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) 2017-2024, 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 "catalog/index.h"
34 : #include "catalog/pg_am.h"
35 : #include "catalog/pg_opfamily_d.h"
36 : #include "commands/tablecmds.h"
37 : #include "common/pg_prng.h"
38 : #include "lib/bloomfilter.h"
39 : #include "miscadmin.h"
40 : #include "storage/lmgr.h"
41 : #include "storage/smgr.h"
42 : #include "utils/guc.h"
43 : #include "utils/memutils.h"
44 : #include "utils/snapmgr.h"
45 :
46 :
47 642 : PG_MODULE_MAGIC;
48 :
49 : /*
50 : * A B-Tree cannot possibly have this many levels, since there must be one
51 : * block per level, which is bound by the range of BlockNumber:
52 : */
53 : #define InvalidBtreeLevel ((uint32) InvalidBlockNumber)
54 : #define BTreeTupleGetNKeyAtts(itup, rel) \
55 : Min(IndexRelationGetNumberOfKeyAttributes(rel), BTreeTupleGetNAtts(itup, rel))
56 :
57 : /*
58 : * State associated with verifying a B-Tree index
59 : *
60 : * target is the point of reference for a verification operation.
61 : *
62 : * Other B-Tree pages may be allocated, but those are always auxiliary (e.g.,
63 : * they are current target's child pages). Conceptually, problems are only
64 : * ever found in the current target page (or for a particular heap tuple during
65 : * heapallindexed verification). Each page found by verification's left/right,
66 : * top/bottom scan becomes the target exactly once.
67 : */
68 : typedef struct BtreeCheckState
69 : {
70 : /*
71 : * Unchanging state, established at start of verification:
72 : */
73 :
74 : /* B-Tree Index Relation and associated heap relation */
75 : Relation rel;
76 : Relation heaprel;
77 : /* rel is heapkeyspace index? */
78 : bool heapkeyspace;
79 : /* ShareLock held on heap/index, rather than AccessShareLock? */
80 : bool readonly;
81 : /* Also verifying heap has no unindexed tuples? */
82 : bool heapallindexed;
83 : /* Also making sure non-pivot tuples can be found by new search? */
84 : bool rootdescend;
85 : /* Also check uniqueness constraint if index is unique */
86 : bool checkunique;
87 : /* Per-page context */
88 : MemoryContext targetcontext;
89 : /* Buffer access strategy */
90 : BufferAccessStrategy checkstrategy;
91 :
92 : /*
93 : * Info for uniqueness checking. Fill these fields once per index check.
94 : */
95 : IndexInfo *indexinfo;
96 : Snapshot snapshot;
97 :
98 : /*
99 : * Mutable state, for verification of particular page:
100 : */
101 :
102 : /* Current target page */
103 : Page target;
104 : /* Target block number */
105 : BlockNumber targetblock;
106 : /* Target page's LSN */
107 : XLogRecPtr targetlsn;
108 :
109 : /*
110 : * Low key: high key of left sibling of target page. Used only for child
111 : * verification. So, 'lowkey' is kept only when 'readonly' is set.
112 : */
113 : IndexTuple lowkey;
114 :
115 : /*
116 : * The rightlink and incomplete split flag of block one level down to the
117 : * target page, which was visited last time via downlink from target page.
118 : * We use it to check for missing downlinks.
119 : */
120 : BlockNumber prevrightlink;
121 : bool previncompletesplit;
122 :
123 : /*
124 : * Mutable state, for optional heapallindexed verification:
125 : */
126 :
127 : /* Bloom filter fingerprints B-Tree index */
128 : bloom_filter *filter;
129 : /* Debug counter */
130 : int64 heaptuplespresent;
131 : } BtreeCheckState;
132 :
133 : /*
134 : * Starting point for verifying an entire B-Tree index level
135 : */
136 : typedef struct BtreeLevel
137 : {
138 : /* Level number (0 is leaf page level). */
139 : uint32 level;
140 :
141 : /* Left most block on level. Scan of level begins here. */
142 : BlockNumber leftmost;
143 :
144 : /* Is this level reported as "true" root level by meta page? */
145 : bool istruerootlevel;
146 : } BtreeLevel;
147 :
148 182 : PG_FUNCTION_INFO_V1(bt_index_check);
149 120 : PG_FUNCTION_INFO_V1(bt_index_parent_check);
150 :
151 : static void bt_index_check_internal(Oid indrelid, bool parentcheck,
152 : bool heapallindexed, bool rootdescend,
153 : bool checkunique);
154 : static inline void btree_index_checkable(Relation rel);
155 : static inline bool btree_index_mainfork_expected(Relation rel);
156 : static void bt_check_every_level(Relation rel, Relation heaprel,
157 : bool heapkeyspace, bool readonly, bool heapallindexed,
158 : bool rootdescend, bool checkunique);
159 : static BtreeLevel bt_check_level_from_leftmost(BtreeCheckState *state,
160 : BtreeLevel level);
161 : static bool bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
162 : BlockNumber start,
163 : BTPageOpaque start_opaque);
164 : static void bt_recheck_sibling_links(BtreeCheckState *state,
165 : BlockNumber btpo_prev_from_target,
166 : BlockNumber leftcurrent);
167 : static bool heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid);
168 : static void bt_report_duplicate(BtreeCheckState *state, ItemPointer tid,
169 : BlockNumber block, OffsetNumber offset,
170 : int posting, ItemPointer nexttid,
171 : BlockNumber nblock, OffsetNumber noffset,
172 : int nposting);
173 : static void bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
174 : BlockNumber targetblock,
175 : OffsetNumber offset, int *lVis_i,
176 : ItemPointer *lVis_tid,
177 : OffsetNumber *lVis_offset,
178 : BlockNumber *lVis_block);
179 : static void bt_target_page_check(BtreeCheckState *state);
180 : static BTScanInsert bt_right_page_check_scankey(BtreeCheckState *state,
181 : OffsetNumber *rightfirstoffset);
182 : static void bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
183 : OffsetNumber downlinkoffnum);
184 : static void bt_child_highkey_check(BtreeCheckState *state,
185 : OffsetNumber target_downlinkoffnum,
186 : Page loaded_child,
187 : uint32 target_level);
188 : static void bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
189 : BlockNumber blkno, Page page);
190 : static void bt_tuple_present_callback(Relation index, ItemPointer tid,
191 : Datum *values, bool *isnull,
192 : bool tupleIsAlive, void *checkstate);
193 : static IndexTuple bt_normalize_tuple(BtreeCheckState *state,
194 : IndexTuple itup);
195 : static inline IndexTuple bt_posting_plain_tuple(IndexTuple itup, int n);
196 : static bool bt_rootdescend(BtreeCheckState *state, IndexTuple itup);
197 : static inline bool offset_is_negative_infinity(BTPageOpaque opaque,
198 : OffsetNumber offset);
199 : static inline bool invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
200 : OffsetNumber upperbound);
201 : static inline bool invariant_leq_offset(BtreeCheckState *state,
202 : BTScanInsert key,
203 : OffsetNumber upperbound);
204 : static inline bool invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
205 : OffsetNumber lowerbound);
206 : static inline bool invariant_l_nontarget_offset(BtreeCheckState *state,
207 : BTScanInsert key,
208 : BlockNumber nontargetblock,
209 : Page nontarget,
210 : OffsetNumber upperbound);
211 : static Page palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum);
212 : static inline BTScanInsert bt_mkscankey_pivotsearch(Relation rel,
213 : IndexTuple itup);
214 : static ItemId PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block,
215 : Page page, OffsetNumber offset);
216 : static inline ItemPointer BTreeTupleGetHeapTIDCareful(BtreeCheckState *state,
217 : IndexTuple itup, bool nonpivot);
218 : static inline ItemPointer BTreeTupleGetPointsToTID(IndexTuple itup);
219 :
220 : /*
221 : * bt_index_check(index regclass, heapallindexed boolean, checkunique boolean)
222 : *
223 : * Verify integrity of B-Tree index.
224 : *
225 : * Acquires AccessShareLock on heap & index relations. Does not consider
226 : * invariants that exist between parent/child pages. Optionally verifies
227 : * that heap does not contain any unindexed or incorrectly indexed tuples.
228 : */
229 : Datum
230 7950 : bt_index_check(PG_FUNCTION_ARGS)
231 : {
232 7950 : Oid indrelid = PG_GETARG_OID(0);
233 7950 : bool heapallindexed = false;
234 7950 : bool checkunique = false;
235 :
236 7950 : if (PG_NARGS() >= 2)
237 7938 : heapallindexed = PG_GETARG_BOOL(1);
238 7950 : if (PG_NARGS() == 3)
239 1372 : checkunique = PG_GETARG_BOOL(2);
240 :
241 7950 : bt_index_check_internal(indrelid, false, heapallindexed, false, checkunique);
242 :
243 7898 : PG_RETURN_VOID();
244 : }
245 :
246 : /*
247 : * bt_index_parent_check(index regclass, heapallindexed boolean, rootdescend boolean, checkunique boolean)
248 : *
249 : * Verify integrity of B-Tree index.
250 : *
251 : * Acquires ShareLock on heap & index relations. Verifies that downlinks in
252 : * parent pages are valid lower bounds on child pages. Optionally verifies
253 : * that heap does not contain any unindexed or incorrectly indexed tuples.
254 : */
255 : Datum
256 120 : bt_index_parent_check(PG_FUNCTION_ARGS)
257 : {
258 120 : Oid indrelid = PG_GETARG_OID(0);
259 120 : bool heapallindexed = false;
260 120 : bool rootdescend = false;
261 120 : bool checkunique = false;
262 :
263 120 : if (PG_NARGS() >= 2)
264 110 : heapallindexed = PG_GETARG_BOOL(1);
265 120 : if (PG_NARGS() >= 3)
266 104 : rootdescend = PG_GETARG_BOOL(2);
267 120 : if (PG_NARGS() == 4)
268 52 : checkunique = PG_GETARG_BOOL(3);
269 :
270 120 : bt_index_check_internal(indrelid, true, heapallindexed, rootdescend, checkunique);
271 :
272 86 : PG_RETURN_VOID();
273 : }
274 :
275 : /*
276 : * Helper for bt_index_[parent_]check, coordinating the bulk of the work.
277 : */
278 : static void
279 8070 : bt_index_check_internal(Oid indrelid, bool parentcheck, bool heapallindexed,
280 : bool rootdescend, bool checkunique)
281 : {
282 : Oid heapid;
283 : Relation indrel;
284 : Relation heaprel;
285 : LOCKMODE lockmode;
286 : Oid save_userid;
287 : int save_sec_context;
288 : int save_nestlevel;
289 :
290 8070 : if (parentcheck)
291 120 : lockmode = ShareLock;
292 : else
293 7950 : lockmode = AccessShareLock;
294 :
295 : /*
296 : * We must lock table before index to avoid deadlocks. However, if the
297 : * passed indrelid isn't an index then IndexGetRelation() will fail.
298 : * Rather than emitting a not-very-helpful error message, postpone
299 : * complaining, expecting that the is-it-an-index test below will fail.
300 : *
301 : * In hot standby mode this will raise an error when parentcheck is true.
302 : */
303 8070 : heapid = IndexGetRelation(indrelid, true);
304 8070 : if (OidIsValid(heapid))
305 : {
306 8062 : heaprel = table_open(heapid, lockmode);
307 :
308 : /*
309 : * Switch to the table owner's userid, so that any index functions are
310 : * run as that user. Also lock down security-restricted operations
311 : * and arrange to make GUC variable changes local to this command.
312 : */
313 8062 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
314 8062 : SetUserIdAndSecContext(heaprel->rd_rel->relowner,
315 : save_sec_context | SECURITY_RESTRICTED_OPERATION);
316 8062 : save_nestlevel = NewGUCNestLevel();
317 8062 : RestrictSearchPath();
318 : }
319 : else
320 : {
321 8 : heaprel = NULL;
322 : /* Set these just to suppress "uninitialized variable" warnings */
323 8 : save_userid = InvalidOid;
324 8 : save_sec_context = -1;
325 8 : save_nestlevel = -1;
326 : }
327 :
328 : /*
329 : * Open the target index relations separately (like relation_openrv(), but
330 : * with heap relation locked first to prevent deadlocking). In hot
331 : * standby mode this will raise an error when parentcheck is true.
332 : *
333 : * There is no need for the usual indcheckxmin usability horizon test
334 : * here, even in the heapallindexed case, because index undergoing
335 : * verification only needs to have entries for a new transaction snapshot.
336 : * (If this is a parentcheck verification, there is no question about
337 : * committed or recently dead heap tuples lacking index entries due to
338 : * concurrent activity.)
339 : */
340 8070 : indrel = index_open(indrelid, lockmode);
341 :
342 : /*
343 : * Since we did the IndexGetRelation call above without any lock, it's
344 : * barely possible that a race against an index drop/recreation could have
345 : * netted us the wrong table.
346 : */
347 8062 : if (heaprel == NULL || heapid != IndexGetRelation(indrelid, false))
348 0 : ereport(ERROR,
349 : (errcode(ERRCODE_UNDEFINED_TABLE),
350 : errmsg("could not open parent table of index \"%s\"",
351 : RelationGetRelationName(indrel))));
352 :
353 : /* Relation suitable for checking as B-Tree? */
354 8062 : btree_index_checkable(indrel);
355 :
356 8060 : if (btree_index_mainfork_expected(indrel))
357 : {
358 : bool heapkeyspace,
359 : allequalimage;
360 :
361 8060 : if (!smgrexists(RelationGetSmgr(indrel), MAIN_FORKNUM))
362 36 : ereport(ERROR,
363 : (errcode(ERRCODE_INDEX_CORRUPTED),
364 : errmsg("index \"%s\" lacks a main relation fork",
365 : RelationGetRelationName(indrel))));
366 :
367 : /* Extract metadata from metapage, and sanitize it in passing */
368 8024 : _bt_metaversion(indrel, &heapkeyspace, &allequalimage);
369 8024 : if (allequalimage && !heapkeyspace)
370 0 : ereport(ERROR,
371 : (errcode(ERRCODE_INDEX_CORRUPTED),
372 : errmsg("index \"%s\" metapage has equalimage field set on unsupported nbtree version",
373 : RelationGetRelationName(indrel))));
374 8024 : if (allequalimage && !_bt_allequalimage(indrel, false))
375 : {
376 0 : bool has_interval_ops = false;
377 :
378 0 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(indrel); i++)
379 0 : if (indrel->rd_opfamily[i] == INTERVAL_BTREE_FAM_OID)
380 0 : has_interval_ops = true;
381 0 : ereport(ERROR,
382 : (errcode(ERRCODE_INDEX_CORRUPTED),
383 : errmsg("index \"%s\" metapage incorrectly indicates that deduplication is safe",
384 : RelationGetRelationName(indrel)),
385 : has_interval_ops
386 : ? errhint("This is known of \"interval\" indexes last built on a version predating 2023-11.")
387 : : 0));
388 : }
389 :
390 : /* Check index, possibly against table it is an index on */
391 8024 : bt_check_every_level(indrel, heaprel, heapkeyspace, parentcheck,
392 : heapallindexed, rootdescend, checkunique);
393 : }
394 :
395 : /* Roll back any GUC changes executed by index functions */
396 7984 : AtEOXact_GUC(false, save_nestlevel);
397 :
398 : /* Restore userid and security context */
399 7984 : SetUserIdAndSecContext(save_userid, save_sec_context);
400 :
401 : /*
402 : * Release locks early. That's ok here because nothing in the called
403 : * routines will trigger shared cache invalidations to be sent, so we can
404 : * relax the usual pattern of only releasing locks after commit.
405 : */
406 7984 : index_close(indrel, lockmode);
407 7984 : if (heaprel)
408 7984 : table_close(heaprel, lockmode);
409 7984 : }
410 :
411 : /*
412 : * Basic checks about the suitability of a relation for checking as a B-Tree
413 : * index.
414 : *
415 : * NB: Intentionally not checking permissions, the function is normally not
416 : * callable by non-superusers. If granted, it's useful to be able to check a
417 : * whole cluster.
418 : */
419 : static inline void
420 8062 : btree_index_checkable(Relation rel)
421 : {
422 8062 : if (rel->rd_rel->relkind != RELKIND_INDEX ||
423 8062 : rel->rd_rel->relam != BTREE_AM_OID)
424 2 : ereport(ERROR,
425 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
426 : errmsg("only B-Tree indexes are supported as targets for verification"),
427 : errdetail("Relation \"%s\" is not a B-Tree index.",
428 : RelationGetRelationName(rel))));
429 :
430 8060 : if (RELATION_IS_OTHER_TEMP(rel))
431 0 : ereport(ERROR,
432 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
433 : errmsg("cannot access temporary tables of other sessions"),
434 : errdetail("Index \"%s\" is associated with temporary relation.",
435 : RelationGetRelationName(rel))));
436 :
437 8060 : if (!rel->rd_index->indisvalid)
438 0 : ereport(ERROR,
439 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
440 : errmsg("cannot check index \"%s\"",
441 : RelationGetRelationName(rel)),
442 : errdetail("Index is not valid.")));
443 8060 : }
444 :
445 : /*
446 : * Check if B-Tree index relation should have a file for its main relation
447 : * fork. Verification uses this to skip unlogged indexes when in hot standby
448 : * mode, where there is simply nothing to verify. We behave as if the
449 : * relation is empty.
450 : *
451 : * NB: Caller should call btree_index_checkable() before calling here.
452 : */
453 : static inline bool
454 8060 : btree_index_mainfork_expected(Relation rel)
455 : {
456 8060 : if (rel->rd_rel->relpersistence != RELPERSISTENCE_UNLOGGED ||
457 0 : !RecoveryInProgress())
458 8060 : return true;
459 :
460 0 : ereport(DEBUG1,
461 : (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
462 : errmsg("cannot verify unlogged index \"%s\" during recovery, skipping",
463 : RelationGetRelationName(rel))));
464 :
465 0 : return false;
466 : }
467 :
468 : /*
469 : * Main entry point for B-Tree SQL-callable functions. Walks the B-Tree in
470 : * logical order, verifying invariants as it goes. Optionally, verification
471 : * checks if the heap relation contains any tuples that are not represented in
472 : * the index but should be.
473 : *
474 : * It is the caller's responsibility to acquire appropriate heavyweight lock on
475 : * the index relation, and advise us if extra checks are safe when a ShareLock
476 : * is held. (A lock of the same type must also have been acquired on the heap
477 : * relation.)
478 : *
479 : * A ShareLock is generally assumed to prevent any kind of physical
480 : * modification to the index structure, including modifications that VACUUM may
481 : * make. This does not include setting of the LP_DEAD bit by concurrent index
482 : * scans, although that is just metadata that is not able to directly affect
483 : * any check performed here. Any concurrent process that might act on the
484 : * LP_DEAD bit being set (recycle space) requires a heavyweight lock that
485 : * cannot be held while we hold a ShareLock. (Besides, even if that could
486 : * happen, the ad-hoc recycling when a page might otherwise split is performed
487 : * per-page, and requires an exclusive buffer lock, which wouldn't cause us
488 : * trouble. _bt_delitems_vacuum() may only delete leaf items, and so the extra
489 : * parent/child check cannot be affected.)
490 : */
491 : static void
492 8024 : bt_check_every_level(Relation rel, Relation heaprel, bool heapkeyspace,
493 : bool readonly, bool heapallindexed, bool rootdescend,
494 : bool checkunique)
495 : {
496 : BtreeCheckState *state;
497 : Page metapage;
498 : BTMetaPageData *metad;
499 : uint32 previouslevel;
500 : BtreeLevel current;
501 8024 : Snapshot snapshot = SnapshotAny;
502 :
503 8024 : if (!readonly)
504 7926 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\"",
505 : RelationGetRelationName(rel));
506 : else
507 98 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\" with cross-level checks",
508 : RelationGetRelationName(rel));
509 :
510 : /*
511 : * This assertion matches the one in index_getnext_tid(). See page
512 : * recycling/"visible to everyone" notes in nbtree README.
