at a truncated attribute (value of attribute in pivot is
3129 : * minus infinity). However, it is not capable of determining that a
3130 : * scankey is _less than_ a tuple on the basis of a comparison resolved at
3131 : * _scankey_ minus infinity attribute. Complete an extra step to simulate
3132 : * having minus infinity values for omitted scankey attribute(s).
3133 : */
3134 2040401 : if (cmp == 0)
3135 : {
3136 : BTPageOpaque topaque;
3137 : IndexTuple ritup;
3138 : int uppnkeyatts;
3139 : ItemPointer rheaptid;
3140 : bool nonpivot;
3141 :
3142 0 : ritup = (IndexTuple) PageGetItem(state->target, itemid);
3143 0 : topaque = BTPageGetOpaque(state->target);
3144 0 : nonpivot = P_ISLEAF(topaque) && upperbound >= P_FIRSTDATAKEY(topaque);
3145 :
3146 : /* Get number of keys + heap TID for item to the right */
3147 0 : uppnkeyatts = BTreeTupleGetNKeyAtts(ritup, state->rel);
3148 0 : rheaptid = BTreeTupleGetHeapTIDCareful(state, ritup, nonpivot);
3149 :
3150 : /* Heap TID is tiebreaker key attribute */
3151 0 : if (key->keysz == uppnkeyatts)
3152 0 : return key->scantid == NULL && rheaptid != NULL;
3153 :
3154 0 : return key->keysz < uppnkeyatts;
3155 : }
3156 :
3157 2040401 : return cmp < 0;
3158 : }
3159 :
3160 : /*
3161 : * Does the invariant hold that the key is less than or equal to a given upper
3162 : * bound offset item?
3163 : *
3164 : * Caller should have verified that upperbound's line pointer is consistent
3165 : * using PageGetItemIdCareful() call.
3166 : *
3167 : * If this function returns false, convention is that caller throws error due
3168 : * to corruption.
3169 : */
3170 : static inline bool
3171 1883143 : invariant_leq_offset(BtreeCheckState *state, BTScanInsert key,
3172 : OffsetNumber upperbound)
3173 : {
3174 : int32 cmp;
3175 :
3176 : Assert(!key->nextkey && key->backward);
3177 :
3178 1883143 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3179 :
3180 1883143 : return cmp <= 0;
3181 : }
3182 :
3183 : /*
3184 : * Does the invariant hold that the key is strictly greater than a given lower
3185 : * bound offset item?
3186 : *
3187 : * Caller should have verified that lowerbound's line pointer is consistent
3188 : * using PageGetItemIdCareful() call.
3189 : *
3190 : * If this function returns false, convention is that caller throws error due
3191 : * to corruption.
3192 : */
3193 : static inline bool
3194 6760 : invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
3195 : OffsetNumber lowerbound)
3196 : {
3197 : int32 cmp;
3198 :
3199 : Assert(!key->nextkey && key->backward);
3200 :
3201 6760 : cmp = _bt_compare(state->rel, key, state->target, lowerbound);
3202 :
3203 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3204 6760 : if (!key->heapkeyspace)
3205 0 : return cmp >= 0;
3206 :
3207 : /*
3208 : * No need to consider the possibility that scankey has attributes that we
3209 : * need to force to be interpreted as negative infinity. _bt_compare() is
3210 : * able to determine that scankey is greater than negative infinity. The
3211 : * distinction between "==" and "<" isn't interesting here, since
3212 : * corruption is indicated either way.
3213 : */
3214 6760 : return cmp > 0;
3215 : }
3216 :
3217 : /*
3218 : * Does the invariant hold that the key is strictly less than a given upper
3219 : * bound offset item, with the offset relating to a caller-supplied page that
3220 : * is not the current target page?
3221 : *
3222 : * Caller's non-target page is a child page of the target, checked as part of
3223 : * checking a property of the target page (i.e. the key comes from the
3224 : * target). Verifies line pointer on behalf of caller.
3225 : *
3226 : * If this function returns false, convention is that caller throws error due
3227 : * to corruption.
