Age Owner TLA 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) 2017-2023, PostgreSQL Global Development Group
18 : *
19 : * IDENTIFICATION
20 : * contrib/amcheck/verify_nbtree.c
21 : *
22 : *-------------------------------------------------------------------------
23 : */
24 : #include "postgres.h"
25 :
26 : #include "access/htup_details.h"
27 : #include "access/nbtree.h"
28 : #include "access/table.h"
29 : #include "access/tableam.h"
30 : #include "access/transam.h"
31 : #include "access/xact.h"
32 : #include "catalog/index.h"
33 : #include "catalog/pg_am.h"
34 : #include "commands/tablecmds.h"
35 : #include "common/pg_prng.h"
36 : #include "lib/bloomfilter.h"
37 : #include "miscadmin.h"
38 : #include "storage/lmgr.h"
39 : #include "storage/smgr.h"
40 : #include "utils/guc.h"
41 : #include "utils/memutils.h"
42 : #include "utils/snapmgr.h"
43 :
44 :
2222 andres 45 GIC 302 : PG_MODULE_MAGIC;
2222 andres 46 ECB :
47 : /*
48 : * A B-Tree cannot possibly have this many levels, since there must be one
49 : * block per level, which is bound by the range of BlockNumber:
50 : */
51 : #define InvalidBtreeLevel ((uint32) InvalidBlockNumber)
52 : #define BTreeTupleGetNKeyAtts(itup, rel) \
53 : Min(IndexRelationGetNumberOfKeyAttributes(rel), BTreeTupleGetNAtts(itup, rel))
54 :
55 : /*
56 : * State associated with verifying a B-Tree index
57 : *
58 : * target is the point of reference for a verification operation.
59 : *
60 : * Other B-Tree pages may be allocated, but those are always auxiliary (e.g.,
61 : * they are current target's child pages). Conceptually, problems are only
62 : * ever found in the current target page (or for a particular heap tuple during
63 : * heapallindexed verification). Each page found by verification's left/right,
64 : * top/bottom scan becomes the target exactly once.
65 : */
66 : typedef struct BtreeCheckState
67 : {
68 : /*
69 : * Unchanging state, established at start of verification:
70 : */
71 :
72 : /* B-Tree Index Relation and associated heap relation */
73 : Relation rel;
74 : Relation heaprel;
75 : /* rel is heapkeyspace index? */
76 : bool heapkeyspace;
77 : /* ShareLock held on heap/index, rather than AccessShareLock? */
78 : bool readonly;
79 : /* Also verifying heap has no unindexed tuples? */
80 : bool heapallindexed;
81 : /* Also making sure non-pivot tuples can be found by new search? */
82 : bool rootdescend;
83 : /* Per-page context */
84 : MemoryContext targetcontext;
85 : /* Buffer access strategy */
86 : BufferAccessStrategy checkstrategy;
87 :
88 : /*
89 : * Mutable state, for verification of particular page:
90 : */
91 :
92 : /* Current target page */
93 : Page target;
94 : /* Target block number */
95 : BlockNumber targetblock;
96 : /* Target page's LSN */
97 : XLogRecPtr targetlsn;
98 :
99 : /*
100 : * Low key: high key of left sibling of target page. Used only for child
101 : * verification. So, 'lowkey' is kept only when 'readonly' is set.
102 : */
103 : IndexTuple lowkey;
104 :
105 : /*
106 : * The rightlink and incomplete split flag of block one level down to the
107 : * target page, which was visited last time via downlink from taget page.
108 : * We use it to check for missing downlinks.