513 : */
514 : Assert(TransactionIdIsValid(RecentXmin));
515 :
516 : /*
517 : * Initialize state for entire verification operation
518 : */
519 8024 : state = palloc0(sizeof(BtreeCheckState));
520 8024 : state->rel = rel;
521 8024 : state->heaprel = heaprel;
522 8024 : state->heapkeyspace = heapkeyspace;
523 8024 : state->readonly = readonly;
524 8024 : state->heapallindexed = heapallindexed;
525 8024 : state->rootdescend = rootdescend;
526 8024 : state->checkunique = checkunique;
527 8024 : state->snapshot = InvalidSnapshot;
528 :
529 8024 : if (state->heapallindexed)
530 : {
531 : int64 total_pages;
532 : int64 total_elems;
533 : uint64 seed;
534 :
535 : /*
536 : * Size Bloom filter based on estimated number of tuples in index,
537 : * while conservatively assuming that each block must contain at least
538 : * MaxTIDsPerBTreePage / 3 "plain" tuples -- see
539 : * bt_posting_plain_tuple() for definition, and details of how posting
540 : * list tuples are handled.
541 : */
542 180 : total_pages = RelationGetNumberOfBlocks(rel);
543 180 : total_elems = Max(total_pages * (MaxTIDsPerBTreePage / 3),
544 : (int64) state->rel->rd_rel->reltuples);
545 : /* Generate a random seed to avoid repetition */
546 180 : seed = pg_prng_uint64(&pg_global_prng_state);
547 : /* Create Bloom filter to fingerprint index */
548 180 : state->filter = bloom_create(total_elems, maintenance_work_mem, seed);
549 180 : state->heaptuplespresent = 0;
550 :
551 : /*
552 : * Register our own snapshot in !readonly case, rather than asking
553 : * table_index_build_scan() to do this for us later. This needs to
554 : * happen before index fingerprinting begins, so we can later be
555 : * certain that index fingerprinting should have reached all tuples
556 : * returned by table_index_build_scan().
557 : */
558 180 : if (!state->readonly)
559 : {
560 126 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
561 :
562 : /*
563 : * GetTransactionSnapshot() always acquires a new MVCC snapshot in
564 : * READ COMMITTED mode. A new snapshot is guaranteed to have all
565 : * the entries it requires in the index.
566 : *
567 : * We must defend against the possibility that an old xact
568 : * snapshot was returned at higher isolation levels when that
569 : * snapshot is not safe for index scans of the target index. This
570 : * is possible when the snapshot sees tuples that are before the
571 : * index's indcheckxmin horizon. Throwing an error here should be
572 : * very rare. It doesn't seem worth using a secondary snapshot to
573 : * avoid this.
574 : */
575 126 : if (IsolationUsesXactSnapshot() && rel->rd_index->indcheckxmin &&
576 0 : !TransactionIdPrecedes(HeapTupleHeaderGetXmin(rel->rd_indextuple->t_data),
577 : snapshot->xmin))
578 0 : ereport(ERROR,
579 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
580 : errmsg("index \"%s\" cannot be verified using transaction snapshot",
581 : RelationGetRelationName(rel))));
582 : }
583 : }
584 :
585 : /*
586 : * We need a snapshot to check the uniqueness of the index. For better
587 : * performance take it once per index check. If snapshot already taken
588 : * reuse it.
589 : */
590 8024 : if (state->checkunique)
591 : {
592 1416 : state->indexinfo = BuildIndexInfo(state->rel);
593 1416 : if (state->indexinfo->ii_Unique)
594 : {
595 1276 : if (snapshot != SnapshotAny)
596 14 : state->snapshot = snapshot;
597 : else
598 1262 : state->snapshot = RegisterSnapshot(GetTransactionSnapshot());
599 : }
600 : }
601 :
602 : Assert(!state->rootdescend || state->readonly);
603 8024 : if (state->rootdescend && !state->heapkeyspace)
604 0 : ereport(ERROR,
605 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
606 : errmsg("cannot verify that tuples from index \"%s\" can each be found by an independent index search",
607 : RelationGetRelationName(rel)),
608 : errhint("Only B-Tree version 4 indexes support rootdescend verification.")));
609 :
610 : /* Create context for page */
611 8024 : state->targetcontext = AllocSetContextCreate(CurrentMemoryContext,
612 : "amcheck context",
613 : ALLOCSET_DEFAULT_SIZES);
614 8024 : state->checkstrategy = GetAccessStrategy(BAS_BULKREAD);
615 :
616 : /* Get true root block from meta-page */
617 8024 : metapage = palloc_btree_page(state, BTREE_METAPAGE);
618 8024 : metad = BTPageGetMeta(metapage);
619 :
620 : /*
621 : * Certain deletion patterns can result in "skinny" B-Tree indexes, where
622 : * the fast root and true root differ.
623 : *
624 : * Start from the true root, not the fast root, unlike conventional index
625 : * scans. This approach is more thorough, and removes the risk of
626 : * following a stale fast root from the meta page.
627 : */
628 8024 : if (metad->btm_fastroot != metad->btm_root)
629 26 : ereport(DEBUG1,
630 : (errcode(ERRCODE_NO_DATA),
631 : errmsg_internal("harmless fast root mismatch in index \"%s\"",
632 : RelationGetRelationName(rel)),
633 : errdetail_internal("Fast root block %u (level %u) differs from true root block %u (level %u).",
634 : metad->btm_fastroot, metad->btm_fastlevel,
635 : metad->btm_root, metad->btm_level)));
636 :
637 : /*
638 : * Starting at the root, verify every level. Move left to right, top to
639 : * bottom. Note that there may be no pages other than the meta page (meta
640 : * page can indicate that root is P_NONE when the index is totally empty).
641 : */
642 8024 : previouslevel = InvalidBtreeLevel;
643 8024 : current.level = metad->btm_level;
644 8024 : current.leftmost = metad->btm_root;
645 8024 : current.istruerootlevel = true;
646 12956 : while (current.leftmost != P_NONE)
647 : {
648 : /*
649 : * Verify this level, and get left most page for next level down, if
650 : * not at leaf level
651 : */
652 4968 : current = bt_check_level_from_leftmost(state, current);
653 :
654 4932 : if (current.leftmost == InvalidBlockNumber)
655 0 : ereport(ERROR,
656 : (errcode(ERRCODE_INDEX_CORRUPTED),
657 : errmsg("index \"%s\" has no valid pages on level below %u or first level",
658 : RelationGetRelationName(rel), previouslevel)));
659 :
660 4932 : previouslevel = current.level;
661 : }
662 :
663 : /*
664 : * * Check whether heap contains unindexed/malformed tuples *
665 : */
666 7988 : if (state->heapallindexed)
667 : {
668 166 : IndexInfo *indexinfo = BuildIndexInfo(state->rel);
669 : TableScanDesc scan;
670 :
671 : /*
672 : * Create our own scan for table_index_build_scan(), rather than
673 : * getting it to do so for us. This is required so that we can
674 : * actually use the MVCC snapshot registered earlier in !readonly
675 : * case.
676 : *
677 : * Note that table_index_build_scan() calls heap_endscan() for us.
678 : */
679 166 : scan = table_beginscan_strat(state->heaprel, /* relation */
680 : snapshot, /* snapshot */
681 : 0, /* number of keys */
682 : NULL, /* scan key */
683 : true, /* buffer access strategy OK */
684 : true); /* syncscan OK? */
685 :
686 : /*
687 : * Scan will behave as the first scan of a CREATE INDEX CONCURRENTLY
688 : * behaves in !readonly case.
689 : *
690 : * It's okay that we don't actually use the same lock strength for the
691 : * heap relation as any other ii_Concurrent caller would in !readonly
692 : * case. We have no reason to care about a concurrent VACUUM
693 : * operation, since there isn't going to be a second scan of the heap
694 : * that needs to be sure that there was no concurrent recycling of
695 : * TIDs.
696 : */
697 162 : indexinfo->ii_Concurrent = !state->readonly;
698 :
699 : /*
700 : * Don't wait for uncommitted tuple xact commit/abort when index is a
701 : * unique index on a catalog (or an index used by an exclusion
702 : * constraint). This could otherwise happen in the readonly case.
703 : */
704 162 : indexinfo->ii_Unique = false;
705 162 : indexinfo->ii_ExclusionOps = NULL;
706 162 : indexinfo->ii_ExclusionProcs = NULL;
707 162 : indexinfo->ii_ExclusionStrats = NULL;
708 :
709 162 : elog(DEBUG1, "verifying that tuples from index \"%s\" are present in \"%s\"",
710 : RelationGetRelationName(state->rel),
711 : RelationGetRelationName(state->heaprel));
712 :
713 162 : table_index_build_scan(state->heaprel, state->rel, indexinfo, true, false,
714 : bt_tuple_present_callback, (void *) state, scan);
715 :
716 162 : ereport(DEBUG1,
717 : (errmsg_internal("finished verifying presence of " INT64_FORMAT " tuples from table \"%s\" with bitset %.2f%% set",
718 : state->heaptuplespresent, RelationGetRelationName(heaprel),
719 : 100.0 * bloom_prop_bits_set(state->filter))));
720 :
721 162 : if (snapshot != SnapshotAny)
722 116 : UnregisterSnapshot(snapshot);
723 :
724 162 : bloom_free(state->filter);
725 : }
726 :
727 : /* Be tidy: */
728 7984 : if (snapshot == SnapshotAny && state->snapshot != InvalidSnapshot)
729 1262 : UnregisterSnapshot(state->snapshot);
730 7984 : MemoryContextDelete(state->targetcontext);
731 7984 : }
732 :
733 : /*
734 : * Given a left-most block at some level, move right, verifying each page
735 : * individually (with more verification across pages for "readonly"
736 : * callers). Caller should pass the true root page as the leftmost initially,
737 : * working their way down by passing what is returned for the last call here
738 : * until level 0 (leaf page level) was reached.
739 : *
740 : * Returns state for next call, if any. This includes left-most block number
741 : * one level lower that should be passed on next level/call, which is set to
742 : * P_NONE on last call here (when leaf level is verified). Level numbers
743 : * follow the nbtree convention: higher levels have higher numbers, because new
744 : * levels are added only due to a root page split. Note that prior to the
745 : * first root page split, the root is also a leaf page, so there is always a
746 : * level 0 (leaf level), and it's always the last level processed.
747 : *
748 : * Note on memory management: State's per-page context is reset here, between
749 : * each call to bt_target_page_check().
750 : */
751 : static BtreeLevel
752 4968 : bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
753 : {
754 : /* State to establish early, concerning entire level */
755 : BTPageOpaque opaque;
756 : MemoryContext oldcontext;
757 : BtreeLevel nextleveldown;
758 :
759 : /* Variables for iterating across level using right links */
760 4968 : BlockNumber leftcurrent = P_NONE;
761 4968 : BlockNumber current = level.leftmost;
762 :
763 : /* Initialize return state */
764 4968 : nextleveldown.leftmost = InvalidBlockNumber;
765 4968 : nextleveldown.level = InvalidBtreeLevel;
766 4968 : nextleveldown.istruerootlevel = false;
767 :
768 : /* Use page-level context for duration of this call */
769 4968 : oldcontext = MemoryContextSwitchTo(state->targetcontext);
770 :
771 4968 : elog(DEBUG1, "verifying level %u%s", level.level,
772 : level.istruerootlevel ?
773 : " (true root level)" : level.level == 0 ? " (leaf level)" : "");
774 :
775 4968 : state->prevrightlink = InvalidBlockNumber;
776 4968 : state->previncompletesplit = false;
777 :
778 : do
779 : {
780 : /* Don't rely on CHECK_FOR_INTERRUPTS() calls at lower level */
781 17862 : CHECK_FOR_INTERRUPTS();
782 :
783 : /* Initialize state for this iteration */
784 17862 : state->targetblock = current;
785 17862 : state->target = palloc_btree_page(state, state->targetblock);
786 17838 : state->targetlsn = PageGetLSN(state->target);
787 :
788 17838 : opaque = BTPageGetOpaque(state->target);
789 :
790 17838 : if (P_IGNORE(opaque))
791 : {
792 : /*
793 : * Since there cannot be a concurrent VACUUM operation in readonly
794 : * mode, and since a page has no links within other pages
795 : * (siblings and parent) once it is marked fully deleted, it
796 : * should be impossible to land on a fully deleted page in
797 : * readonly mode. See bt_child_check() for further details.
798 : *
799 : * The bt_child_check() P_ISDELETED() check is repeated here so
800 : * that pages that are only reachable through sibling links get
801 : * checked.
802 : */
803 0 : if (state->readonly && P_ISDELETED(opaque))
804 0 : ereport(ERROR,
805 : (errcode(ERRCODE_INDEX_CORRUPTED),
806 : errmsg("downlink or sibling link points to deleted block in index \"%s\"",
807 : RelationGetRelationName(state->rel)),
808 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
809 : current, leftcurrent, opaque->btpo_prev)));
810 :
811 0 : if (P_RIGHTMOST(opaque))
812 0 : ereport(ERROR,
813 : (errcode(ERRCODE_INDEX_CORRUPTED),
814 : errmsg("block %u fell off the end of index \"%s\"",
815 : current, RelationGetRelationName(state->rel))));
816 : else
817 0 : ereport(DEBUG1,
818 : (errcode(ERRCODE_NO_DATA),
819 : errmsg_internal("block %u of index \"%s\" concurrently deleted",
820 : current, RelationGetRelationName(state->rel))));
821 0 : goto nextpage;
822 : }
823 17838 : else if (nextleveldown.leftmost == InvalidBlockNumber)
824 : {
825 : /*
826 : * A concurrent page split could make the caller supplied leftmost
827 : * block no longer contain the leftmost page, or no longer be the
828 : * true root, but where that isn't possible due to heavyweight
829 : * locking, check that the first valid page meets caller's
830 : * expectations.
831 : */
832 4944 : if (state->readonly)
833 : {
834 88 : if (!bt_leftmost_ignoring_half_dead(state, current, opaque))
835 0 : ereport(ERROR,
836 : (errcode(ERRCODE_INDEX_CORRUPTED),
837 : errmsg("block %u is not leftmost in index \"%s\"",
838 : current, RelationGetRelationName(state->rel))));
839 :
840 88 : if (level.istruerootlevel && !P_ISROOT(opaque))
841 0 : ereport(ERROR,
842 : (errcode(ERRCODE_INDEX_CORRUPTED),
843 : errmsg("block %u is not true root in index \"%s\"",
844 : current, RelationGetRelationName(state->rel))));
845 : }
846 :
847 : /*
848 : * Before beginning any non-trivial examination of level, prepare
849 : * state for next bt_check_level_from_leftmost() invocation for
850 : * the next level for the next level down (if any).
851 : *
852 : * There should be at least one non-ignorable page per level,
853 : * unless this is the leaf level, which is assumed by caller to be
854 : * final level.
855 : */
856 4944 : if (!P_ISLEAF(opaque))
857 : {
858 : IndexTuple itup;
859 : ItemId itemid;
860 :
861 : /* Internal page -- downlink gets leftmost on next level */
862 1026 : itemid = PageGetItemIdCareful(state, state->targetblock,
863 : state->target,
864 1026 : P_FIRSTDATAKEY(opaque));
865 1026 : itup = (IndexTuple) PageGetItem(state->target, itemid);
866 1026 : nextleveldown.leftmost = BTreeTupleGetDownLink(itup);
867 1026 : nextleveldown.level = opaque->btpo_level - 1;
868 : }
869 : else
870 : {
871 : /*
872 : * Leaf page -- final level caller must process.
873 : *
874 : * Note that this could also be the root page, if there has
875 : * been no root page split yet.
876 : */
877 3918 : nextleveldown.leftmost = P_NONE;
878 3918 : nextleveldown.level = InvalidBtreeLevel;
879 : }
880 :
881 : /*
882 : * Finished setting up state for this call/level. Control will
883 : * never end up back here in any future loop iteration for this
884 : * level.
885 : */
886 : }
887 :
888 : /*
889 : * Sibling links should be in mutual agreement. There arises
890 : * leftcurrent == P_NONE && btpo_prev != P_NONE when the left sibling
891 : * of the parent's low-key downlink is half-dead. (A half-dead page
892 : * has no downlink from its parent.) Under heavyweight locking, the
893 : * last bt_leftmost_ignoring_half_dead() validated this btpo_prev.
894 : * Without heavyweight locking, validation of the P_NONE case remains
895 : * unimplemented.
896 : */
897 17838 : if (opaque->btpo_prev != leftcurrent && leftcurrent != P_NONE)
898 0 : bt_recheck_sibling_links(state, opaque->btpo_prev, leftcurrent);
899 :
900 : /* Check level */
901 17838 : if (level.level != opaque->btpo_level)
902 0 : ereport(ERROR,
903 : (errcode(ERRCODE_INDEX_CORRUPTED),
904 : errmsg("leftmost down link for level points to block in index \"%s\" whose level is not one level down",
905 : RelationGetRelationName(state->rel)),
906 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
907 : current, level.level, opaque->btpo_level)));
908 :
909 : /* Verify invariants for page */
910 17838 : bt_target_page_check(state);
911 :
912 17826 : nextpage:
913 :
914 : /* Try to detect circular links */
915 17826 : if (current == leftcurrent || current == opaque->btpo_prev)
916 0 : ereport(ERROR,
917 : (errcode(ERRCODE_INDEX_CORRUPTED),
918 : errmsg("circular link chain found in block %u of index \"%s\"",
919 : current, RelationGetRelationName(state->rel))));
920 :
921 17826 : leftcurrent = current;
922 17826 : current = opaque->btpo_next;
923 :
924 17826 : if (state->lowkey)
925 : {
926 : Assert(state->readonly);
927 3724 : pfree(state->lowkey);
928 3724 : state->lowkey = NULL;
929 : }
930 :
931 : /*
932 : * Copy current target high key as the low key of right sibling.
933 : * Allocate memory in upper level context, so it would be cleared
934 : * after reset of target context.
935 : *
936 : * We only need the low key in corner cases of checking child high
937 : * keys. We use high key only when incomplete split on the child level
938 : * falls to the boundary of pages on the target level. See
939 : * bt_child_highkey_check() for details. So, typically we won't end
940 : * up doing anything with low key, but it's simpler for general case
941 : * high key verification to always have it available.
942 : *
943 : * The correctness of managing low key in the case of concurrent
944 : * splits wasn't investigated yet. Thankfully we only need low key
945 : * for readonly verification and concurrent splits won't happen.
946 : */
947 17826 : if (state->readonly && !P_RIGHTMOST(opaque))
948 : {
949 : IndexTuple itup;
950 : ItemId itemid;
951 :
952 3724 : itemid = PageGetItemIdCareful(state, state->targetblock,
953 : state->target, P_HIKEY);
954 3724 : itup = (IndexTuple) PageGetItem(state->target, itemid);
955 :
956 3724 : state->lowkey = MemoryContextAlloc(oldcontext, IndexTupleSize(itup));
957 3724 : memcpy(state->lowkey, itup, IndexTupleSize(itup));
958 : }
959 :
960 : /* Free page and associated memory for this iteration */
961 17826 : MemoryContextReset(state->targetcontext);
962 : }
963 17826 : while (current != P_NONE);
964 :
965 4932 : if (state->lowkey)
966 : {
967 : Assert(state->readonly);
968 0 : pfree(state->lowkey);
969 0 : state->lowkey = NULL;
970 : }
971 :
972 : /* Don't change context for caller */
973 4932 : MemoryContextSwitchTo(oldcontext);
974 :
975 4932 : return nextleveldown;
976 : }
977 :
978 : /* Check visibility of the table entry referenced by nbtree index */
979 : static bool
980 278932 : heap_entry_is_visible(BtreeCheckState *state, ItemPointer tid)
981 : {
982 : bool tid_visible;
983 :
984 278932 : TupleTableSlot *slot = table_slot_create(state->heaprel, NULL);
985 :
986 278932 : tid_visible = table_tuple_fetch_row_version(state->heaprel,
987 : tid, state->snapshot, slot);
988 278932 : if (slot != NULL)
989 278932 : ExecDropSingleTupleTableSlot(slot);
990 :
991 278932 : return tid_visible;
992 : }
993 :
994 : /*
995 : * Prepare an error message for unique constrain violation in
996 : * a btree index and report ERROR.