3228 : */
3229 : static inline bool
3230 598488 : invariant_l_nontarget_offset(BtreeCheckState *state, BTScanInsert key,
3231 : BlockNumber nontargetblock, Page nontarget,
3232 : OffsetNumber upperbound)
3233 : {
3234 : ItemId itemid;
3235 : int32 cmp;
3236 :
3237 : Assert(!key->nextkey && key->backward);
3238 :
3239 : /* Verify line pointer before checking tuple */
3240 598488 : itemid = PageGetItemIdCareful(state, nontargetblock, nontarget,
3241 : upperbound);
3242 598488 : cmp = _bt_compare(state->rel, key, nontarget, upperbound);
3243 :
3244 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3245 598488 : if (!key->heapkeyspace)
3246 0 : return cmp <= 0;
3247 :
3248 : /* See invariant_l_offset() for an explanation of this extra step */
3249 598488 : if (cmp == 0)
3250 : {
3251 : IndexTuple child;
3252 : int uppnkeyatts;
3253 : ItemPointer childheaptid;
3254 : BTPageOpaque copaque;
3255 : bool nonpivot;
3256 :
3257 1860 : child = (IndexTuple) PageGetItem(nontarget, itemid);
3258 1860 : copaque = BTPageGetOpaque(nontarget);
3259 1860 : nonpivot = P_ISLEAF(copaque) && upperbound >= P_FIRSTDATAKEY(copaque);
3260 :
3261 : /* Get number of keys + heap TID for child/non-target item */
3262 1860 : uppnkeyatts = BTreeTupleGetNKeyAtts(child, state->rel);
3263 1860 : childheaptid = BTreeTupleGetHeapTIDCareful(state, child, nonpivot);
3264 :
3265 : /* Heap TID is tiebreaker key attribute */
3266 1860 : if (key->keysz == uppnkeyatts)
3267 1860 : return key->scantid == NULL && childheaptid != NULL;
3268 :
3269 0 : return key->keysz < uppnkeyatts;
3270 : }
3271 :
3272 596628 : return cmp < 0;
3273 : }
3274 :
3275 : /*
3276 : * Given a block number of a B-Tree page, return page in palloc()'d memory.
3277 : * While at it, perform some basic checks of the page.
3278 : *
3279 : * There is never an attempt to get a consistent view of multiple pages using
3280 : * multiple concurrent buffer locks; in general, we only acquire a single pin
3281 : * and buffer lock at a time, which is often all that the nbtree code requires.
3282 : * (Actually, bt_recheck_sibling_links couples buffer locks, which is the only
3283 : * exception to this general rule.)
3284 : *
3285 : * Operating on a copy of the page is useful because it prevents control
3286 : * getting stuck in an uninterruptible state when an underlying operator class
3287 : * misbehaves.
3288 : */
3289 : static Page
3290 22149 : palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum)
3291 : {
3292 : Buffer buffer;
3293 : Page page;
3294 : BTPageOpaque opaque;
3295 : OffsetNumber maxoffset;
3296 :
3297 22149 : page = palloc(BLCKSZ);
3298 :
3299 : /*
3300 : * We copy the page into local storage to avoid holding pin on the buffer
3301 : * longer than we must.
3302 : */
3303 22149 : buffer = ReadBufferExtended(state->rel, MAIN_FORKNUM, blocknum, RBM_NORMAL,
3304 : state->checkstrategy);
3305 22137 : LockBuffer(buffer, BT_READ);
3306 :
3307 : /*
3308 : * Perform the same basic sanity checking that nbtree itself performs for
3309 : * every page:
3310 : */
3311 22137 : _bt_checkpage(state->rel, buffer);
3312 :
3313 : /* Only use copy of page in palloc()'d memory */
3314 22137 : memcpy(page, BufferGetPage(buffer), BLCKSZ);
3315 22137 : UnlockReleaseBuffer(buffer);
3316 :
3317 22137 : opaque = BTPageGetOpaque(page);
3318 :
3319 22137 : if (P_ISMETA(opaque) && blocknum != BTREE_METAPAGE)
3320 0 : ereport(ERROR,
3321 : (errcode(ERRCODE_INDEX_CORRUPTED),
3322 : errmsg("invalid meta page found at block %u in index \"%s\"",
3323 : blocknum, RelationGetRelationName(state->rel))));
3324 :
3325 : /* Check page from block that ought to be meta page */
3326 22137 : if (blocknum == BTREE_METAPAGE)
3327 : {
3328 4246 : BTMetaPageData *metad = BTPageGetMeta(page);
3329 :
3330 4246 : if (!P_ISMETA(opaque) ||
3331 4246 : metad->btm_magic != BTREE_MAGIC)
3332 0 : ereport(ERROR,
3333 : (errcode(ERRCODE_INDEX_CORRUPTED),
3334 : errmsg("index \"%s\" meta page is corrupt",
3335 : RelationGetRelationName(state->rel))));
3336 :
3337 4246 : if (metad->btm_version < BTREE_MIN_VERSION ||
3338 4246 : metad->btm_version > BTREE_VERSION)
3339 0 : ereport(ERROR,
3340 : (errcode(ERRCODE_INDEX_CORRUPTED),
3341 : errmsg("version mismatch in index \"%s\": file version %d, "
3342 : "current version %d, minimum supported version %d",
3343 : RelationGetRelationName(state->rel),
3344 : metad->btm_version, BTREE_VERSION,
3345 : BTREE_MIN_VERSION)));
3346 :
3347 : /* Finished with metapage checks */
3348 4246 : return page;
3349 : }
3350 :
3351 : /*
3352 : * Deleted pages that still use the old 32-bit XID representation have no
3353 : * sane "level" field because they type pun the field, but all other pages
3354 : * (including pages deleted on Postgres 14+) have a valid value.
3355 : */
3356 17891 : if (!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque))
3357 : {
3358 : /* Okay, no reason not to trust btpo_level field from page */
3359 :
3360 17891 : if (P_ISLEAF(opaque) && opaque->btpo_level != 0)
3361 0 : ereport(ERROR,
3362 : (errcode(ERRCODE_INDEX_CORRUPTED),
3363 : errmsg_internal("invalid leaf page level %u for block %u in index \"%s\"",
3364 : opaque->btpo_level, blocknum,
3365 : RelationGetRelationName(state->rel))));
3366 :
3367 17891 : if (!P_ISLEAF(opaque) && opaque->btpo_level == 0)
3368 0 : ereport(ERROR,
3369 : (errcode(ERRCODE_INDEX_CORRUPTED),
3370 : errmsg_internal("invalid internal page level 0 for block %u in index \"%s\"",
3371 : blocknum,
3372 : RelationGetRelationName(state->rel))));
3373 : }
3374 :
3375 : /*
3376 : * Sanity checks for number of items on page.
3377 : *
3378 : * As noted at the beginning of _bt_binsrch(), an internal page must have
3379 : * children, since there must always be a negative infinity downlink
3380 : * (there may also be a highkey). In the case of non-rightmost leaf
3381 : * pages, there must be at least a highkey. The exceptions are deleted
3382 : * pages, which contain no items.
3383 : *
3384 : * This is correct when pages are half-dead, since internal pages are
3385 : * never half-dead, and leaf pages must have a high key when half-dead
3386 : * (the rightmost page can never be deleted). It's also correct with
3387 : * fully deleted pages: _bt_unlink_halfdead_page() doesn't change anything
3388 : * about the target page other than setting the page as fully dead, and
3389 : * setting its xact field. In particular, it doesn't change the sibling
3390 : * links in the deletion target itself, since they're required when index
3391 : * scans land on the deletion target, and then need to move right (or need
3392 : * to move left, in the case of backward index scans).
3393 : */
3394 17891 : maxoffset = PageGetMaxOffsetNumber(page);
3395 17891 : if (maxoffset > MaxIndexTuplesPerPage)
3396 0 : ereport(ERROR,
3397 : (errcode(ERRCODE_INDEX_CORRUPTED),
3398 : errmsg("Number of items on block %u of index \"%s\" exceeds MaxIndexTuplesPerPage (%u)",
3399 : blocknum, RelationGetRelationName(state->rel),
3400 : MaxIndexTuplesPerPage)));
3401 :
3402 17891 : if (!P_ISLEAF(opaque) && !P_ISDELETED(opaque) && maxoffset < P_FIRSTDATAKEY(opaque))
3403 0 : ereport(ERROR,
3404 : (errcode(ERRCODE_INDEX_CORRUPTED),
3405 : errmsg("internal block %u in index \"%s\" lacks high key and/or at least one downlink",
3406 : blocknum, RelationGetRelationName(state->rel))));
3407 :
3408 17891 : if (P_ISLEAF(opaque) && !P_ISDELETED(opaque) && !P_RIGHTMOST(opaque) && maxoffset < P_HIKEY)
3409 0 : ereport(ERROR,
3410 : (errcode(ERRCODE_INDEX_CORRUPTED),
3411 : errmsg("non-rightmost leaf block %u in index \"%s\" lacks high key item",
3412 : blocknum, RelationGetRelationName(state->rel))));
3413 :
3414 : /*
3415 : * In general, internal pages are never marked half-dead, except on
3416 : * versions of Postgres prior to 9.4, where it can be valid transient
3417 : * state. This state is nonetheless treated as corruption by VACUUM on
3418 : * from version 9.4 on, so do the same here. See _bt_pagedel() for full
3419 : * details.