997 : */
998 : static void
999 6 : bt_report_duplicate(BtreeCheckState *state,
1000 : ItemPointer tid, BlockNumber block, OffsetNumber offset,
1001 : int posting,
1002 : ItemPointer nexttid, BlockNumber nblock, OffsetNumber noffset,
1003 : int nposting)
1004 : {
1005 : char *htid,
1006 : *nhtid,
1007 : *itid,
1008 6 : *nitid = "",
1009 6 : *pposting = "",
1010 6 : *pnposting = "";
1011 :
1012 6 : htid = psprintf("tid=(%u,%u)",
1013 : ItemPointerGetBlockNumberNoCheck(tid),
1014 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1015 6 : nhtid = psprintf("tid=(%u,%u)",
1016 : ItemPointerGetBlockNumberNoCheck(nexttid),
1017 6 : ItemPointerGetOffsetNumberNoCheck(nexttid));
1018 6 : itid = psprintf("tid=(%u,%u)", block, offset);
1019 :
1020 6 : if (nblock != block || noffset != offset)
1021 6 : nitid = psprintf(" tid=(%u,%u)", nblock, noffset);
1022 :
1023 6 : if (posting >= 0)
1024 0 : pposting = psprintf(" posting %u", posting);
1025 :
1026 6 : if (nposting >= 0)
1027 0 : pnposting = psprintf(" posting %u", nposting);
1028 :
1029 6 : ereport(ERROR,
1030 : (errcode(ERRCODE_INDEX_CORRUPTED),
1031 : errmsg("index uniqueness is violated for index \"%s\"",
1032 : RelationGetRelationName(state->rel)),
1033 : errdetail("Index %s%s and%s%s (point to heap %s and %s) page lsn=%X/%X.",
1034 : itid, pposting, nitid, pnposting, htid, nhtid,
1035 : LSN_FORMAT_ARGS(state->targetlsn))));
1036 : }
1037 :
1038 : /* Check if current nbtree leaf entry complies with UNIQUE constraint */
1039 : static void
1040 278932 : bt_entry_unique_check(BtreeCheckState *state, IndexTuple itup,
1041 : BlockNumber targetblock, OffsetNumber offset, int *lVis_i,
1042 : ItemPointer *lVis_tid, OffsetNumber *lVis_offset,
1043 : BlockNumber *lVis_block)
1044 : {
1045 : ItemPointer tid;
1046 278932 : bool has_visible_entry = false;
1047 :
1048 : Assert(targetblock != P_NONE);
1049 :
1050 : /*
1051 : * Current tuple has posting list. Report duplicate if TID of any posting
1052 : * list entry is visible and lVis_tid is valid.
1053 : */
1054 278932 : if (BTreeTupleIsPosting(itup))
1055 : {
1056 0 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
1057 : {
1058 0 : tid = BTreeTupleGetPostingN(itup, i);
1059 0 : if (heap_entry_is_visible(state, tid))
1060 : {
1061 0 : has_visible_entry = true;
1062 0 : if (ItemPointerIsValid(*lVis_tid))
1063 : {
1064 0 : bt_report_duplicate(state,
1065 : *lVis_tid, *lVis_block,
1066 0 : *lVis_offset, *lVis_i,
1067 : tid, targetblock,
1068 : offset, i);
1069 : }
1070 :
1071 : /*
1072 : * Prevent double reporting unique constraint violation
1073 : * between the posting list entries of the first tuple on the
1074 : * page after cross-page check.
1075 : */
1076 0 : if (*lVis_block != targetblock && ItemPointerIsValid(*lVis_tid))
1077 0 : return;
1078 :
1079 0 : *lVis_i = i;
1080 0 : *lVis_tid = tid;
1081 0 : *lVis_offset = offset;
1082 0 : *lVis_block = targetblock;
1083 : }
1084 : }
1085 : }
1086 :
1087 : /*
1088 : * Current tuple has no posting list. If TID is visible save info about it
1089 : * for the next comparisons in the loop in bt_page_check(). Report
1090 : * duplicate if lVis_tid is already valid.
1091 : */
1092 : else
1093 : {
1094 278932 : tid = BTreeTupleGetHeapTID(itup);
1095 278932 : if (heap_entry_is_visible(state, tid))
1096 : {
1097 275190 : has_visible_entry = true;
1098 275190 : if (ItemPointerIsValid(*lVis_tid))
1099 : {
1100 6 : bt_report_duplicate(state,
1101 : *lVis_tid, *lVis_block,
1102 6 : *lVis_offset, *lVis_i,
1103 : tid, targetblock,
1104 : offset, -1);
1105 : }
1106 275184 : *lVis_i = -1;
1107 275184 : *lVis_tid = tid;
1108 275184 : *lVis_offset = offset;
1109 275184 : *lVis_block = targetblock;
1110 : }
1111 : }
1112 :
1113 278926 : if (!has_visible_entry && *lVis_block != InvalidBlockNumber &&
1114 0 : *lVis_block != targetblock)
1115 : {
1116 0 : char *posting = "";
1117 :
1118 0 : if (*lVis_i >= 0)
1119 0 : posting = psprintf(" posting %u", *lVis_i);
1120 0 : ereport(DEBUG1,
1121 : (errcode(ERRCODE_NO_DATA),
1122 : errmsg("index uniqueness can not be checked for index tid=(%u,%u) in index \"%s\"",
1123 : targetblock, offset,
1124 : RelationGetRelationName(state->rel)),
1125 : errdetail("It doesn't have visible heap tids and key is equal to the tid=(%u,%u)%s (points to heap tid=(%u,%u)).",
1126 : *lVis_block, *lVis_offset, posting,
1127 : ItemPointerGetBlockNumberNoCheck(*lVis_tid),
1128 : ItemPointerGetOffsetNumberNoCheck(*lVis_tid)),
1129 : errhint("VACUUM the table and repeat the check.")));
1130 : }
1131 : }
1132 :
1133 : /*
1134 : * Like P_LEFTMOST(start_opaque), but accept an arbitrarily-long chain of
1135 : * half-dead, sibling-linked pages to the left. If a half-dead page appears
1136 : * under state->readonly, the database exited recovery between the first-stage
1137 : * and second-stage WAL records of a deletion.
1138 : */
1139 : static bool
1140 110 : bt_leftmost_ignoring_half_dead(BtreeCheckState *state,
1141 : BlockNumber start,
1142 : BTPageOpaque start_opaque)
1143 : {
1144 110 : BlockNumber reached = start_opaque->btpo_prev,
1145 110 : reached_from = start;
1146 110 : bool all_half_dead = true;
1147 :
1148 : /*
1149 : * To handle the !readonly case, we'd need to accept BTP_DELETED pages and
1150 : * potentially observe nbtree/README "Page deletion and backwards scans".
1151 : */
1152 : Assert(state->readonly);
1153 :
1154 114 : while (reached != P_NONE && all_half_dead)
1155 : {
1156 4 : Page page = palloc_btree_page(state, reached);
1157 4 : BTPageOpaque reached_opaque = BTPageGetOpaque(page);
1158 :
1159 4 : CHECK_FOR_INTERRUPTS();
1160 :
1161 : /*
1162 : * Try to detect btpo_prev circular links. _bt_unlink_halfdead_page()
1163 : * writes that side-links will continue to point to the siblings.
1164 : * Check btpo_next for that property.
1165 : */
1166 4 : all_half_dead = P_ISHALFDEAD(reached_opaque) &&
1167 4 : reached != start &&
1168 8 : reached != reached_from &&
1169 4 : reached_opaque->btpo_next == reached_from;
1170 4 : if (all_half_dead)
1171 : {
1172 4 : XLogRecPtr pagelsn = PageGetLSN(page);
1173 :
1174 : /* pagelsn should point to an XLOG_BTREE_MARK_PAGE_HALFDEAD */
1175 4 : ereport(DEBUG1,
1176 : (errcode(ERRCODE_NO_DATA),
1177 : errmsg_internal("harmless interrupted page deletion detected in index \"%s\"",
1178 : RelationGetRelationName(state->rel)),
1179 : errdetail_internal("Block=%u right block=%u page lsn=%X/%X.",
1180 : reached, reached_from,
1181 : LSN_FORMAT_ARGS(pagelsn))));
1182 :
1183 4 : reached_from = reached;
1184 4 : reached = reached_opaque->btpo_prev;
1185 : }
1186 :
1187 4 : pfree(page);
1188 : }
1189 :
1190 110 : return all_half_dead;
1191 : }
1192 :
1193 : /*
1194 : * Raise an error when target page's left link does not point back to the
1195 : * previous target page, called leftcurrent here. The leftcurrent page's
1196 : * right link was followed to get to the current target page, and we expect
1197 : * mutual agreement among leftcurrent and the current target page. Make sure
1198 : * that this condition has definitely been violated in the !readonly case,
1199 : * where concurrent page splits are something that we need to deal with.
1200 : *
1201 : * Cross-page inconsistencies involving pages that don't agree about being
1202 : * siblings are known to be a particularly good indicator of corruption
1203 : * involving partial writes/lost updates. The bt_right_page_check_scankey
1204 : * check also provides a way of detecting cross-page inconsistencies for
1205 : * !readonly callers, but it can only detect sibling pages that have an
1206 : * out-of-order keyspace, which can't catch many of the problems that we
1207 : * expect to catch here.
1208 : *
1209 : * The classic example of the kind of inconsistency that we can only catch
1210 : * with this check (when in !readonly mode) involves three sibling pages that
1211 : * were affected by a faulty page split at some point in the past. The
1212 : * effects of the split are reflected in the original page and its new right
1213 : * sibling page, with a lack of any accompanying changes for the _original_
1214 : * right sibling page. The original right sibling page's left link fails to
1215 : * point to the new right sibling page (its left link still points to the
1216 : * original page), even though the first phase of a page split is supposed to
1217 : * work as a single atomic action. This subtle inconsistency will probably
1218 : * only break backwards scans in practice.
1219 : *
1220 : * Note that this is the only place where amcheck will "couple" buffer locks
1221 : * (and only for !readonly callers). In general we prefer to avoid more
1222 : * thorough cross-page checks in !readonly mode, but it seems worth the
1223 : * complexity here. Also, the performance overhead of performing lock
1224 : * coupling here is negligible in practice. Control only reaches here with a
1225 : * non-corrupt index when there is a concurrent page split at the instant
1226 : * caller crossed over to target page from leftcurrent page.
1227 : */
1228 : static void
1229 0 : bt_recheck_sibling_links(BtreeCheckState *state,
1230 : BlockNumber btpo_prev_from_target,
1231 : BlockNumber leftcurrent)
1232 : {
1233 : /* passing metapage to BTPageGetOpaque() would give irrelevant findings */
1234 : Assert(leftcurrent != P_NONE);
1235 :
1236 0 : if (!state->readonly)
1237 : {
1238 : Buffer lbuf;
1239 : Buffer newtargetbuf;
1240 : Page page;
1241 : BTPageOpaque opaque;
1242 : BlockNumber newtargetblock;
1243 :
1244 : /* Couple locks in the usual order for nbtree: Left to right */
1245 0 : lbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM, leftcurrent,
1246 : RBM_NORMAL, state->checkstrategy);
1247 0 : LockBuffer(lbuf, BT_READ);
1248 0 : _bt_checkpage(state->rel, lbuf);
1249 0 : page = BufferGetPage(lbuf);
1250 0 : opaque = BTPageGetOpaque(page);
1251 0 : if (P_ISDELETED(opaque))
1252 : {
1253 : /*
1254 : * Cannot reason about concurrently deleted page -- the left link
1255 : * in the page to the right is expected to point to some other
1256 : * page to the left (not leftcurrent page).
1257 : *
1258 : * Note that we deliberately don't give up with a half-dead page.
1259 : */
1260 0 : UnlockReleaseBuffer(lbuf);
1261 0 : return;
1262 : }
1263 :
1264 0 : newtargetblock = opaque->btpo_next;
1265 : /* Avoid self-deadlock when newtargetblock == leftcurrent */
1266 0 : if (newtargetblock != leftcurrent)
1267 : {
1268 0 : newtargetbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM,
1269 : newtargetblock, RBM_NORMAL,
1270 : state->checkstrategy);
1271 0 : LockBuffer(newtargetbuf, BT_READ);
1272 0 : _bt_checkpage(state->rel, newtargetbuf);
1273 0 : page = BufferGetPage(newtargetbuf);
1274 0 : opaque = BTPageGetOpaque(page);
1275 : /* btpo_prev_from_target may have changed; update it */
1276 0 : btpo_prev_from_target = opaque->btpo_prev;
1277 : }
1278 : else
1279 : {
1280 : /*
1281 : * leftcurrent right sibling points back to leftcurrent block.
1282 : * Index is corrupt. Easiest way to handle this is to pretend
1283 : * that we actually read from a distinct page that has an invalid
1284 : * block number in its btpo_prev.
1285 : */
1286 0 : newtargetbuf = InvalidBuffer;
1287 0 : btpo_prev_from_target = InvalidBlockNumber;
1288 : }
1289 :
1290 : /*
1291 : * No need to check P_ISDELETED here, since new target block cannot be
1292 : * marked deleted as long as we hold a lock on lbuf
1293 : */
1294 0 : if (BufferIsValid(newtargetbuf))
1295 0 : UnlockReleaseBuffer(newtargetbuf);
1296 0 : UnlockReleaseBuffer(lbuf);
1297 :
1298 0 : if (btpo_prev_from_target == leftcurrent)
1299 : {
1300 : /* Report split in left sibling, not target (or new target) */
1301 0 : ereport(DEBUG1,
1302 : (errcode(ERRCODE_INTERNAL_ERROR),
1303 : errmsg_internal("harmless concurrent page split detected in index \"%s\"",
1304 : RelationGetRelationName(state->rel)),
1305 : errdetail_internal("Block=%u new right sibling=%u original right sibling=%u.",
1306 : leftcurrent, newtargetblock,
1307 : state->targetblock)));
1308 0 : return;
1309 : }
1310 :
1311 : /*
1312 : * Index is corrupt. Make sure that we report correct target page.
1313 : *
1314 : * This could have changed in cases where there was a concurrent page
1315 : * split, as well as index corruption (at least in theory). Note that
1316 : * btpo_prev_from_target was already updated above.
1317 : */
1318 0 : state->targetblock = newtargetblock;
1319 : }
1320 :
1321 0 : ereport(ERROR,
1322 : (errcode(ERRCODE_INDEX_CORRUPTED),
1323 : errmsg("left link/right link pair in index \"%s\" not in agreement",
1324 : RelationGetRelationName(state->rel)),
1325 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
1326 : state->targetblock, leftcurrent,
1327 : btpo_prev_from_target)));
1328 : }
1329 :
1330 : /*
1331 : * Function performs the following checks on target page, or pages ancillary to
1332 : * target page:
1333 : *
1334 : * - That every "real" data item is less than or equal to the high key, which
1335 : * is an upper bound on the items on the page. Data items should be
1336 : * strictly less than the high key when the page is an internal page.
1337 : *
1338 : * - That within the page, every data item is strictly less than the item
1339 : * immediately to its right, if any (i.e., that the items are in order
1340 : * within the page, so that the binary searches performed by index scans are
1341 : * sane).
1342 : *
1343 : * - That the last data item stored on the page is strictly less than the
1344 : * first data item on the page to the right (when such a first item is
1345 : * available).
1346 : *
1347 : * - Various checks on the structure of tuples themselves. For example, check
1348 : * that non-pivot tuples have no truncated attributes.
1349 : *
1350 : * - For index with unique constraint make sure that only one of table entries
1351 : * for equal keys is visible.
1352 : *
1353 : * Furthermore, when state passed shows ShareLock held, function also checks:
1354 : *
1355 : * - That all child pages respect strict lower bound from parent's pivot
1356 : * tuple.
1357 : *
1358 : * - That downlink to block was encountered in parent where that's expected.
1359 : *
1360 : * - That high keys of child pages matches corresponding pivot keys in parent.
1361 : *
1362 : * This is also where heapallindexed callers use their Bloom filter to
1363 : * fingerprint IndexTuples for later table_index_build_scan() verification.
1364 : *
1365 : * Note: Memory allocated in this routine is expected to be released by caller
1366 : * resetting state->targetcontext.
1367 : */
1368 : static void
1369 17838 : bt_target_page_check(BtreeCheckState *state)
1370 : {
1371 : OffsetNumber offset;
1372 : OffsetNumber max;
1373 : BTPageOpaque topaque;
1374 :
1375 : /* last visible entry info for checking indexes with unique constraint */
1376 17838 : int lVis_i = -1; /* the position of last visible item for
1377 : * posting tuple. for non-posting tuple (-1) */
1378 17838 : ItemPointer lVis_tid = NULL;
1379 17838 : BlockNumber lVis_block = InvalidBlockNumber;
1380 17838 : OffsetNumber lVis_offset = InvalidOffsetNumber;
1381 :
1382 17838 : topaque = BTPageGetOpaque(state->target);
1383 17838 : max = PageGetMaxOffsetNumber(state->target);
1384 :
1385 17838 : elog(DEBUG2, "verifying %u items on %s block %u", max,
1386 : P_ISLEAF(topaque) ? "leaf" : "internal", state->targetblock);
1387 :
1388 : /*
1389 : * Check the number of attributes in high key. Note, rightmost page
1390 : * doesn't contain a high key, so nothing to check
1391 : */
1392 17838 : if (!P_RIGHTMOST(topaque))
1393 : {
1394 : ItemId itemid;
1395 : IndexTuple itup;
1396 :
1397 : /* Verify line pointer before checking tuple */
1398 12902 : itemid = PageGetItemIdCareful(state, state->targetblock,
1399 : state->target, P_HIKEY);
1400 12902 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1401 : P_HIKEY))
1402 : {
1403 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1404 0 : ereport(ERROR,
1405 : (errcode(ERRCODE_INDEX_CORRUPTED),
1406 : errmsg("wrong number of high key index tuple attributes in index \"%s\"",
1407 : RelationGetRelationName(state->rel)),
1408 : errdetail_internal("Index block=%u natts=%u block type=%s page lsn=%X/%X.",
1409 : state->targetblock,
1410 : BTreeTupleGetNAtts(itup, state->rel),
1411 : P_ISLEAF(topaque) ? "heap" : "index",
1412 : LSN_FORMAT_ARGS(state->targetlsn))));
1413 : }
1414 : }
1415 :
1416 : /*
1417 : * Loop over page items, starting from first non-highkey item, not high
1418 : * key (if any). Most tests are not performed for the "negative infinity"
1419 : * real item (if any).
1420 : */
1421 3998676 : for (offset = P_FIRSTDATAKEY(topaque);
1422 : offset <= max;
1423 3980838 : offset = OffsetNumberNext(offset))
1424 : {
1425 : ItemId itemid;
1426 : IndexTuple itup;
1427 : size_t tupsize;
1428 : BTScanInsert skey;
1429 : bool lowersizelimit;
1430 : ItemPointer scantid;
1431 :
1432 3980850 : CHECK_FOR_INTERRUPTS();
1433 :
1434 3980850 : itemid = PageGetItemIdCareful(state, state->targetblock,
1435 : state->target, offset);
1436 3980850 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1437 3980850 : tupsize = IndexTupleSize(itup);
1438 :
1439 : /*
1440 : * lp_len should match the IndexTuple reported length exactly, since
1441 : * lp_len is completely redundant in indexes, and both sources of
1442 : * tuple length are MAXALIGN()'d. nbtree does not use lp_len all that
1443 : * frequently, and is surprisingly tolerant of corrupt lp_len fields.
1444 : */
1445 3980850 : if (tupsize != ItemIdGetLength(itemid))
1446 0 : ereport(ERROR,
1447 : (errcode(ERRCODE_INDEX_CORRUPTED),
1448 : errmsg("index tuple size does not equal lp_len in index \"%s\"",
1449 : RelationGetRelationName(state->rel)),
1450 : errdetail_internal("Index tid=(%u,%u) tuple size=%zu lp_len=%u page lsn=%X/%X.",
1451 : state->targetblock, offset,
1452 : tupsize, ItemIdGetLength(itemid),
1453 : LSN_FORMAT_ARGS(state->targetlsn)),
1454 : errhint("This could be a torn page problem.")));
1455 :
1456 : /* Check the number of index tuple attributes */
1457 3980850 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1458 : offset))
1459 : {
1460 : ItemPointer tid;
1461 : char *itid,
1462 : *htid;
1463 :
1464 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1465 0 : tid = BTreeTupleGetPointsToTID(itup);
1466 0 : htid = psprintf("(%u,%u)",
1467 : ItemPointerGetBlockNumberNoCheck(tid),
1468 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1469 :
1470 0 : ereport(ERROR,
1471 : (errcode(ERRCODE_INDEX_CORRUPTED),
1472 : errmsg("wrong number of index tuple attributes in index \"%s\"",
1473 : RelationGetRelationName(state->rel)),
1474 : errdetail_internal("Index tid=%s natts=%u points to %s tid=%s page lsn=%X/%X.",
1475 : itid,
1476 : BTreeTupleGetNAtts(itup, state->rel),
1477 : P_ISLEAF(topaque) ? "heap" : "index",
1478 : htid,
1479 : LSN_FORMAT_ARGS(state->targetlsn))));
1480 : }
1481 :
1482 : /*
1483 : * Don't try to generate scankey using "negative infinity" item on
1484 : * internal pages. They are always truncated to zero attributes.
1485 : */
1486 3980850 : if (offset_is_negative_infinity(topaque, offset))
1487 : {
1488 : /*
1489 : * We don't call bt_child_check() for "negative infinity" items.
1490 : * But if we're performing downlink connectivity check, we do it
1491 : * for every item including "negative infinity" one.
1492 : */
1493 1030 : if (!P_ISLEAF(topaque) && state->readonly)
1494 : {
1495 24 : bt_child_highkey_check(state,
1496 : offset,
1497 : NULL,
1498 : topaque->btpo_level);
1499 : }
1500 1030 : continue;
1501 : }
1502 :
1503 : /*
1504 : * Readonly callers may optionally verify that non-pivot tuples can
1505 : * each be found by an independent search that starts from the root.
1506 : * Note that we deliberately don't do individual searches for each
1507 : * TID, since the posting list itself is validated by other checks.
1508 : */
1509 3979820 : if (state->rootdescend && P_ISLEAF(topaque) &&
1510 402196 : !bt_rootdescend(state, itup))
1511 : {
1512 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1513 : char *itid,
1514 : *htid;
1515 :
1516 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1517 0 : htid = psprintf("(%u,%u)", ItemPointerGetBlockNumber(tid),
1518 0 : ItemPointerGetOffsetNumber(tid));
1519 :
1520 0 : ereport(ERROR,
1521 : (errcode(ERRCODE_INDEX_CORRUPTED),
1522 : errmsg("could not find tuple using search from root page in index \"%s\"",
1523 : RelationGetRelationName(state->rel)),
1524 : errdetail_internal("Index tid=%s points to heap tid=%s page lsn=%X/%X.",
1525 : itid, htid,
1526 : LSN_FORMAT_ARGS(state->targetlsn))));
1527 : }
1528 :
1529 : /*
1530 : * If tuple is a posting list tuple, make sure posting list TIDs are
1531 : * in order
1532 : */
1533 3979820 : if (BTreeTupleIsPosting(itup))
1534 : {
1535 : ItemPointerData last;
1536 : ItemPointer current;
1537 :
1538 20600 : ItemPointerCopy(BTreeTupleGetHeapTID(itup), &last);
1539 :
1540 167292 : for (int i = 1; i < BTreeTupleGetNPosting(itup); i++)
1541 : {
1542 :
1543 146692 : current = BTreeTupleGetPostingN(itup, i);
1544 :
1545 146692 : if (ItemPointerCompare(current, &last) <= 0)
1546 : {
1547 0 : char *itid = psprintf("(%u,%u)", state->targetblock, offset);
1548 :
1549 0 : ereport(ERROR,
1550 : (errcode(ERRCODE_INDEX_CORRUPTED),
1551 : errmsg_internal("posting list contains misplaced TID in index \"%s\"",
1552 : RelationGetRelationName(state->rel)),
1553 : errdetail_internal("Index tid=%s posting list offset=%d page lsn=%X/%X.",
1554 : itid, i,
1555 : LSN_FORMAT_ARGS(state->targetlsn))));
1556 : }
1557 :
1558 146692 : ItemPointerCopy(current, &last);
1559 : }
1560 : }
1561 :
1562 : /* Build insertion scankey for current page offset */
1563 3979820 : skey = bt_mkscankey_pivotsearch(state->rel, itup);
1564 :
1565 : /*
1566 : * Make sure tuple size does not exceed the relevant BTREE_VERSION
1567 : * specific limit.
1568 : *
1569 : * BTREE_VERSION 4 (which introduced heapkeyspace rules) requisitioned
1570 : * a small amount of space from BTMaxItemSize() in order to ensure
1571 : * that suffix truncation always has enough space to add an explicit
1572 : * heap TID back to a tuple -- we pessimistically assume that every
1573 : * newly inserted tuple will eventually need to have a heap TID
1574 : * appended during a future leaf page split, when the tuple becomes
1575 : * the basis of the new high key (pivot tuple) for the leaf page.
1576 : *
1577 : * Since the reclaimed space is reserved for that purpose, we must not
1578 : * enforce the slightly lower limit when the extra space has been used
1579 : * as intended. In other words, there is only a cross-version
1580 : * difference in the limit on tuple size within leaf pages.
1581 : *
1582 : * Still, we're particular about the details within BTREE_VERSION 4
1583 : * internal pages. Pivot tuples may only use the extra space for its
1584 : * designated purpose. Enforce the lower limit for pivot tuples when
1585 : * an explicit heap TID isn't actually present. (In all other cases
1586 : * suffix truncation is guaranteed to generate a pivot tuple that's no
1587 : * larger than the firstright tuple provided to it by its caller.)
1588 : */
1589 7959640 : lowersizelimit = skey->heapkeyspace &&
1590 3979820 : (P_ISLEAF(topaque) || BTreeTupleGetHeapTID(itup) == NULL);
1591 3979852 : if (tupsize > (lowersizelimit ? BTMaxItemSize(state->target) :
1592 32 : BTMaxItemSizeNoHeapTid(state->target)))
1593 : {
1594 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1595 : char *itid,
1596 : *htid;
1597 :
1598 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1599 0 : htid = psprintf("(%u,%u)",
1600 : ItemPointerGetBlockNumberNoCheck(tid),
1601 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1602 :
1603 0 : ereport(ERROR,
1604 : (errcode(ERRCODE_INDEX_CORRUPTED),
1605 : errmsg("index row size %zu exceeds maximum for index \"%s\"",
1606 : tupsize, RelationGetRelationName(state->rel)),
1607 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%X.",
1608 : itid,
1609 : P_ISLEAF(topaque) ? "heap" : "index",
1610 : htid,
1611 : LSN_FORMAT_ARGS(state->targetlsn))));
1612 : }
1613 :
1614 : /* Fingerprint leaf page tuples (those that point to the heap) */
1615 3979820 : if (state->heapallindexed && P_ISLEAF(topaque) && !ItemIdIsDead(itemid))
1616 : {
1617 : IndexTuple norm;
1618 :
1619 1014028 : if (BTreeTupleIsPosting(itup))
1620 : {
1621 : /* Fingerprint all elements as distinct "plain" tuples */
1622 64410 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
1623 : {
1624 : IndexTuple logtuple;
1625 :
1626 63946 : logtuple = bt_posting_plain_tuple(itup, i);
1627 63946 : norm = bt_normalize_tuple(state, logtuple);
1628 63946 : bloom_add_element(state->filter, (unsigned char *) norm,
1629 63946 : IndexTupleSize(norm));
1630 : /* Be tidy */
1631 63946 : if (norm != logtuple)
1632 4 : pfree(norm);
1633 63946 : pfree(logtuple);
1634 : }
1635 : }
1636 : else
1637 : {
1638 1013564 : norm = bt_normalize_tuple(state, itup);
1639 1013564 : bloom_add_element(state->filter, (unsigned char *) norm,
1640 1013564 : IndexTupleSize(norm));
1641 : /* Be tidy */
1642 1013564 : if (norm != itup)
1643 2 : pfree(norm);
1644 : }
1645 : }
1646 :
1647 : /*
1648 : * * High key check *
1649 : *
1650 : * If there is a high key (if this is not the rightmost page on its
1651 : * entire level), check that high key actually is upper bound on all
1652 : * page items. If this is a posting list tuple, we'll need to set
1653 : * scantid to be highest TID in posting list.
1654 : *
1655 : * We prefer to check all items against high key rather than checking
1656 : * just the last and trusting that the operator class obeys the
1657 : * transitive law (which implies that all previous items also
1658 : * respected the high key invariant if they pass the item order
1659 : * check).
1660 : *
1661 : * Ideally, we'd compare every item in the index against every other
1662 : * item in the index, and not trust opclass obedience of the
1663 : * transitive law to bridge the gap between children and their
1664 : * grandparents (as well as great-grandparents, and so on). We don't
1665 : * go to those lengths because that would be prohibitively expensive,
1666 : * and probably not markedly more effective in practice.
1667 : *
1668 : * On the leaf level, we check that the key is <= the highkey.
1669 : * However, on non-leaf levels we check that the key is < the highkey,
1670 : * because the high key is "just another separator" rather than a copy
1671 : * of some existing key item; we expect it to be unique among all keys
1672 : * on the same level. (Suffix truncation will sometimes produce a
1673 : * leaf highkey that is an untruncated copy of the lastleft item, but
1674 : * never any other item, which necessitates weakening the leaf level
1675 : * check to <=.)
1676 : *
1677 : * Full explanation for why a highkey is never truly a copy of another
1678 : * item from the same level on internal levels:
1679 : *
1680 : * While the new left page's high key is copied from the first offset
1681 : * on the right page during an internal page split, that's not the
1682 : * full story. In effect, internal pages are split in the middle of
1683 : * the firstright tuple, not between the would-be lastleft and
1684 : * firstright tuples: the firstright key ends up on the left side as
1685 : * left's new highkey, and the firstright downlink ends up on the
1686 : * right side as right's new "negative infinity" item. The negative
1687 : * infinity tuple is truncated to zero attributes, so we're only left
1688 : * with the downlink. In other words, the copying is just an
1689 : * implementation detail of splitting in the middle of a (pivot)
1690 : * tuple. (See also: "Notes About Data Representation" in the nbtree
1691 : * README.)
1692 : */
1693 3979820 : scantid = skey->scantid;
1694 3979820 : if (state->heapkeyspace && BTreeTupleIsPosting(itup))
1695 20600 : skey->scantid = BTreeTupleGetMaxHeapTID(itup);
1696 :
1697 7633932 : if (!P_RIGHTMOST(topaque) &&
1698 3654112 : !(P_ISLEAF(topaque) ? invariant_leq_offset(state, skey, P_HIKEY) :
1699 1132 : invariant_l_offset(state, skey, P_HIKEY)))
1700 : {
1701 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1702 : char *itid,
1703 : *htid;
1704 :
1705 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1706 0 : htid = psprintf("(%u,%u)",
1707 : ItemPointerGetBlockNumberNoCheck(tid),
1708 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1709 :
1710 0 : ereport(ERROR,
1711 : (errcode(ERRCODE_INDEX_CORRUPTED),
1712 : errmsg("high key invariant violated for index \"%s\"",
1713 : RelationGetRelationName(state->rel)),
1714 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%X.",
1715 : itid,
1716 : P_ISLEAF(topaque) ? "heap" : "index",
1717 : htid,
1718 : LSN_FORMAT_ARGS(state->targetlsn))));
1719 : }
1720 : /* Reset, in case scantid was set to (itup) posting tuple's max TID */
1721 3979820 : skey->scantid = scantid;
1722 :
1723 : /*
1724 : * * Item order check *
1725 : *
1726 : * Check that items are stored on page in logical order, by checking
1727 : * current item is strictly less than next item (if any).
1728 : */
1729 3979820 : if (OffsetNumberNext(offset) <= max &&
1730 3961998 : !invariant_l_offset(state, skey, OffsetNumberNext(offset)))
1731 : {
1732 : ItemPointer tid;
1733 : char *itid,
1734 : *htid,
1735 : *nitid,
1736 : *nhtid;
1737 :
1738 6 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1739 6 : tid = BTreeTupleGetPointsToTID(itup);
1740 6 : htid = psprintf("(%u,%u)",
1741 : ItemPointerGetBlockNumberNoCheck(tid),
1742 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1743 6 : nitid = psprintf("(%u,%u)", state->targetblock,
1744 6 : OffsetNumberNext(offset));
1745 :
1746 : /* Reuse itup to get pointed-to heap location of second item */
1747 6 : itemid = PageGetItemIdCareful(state, state->targetblock,
1748 : state->target,
1749 6 : OffsetNumberNext(offset));
1750 6 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1751 6 : tid = BTreeTupleGetPointsToTID(itup);
1752 6 : nhtid = psprintf("(%u,%u)",
1753 : ItemPointerGetBlockNumberNoCheck(tid),
1754 6 : ItemPointerGetOffsetNumberNoCheck(tid));
1755 :
1756 6 : ereport(ERROR,
1757 : (errcode(ERRCODE_INDEX_CORRUPTED),
1758 : errmsg("item order invariant violated for index \"%s\"",
1759 : RelationGetRelationName(state->rel)),
1760 : errdetail_internal("Lower index tid=%s (points to %s tid=%s) "
1761 : "higher index tid=%s (points to %s tid=%s) "
1762 : "page lsn=%X/%X.",
1763 : itid,
1764 : P_ISLEAF(topaque) ? "heap" : "index",
1765 : htid,
1766 : nitid,
1767 : P_ISLEAF(topaque) ? "heap" : "index",
1768 : nhtid,
1769 : LSN_FORMAT_ARGS(state->targetlsn))));
1770 : }
1771 :
1772 : /*
1773 : * If the index is unique verify entries uniqueness by checking the
1774 : * heap tuples visibility.
1775 : */
1776 3979814 : if (state->checkunique && state->indexinfo->ii_Unique &&
1777 280084 : P_ISLEAF(topaque) && !skey->anynullkeys)
1778 278932 : bt_entry_unique_check(state, itup, state->targetblock, offset,
1779 : &lVis_i, &lVis_tid, &lVis_offset,
1780 : &lVis_block);
1781 :
1782 3979808 : if (state->checkunique && state->indexinfo->ii_Unique &&
1783 280078 : P_ISLEAF(topaque) && OffsetNumberNext(offset) <= max)
1784 : {
1785 : /* Save current scankey tid */
1786 277522 : scantid = skey->scantid;
1787 :
1788 : /*
1789 : * Invalidate scankey tid to make _bt_compare compare only keys in
1790 : * the item to report equality even if heap TIDs are different
1791 : */
1792 277522 : skey->scantid = NULL;
1793 :
1794 : /*
1795 : * If next key tuple is different, invalidate last visible entry
1796 : * data (whole index tuple or last posting in index tuple). Key
1797 : * containing null value does not violate unique constraint and
1798 : * treated as different to any other key.
1799 : */
1800 277522 : if (_bt_compare(state->rel, skey, state->target,
1801 277838 : OffsetNumberNext(offset)) != 0 || skey->anynullkeys)
1802 : {
1803 277346 : lVis_i = -1;
1804 277346 : lVis_tid = NULL;
1805 277346 : lVis_block = InvalidBlockNumber;
1806 277346 : lVis_offset = InvalidOffsetNumber;
1807 : }
1808 277522 : skey->scantid = scantid; /* Restore saved scan key state */
1809 : }
1810 :
1811 : /*
1812 : * * Last item check *
1813 : *
1814 : * Check last item against next/right page's first data item's when
1815 : * last item on page is reached. This additional check will detect
1816 : * transposed pages iff the supposed right sibling page happens to
1817 : * belong before target in the key space. (Otherwise, a subsequent
1818 : * heap verification will probably detect the problem.)
1819 : *
1820 : * This check is similar to the item order check that will have
1821 : * already been performed for every other "real" item on target page
1822 : * when last item is checked. The difference is that the next item
1823 : * (the item that is compared to target's last item) needs to come
1824 : * from the next/sibling page. There may not be such an item
1825 : * available from sibling for various reasons, though (e.g., target is
1826 : * the rightmost page on level).
1827 : */
1828 3979808 : if (offset == max)
1829 : {
1830 : BTScanInsert rightkey;
1831 : BlockNumber rightblock_number;
1832 :
1833 : /* first offset on a right index page (log only) */
1834 17822 : OffsetNumber rightfirstoffset = InvalidOffsetNumber;
1835 :
1836 : /* Get item in next/right page */
1837 17822 : rightkey = bt_right_page_check_scankey(state, &rightfirstoffset);
1838 :
1839 17822 : if (rightkey &&
1840 12894 : !invariant_g_offset(state, rightkey, max))
1841 : {
1842 : /*
1843 : * As explained at length in bt_right_page_check_scankey(),
1844 : * there is a known !readonly race that could account for
1845 : * apparent violation of invariant, which we must check for
1846 : * before actually proceeding with raising error. Our canary
1847 : * condition is that target page was deleted.
1848 : */
1849 0 : if (!state->readonly)
1850 : {
1851 : /* Get fresh copy of target page */
1852 0 : state->target = palloc_btree_page(state, state->targetblock);
1853 : /* Note that we deliberately do not update target LSN */
1854 0 : topaque = BTPageGetOpaque(state->target);
1855 :
1856 : /*
1857 : * All !readonly checks now performed; just return
1858 : */
1859 0 : if (P_IGNORE(topaque))
1860 0 : return;
1861 : }
1862 :
1863 0 : ereport(ERROR,
1864 : (errcode(ERRCODE_INDEX_CORRUPTED),
1865 : errmsg("cross page item order invariant violated for index \"%s\"",
1866 : RelationGetRelationName(state->rel)),
1867 : errdetail_internal("Last item on page tid=(%u,%u) page lsn=%X/%X.",
1868 : state->targetblock, offset,
1869 : LSN_FORMAT_ARGS(state->targetlsn))));
1870 : }
1871 :
1872 : /*
1873 : * If index has unique constraint make sure that no more than one
1874 : * found equal items is visible.
1875 : */
1876 17822 : rightblock_number = topaque->btpo_next;
1877 17822 : if (state->checkunique && state->indexinfo->ii_Unique &&
1878 988 : rightkey && P_ISLEAF(topaque) && rightblock_number != P_NONE)
1879 : {
1880 988 : elog(DEBUG2, "check cross page unique condition");
1881 :
1882 : /*
1883 : * Make _bt_compare compare only index keys without heap TIDs.
1884 : * rightkey->scantid is modified destructively but it is ok
1885 : * for it is not used later.
1886 : */
1887 988 : rightkey->scantid = NULL;
1888 :
1889 : /* The first key on the next page is the same */
1890 988 : if (_bt_compare(state->rel, rightkey, state->target, max) == 0 && !rightkey->anynullkeys)
1891 : {
1892 0 : elog(DEBUG2, "cross page equal keys");
1893 0 : state->target = palloc_btree_page(state,
1894 : rightblock_number);
1895 0 : topaque = BTPageGetOpaque(state->target);
1896 :
1897 0 : if (P_IGNORE(topaque) || !P_ISLEAF(topaque))
1898 : break;
1899 :
1900 0 : itemid = PageGetItemIdCareful(state, rightblock_number,
1901 : state->target,
1902 : rightfirstoffset);
1903 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1904 :
1905 0 : bt_entry_unique_check(state, itup, rightblock_number, rightfirstoffset,
1906 : &lVis_i, &lVis_tid, &lVis_offset,
1907 : &lVis_block);
1908 : }
1909 : }
1910 : }
1911 :
1912 : /*
1913 : * * Downlink check *
1914 : *
1915 : * Additional check of child items iff this is an internal page and
1916 : * caller holds a ShareLock. This happens for every downlink (item)
1917 : * in target excluding the negative-infinity downlink (again, this is
1918 : * because it has no useful value to compare).
1919 : */
1920 3979808 : if (!P_ISLEAF(topaque) && state->readonly)
1921 3722 : bt_child_check(state, skey, offset);
1922 : }
1923 :
1924 : /*
1925 : * Special case bt_child_highkey_check() call
1926 : *
1927 : * We don't pass a real downlink, but we've to finish the level
1928 : * processing. If condition is satisfied, we've already processed all the
1929 : * downlinks from the target level. But there still might be pages to the
1930 : * right of the child page pointer to by our rightmost downlink. And they
1931 : * might have missing downlinks. This final call checks for them.