3420 : */
3421 17891 : if (!P_ISLEAF(opaque) && P_ISHALFDEAD(opaque))
3422 0 : ereport(ERROR,
3423 : (errcode(ERRCODE_INDEX_CORRUPTED),
3424 : errmsg("internal page block %u in index \"%s\" is half-dead",
3425 : blocknum, RelationGetRelationName(state->rel)),
3426 : errhint("This can be caused by an interrupted VACUUM in version 9.3 or older, before upgrade. Please REINDEX it.")));
3427 :
3428 : /*
3429 : * Check that internal pages have no garbage items, and that no page has
3430 : * an invalid combination of deletion-related page level flags
3431 : */
3432 17891 : if (!P_ISLEAF(opaque) && P_HAS_GARBAGE(opaque))
3433 0 : ereport(ERROR,
3434 : (errcode(ERRCODE_INDEX_CORRUPTED),
3435 : errmsg_internal("internal page block %u in index \"%s\" has garbage items",
3436 : blocknum, RelationGetRelationName(state->rel))));
3437 :
3438 17891 : if (P_HAS_FULLXID(opaque) && !P_ISDELETED(opaque))
3439 0 : ereport(ERROR,
3440 : (errcode(ERRCODE_INDEX_CORRUPTED),
3441 : errmsg_internal("full transaction id page flag appears in non-deleted block %u in index \"%s\"",
3442 : blocknum, RelationGetRelationName(state->rel))));
3443 :
3444 17891 : if (P_ISDELETED(opaque) && P_ISHALFDEAD(opaque))
3445 0 : ereport(ERROR,
3446 : (errcode(ERRCODE_INDEX_CORRUPTED),
3447 : errmsg_internal("deleted page block %u in index \"%s\" is half-dead",
3448 : blocknum, RelationGetRelationName(state->rel))));
3449 :
3450 17891 : return page;
3451 : }
3452 :
3453 : /*
3454 : * _bt_mkscankey() wrapper that automatically prevents insertion scankey from
3455 : * being considered greater than the pivot tuple that its values originated
3456 : * from (or some other identical pivot tuple) in the common case where there
3457 : * are truncated/minus infinity attributes. Without this extra step, there
3458 : * are forms of corruption that amcheck could theoretically fail to report.
3459 : *
3460 : * For example, invariant_g_offset() might miss a cross-page invariant failure
3461 : * on an internal level if the scankey built from the first item on the
3462 : * target's right sibling page happened to be equal to (not greater than) the
3463 : * last item on target page. The !backward tiebreaker in _bt_compare() might
3464 : * otherwise cause amcheck to assume (rather than actually verify) that the
3465 : * scankey is greater.
3466 : */
3467 : static inline BTScanInsert
3468 2055843 : bt_mkscankey_pivotsearch(Relation rel, IndexTuple itup)
3469 : {
3470 : BTScanInsert skey;
3471 :
3472 2055843 : skey = _bt_mkscankey(rel, itup);
3473 2055843 : skey->backward = true;
3474 :
3475 2055843 : return skey;
3476 : }
3477 :
3478 : /*
3479 : * PageGetItemId() wrapper that validates returned line pointer.
3480 : *
3481 : * Buffer page/page item access macros generally trust that line pointers are
3482 : * not corrupt, which might cause problems for verification itself. For
3483 : * example, there is no bounds checking in PageGetItem(). Passing it a
3484 : * corrupt line pointer can cause it to return a tuple/pointer that is unsafe
3485 : * to dereference.
3486 : *
3487 : * Validating line pointers before tuples avoids undefined behavior and
3488 : * assertion failures with corrupt indexes, making the verification process
3489 : * more robust and predictable.