1932 : */
1933 17826 : if (!P_ISLEAF(topaque) && P_RIGHTMOST(topaque) && state->readonly)
1934 : {
1935 22 : bt_child_highkey_check(state, InvalidOffsetNumber,
1936 : NULL, topaque->btpo_level);
1937 : }
1938 : }
1939 :
1940 : /*
1941 : * Return a scankey for an item on page to right of current target (or the
1942 : * first non-ignorable page), sufficient to check ordering invariant on last
1943 : * item in current target page. Returned scankey relies on local memory
1944 : * allocated for the child page, which caller cannot pfree(). Caller's memory
1945 : * context should be reset between calls here.
1946 : *
1947 : * This is the first data item, and so all adjacent items are checked against
1948 : * their immediate sibling item (which may be on a sibling page, or even a
1949 : * "cousin" page at parent boundaries where target's rightlink points to page
1950 : * with different parent page). If no such valid item is available, return
1951 : * NULL instead.
1952 : *
1953 : * Note that !readonly callers must reverify that target page has not
1954 : * been concurrently deleted.
1955 : *
1956 : * Save rightfirstdataoffset for detailed error message.
1957 : */
1958 : static BTScanInsert
1959 17822 : bt_right_page_check_scankey(BtreeCheckState *state, OffsetNumber *rightfirstoffset)
1960 : {
1961 : BTPageOpaque opaque;
1962 : ItemId rightitem;
1963 : IndexTuple firstitup;
1964 : BlockNumber targetnext;
1965 : Page rightpage;
1966 : OffsetNumber nline;
1967 :
1968 : /* Determine target's next block number */
1969 17822 : opaque = BTPageGetOpaque(state->target);
1970 :
1971 : /* If target is already rightmost, no right sibling; nothing to do here */
1972 17822 : if (P_RIGHTMOST(opaque))
1973 4928 : return NULL;
1974 :
1975 : /*
1976 : * General notes on concurrent page splits and page deletion:
1977 : *
1978 : * Routines like _bt_search() don't require *any* page split interlock
1979 : * when descending the tree, including something very light like a buffer
1980 : * pin. That's why it's okay that we don't either. This avoidance of any
1981 : * need to "couple" buffer locks is the raison d' etre of the Lehman & Yao
1982 : * algorithm, in fact.
1983 : *
1984 : * That leaves deletion. A deleted page won't actually be recycled by
1985 : * VACUUM early enough for us to fail to at least follow its right link
1986 : * (or left link, or downlink) and find its sibling, because recycling
1987 : * does not occur until no possible index scan could land on the page.
1988 : * Index scans can follow links with nothing more than their snapshot as
1989 : * an interlock and be sure of at least that much. (See page
1990 : * recycling/"visible to everyone" notes in nbtree README.)
1991 : *
1992 : * Furthermore, it's okay if we follow a rightlink and find a half-dead or
1993 : * dead (ignorable) page one or more times. There will either be a
1994 : * further right link to follow that leads to a live page before too long
1995 : * (before passing by parent's rightmost child), or we will find the end
1996 : * of the entire level instead (possible when parent page is itself the
1997 : * rightmost on its level).
1998 : */
1999 12894 : targetnext = opaque->btpo_next;
2000 : for (;;)
2001 : {
2002 12894 : CHECK_FOR_INTERRUPTS();
2003 :
2004 12894 : rightpage = palloc_btree_page(state, targetnext);
2005 12894 : opaque = BTPageGetOpaque(rightpage);
2006 :
2007 12894 : if (!P_IGNORE(opaque) || P_RIGHTMOST(opaque))
2008 : break;
2009 :
2010 : /*
2011 : * We landed on a deleted or half-dead sibling page. Step right until
2012 : * we locate a live sibling page.
2013 : */
2014 0 : ereport(DEBUG2,
2015 : (errcode(ERRCODE_NO_DATA),
2016 : errmsg_internal("level %u sibling page in block %u of index \"%s\" was found deleted or half dead",
2017 : opaque->btpo_level, targetnext, RelationGetRelationName(state->rel)),
2018 : errdetail_internal("Deleted page found when building scankey from right sibling.")));
2019 :
2020 0 : targetnext = opaque->btpo_next;
2021 :
2022 : /* Be slightly more pro-active in freeing this memory, just in case */
2023 0 : pfree(rightpage);
2024 : }
2025 :
2026 : /*
2027 : * No ShareLock held case -- why it's safe to proceed.
2028 : *
2029 : * Problem:
2030 : *
2031 : * We must avoid false positive reports of corruption when caller treats
2032 : * item returned here as an upper bound on target's last item. In
2033 : * general, false positives are disallowed. Avoiding them here when
2034 : * caller is !readonly is subtle.
2035 : *
2036 : * A concurrent page deletion by VACUUM of the target page can result in
2037 : * the insertion of items on to this right sibling page that would
2038 : * previously have been inserted on our target page. There might have
2039 : * been insertions that followed the target's downlink after it was made
2040 : * to point to right sibling instead of target by page deletion's first
2041 : * phase. The inserters insert items that would belong on target page.
2042 : * This race is very tight, but it's possible. This is our only problem.
2043 : *
2044 : * Non-problems:
2045 : *
2046 : * We are not hindered by a concurrent page split of the target; we'll
2047 : * never land on the second half of the page anyway. A concurrent split
2048 : * of the right page will also not matter, because the first data item
2049 : * remains the same within the left half, which we'll reliably land on. If
2050 : * we had to skip over ignorable/deleted pages, it cannot matter because
2051 : * their key space has already been atomically merged with the first
2052 : * non-ignorable page we eventually find (doesn't matter whether the page
2053 : * we eventually find is a true sibling or a cousin of target, which we go
2054 : * into below).
2055 : *
2056 : * Solution:
2057 : *
2058 : * Caller knows that it should reverify that target is not ignorable
2059 : * (half-dead or deleted) when cross-page sibling item comparison appears
2060 : * to indicate corruption (invariant fails). This detects the single race
2061 : * condition that exists for caller. This is correct because the
2062 : * continued existence of target block as non-ignorable (not half-dead or
2063 : * deleted) implies that target page was not merged into from the right by
2064 : * deletion; the key space at or after target never moved left. Target's
2065 : * parent either has the same downlink to target as before, or a <
2066 : * downlink due to deletion at the left of target. Target either has the
2067 : * same highkey as before, or a highkey < before when there is a page
2068 : * split. (The rightmost concurrently-split-from-target-page page will
2069 : * still have the same highkey as target was originally found to have,
2070 : * which for our purposes is equivalent to target's highkey itself never
2071 : * changing, since we reliably skip over
2072 : * concurrently-split-from-target-page pages.)
2073 : *
2074 : * In simpler terms, we allow that the key space of the target may expand
2075 : * left (the key space can move left on the left side of target only), but
2076 : * the target key space cannot expand right and get ahead of us without
2077 : * our detecting it. The key space of the target cannot shrink, unless it
2078 : * shrinks to zero due to the deletion of the original page, our canary
2079 : * condition. (To be very precise, we're a bit stricter than that because
2080 : * it might just have been that the target page split and only the
2081 : * original target page was deleted. We can be more strict, just not more
2082 : * lax.)
2083 : *
2084 : * Top level tree walk caller moves on to next page (makes it the new
2085 : * target) following recovery from this race. (cf. The rationale for
2086 : * child/downlink verification needing a ShareLock within
2087 : * bt_child_check(), where page deletion is also the main source of
2088 : * trouble.)
2089 : *
2090 : * Note that it doesn't matter if right sibling page here is actually a
2091 : * cousin page, because in order for the key space to be readjusted in a
2092 : * way that causes us issues in next level up (guiding problematic
2093 : * concurrent insertions to the cousin from the grandparent rather than to
2094 : * the sibling from the parent), there'd have to be page deletion of
2095 : * target's parent page (affecting target's parent's downlink in target's
2096 : * grandparent page). Internal page deletion only occurs when there are
2097 : * no child pages (they were all fully deleted), and caller is checking
2098 : * that the target's parent has at least one non-deleted (so
2099 : * non-ignorable) child: the target page. (Note that the first phase of
2100 : * deletion atomically marks the page to be deleted half-dead/ignorable at
2101 : * the same time downlink in its parent is removed, so caller will
2102 : * definitely not fail to detect that this happened.)
2103 : *
2104 : * This trick is inspired by the method backward scans use for dealing
2105 : * with concurrent page splits; concurrent page deletion is a problem that
2106 : * similarly receives special consideration sometimes (it's possible that
2107 : * the backwards scan will re-read its "original" block after failing to
2108 : * find a right-link to it, having already moved in the opposite direction
2109 : * (right/"forwards") a few times to try to locate one). Just like us,
2110 : * that happens only to determine if there was a concurrent page deletion
2111 : * of a reference page, and just like us if there was a page deletion of
2112 : * that reference page it means we can move on from caring about the
2113 : * reference page. See the nbtree README for a full description of how
2114 : * that works.
2115 : */
2116 12894 : nline = PageGetMaxOffsetNumber(rightpage);
2117 :
2118 : /*
2119 : * Get first data item, if any
2120 : */
2121 12894 : if (P_ISLEAF(opaque) && nline >= P_FIRSTDATAKEY(opaque))
2122 : {
2123 : /* Return first data item (if any) */
2124 12890 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2125 12890 : P_FIRSTDATAKEY(opaque));
2126 12890 : *rightfirstoffset = P_FIRSTDATAKEY(opaque);
2127 : }
2128 8 : else if (!P_ISLEAF(opaque) &&
2129 4 : nline >= OffsetNumberNext(P_FIRSTDATAKEY(opaque)))
2130 : {
2131 : /*
2132 : * Return first item after the internal page's "negative infinity"
2133 : * item
2134 : */
2135 4 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
2136 4 : OffsetNumberNext(P_FIRSTDATAKEY(opaque)));
2137 : }
2138 : else
2139 : {
2140 : /*
2141 : * No first item. Page is probably empty leaf page, but it's also
2142 : * possible that it's an internal page with only a negative infinity
2143 : * item.
2144 : */
2145 0 : ereport(DEBUG2,
2146 : (errcode(ERRCODE_NO_DATA),
2147 : errmsg_internal("%s block %u of index \"%s\" has no first data item",
2148 : P_ISLEAF(opaque) ? "leaf" : "internal", targetnext,
2149 : RelationGetRelationName(state->rel))));
2150 0 : return NULL;
2151 : }
2152 :
2153 : /*
2154 : * Return first real item scankey. Note that this relies on right page
2155 : * memory remaining allocated.
2156 : */
2157 12894 : firstitup = (IndexTuple) PageGetItem(rightpage, rightitem);
2158 12894 : return bt_mkscankey_pivotsearch(state->rel, firstitup);
2159 : }
2160 :
2161 : /*
2162 : * Check if two tuples are binary identical except the block number. So,
2163 : * this function is capable to compare pivot keys on different levels.
2164 : */
2165 : static bool
2166 3724 : bt_pivot_tuple_identical(bool heapkeyspace, IndexTuple itup1, IndexTuple itup2)
2167 : {
2168 3724 : if (IndexTupleSize(itup1) != IndexTupleSize(itup2))
2169 0 : return false;
2170 :
2171 3724 : if (heapkeyspace)
2172 : {
2173 : /*
2174 : * Offset number will contain important information in heapkeyspace
2175 : * indexes: the number of attributes left in the pivot tuple following
2176 : * suffix truncation. Don't skip over it (compare it too).
2177 : */
2178 3724 : if (memcmp(&itup1->t_tid.ip_posid, &itup2->t_tid.ip_posid,
2179 3724 : IndexTupleSize(itup1) -
2180 : offsetof(ItemPointerData, ip_posid)) != 0)
2181 0 : return false;
2182 : }
2183 : else
2184 : {
2185 : /*
2186 : * Cannot rely on offset number field having consistent value across
2187 : * levels on pg_upgrade'd !heapkeyspace indexes. Compare contents of
2188 : * tuple starting from just after item pointer (i.e. after block
2189 : * number and offset number).
2190 : */
2191 0 : if (memcmp(&itup1->t_info, &itup2->t_info,
2192 0 : IndexTupleSize(itup1) -
2193 : offsetof(IndexTupleData, t_info)) != 0)
2194 0 : return false;
2195 : }
2196 :
2197 3724 : return true;
2198 : }
2199 :
2200 : /*---
2201 : * Check high keys on the child level. Traverse rightlinks from previous
2202 : * downlink to the current one. Check that there are no intermediate pages
2203 : * with missing downlinks.
2204 : *
2205 : * If 'loaded_child' is given, it's assumed to be the page pointed to by the
2206 : * downlink referenced by 'downlinkoffnum' of the target page.
2207 : *
2208 : * Basically this function is called for each target downlink and checks two
2209 : * invariants:
2210 : *
2211 : * 1) You can reach the next child from previous one via rightlinks;
2212 : * 2) Each child high key have matching pivot key on target level.
2213 : *
2214 : * Consider the sample tree picture.
2215 : *
2216 : * 1
2217 : * / \
2218 : * 2 <-> 3
2219 : * / \ / \
2220 : * 4 <> 5 <> 6 <> 7 <> 8
2221 : *
2222 : * This function will be called for blocks 4, 5, 6 and 8. Consider what is
2223 : * happening for each function call.
2224 : *
2225 : * - The function call for block 4 initializes data structure and matches high
2226 : * key of block 4 to downlink's pivot key of block 2.
2227 : * - The high key of block 5 is matched to the high key of block 2.
2228 : * - The block 6 has an incomplete split flag set, so its high key isn't
2229 : * matched to anything.
2230 : * - The function call for block 8 checks that block 8 can be found while
2231 : * following rightlinks from block 6. The high key of block 7 will be
2232 : * matched to downlink's pivot key in block 3.
2233 : *
2234 : * There is also final call of this function, which checks that there is no
2235 : * missing downlinks for children to the right of the child referenced by
2236 : * rightmost downlink in target level.
2237 : */
2238 : static void
2239 3768 : bt_child_highkey_check(BtreeCheckState *state,
2240 : OffsetNumber target_downlinkoffnum,
2241 : Page loaded_child,
2242 : uint32 target_level)
2243 : {
2244 3768 : BlockNumber blkno = state->prevrightlink;
2245 : Page page;
2246 : BTPageOpaque opaque;
2247 3768 : bool rightsplit = state->previncompletesplit;
2248 3768 : bool first = true;
2249 : ItemId itemid;
2250 : IndexTuple itup;
2251 : BlockNumber downlink;
2252 :
2253 3768 : if (OffsetNumberIsValid(target_downlinkoffnum))
2254 : {
2255 3746 : itemid = PageGetItemIdCareful(state, state->targetblock,
2256 : state->target, target_downlinkoffnum);
2257 3746 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2258 3746 : downlink = BTreeTupleGetDownLink(itup);
2259 : }
2260 : else
2261 : {
2262 22 : downlink = P_NONE;
2263 : }
2264 :
2265 : /*
2266 : * If no previous rightlink is memorized for current level just below
2267 : * target page's level, we are about to start from the leftmost page. We
2268 : * can't follow rightlinks from previous page, because there is no
2269 : * previous page. But we still can match high key.
2270 : *
2271 : * So we initialize variables for the loop above like there is previous
2272 : * page referencing current child. Also we imply previous page to not
2273 : * have incomplete split flag, that would make us require downlink for
2274 : * current child. That's correct, because leftmost page on the level
2275 : * should always have parent downlink.
2276 : */
2277 3768 : if (!BlockNumberIsValid(blkno))
2278 : {
2279 22 : blkno = downlink;
2280 22 : rightsplit = false;
2281 : }
2282 :
2283 : /* Move to the right on the child level */
2284 : while (true)
2285 : {
2286 : /*
2287 : * Did we traverse the whole tree level and this is check for pages to
2288 : * the right of rightmost downlink?
2289 : */
2290 3768 : if (blkno == P_NONE && downlink == P_NONE)
2291 : {
2292 22 : state->prevrightlink = InvalidBlockNumber;
2293 22 : state->previncompletesplit = false;
2294 22 : return;
2295 : }
2296 :
2297 : /* Did we traverse the whole tree level and don't find next downlink? */
2298 3746 : if (blkno == P_NONE)
2299 0 : ereport(ERROR,
2300 : (errcode(ERRCODE_INDEX_CORRUPTED),
2301 : errmsg("can't traverse from downlink %u to downlink %u of index \"%s\"",
2302 : state->prevrightlink, downlink,
2303 : RelationGetRelationName(state->rel))));
2304 :
2305 : /* Load page contents */
2306 3746 : if (blkno == downlink && loaded_child)
2307 3722 : page = loaded_child;
2308 : else
2309 24 : page = palloc_btree_page(state, blkno);
2310 :
2311 3746 : opaque = BTPageGetOpaque(page);
2312 :
2313 : /* The first page we visit at the level should be leftmost */
2314 3746 : if (first && !BlockNumberIsValid(state->prevrightlink) &&
2315 22 : !bt_leftmost_ignoring_half_dead(state, blkno, opaque))
2316 0 : ereport(ERROR,
2317 : (errcode(ERRCODE_INDEX_CORRUPTED),
2318 : errmsg("the first child of leftmost target page is not leftmost of its level in index \"%s\"",
2319 : RelationGetRelationName(state->rel)),
2320 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2321 : state->targetblock, blkno,
2322 : LSN_FORMAT_ARGS(state->targetlsn))));
2323 :
2324 : /* Do level sanity check */
2325 3746 : if ((!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque)) &&
2326 3746 : opaque->btpo_level != target_level - 1)
2327 0 : ereport(ERROR,
2328 : (errcode(ERRCODE_INDEX_CORRUPTED),
2329 : errmsg("block found while following rightlinks from child of index \"%s\" has invalid level",
2330 : RelationGetRelationName(state->rel)),
2331 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
2332 : blkno, target_level - 1, opaque->btpo_level)));
2333 :
2334 : /* Try to detect circular links */
2335 3746 : if ((!first && blkno == state->prevrightlink) || blkno == opaque->btpo_prev)
2336 0 : ereport(ERROR,
2337 : (errcode(ERRCODE_INDEX_CORRUPTED),
2338 : errmsg("circular link chain found in block %u of index \"%s\"",
2339 : blkno, RelationGetRelationName(state->rel))));
2340 :
2341 3746 : if (blkno != downlink && !P_IGNORE(opaque))
2342 : {
2343 : /* blkno probably has missing parent downlink */
2344 0 : bt_downlink_missing_check(state, rightsplit, blkno, page);
2345 : }
2346 :
2347 3746 : rightsplit = P_INCOMPLETE_SPLIT(opaque);
2348 :
2349 : /*
2350 : * If we visit page with high key, check that it is equal to the
2351 : * target key next to corresponding downlink.
2352 : */
2353 3746 : if (!rightsplit && !P_RIGHTMOST(opaque))
2354 : {
2355 : BTPageOpaque topaque;
2356 : IndexTuple highkey;
2357 : OffsetNumber pivotkey_offset;
2358 :
2359 : /* Get high key */
2360 3724 : itemid = PageGetItemIdCareful(state, blkno, page, P_HIKEY);
2361 3724 : highkey = (IndexTuple) PageGetItem(page, itemid);
2362 :
2363 : /*
2364 : * There might be two situations when we examine high key. If
2365 : * current child page is referenced by given target downlink, we
2366 : * should look to the next offset number for matching key from
2367 : * target page.
2368 : *
2369 : * Alternatively, we're following rightlinks somewhere in the
2370 : * middle between page referenced by previous target's downlink
2371 : * and the page referenced by current target's downlink. If
2372 : * current child page hasn't incomplete split flag set, then its
2373 : * high key should match to the target's key of current offset
2374 : * number. This happens when a previous call here (to
2375 : * bt_child_highkey_check()) found an incomplete split, and we
2376 : * reach a right sibling page without a downlink -- the right
2377 : * sibling page's high key still needs to be matched to a
2378 : * separator key on the parent/target level.
2379 : *
2380 : * Don't apply OffsetNumberNext() to target_downlinkoffnum when we
2381 : * already had to step right on the child level. Our traversal of
2382 : * the child level must try to move in perfect lockstep behind (to
2383 : * the left of) the target/parent level traversal.
2384 : */
2385 3724 : if (blkno == downlink)
2386 3724 : pivotkey_offset = OffsetNumberNext(target_downlinkoffnum);
2387 : else
2388 0 : pivotkey_offset = target_downlinkoffnum;
2389 :
2390 3724 : topaque = BTPageGetOpaque(state->target);
2391 :
2392 3724 : if (!offset_is_negative_infinity(topaque, pivotkey_offset))
2393 : {
2394 : /*
2395 : * If we're looking for the next pivot tuple in target page,
2396 : * but there is no more pivot tuples, then we should match to
2397 : * high key instead.
2398 : */
2399 3724 : if (pivotkey_offset > PageGetMaxOffsetNumber(state->target))
2400 : {
2401 2 : if (P_RIGHTMOST(topaque))
2402 0 : ereport(ERROR,
2403 : (errcode(ERRCODE_INDEX_CORRUPTED),
2404 : errmsg("child high key is greater than rightmost pivot key on target level in index \"%s\"",
2405 : RelationGetRelationName(state->rel)),
2406 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2407 : state->targetblock, blkno,
2408 : LSN_FORMAT_ARGS(state->targetlsn))));
2409 2 : pivotkey_offset = P_HIKEY;
2410 : }
2411 3724 : itemid = PageGetItemIdCareful(state, state->targetblock,
2412 : state->target, pivotkey_offset);
2413 3724 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2414 : }
2415 : else
2416 : {
2417 : /*
2418 : * We cannot try to match child's high key to a negative
2419 : * infinity key in target, since there is nothing to compare.
2420 : * However, it's still possible to match child's high key
2421 : * outside of target page. The reason why we're are is that
2422 : * bt_child_highkey_check() was previously called for the
2423 : * cousin page of 'loaded_child', which is incomplete split.
2424 : * So, now we traverse to the right of that cousin page and
2425 : * current child level page under consideration still belongs
2426 : * to the subtree of target's left sibling. Thus, we need to
2427 : * match child's high key to it's left uncle page high key.
2428 : * Thankfully we saved it, it's called a "low key" of target
2429 : * page.
2430 : */
2431 0 : if (!state->lowkey)
2432 0 : ereport(ERROR,
2433 : (errcode(ERRCODE_INDEX_CORRUPTED),
2434 : errmsg("can't find left sibling high key in index \"%s\"",
2435 : RelationGetRelationName(state->rel)),
2436 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2437 : state->targetblock, blkno,
2438 : LSN_FORMAT_ARGS(state->targetlsn))));
2439 0 : itup = state->lowkey;
2440 : }
2441 :
2442 3724 : if (!bt_pivot_tuple_identical(state->heapkeyspace, highkey, itup))
2443 : {
2444 0 : ereport(ERROR,
2445 : (errcode(ERRCODE_INDEX_CORRUPTED),
2446 : errmsg("mismatch between parent key and child high key in index \"%s\"",
2447 : RelationGetRelationName(state->rel)),
2448 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2449 : state->targetblock, blkno,
2450 : LSN_FORMAT_ARGS(state->targetlsn))));
2451 : }
2452 : }
2453 :
2454 : /* Exit if we already found next downlink */
2455 3746 : if (blkno == downlink)
2456 : {
2457 3746 : state->prevrightlink = opaque->btpo_next;
2458 3746 : state->previncompletesplit = rightsplit;
2459 3746 : return;
2460 : }
2461 :
2462 : /* Traverse to the next page using rightlink */
2463 0 : blkno = opaque->btpo_next;
2464 :
2465 : /* Free page contents if it's allocated by us */
2466 0 : if (page != loaded_child)
2467 0 : pfree(page);
2468 0 : first = false;
2469 : }
2470 : }
2471 :
2472 : /*
2473 : * Checks one of target's downlink against its child page.
2474 : *
2475 : * Conceptually, the target page continues to be what is checked here. The
2476 : * target block is still blamed in the event of finding an invariant violation.
2477 : * The downlink insertion into the target is probably where any problem raised
2478 : * here arises, and there is no such thing as a parent link, so doing the
2479 : * verification this way around is much more practical.
2480 : *
2481 : * This function visits child page and it's sequentially called for each
2482 : * downlink of target page. Assuming this we also check downlink connectivity
2483 : * here in order to save child page visits.
2484 : */
2485 : static void
2486 3722 : bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
2487 : OffsetNumber downlinkoffnum)
2488 : {
2489 : ItemId itemid;
2490 : IndexTuple itup;
2491 : BlockNumber childblock;
2492 : OffsetNumber offset;
2493 : OffsetNumber maxoffset;
2494 : Page child;
2495 : BTPageOpaque copaque;
2496 : BTPageOpaque topaque;
2497 :
2498 3722 : itemid = PageGetItemIdCareful(state, state->targetblock,
2499 : state->target, downlinkoffnum);
2500 3722 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2501 3722 : childblock = BTreeTupleGetDownLink(itup);
2502 :
2503 : /*
2504 : * Caller must have ShareLock on target relation, because of
2505 : * considerations around page deletion by VACUUM.
2506 : *
2507 : * NB: In general, page deletion deletes the right sibling's downlink, not
2508 : * the downlink of the page being deleted; the deleted page's downlink is
2509 : * reused for its sibling. The key space is thereby consolidated between
2510 : * the deleted page and its right sibling. (We cannot delete a parent
2511 : * page's rightmost child unless it is the last child page, and we intend
2512 : * to also delete the parent itself.)
2513 : *
2514 : * If this verification happened without a ShareLock, the following race
2515 : * condition could cause false positives:
2516 : *
2517 : * In general, concurrent page deletion might occur, including deletion of
2518 : * the left sibling of the child page that is examined here. If such a
2519 : * page deletion were to occur, closely followed by an insertion into the
2520 : * newly expanded key space of the child, a window for the false positive
2521 : * opens up: the stale parent/target downlink originally followed to get
2522 : * to the child legitimately ceases to be a lower bound on all items in
2523 : * the page, since the key space was concurrently expanded "left".
2524 : * (Insertion followed the "new" downlink for the child, not our now-stale
2525 : * downlink, which was concurrently physically removed in target/parent as
2526 : * part of deletion's first phase.)
2527 : *
2528 : * While we use various techniques elsewhere to perform cross-page
2529 : * verification for !readonly callers, a similar trick seems difficult
2530 : * here. The tricks used by bt_recheck_sibling_links and by
2531 : * bt_right_page_check_scankey both involve verification of a same-level,
2532 : * cross-sibling invariant. Cross-level invariants are far more squishy,
2533 : * though. The nbtree REDO routines do not actually couple buffer locks
2534 : * across levels during page splits, so making any cross-level check work
2535 : * reliably in !readonly mode may be impossible.
2536 : */
2537 : Assert(state->readonly);
2538 :
2539 : /*
2540 : * Verify child page has the downlink key from target page (its parent) as
2541 : * a lower bound; downlink must be strictly less than all keys on the
2542 : * page.
2543 : *
2544 : * Check all items, rather than checking just the first and trusting that
2545 : * the operator class obeys the transitive law.
2546 : */
2547 3722 : topaque = BTPageGetOpaque(state->target);
2548 3722 : child = palloc_btree_page(state, childblock);
2549 3722 : copaque = BTPageGetOpaque(child);
2550 3722 : maxoffset = PageGetMaxOffsetNumber(child);
2551 :
2552 : /*
2553 : * Since we've already loaded the child block, combine this check with
2554 : * check for downlink connectivity.
2555 : */
2556 3722 : bt_child_highkey_check(state, downlinkoffnum,
2557 : child, topaque->btpo_level);
2558 :
2559 : /*
2560 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2561 : * and since a page has no links within other pages (siblings and parent)
2562 : * once it is marked fully deleted, it should be impossible to land on a
2563 : * fully deleted page.
2564 : *
2565 : * It does not quite make sense to enforce that the page cannot even be
2566 : * half-dead, despite the fact the downlink is modified at the same stage
2567 : * that the child leaf page is marked half-dead. That's incorrect because
2568 : * there may occasionally be multiple downlinks from a chain of pages
2569 : * undergoing deletion, where multiple successive calls are made to
2570 : * _bt_unlink_halfdead_page() by VACUUM before it can finally safely mark
2571 : * the leaf page as fully dead. While _bt_mark_page_halfdead() usually
2572 : * removes the downlink to the leaf page that is marked half-dead, that's
2573 : * not guaranteed, so it's possible we'll land on a half-dead page with a
2574 : * downlink due to an interrupted multi-level page deletion.
2575 : *
2576 : * We go ahead with our checks if the child page is half-dead. It's safe
2577 : * to do so because we do not test the child's high key, so it does not
2578 : * matter that the original high key will have been replaced by a dummy
2579 : * truncated high key within _bt_mark_page_halfdead(). All other page
2580 : * items are left intact on a half-dead page, so there is still something
2581 : * to test.
2582 : */
2583 3722 : if (P_ISDELETED(copaque))
2584 0 : ereport(ERROR,
2585 : (errcode(ERRCODE_INDEX_CORRUPTED),
2586 : errmsg("downlink to deleted page found in index \"%s\"",
2587 : RelationGetRelationName(state->rel)),
2588 : errdetail_internal("Parent block=%u child block=%u parent page lsn=%X/%X.",
2589 : state->targetblock, childblock,
2590 : LSN_FORMAT_ARGS(state->targetlsn))));
2591 :
2592 1200700 : for (offset = P_FIRSTDATAKEY(copaque);
2593 : offset <= maxoffset;
2594 1196978 : offset = OffsetNumberNext(offset))
2595 : {
2596 : /*
2597 : * Skip comparison of target page key against "negative infinity"
2598 : * item, if any. Checking it would indicate that it's not a strict
2599 : * lower bound, but that's only because of the hard-coding for
2600 : * negative infinity items within _bt_compare().
2601 : *
2602 : * If nbtree didn't truncate negative infinity tuples during internal
2603 : * page splits then we'd expect child's negative infinity key to be
2604 : * equal to the scankey/downlink from target/parent (it would be a
2605 : * "low key" in this hypothetical scenario, and so it would still need
2606 : * to be treated as a special case here).
2607 : *
2608 : * Negative infinity items can be thought of as a strict lower bound
2609 : * that works transitively, with the last non-negative-infinity pivot
2610 : * followed during a descent from the root as its "true" strict lower
2611 : * bound. Only a small number of negative infinity items are truly
2612 : * negative infinity; those that are the first items of leftmost
2613 : * internal pages. In more general terms, a negative infinity item is
2614 : * only negative infinity with respect to the subtree that the page is
2615 : * at the root of.
2616 : *
2617 : * See also: bt_rootdescend(), which can even detect transitive
2618 : * inconsistencies on cousin leaf pages.
2619 : */
2620 1196978 : if (offset_is_negative_infinity(copaque, offset))
2621 2 : continue;
2622 :
2623 1196976 : if (!invariant_l_nontarget_offset(state, targetkey, childblock, child,
2624 : offset))
2625 0 : ereport(ERROR,
2626 : (errcode(ERRCODE_INDEX_CORRUPTED),
2627 : errmsg("down-link lower bound invariant violated for index \"%s\"",
2628 : RelationGetRelationName(state->rel)),
2629 : errdetail_internal("Parent block=%u child index tid=(%u,%u) parent page lsn=%X/%X.",
2630 : state->targetblock, childblock, offset,
2631 : LSN_FORMAT_ARGS(state->targetlsn))));
2632 : }
2633 :
2634 3722 : pfree(child);
2635 3722 : }
2636 :
2637 : /*
2638 : * Checks if page is missing a downlink that it should have.
2639 : *
2640 : * A page that lacks a downlink/parent may indicate corruption. However, we
2641 : * must account for the fact that a missing downlink can occasionally be
2642 : * encountered in a non-corrupt index. This can be due to an interrupted page
2643 : * split, or an interrupted multi-level page deletion (i.e. there was a hard
2644 : * crash or an error during a page split, or while VACUUM was deleting a
2645 : * multi-level chain of pages).
2646 : *
2647 : * Note that this can only be called in readonly mode, so there is no need to
2648 : * be concerned about concurrent page splits or page deletions.
2649 : */
2650 : static void
2651 0 : bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
2652 : BlockNumber blkno, Page page)
2653 : {
2654 0 : BTPageOpaque opaque = BTPageGetOpaque(page);
2655 : ItemId itemid;
2656 : IndexTuple itup;
2657 : Page child;
2658 : BTPageOpaque copaque;
2659 : uint32 level;
2660 : BlockNumber childblk;
2661 : XLogRecPtr pagelsn;
2662 :
2663 : Assert(state->readonly);
2664 : Assert(!P_IGNORE(opaque));
2665 :
2666 : /* No next level up with downlinks to fingerprint from the true root */
2667 0 : if (P_ISROOT(opaque))
2668 0 : return;
2669 :
2670 0 : pagelsn = PageGetLSN(page);
2671 :
2672 : /*
2673 : * Incomplete (interrupted) page splits can account for the lack of a
2674 : * downlink. Some inserting transaction should eventually complete the
2675 : * page split in passing, when it notices that the left sibling page is
2676 : * P_INCOMPLETE_SPLIT().
2677 : *
2678 : * In general, VACUUM is not prepared for there to be no downlink to a
2679 : * page that it deletes. This is the main reason why the lack of a
2680 : * downlink can be reported as corruption here. It's not obvious that an
2681 : * invalid missing downlink can result in wrong answers to queries,
2682 : * though, since index scans that land on the child may end up
2683 : * consistently moving right. The handling of concurrent page splits (and
2684 : * page deletions) within _bt_moveright() cannot distinguish
2685 : * inconsistencies that last for a moment from inconsistencies that are
2686 : * permanent and irrecoverable.
2687 : *
2688 : * VACUUM isn't even prepared to delete pages that have no downlink due to
2689 : * an incomplete page split, but it can detect and reason about that case
2690 : * by design, so it shouldn't be taken to indicate corruption. See
2691 : * _bt_pagedel() for full details.
2692 : */
2693 0 : if (rightsplit)
2694 : {
2695 0 : ereport(DEBUG1,
2696 : (errcode(ERRCODE_NO_DATA),
2697 : errmsg_internal("harmless interrupted page split detected in index \"%s\"",
2698 : RelationGetRelationName(state->rel)),
2699 : errdetail_internal("Block=%u level=%u left sibling=%u page lsn=%X/%X.",
2700 : blkno, opaque->btpo_level,
2701 : opaque->btpo_prev,
2702 : LSN_FORMAT_ARGS(pagelsn))));
2703 0 : return;
2704 : }
2705 :
2706 : /*
2707 : * Page under check is probably the "top parent" of a multi-level page
2708 : * deletion. We'll need to descend the subtree to make sure that
2709 : * descendant pages are consistent with that, though.
2710 : *
2711 : * If the page (which must be non-ignorable) is a leaf page, then clearly
2712 : * it can't be the top parent. The lack of a downlink is probably a
2713 : * symptom of a broad problem that could just as easily cause
2714 : * inconsistencies anywhere else.
2715 : */
2716 0 : if (P_ISLEAF(opaque))
2717 0 : ereport(ERROR,
2718 : (errcode(ERRCODE_INDEX_CORRUPTED),
2719 : errmsg("leaf index block lacks downlink in index \"%s\"",
2720 : RelationGetRelationName(state->rel)),
2721 : errdetail_internal("Block=%u page lsn=%X/%X.",
2722 : blkno,
2723 : LSN_FORMAT_ARGS(pagelsn))));
2724 :
2725 : /* Descend from the given page, which is an internal page */
2726 0 : elog(DEBUG1, "checking for interrupted multi-level deletion due to missing downlink in index \"%s\"",
2727 : RelationGetRelationName(state->rel));
2728 :
2729 0 : level = opaque->btpo_level;
2730 0 : itemid = PageGetItemIdCareful(state, blkno, page, P_FIRSTDATAKEY(opaque));
2731 0 : itup = (IndexTuple) PageGetItem(page, itemid);
2732 0 : childblk = BTreeTupleGetDownLink(itup);
2733 : for (;;)
2734 : {
2735 0 : CHECK_FOR_INTERRUPTS();
2736 :
2737 0 : child = palloc_btree_page(state, childblk);
2738 0 : copaque = BTPageGetOpaque(child);
2739 :
2740 0 : if (P_ISLEAF(copaque))
2741 0 : break;
2742 :
2743 : /* Do an extra sanity check in passing on internal pages */
2744 0 : if (copaque->btpo_level != level - 1)
2745 0 : ereport(ERROR,
2746 : (errcode(ERRCODE_INDEX_CORRUPTED),
2747 : errmsg_internal("downlink points to block in index \"%s\" whose level is not one level down",
2748 : RelationGetRelationName(state->rel)),
2749 : errdetail_internal("Top parent/under check block=%u block pointed to=%u expected level=%u level in pointed to block=%u.",
2750 : blkno, childblk,
2751 : level - 1, copaque->btpo_level)));
2752 :
2753 0 : level = copaque->btpo_level;
2754 0 : itemid = PageGetItemIdCareful(state, childblk, child,
2755 0 : P_FIRSTDATAKEY(copaque));
2756 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2757 0 : childblk = BTreeTupleGetDownLink(itup);
2758 : /* Be slightly more pro-active in freeing this memory, just in case */
2759 0 : pfree(child);
2760 : }
2761 :
2762 : /*
2763 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2764 : * and since a page has no links within other pages (siblings and parent)
2765 : * once it is marked fully deleted, it should be impossible to land on a
2766 : * fully deleted page. See bt_child_check() for further details.
2767 : *
2768 : * The bt_child_check() P_ISDELETED() check is repeated here because
2769 : * bt_child_check() does not visit pages reachable through negative
2770 : * infinity items. Besides, bt_child_check() is unwilling to descend
2771 : * multiple levels. (The similar bt_child_check() P_ISDELETED() check
2772 : * within bt_check_level_from_leftmost() won't reach the page either,
2773 : * since the leaf's live siblings should have their sibling links updated
2774 : * to bypass the deletion target page when it is marked fully dead.)
2775 : *
2776 : * If this error is raised, it might be due to a previous multi-level page
2777 : * deletion that failed to realize that it wasn't yet safe to mark the
2778 : * leaf page as fully dead. A "dangling downlink" will still remain when
2779 : * this happens. The fact that the dangling downlink's page (the leaf's
2780 : * parent/ancestor page) lacked a downlink is incidental.
2781 : */
2782 0 : if (P_ISDELETED(copaque))
2783 0 : ereport(ERROR,
2784 : (errcode(ERRCODE_INDEX_CORRUPTED),
2785 : errmsg_internal("downlink to deleted leaf page found in index \"%s\"",
2786 : RelationGetRelationName(state->rel)),
2787 : errdetail_internal("Top parent/target block=%u leaf block=%u top parent/under check lsn=%X/%X.",
2788 : blkno, childblk,
2789 : LSN_FORMAT_ARGS(pagelsn))));
2790 :
2791 : /*
2792 : * Iff leaf page is half-dead, its high key top parent link should point
2793 : * to what VACUUM considered to be the top parent page at the instant it
2794 : * was interrupted. Provided the high key link actually points to the
2795 : * page under check, the missing downlink we detected is consistent with
2796 : * there having been an interrupted multi-level page deletion. This means
2797 : * that the subtree with the page under check at its root (a page deletion
2798 : * chain) is in a consistent state, enabling VACUUM to resume deleting the
2799 : * entire chain the next time it encounters the half-dead leaf page.
2800 : */
2801 0 : if (P_ISHALFDEAD(copaque) && !P_RIGHTMOST(copaque))
2802 : {
2803 0 : itemid = PageGetItemIdCareful(state, childblk, child, P_HIKEY);
2804 0 : itup = (IndexTuple) PageGetItem(child, itemid);
2805 0 : if (BTreeTupleGetTopParent(itup) == blkno)
2806 0 : return;
2807 : }
2808 :
2809 0 : ereport(ERROR,
2810 : (errcode(ERRCODE_INDEX_CORRUPTED),
2811 : errmsg("internal index block lacks downlink in index \"%s\"",
2812 : RelationGetRelationName(state->rel)),
2813 : errdetail_internal("Block=%u level=%u page lsn=%X/%X.",
2814 : blkno, opaque->btpo_level,
2815 : LSN_FORMAT_ARGS(pagelsn))));
2816 : }
2817 :
2818 : /*
2819 : * Per-tuple callback from table_index_build_scan, used to determine if index has
2820 : * all the entries that definitely should have been observed in leaf pages of
2821 : * the target index (that is, all IndexTuples that were fingerprinted by our
2822 : * Bloom filter). All heapallindexed checks occur here.