3490 : */
3491 : static ItemId
3492 4711931 : PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block, Page page,
3493 : OffsetNumber offset)
3494 : {
3495 4711931 : ItemId itemid = PageGetItemId(page, offset);
3496 :
3497 4711931 : if (ItemIdGetOffset(itemid) + ItemIdGetLength(itemid) >
3498 : BLCKSZ - MAXALIGN(sizeof(BTPageOpaqueData)))
3499 0 : ereport(ERROR,
3500 : (errcode(ERRCODE_INDEX_CORRUPTED),
3501 : errmsg("line pointer points past end of tuple space in index \"%s\"",
3502 : RelationGetRelationName(state->rel)),
3503 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3504 : block, offset, ItemIdGetOffset(itemid),
3505 : ItemIdGetLength(itemid),
3506 : ItemIdGetFlags(itemid))));
3507 :
3508 : /*
3509 : * Verify that line pointer isn't LP_REDIRECT or LP_UNUSED, since nbtree
3510 : * never uses either. Verify that line pointer has storage, too, since
3511 : * even LP_DEAD items should within nbtree.
3512 : */
3513 4711931 : if (ItemIdIsRedirected(itemid) || !ItemIdIsUsed(itemid) ||
3514 4711931 : ItemIdGetLength(itemid) == 0)
3515 0 : ereport(ERROR,
3516 : (errcode(ERRCODE_INDEX_CORRUPTED),
3517 : errmsg("invalid line pointer storage in index \"%s\"",
3518 : RelationGetRelationName(state->rel)),
3519 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3520 : block, offset, ItemIdGetOffset(itemid),
3521 : ItemIdGetLength(itemid),
3522 : ItemIdGetFlags(itemid))));
3523 :
3524 4711931 : return itemid;
3525 : }
3526 :
3527 : /*
3528 : * BTreeTupleGetHeapTID() wrapper that enforces that a heap TID is present in
3529 : * cases where that is mandatory (i.e. for non-pivot tuples)
3530 : */
3531 : static inline ItemPointer
3532 1860 : BTreeTupleGetHeapTIDCareful(BtreeCheckState *state, IndexTuple itup,
3533 : bool nonpivot)
3534 : {
3535 : ItemPointer htid;
3536 :
3537 : /*
3538 : * Caller determines whether this is supposed to be a pivot or non-pivot
3539 : * tuple using page type and item offset number. Verify that tuple
3540 : * metadata agrees with this.
3541 : */
3542 : Assert(state->heapkeyspace);
3543 1860 : if (BTreeTupleIsPivot(itup) && nonpivot)
3544 0 : ereport(ERROR,
3545 : (errcode(ERRCODE_INDEX_CORRUPTED),
3546 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected pivot tuple",
3547 : state->targetblock,
3548 : RelationGetRelationName(state->rel))));
3549 :
3550 1860 : if (!BTreeTupleIsPivot(itup) && !nonpivot)
3551 0 : ereport(ERROR,
3552 : (errcode(ERRCODE_INDEX_CORRUPTED),
3553 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected non-pivot tuple",
3554 : state->targetblock,
3555 : RelationGetRelationName(state->rel))));
3556 :
3557 1860 : htid = BTreeTupleGetHeapTID(itup);
3558 1860 : if (!ItemPointerIsValid(htid) && nonpivot)
3559 0 : ereport(ERROR,
3560 : (errcode(ERRCODE_INDEX_CORRUPTED),
3561 : errmsg("block %u or its right sibling block or child block in index \"%s\" contains non-pivot tuple that lacks a heap TID",
3562 : state->targetblock,
3563 : RelationGetRelationName(state->rel))));
3564 :
3565 1860 : return htid;
3566 : }
3567 :
3568 : /*
3569 : * Return the "pointed to" TID for itup, which is used to generate a
3570 : * descriptive error message. itup must be a "data item" tuple (it wouldn't
3571 : * make much sense to call here with a high key tuple, since there won't be a
3572 : * valid downlink/block number to display).
3573 : *
3574 : * Returns either a heap TID (which will be the first heap TID in posting list
3575 : * if itup is posting list tuple), or a TID that contains downlink block
3576 : * number, plus some encoded metadata (e.g., the number of attributes present
3577 : * in itup).
3578 : */
3579 : static inline ItemPointer
3580 6 : BTreeTupleGetPointsToTID(IndexTuple itup)
3581 : {
3582 : /*
3583 : * Rely on the assumption that !heapkeyspace internal page data items will
3584 : * correctly return TID with downlink here -- BTreeTupleGetHeapTID() won't
3585 : * recognize it as a pivot tuple, but everything still works out because
3586 : * the t_tid field is still returned
3587 : */
3588 6 : if (!BTreeTupleIsPivot(itup))
3589 4 : return BTreeTupleGetHeapTID(itup);
3590 :
3591 : /* Pivot tuple returns TID with downlink block (heapkeyspace variant) */
3592 2 : return &itup->t_tid;
3593 : }
|