2823 : *
2824 : * The redundancy between an index and the table it indexes provides a good
2825 : * opportunity to detect corruption, especially corruption within the table.
2826 : * The high level principle behind the verification performed here is that any
2827 : * IndexTuple that should be in an index following a fresh CREATE INDEX (based
2828 : * on the same index definition) should also have been in the original,
2829 : * existing index, which should have used exactly the same representation
2830 : *
2831 : * Since the overall structure of the index has already been verified, the most
2832 : * likely explanation for error here is a corrupt heap page (could be logical
2833 : * or physical corruption). Index corruption may still be detected here,
2834 : * though. Only readonly callers will have verified that left links and right
2835 : * links are in agreement, and so it's possible that a leaf page transposition
2836 : * within index is actually the source of corruption detected here (for
2837 : * !readonly callers). The checks performed only for readonly callers might
2838 : * more accurately frame the problem as a cross-page invariant issue (this
2839 : * could even be due to recovery not replaying all WAL records). The !readonly
2840 : * ERROR message raised here includes a HINT about retrying with readonly
2841 : * verification, just in case it's a cross-page invariant issue, though that
2842 : * isn't particularly likely.
2843 : *
2844 : * table_index_build_scan() expects to be able to find the root tuple when a
2845 : * heap-only tuple (the live tuple at the end of some HOT chain) needs to be
2846 : * indexed, in order to replace the actual tuple's TID with the root tuple's
2847 : * TID (which is what we're actually passed back here). The index build heap
2848 : * scan code will raise an error when a tuple that claims to be the root of the
2849 : * heap-only tuple's HOT chain cannot be located. This catches cases where the
2850 : * original root item offset/root tuple for a HOT chain indicates (for whatever
2851 : * reason) that the entire HOT chain is dead, despite the fact that the latest
2852 : * heap-only tuple should be indexed. When this happens, sequential scans may
2853 : * always give correct answers, and all indexes may be considered structurally
2854 : * consistent (i.e. the nbtree structural checks would not detect corruption).
2855 : * It may be the case that only index scans give wrong answers, and yet heap or
2856 : * SLRU corruption is the real culprit. (While it's true that LP_DEAD bit
2857 : * setting will probably also leave the index in a corrupt state before too
2858 : * long, the problem is nonetheless that there is heap corruption.)
2859 : *
2860 : * Heap-only tuple handling within table_index_build_scan() works in a way that
2861 : * helps us to detect index tuples that contain the wrong values (values that
2862 : * don't match the latest tuple in the HOT chain). This can happen when there
2863 : * is no superseding index tuple due to a faulty assessment of HOT safety,
2864 : * perhaps during the original CREATE INDEX. Because the latest tuple's
2865 : * contents are used with the root TID, an error will be raised when a tuple
2866 : * with the same TID but non-matching attribute values is passed back to us.
2867 : * Faulty assessment of HOT-safety was behind at least two distinct CREATE
2868 : * INDEX CONCURRENTLY bugs that made it into stable releases, one of which was
2869 : * undetected for many years. In short, the same principle that allows a
2870 : * REINDEX to repair corruption when there was an (undetected) broken HOT chain
2871 : * also allows us to detect the corruption in many cases.
2872 : */
2873 : static void
2874 1072692 : bt_tuple_present_callback(Relation index, ItemPointer tid, Datum *values,
2875 : bool *isnull, bool tupleIsAlive, void *checkstate)
2876 : {
2877 1072692 : BtreeCheckState *state = (BtreeCheckState *) checkstate;
2878 : IndexTuple itup,
2879 : norm;
2880 :
2881 : Assert(state->heapallindexed);
2882 :
2883 : /* Generate a normalized index tuple for fingerprinting */
2884 1072692 : itup = index_form_tuple(RelationGetDescr(index), values, isnull);
2885 1072692 : itup->t_tid = *tid;
2886 1072692 : norm = bt_normalize_tuple(state, itup);
2887 :
2888 : /* Probe Bloom filter -- tuple should be present */
2889 1072692 : if (bloom_lacks_element(state->filter, (unsigned char *) norm,
2890 1072692 : IndexTupleSize(norm)))
2891 0 : ereport(ERROR,
2892 : (errcode(ERRCODE_DATA_CORRUPTED),
2893 : errmsg("heap tuple (%u,%u) from table \"%s\" lacks matching index tuple within index \"%s\"",
2894 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2895 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2896 : RelationGetRelationName(state->heaprel),
2897 : RelationGetRelationName(state->rel)),
2898 : !state->readonly
2899 : ? errhint("Retrying verification using the function bt_index_parent_check() might provide a more specific error.")
2900 : : 0));
2901 :
2902 1072692 : state->heaptuplespresent++;
2903 1072692 : pfree(itup);
2904 : /* Cannot leak memory here */
2905 1072692 : if (norm != itup)
2906 10 : pfree(norm);
2907 1072692 : }
2908 :
2909 : /*
2910 : * Normalize an index tuple for fingerprinting.
2911 : *
2912 : * In general, index tuple formation is assumed to be deterministic by
2913 : * heapallindexed verification, and IndexTuples are assumed immutable. While
2914 : * the LP_DEAD bit is mutable in leaf pages, that's ItemId metadata, which is
2915 : * not fingerprinted. Normalization is required to compensate for corner
2916 : * cases where the determinism assumption doesn't quite work.
2917 : *
2918 : * There is currently one such case: index_form_tuple() does not try to hide
2919 : * the source TOAST state of input datums. The executor applies TOAST
2920 : * compression for heap tuples based on different criteria to the compression
2921 : * applied within btinsert()'s call to index_form_tuple(): it sometimes
2922 : * compresses more aggressively, resulting in compressed heap tuple datums but
2923 : * uncompressed corresponding index tuple datums. A subsequent heapallindexed
2924 : * verification will get a logically equivalent though bitwise unequal tuple
2925 : * from index_form_tuple(). False positive heapallindexed corruption reports
2926 : * could occur without normalizing away the inconsistency.
2927 : *
2928 : * Returned tuple is often caller's own original tuple. Otherwise, it is a
2929 : * new representation of caller's original index tuple, palloc()'d in caller's
2930 : * memory context.
2931 : *
2932 : * Note: This routine is not concerned with distinctions about the
2933 : * representation of tuples beyond those that might break heapallindexed
2934 : * verification. In particular, it won't try to normalize opclass-equal
2935 : * datums with potentially distinct representations (e.g., btree/numeric_ops
2936 : * index datums will not get their display scale normalized-away here).
2937 : * Caller does normalization for non-pivot tuples that have a posting list,
2938 : * since dummy CREATE INDEX callback code generates new tuples with the same
2939 : * normalized representation.
2940 : */
2941 : static IndexTuple
2942 2150202 : bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
2943 : {
2944 2150202 : TupleDesc tupleDescriptor = RelationGetDescr(state->rel);
2945 : Datum normalized[INDEX_MAX_KEYS];
2946 : bool isnull[INDEX_MAX_KEYS];
2947 : bool need_free[INDEX_MAX_KEYS];
2948 2150202 : bool formnewtup = false;
2949 : IndexTuple reformed;
2950 : int i;
2951 :
2952 : /* Caller should only pass "logical" non-pivot tuples here */
2953 : Assert(!BTreeTupleIsPosting(itup) && !BTreeTupleIsPivot(itup));
2954 :
2955 : /* Easy case: It's immediately clear that tuple has no varlena datums */
2956 2150202 : if (!IndexTupleHasVarwidths(itup))
2957 2150154 : return itup;
2958 :
2959 96 : for (i = 0; i < tupleDescriptor->natts; i++)
2960 : {
2961 : Form_pg_attribute att;
2962 :
2963 48 : att = TupleDescAttr(tupleDescriptor, i);
2964 :
2965 : /* Assume untoasted/already normalized datum initially */
2966 48 : need_free[i] = false;
2967 48 : normalized[i] = index_getattr(itup, att->attnum,
2968 : tupleDescriptor,
2969 : &isnull[i]);
2970 48 : if (att->attbyval || att->attlen != -1 || isnull[i])
2971 0 : continue;
2972 :
2973 : /*
2974 : * Callers always pass a tuple that could safely be inserted into the
2975 : * index without further processing, so an external varlena header
2976 : * should never be encountered here
2977 : */
2978 48 : if (VARATT_IS_EXTERNAL(DatumGetPointer(normalized[i])))
2979 0 : ereport(ERROR,
2980 : (errcode(ERRCODE_INDEX_CORRUPTED),
2981 : errmsg("external varlena datum in tuple that references heap row (%u,%u) in index \"%s\"",
2982 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2983 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2984 : RelationGetRelationName(state->rel))));
2985 48 : else if (!VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])) &&
2986 44 : VARSIZE(DatumGetPointer(normalized[i])) > TOAST_INDEX_TARGET &&
2987 42 : (att->attstorage == TYPSTORAGE_EXTENDED ||
2988 32 : att->attstorage == TYPSTORAGE_MAIN))
2989 : {
2990 : /*
2991 : * This value will be compressed by index_form_tuple() with the
2992 : * current storage settings. We may be here because this tuple
2993 : * was formed with different storage settings. So, force forming.
2994 : */
2995 10 : formnewtup = true;
2996 : }
2997 38 : else if (VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])))
2998 : {
2999 4 : formnewtup = true;
3000 4 : normalized[i] = PointerGetDatum(PG_DETOAST_DATUM(normalized[i]));
3001 4 : need_free[i] = true;
3002 : }
3003 :
3004 : /*
3005 : * Short tuples may have 1B or 4B header. Convert 4B header of short
3006 : * tuples to 1B
3007 : */
3008 34 : else if (VARATT_CAN_MAKE_SHORT(DatumGetPointer(normalized[i])))
3009 : {
3010 : /* convert to short varlena */
3011 2 : Size len = VARATT_CONVERTED_SHORT_SIZE(DatumGetPointer(normalized[i]));
3012 2 : char *data = palloc(len);
3013 :
3014 2 : SET_VARSIZE_SHORT(data, len);
3015 2 : memcpy(data + 1, VARDATA(DatumGetPointer(normalized[i])), len - 1);
3016 :
3017 2 : formnewtup = true;
3018 2 : normalized[i] = PointerGetDatum(data);
3019 2 : need_free[i] = true;
3020 : }
3021 : }
3022 :
3023 : /*
3024 : * Easier case: Tuple has varlena datums, none of which are compressed or
3025 : * short with 4B header
3026 : */
3027 48 : if (!formnewtup)
3028 32 : return itup;
3029 :
3030 : /*
3031 : * Hard case: Tuple had compressed varlena datums that necessitate
3032 : * creating normalized version of the tuple from uncompressed input datums
3033 : * (normalized input datums). This is rather naive, but shouldn't be
3034 : * necessary too often.
3035 : *
3036 : * In the heap, tuples may contain short varlena datums with both 1B
3037 : * header and 4B headers. But the corresponding index tuple should always
3038 : * have such varlena's with 1B headers. So, if there is a short varlena
3039 : * with 4B header, we need to convert it for fingerprinting.
3040 : *
3041 : * Note that we rely on deterministic index_form_tuple() TOAST compression
3042 : * of normalized input.
3043 : */
3044 16 : reformed = index_form_tuple(tupleDescriptor, normalized, isnull);
3045 16 : reformed->t_tid = itup->t_tid;
3046 :
3047 : /* Cannot leak memory here */
3048 32 : for (i = 0; i < tupleDescriptor->natts; i++)
3049 16 : if (need_free[i])
3050 6 : pfree(DatumGetPointer(normalized[i]));
3051 :
3052 16 : return reformed;
3053 : }
3054 :
3055 : /*
3056 : * Produce palloc()'d "plain" tuple for nth posting list entry/TID.
3057 : *
3058 : * In general, deduplication is not supposed to change the logical contents of
3059 : * an index. Multiple index tuples are merged together into one equivalent
3060 : * posting list index tuple when convenient.
3061 : *
3062 : * heapallindexed verification must normalize-away this variation in
3063 : * representation by converting posting list tuples into two or more "plain"
3064 : * tuples. Each tuple must be fingerprinted separately -- there must be one
3065 : * tuple for each corresponding Bloom filter probe during the heap scan.
3066 : *
3067 : * Note: Caller still needs to call bt_normalize_tuple() with returned tuple.
3068 : */
3069 : static inline IndexTuple
3070 63946 : bt_posting_plain_tuple(IndexTuple itup, int n)
3071 : {
3072 : Assert(BTreeTupleIsPosting(itup));
3073 :
3074 : /* Returns non-posting-list tuple */
3075 63946 : return _bt_form_posting(itup, BTreeTupleGetPostingN(itup, n), 1);
3076 : }
3077 :
3078 : /*
3079 : * Search for itup in index, starting from fast root page. itup must be a
3080 : * non-pivot tuple. This is only supported with heapkeyspace indexes, since
3081 : * we rely on having fully unique keys to find a match with only a single
3082 : * visit to a leaf page, barring an interrupted page split, where we may have
3083 : * to move right. (A concurrent page split is impossible because caller must
3084 : * be readonly caller.)
3085 : *
3086 : * This routine can detect very subtle transitive consistency issues across
3087 : * more than one level of the tree. Leaf pages all have a high key (even the
3088 : * rightmost page has a conceptual positive infinity high key), but not a low
3089 : * key. Their downlink in parent is a lower bound, which along with the high
3090 : * key is almost enough to detect every possible inconsistency. A downlink
3091 : * separator key value won't always be available from parent, though, because
3092 : * the first items of internal pages are negative infinity items, truncated
3093 : * down to zero attributes during internal page splits. While it's true that
3094 : * bt_child_check() and the high key check can detect most imaginable key
3095 : * space problems, there are remaining problems it won't detect with non-pivot
3096 : * tuples in cousin leaf pages. Starting a search from the root for every
3097 : * existing leaf tuple detects small inconsistencies in upper levels of the
3098 : * tree that cannot be detected any other way. (Besides all this, this is
3099 : * probably also useful as a direct test of the code used by index scans
3100 : * themselves.)
3101 : */
3102 : static bool
3103 402196 : bt_rootdescend(BtreeCheckState *state, IndexTuple itup)
3104 : {
3105 : BTScanInsert key;
3106 : BTStack stack;
3107 : Buffer lbuf;
3108 : bool exists;
3109 :
3110 402196 : key = _bt_mkscankey(state->rel, itup);
3111 : Assert(key->heapkeyspace && key->scantid != NULL);
3112 :
3113 : /*
3114 : * Search from root.
3115 : *
3116 : * Ideally, we would arrange to only move right within _bt_search() when
3117 : * an interrupted page split is detected (i.e. when the incomplete split
3118 : * bit is found to be set), but for now we accept the possibility that
3119 : * that could conceal an inconsistency.
3120 : */
3121 : Assert(state->readonly && state->rootdescend);
3122 402196 : exists = false;
3123 402196 : stack = _bt_search(state->rel, NULL, key, &lbuf, BT_READ);
3124 :
3125 402196 : if (BufferIsValid(lbuf))
3126 : {
3127 : BTInsertStateData insertstate;
3128 : OffsetNumber offnum;
3129 : Page page;
3130 :
3131 402196 : insertstate.itup = itup;
3132 402196 : insertstate.itemsz = MAXALIGN(IndexTupleSize(itup));
3133 402196 : insertstate.itup_key = key;
3134 402196 : insertstate.postingoff = 0;
3135 402196 : insertstate.bounds_valid = false;
3136 402196 : insertstate.buf = lbuf;
3137 :
3138 : /* Get matching tuple on leaf page */
3139 402196 : offnum = _bt_binsrch_insert(state->rel, &insertstate);
3140 : /* Compare first >= matching item on leaf page, if any */
3141 402196 : page = BufferGetPage(lbuf);
3142 : /* Should match on first heap TID when tuple has a posting list */
3143 402196 : if (offnum <= PageGetMaxOffsetNumber(page) &&
3144 804392 : insertstate.postingoff <= 0 &&
3145 402196 : _bt_compare(state->rel, key, page, offnum) == 0)
3146 402196 : exists = true;
3147 402196 : _bt_relbuf(state->rel, lbuf);
3148 : }
3149 :
3150 402196 : _bt_freestack(stack);
3151 402196 : pfree(key);
3152 :
3153 402196 : return exists;
3154 : }
3155 :
3156 : /*
3157 : * Is particular offset within page (whose special state is passed by caller)
3158 : * the page negative-infinity item?
3159 : *
3160 : * As noted in comments above _bt_compare(), there is special handling of the
3161 : * first data item as a "negative infinity" item. The hard-coding within
3162 : * _bt_compare() makes comparing this item for the purposes of verification
3163 : * pointless at best, since the IndexTuple only contains a valid TID (a
3164 : * reference TID to child page).
3165 : */
3166 : static inline bool
3167 5181552 : offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
3168 : {
3169 : /*
3170 : * For internal pages only, the first item after high key, if any, is
3171 : * negative infinity item. Internal pages always have a negative infinity
3172 : * item, whereas leaf pages never have one. This implies that negative
3173 : * infinity item is either first or second line item, or there is none
3174 : * within page.
3175 : *
3176 : * Negative infinity items are a special case among pivot tuples. They
3177 : * always have zero attributes, while all other pivot tuples always have
3178 : * nkeyatts attributes.
3179 : *
3180 : * Right-most pages don't have a high key, but could be said to
3181 : * conceptually have a "positive infinity" high key. Thus, there is a
3182 : * symmetry between down link items in parent pages, and high keys in
3183 : * children. Together, they represent the part of the key space that
3184 : * belongs to each page in the index. For example, all children of the
3185 : * root page will have negative infinity as a lower bound from root
3186 : * negative infinity downlink, and positive infinity as an upper bound
3187 : * (implicitly, from "imaginary" positive infinity high key in root).
3188 : */
3189 5181552 : return !P_ISLEAF(opaque) && offset == P_FIRSTDATAKEY(opaque);
3190 : }
3191 :
3192 : /*
3193 : * Does the invariant hold that the key is strictly less than a given upper
3194 : * bound offset item?
3195 : *
3196 : * Verifies line pointer on behalf of caller.
3197 : *
3198 : * If this function returns false, convention is that caller throws error due
3199 : * to corruption.
3200 : */
3201 : static inline bool
3202 3963130 : invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
3203 : OffsetNumber upperbound)
3204 : {
3205 : ItemId itemid;
3206 : int32 cmp;
3207 :
3208 : Assert(!key->nextkey && key->backward);
3209 :
3210 : /* Verify line pointer before checking tuple */
3211 3963130 : itemid = PageGetItemIdCareful(state, state->targetblock, state->target,
3212 : upperbound);
3213 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3214 3963130 : if (!key->heapkeyspace)
3215 0 : return invariant_leq_offset(state, key, upperbound);
3216 :
3217 3963130 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3218 :
3219 : /*
3220 : * _bt_compare() is capable of determining that a scankey with a
3221 : * filled-out attribute is greater than pivot tuples where the comparison
3222 : * is resolved at a truncated attribute (value of attribute in pivot is
3223 : * minus infinity). However, it is not capable of determining that a
3224 : * scankey is _less than_ a tuple on the basis of a comparison resolved at
3225 : * _scankey_ minus infinity attribute. Complete an extra step to simulate
3226 : * having minus infinity values for omitted scankey attribute(s).
3227 : */
3228 3963130 : if (cmp == 0)
3229 : {
3230 : BTPageOpaque topaque;
3231 : IndexTuple ritup;
3232 : int uppnkeyatts;
3233 : ItemPointer rheaptid;
3234 : bool nonpivot;
3235 :
3236 0 : ritup = (IndexTuple) PageGetItem(state->target, itemid);
3237 0 : topaque = BTPageGetOpaque(state->target);
3238 0 : nonpivot = P_ISLEAF(topaque) && upperbound >= P_FIRSTDATAKEY(topaque);
3239 :
3240 : /* Get number of keys + heap TID for item to the right */
3241 0 : uppnkeyatts = BTreeTupleGetNKeyAtts(ritup, state->rel);
3242 0 : rheaptid = BTreeTupleGetHeapTIDCareful(state, ritup, nonpivot);
3243 :
3244 : /* Heap TID is tiebreaker key attribute */
3245 0 : if (key->keysz == uppnkeyatts)
3246 0 : return key->scantid == NULL && rheaptid != NULL;
3247 :
3248 0 : return key->keysz < uppnkeyatts;
3249 : }
3250 :
3251 3963130 : return cmp < 0;
3252 : }
3253 :
3254 : /*
3255 : * Does the invariant hold that the key is less than or equal to a given upper
3256 : * bound offset item?
3257 : *
3258 : * Caller should have verified that upperbound's line pointer is consistent
3259 : * using PageGetItemIdCareful() call.
3260 : *
3261 : * If this function returns false, convention is that caller throws error due
3262 : * to corruption.
3263 : */
3264 : static inline bool
3265 3652980 : invariant_leq_offset(BtreeCheckState *state, BTScanInsert key,
3266 : OffsetNumber upperbound)
3267 : {
3268 : int32 cmp;
3269 :
3270 : Assert(!key->nextkey && key->backward);
3271 :
3272 3652980 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
3273 :
3274 3652980 : return cmp <= 0;
3275 : }
3276 :
3277 : /*
3278 : * Does the invariant hold that the key is strictly greater than a given lower
3279 : * bound offset item?
3280 : *
3281 : * Caller should have verified that lowerbound's line pointer is consistent
3282 : * using PageGetItemIdCareful() call.
3283 : *
3284 : * If this function returns false, convention is that caller throws error due
3285 : * to corruption.
3286 : */
3287 : static inline bool
3288 12894 : invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
3289 : OffsetNumber lowerbound)
3290 : {
3291 : int32 cmp;
3292 :
3293 : Assert(!key->nextkey && key->backward);
3294 :
3295 12894 : cmp = _bt_compare(state->rel, key, state->target, lowerbound);
3296 :
3297 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3298 12894 : if (!key->heapkeyspace)
3299 0 : return cmp >= 0;
3300 :
3301 : /*
3302 : * No need to consider the possibility that scankey has attributes that we
3303 : * need to force to be interpreted as negative infinity. _bt_compare() is
3304 : * able to determine that scankey is greater than negative infinity. The
3305 : * distinction between "==" and "<" isn't interesting here, since
3306 : * corruption is indicated either way.
3307 : */
3308 12894 : return cmp > 0;
3309 : }
3310 :
3311 : /*
3312 : * Does the invariant hold that the key is strictly less than a given upper
3313 : * bound offset item, with the offset relating to a caller-supplied page that
3314 : * is not the current target page?
3315 : *
3316 : * Caller's non-target page is a child page of the target, checked as part of
3317 : * checking a property of the target page (i.e. the key comes from the
3318 : * target). Verifies line pointer on behalf of caller.
3319 : *
3320 : * If this function returns false, convention is that caller throws error due
3321 : * to corruption.
3322 : */
3323 : static inline bool
3324 1196976 : invariant_l_nontarget_offset(BtreeCheckState *state, BTScanInsert key,
3325 : BlockNumber nontargetblock, Page nontarget,
3326 : OffsetNumber upperbound)
3327 : {
3328 : ItemId itemid;
3329 : int32 cmp;
3330 :
3331 : Assert(!key->nextkey && key->backward);
3332 :
3333 : /* Verify line pointer before checking tuple */
3334 1196976 : itemid = PageGetItemIdCareful(state, nontargetblock, nontarget,
3335 : upperbound);
3336 1196976 : cmp = _bt_compare(state->rel, key, nontarget, upperbound);
3337 :
3338 : /* pg_upgrade'd indexes may legally have equal sibling tuples */
3339 1196976 : if (!key->heapkeyspace)
3340 0 : return cmp <= 0;
3341 :
3342 : /* See invariant_l_offset() for an explanation of this extra step */
3343 1196976 : if (cmp == 0)
3344 : {
3345 : IndexTuple child;
3346 : int uppnkeyatts;
3347 : ItemPointer childheaptid;
3348 : BTPageOpaque copaque;
3349 : bool nonpivot;
3350 :
3351 3720 : child = (IndexTuple) PageGetItem(nontarget, itemid);
3352 3720 : copaque = BTPageGetOpaque(nontarget);
3353 3720 : nonpivot = P_ISLEAF(copaque) && upperbound >= P_FIRSTDATAKEY(copaque);
3354 :
3355 : /* Get number of keys + heap TID for child/non-target item */
3356 3720 : uppnkeyatts = BTreeTupleGetNKeyAtts(child, state->rel);
3357 3720 : childheaptid = BTreeTupleGetHeapTIDCareful(state, child, nonpivot);
3358 :
3359 : /* Heap TID is tiebreaker key attribute */
3360 3720 : if (key->keysz == uppnkeyatts)
3361 3720 : return key->scantid == NULL && childheaptid != NULL;
3362 :
3363 0 : return key->keysz < uppnkeyatts;
3364 : }
3365 :
3366 1193256 : return cmp < 0;
3367 : }
3368 :
3369 : /*
3370 : * Given a block number of a B-Tree page, return page in palloc()'d memory.
3371 : * While at it, perform some basic checks of the page.
3372 : *
3373 : * There is never an attempt to get a consistent view of multiple pages using
3374 : * multiple concurrent buffer locks; in general, we only acquire a single pin
3375 : * and buffer lock at a time, which is often all that the nbtree code requires.
3376 : * (Actually, bt_recheck_sibling_links couples buffer locks, which is the only
3377 : * exception to this general rule.)
3378 : *
3379 : * Operating on a copy of the page is useful because it prevents control
3380 : * getting stuck in an uninterruptible state when an underlying operator class
3381 : * misbehaves.
3382 : */
3383 : static Page
3384 42530 : palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum)
3385 : {
3386 : Buffer buffer;
3387 : Page page;
3388 : BTPageOpaque opaque;
3389 : OffsetNumber maxoffset;
3390 :
3391 42530 : page = palloc(BLCKSZ);
3392 :
3393 : /*
3394 : * We copy the page into local storage to avoid holding pin on the buffer
3395 : * longer than we must.
3396 : */
3397 42530 : buffer = ReadBufferExtended(state->rel, MAIN_FORKNUM, blocknum, RBM_NORMAL,
3398 : state->checkstrategy);
3399 42506 : LockBuffer(buffer, BT_READ);
3400 :
3401 : /*
3402 : * Perform the same basic sanity checking that nbtree itself performs for
3403 : * every page:
3404 : */
3405 42506 : _bt_checkpage(state->rel, buffer);
3406 :
3407 : /* Only use copy of page in palloc()'d memory */
3408 42506 : memcpy(page, BufferGetPage(buffer), BLCKSZ);
3409 42506 : UnlockReleaseBuffer(buffer);
3410 :
3411 42506 : opaque = BTPageGetOpaque(page);
3412 :
3413 42506 : if (P_ISMETA(opaque) && blocknum != BTREE_METAPAGE)
3414 0 : ereport(ERROR,
3415 : (errcode(ERRCODE_INDEX_CORRUPTED),
3416 : errmsg("invalid meta page found at block %u in index \"%s\"",
3417 : blocknum, RelationGetRelationName(state->rel))));
3418 :
3419 : /* Check page from block that ought to be meta page */
3420 42506 : if (blocknum == BTREE_METAPAGE)
3421 : {
3422 8024 : BTMetaPageData *metad = BTPageGetMeta(page);
3423 :
3424 8024 : if (!P_ISMETA(opaque) ||
3425 8024 : metad->btm_magic != BTREE_MAGIC)
3426 0 : ereport(ERROR,
3427 : (errcode(ERRCODE_INDEX_CORRUPTED),
3428 : errmsg("index \"%s\" meta page is corrupt",
3429 : RelationGetRelationName(state->rel))));
3430 :
3431 8024 : if (metad->btm_version < BTREE_MIN_VERSION ||
3432 8024 : metad->btm_version > BTREE_VERSION)
3433 0 : ereport(ERROR,
3434 : (errcode(ERRCODE_INDEX_CORRUPTED),
3435 : errmsg("version mismatch in index \"%s\": file version %d, "
3436 : "current version %d, minimum supported version %d",
3437 : RelationGetRelationName(state->rel),
3438 : metad->btm_version, BTREE_VERSION,
3439 : BTREE_MIN_VERSION)));
3440 :
3441 : /* Finished with metapage checks */
3442 8024 : return page;
3443 : }
3444 :
3445 : /*
3446 : * Deleted pages that still use the old 32-bit XID representation have no
3447 : * sane "level" field because they type pun the field, but all other pages
3448 : * (including pages deleted on Postgres 14+) have a valid value.
3449 : */
3450 34482 : if (!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque))
3451 : {
3452 : /* Okay, no reason not to trust btpo_level field from page */
3453 :
3454 34482 : if (P_ISLEAF(opaque) && opaque->btpo_level != 0)
3455 0 : ereport(ERROR,
3456 : (errcode(ERRCODE_INDEX_CORRUPTED),
3457 : errmsg_internal("invalid leaf page level %u for block %u in index \"%s\"",
3458 : opaque->btpo_level, blocknum,
3459 : RelationGetRelationName(state->rel))));
3460 :
3461 34482 : if (!P_ISLEAF(opaque) && opaque->btpo_level == 0)
3462 0 : ereport(ERROR,
3463 : (errcode(ERRCODE_INDEX_CORRUPTED),
3464 : errmsg_internal("invalid internal page level 0 for block %u in index \"%s\"",
3465 : blocknum,
3466 : RelationGetRelationName(state->rel))));
3467 : }
3468 :
3469 : /*
3470 : * Sanity checks for number of items on page.
3471 : *
3472 : * As noted at the beginning of _bt_binsrch(), an internal page must have
3473 : * children, since there must always be a negative infinity downlink
3474 : * (there may also be a highkey). In the case of non-rightmost leaf
3475 : * pages, there must be at least a highkey. The exceptions are deleted
3476 : * pages, which contain no items.
3477 : *
3478 : * This is correct when pages are half-dead, since internal pages are
3479 : * never half-dead, and leaf pages must have a high key when half-dead
3480 : * (the rightmost page can never be deleted). It's also correct with
3481 : * fully deleted pages: _bt_unlink_halfdead_page() doesn't change anything
3482 : * about the target page other than setting the page as fully dead, and
3483 : * setting its xact field. In particular, it doesn't change the sibling
3484 : * links in the deletion target itself, since they're required when index
3485 : * scans land on the deletion target, and then need to move right (or need
3486 : * to move left, in the case of backward index scans).
3487 : */
3488 34482 : maxoffset = PageGetMaxOffsetNumber(page);
3489 34482 : if (maxoffset > MaxIndexTuplesPerPage)
3490 0 : ereport(ERROR,
3491 : (errcode(ERRCODE_INDEX_CORRUPTED),
3492 : errmsg("Number of items on block %u of index \"%s\" exceeds MaxIndexTuplesPerPage (%u)",
3493 : blocknum, RelationGetRelationName(state->rel),
3494 : MaxIndexTuplesPerPage)));
3495 :
3496 34482 : if (!P_ISLEAF(opaque) && !P_ISDELETED(opaque) && maxoffset < P_FIRSTDATAKEY(opaque))
3497 0 : ereport(ERROR,
3498 : (errcode(ERRCODE_INDEX_CORRUPTED),
3499 : errmsg("internal block %u in index \"%s\" lacks high key and/or at least one downlink",
3500 : blocknum, RelationGetRelationName(state->rel))));
3501 :
3502 34482 : if (P_ISLEAF(opaque) && !P_ISDELETED(opaque) && !P_RIGHTMOST(opaque) && maxoffset < P_HIKEY)
3503 0 : ereport(ERROR,
3504 : (errcode(ERRCODE_INDEX_CORRUPTED),
3505 : errmsg("non-rightmost leaf block %u in index \"%s\" lacks high key item",
3506 : blocknum, RelationGetRelationName(state->rel))));
3507 :
3508 : /*
3509 : * In general, internal pages are never marked half-dead, except on
3510 : * versions of Postgres prior to 9.4, where it can be valid transient
3511 : * state. This state is nonetheless treated as corruption by VACUUM on
3512 : * from version 9.4 on, so do the same here. See _bt_pagedel() for full
3513 : * details.
3514 : */
3515 34482 : if (!P_ISLEAF(opaque) && P_ISHALFDEAD(opaque))
3516 0 : ereport(ERROR,
3517 : (errcode(ERRCODE_INDEX_CORRUPTED),
3518 : errmsg("internal page block %u in index \"%s\" is half-dead",
3519 : blocknum, RelationGetRelationName(state->rel)),
3520 : errhint("This can be caused by an interrupted VACUUM in version 9.3 or older, before upgrade. Please REINDEX it.")));
3521 :
3522 : /*
3523 : * Check that internal pages have no garbage items, and that no page has
3524 : * an invalid combination of deletion-related page level flags
3525 : */
3526 34482 : if (!P_ISLEAF(opaque) && P_HAS_GARBAGE(opaque))
3527 0 : ereport(ERROR,
3528 : (errcode(ERRCODE_INDEX_CORRUPTED),
3529 : errmsg_internal("internal page block %u in index \"%s\" has garbage items",
3530 : blocknum, RelationGetRelationName(state->rel))));
3531 :
3532 34482 : if (P_HAS_FULLXID(opaque) && !P_ISDELETED(opaque))
3533 0 : ereport(ERROR,
3534 : (errcode(ERRCODE_INDEX_CORRUPTED),
3535 : errmsg_internal("full transaction id page flag appears in non-deleted block %u in index \"%s\"",
3536 : blocknum, RelationGetRelationName(state->rel))));
3537 :
3538 34482 : if (P_ISDELETED(opaque) && P_ISHALFDEAD(opaque))
3539 0 : ereport(ERROR,
3540 : (errcode(ERRCODE_INDEX_CORRUPTED),
3541 : errmsg_internal("deleted page block %u in index \"%s\" is half-dead",
3542 : blocknum, RelationGetRelationName(state->rel))));
3543 :
3544 34482 : return page;
3545 : }
3546 :
3547 : /*
3548 : * _bt_mkscankey() wrapper that automatically prevents insertion scankey from
3549 : * being considered greater than the pivot tuple that its values originated
3550 : * from (or some other identical pivot tuple) in the common case where there
3551 : * are truncated/minus infinity attributes. Without this extra step, there
3552 : * are forms of corruption that amcheck could theoretically fail to report.
3553 : *
3554 : * For example, invariant_g_offset() might miss a cross-page invariant failure
3555 : * on an internal level if the scankey built from the first item on the
3556 : * target's right sibling page happened to be equal to (not greater than) the
3557 : * last item on target page. The !backward tiebreaker in _bt_compare() might
3558 : * otherwise cause amcheck to assume (rather than actually verify) that the
3559 : * scankey is greater.
3560 : */
3561 : static inline BTScanInsert
3562 3992714 : bt_mkscankey_pivotsearch(Relation rel, IndexTuple itup)
3563 : {
3564 : BTScanInsert skey;
3565 :
3566 3992714 : skey = _bt_mkscankey(rel, itup);
3567 3992714 : skey->backward = true;
3568 :
3569 3992714 : return skey;
3570 : }
3571 :
3572 : /*
3573 : * PageGetItemId() wrapper that validates returned line pointer.
3574 : *
3575 : * Buffer page/page item access macros generally trust that line pointers are
3576 : * not corrupt, which might cause problems for verification itself. For
3577 : * example, there is no bounds checking in PageGetItem(). Passing it a
3578 : * corrupt line pointer can cause it to return a tuple/pointer that is unsafe
3579 : * to dereference.
3580 : *
3581 : * Validating line pointers before tuples avoids undefined behavior and
3582 : * assertion failures with corrupt indexes, making the verification process
3583 : * more robust and predictable.
3584 : */
3585 : static ItemId
3586 9186424 : PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block, Page page,
3587 : OffsetNumber offset)
3588 : {
3589 9186424 : ItemId itemid = PageGetItemId(page, offset);
3590 :
3591 9186424 : if (ItemIdGetOffset(itemid) + ItemIdGetLength(itemid) >
3592 : BLCKSZ - MAXALIGN(sizeof(BTPageOpaqueData)))
3593 0 : ereport(ERROR,
3594 : (errcode(ERRCODE_INDEX_CORRUPTED),
3595 : errmsg("line pointer points past end of tuple space in index \"%s\"",
3596 : RelationGetRelationName(state->rel)),
3597 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3598 : block, offset, ItemIdGetOffset(itemid),
3599 : ItemIdGetLength(itemid),
3600 : ItemIdGetFlags(itemid))));
3601 :
3602 : /*
3603 : * Verify that line pointer isn't LP_REDIRECT or LP_UNUSED, since nbtree
3604 : * never uses either. Verify that line pointer has storage, too, since
3605 : * even LP_DEAD items should within nbtree.
3606 : */
3607 9186424 : if (ItemIdIsRedirected(itemid) || !ItemIdIsUsed(itemid) ||
3608 9186424 : ItemIdGetLength(itemid) == 0)
3609 0 : ereport(ERROR,
3610 : (errcode(ERRCODE_INDEX_CORRUPTED),
3611 : errmsg("invalid line pointer storage in index \"%s\"",
3612 : RelationGetRelationName(state->rel)),
3613 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3614 : block, offset, ItemIdGetOffset(itemid),
3615 : ItemIdGetLength(itemid),
3616 : ItemIdGetFlags(itemid))));
3617 :
3618 9186424 : return itemid;
3619 : }
3620 :
3621 : /*
3622 : * BTreeTupleGetHeapTID() wrapper that enforces that a heap TID is present in
3623 : * cases where that is mandatory (i.e. for non-pivot tuples)
3624 : */
3625 : static inline ItemPointer
3626 3720 : BTreeTupleGetHeapTIDCareful(BtreeCheckState *state, IndexTuple itup,
3627 : bool nonpivot)
3628 : {
3629 : ItemPointer htid;
3630 :
3631 : /*
3632 : * Caller determines whether this is supposed to be a pivot or non-pivot
3633 : * tuple using page type and item offset number. Verify that tuple
3634 : * metadata agrees with this.
3635 : */
3636 : Assert(state->heapkeyspace);
3637 3720 : if (BTreeTupleIsPivot(itup) && nonpivot)
3638 0 : ereport(ERROR,
3639 : (errcode(ERRCODE_INDEX_CORRUPTED),
3640 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected pivot tuple",
3641 : state->targetblock,
3642 : RelationGetRelationName(state->rel))));
3643 :
3644 3720 : if (!BTreeTupleIsPivot(itup) && !nonpivot)
3645 0 : ereport(ERROR,
3646 : (errcode(ERRCODE_INDEX_CORRUPTED),
3647 : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected non-pivot tuple",
3648 : state->targetblock,
3649 : RelationGetRelationName(state->rel))));
3650 :
3651 3720 : htid = BTreeTupleGetHeapTID(itup);
3652 3720 : if (!ItemPointerIsValid(htid) && nonpivot)
3653 0 : ereport(ERROR,
3654 : (errcode(ERRCODE_INDEX_CORRUPTED),
3655 : errmsg("block %u or its right sibling block or child block in index \"%s\" contains non-pivot tuple that lacks a heap TID",
3656 : state->targetblock,
3657 : RelationGetRelationName(state->rel))));
3658 :
3659 3720 : return htid;
3660 : }
3661 :
3662 : /*
3663 : * Return the "pointed to" TID for itup, which is used to generate a
3664 : * descriptive error message. itup must be a "data item" tuple (it wouldn't
3665 : * make much sense to call here with a high key tuple, since there won't be a
3666 : * valid downlink/block number to display).
3667 : *
3668 : * Returns either a heap TID (which will be the first heap TID in posting list
3669 : * if itup is posting list tuple), or a TID that contains downlink block
3670 : * number, plus some encoded metadata (e.g., the number of attributes present
3671 : * in itup).
3672 : */
3673 : static inline ItemPointer
3674 12 : BTreeTupleGetPointsToTID(IndexTuple itup)
3675 : {
3676 : /*
3677 : * Rely on the assumption that !heapkeyspace internal page data items will
3678 : * correctly return TID with downlink here -- BTreeTupleGetHeapTID() won't
3679 : * recognize it as a pivot tuple, but everything still works out because
3680 : * the t_tid field is still returned
3681 : */
3682 12 : if (!BTreeTupleIsPivot(itup))
3683 8 : return BTreeTupleGetHeapTID(itup);
3684 :
3685 : /* Pivot tuple returns TID with downlink block (heapkeyspace variant) */
3686 4 : return &itup->t_tid;
3687 : }
|