109 : */
110 : BlockNumber prevrightlink;
111 : bool previncompletesplit;
112 :
113 : /*
114 : * Mutable state, for optional heapallindexed verification:
115 : */
116 :
117 : /* Bloom filter fingerprints B-Tree index */
118 : bloom_filter *filter;
119 : /* Debug counter */
120 : int64 heaptuplespresent;
121 : } BtreeCheckState;
122 :
123 : /*
124 : * Starting point for verifying an entire B-Tree index level
125 : */
126 : typedef struct BtreeLevel
127 : {
128 : /* Level number (0 is leaf page level). */
129 : uint32 level;
130 :
131 : /* Left most block on level. Scan of level begins here. */
132 : BlockNumber leftmost;
133 :
134 : /* Is this level reported as "true" root level by meta page? */
135 : bool istruerootlevel;
136 : } BtreeLevel;
137 :
2222 andres 138 GIC 60 : PG_FUNCTION_INFO_V1(bt_index_check);
2222 andres 139 CBC 35 : PG_FUNCTION_INFO_V1(bt_index_parent_check);
2222 andres 140 ECB :
141 : static void bt_index_check_internal(Oid indrelid, bool parentcheck,
142 : bool heapallindexed, bool rootdescend);
143 : static inline void btree_index_checkable(Relation rel);
144 : static inline bool btree_index_mainfork_expected(Relation rel);
145 : static void bt_check_every_level(Relation rel, Relation heaprel,
146 : bool heapkeyspace, bool readonly, bool heapallindexed,
147 : bool rootdescend);
148 : static BtreeLevel bt_check_level_from_leftmost(BtreeCheckState *state,
149 : BtreeLevel level);
150 : static void bt_recheck_sibling_links(BtreeCheckState *state,
151 : BlockNumber btpo_prev_from_target,
152 : BlockNumber leftcurrent);
153 : static void bt_target_page_check(BtreeCheckState *state);
154 : static BTScanInsert bt_right_page_check_scankey(BtreeCheckState *state);
155 : static void bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
156 : OffsetNumber downlinkoffnum);
157 : static void bt_child_highkey_check(BtreeCheckState *state,
158 : OffsetNumber target_downlinkoffnum,
159 : Page loaded_child,
160 : uint32 target_level);
161 : static void bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
162 : BlockNumber blkno, Page page);
163 : static void bt_tuple_present_callback(Relation index, ItemPointer tid,
164 : Datum *values, bool *isnull,
165 : bool tupleIsAlive, void *checkstate);
166 : static IndexTuple bt_normalize_tuple(BtreeCheckState *state,
167 : IndexTuple itup);
168 : static inline IndexTuple bt_posting_plain_tuple(IndexTuple itup, int n);
169 : static bool bt_rootdescend(BtreeCheckState *state, IndexTuple itup);
170 : static inline bool offset_is_negative_infinity(BTPageOpaque opaque,
171 : OffsetNumber offset);
172 : static inline bool invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
173 : OffsetNumber upperbound);
174 : static inline bool invariant_leq_offset(BtreeCheckState *state,
175 : BTScanInsert key,
176 : OffsetNumber upperbound);
177 : static inline bool invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
178 : OffsetNumber lowerbound);
179 : static inline bool invariant_l_nontarget_offset(BtreeCheckState *state,
180 : BTScanInsert key,
181 : BlockNumber nontargetblock,
182 : Page nontarget,
183 : OffsetNumber upperbound);
184 : static Page palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum);
185 : static inline BTScanInsert bt_mkscankey_pivotsearch(Relation rel,
186 : Relation heaprel,
187 : IndexTuple itup);
188 : static ItemId PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block,
189 : Page page, OffsetNumber offset);
190 : static inline ItemPointer BTreeTupleGetHeapTIDCareful(BtreeCheckState *state,
191 : IndexTuple itup, bool nonpivot);
192 : static inline ItemPointer BTreeTupleGetPointsToTID(IndexTuple itup);
193 :
194 : /*
195 : * bt_index_check(index regclass, heapallindexed boolean)
196 : *
197 : * Verify integrity of B-Tree index.
198 : *
199 : * Acquires AccessShareLock on heap & index relations. Does not consider
200 : * invariants that exist between parent/child pages. Optionally verifies
201 : * that heap does not contain any unindexed or incorrectly indexed tuples.
202 : */
203 : Datum
2222 andres 204 GIC 2852 : bt_index_check(PG_FUNCTION_ARGS)
205 : {
2222 andres 206 CBC 2852 : Oid indrelid = PG_GETARG_OID(0);
1835 andres 207 GIC 2852 : bool heapallindexed = false;
1835 andres 208 ECB :
1835 andres 209 CBC 2852 : if (PG_NARGS() == 2)
1835 andres 210 GIC 2846 : heapallindexed = PG_GETARG_BOOL(1);
2222 andres 211 ECB :
1481 pg 212 CBC 2852 : bt_index_check_internal(indrelid, false, heapallindexed, false);
213 :
2222 andres 214 2832 : PG_RETURN_VOID();
215 : }
2222 andres 216 ECB :
217 : /*
218 : * bt_index_parent_check(index regclass, heapallindexed boolean)
219 : *
220 : * Verify integrity of B-Tree index.
221 : *
222 : * Acquires ShareLock on heap & index relations. Verifies that downlinks in
223 : * parent pages are valid lower bounds on child pages. Optionally verifies
224 : * that heap does not contain any unindexed or incorrectly indexed tuples.
225 : */
226 : Datum
2222 andres 227 GIC 29 : bt_index_parent_check(PG_FUNCTION_ARGS)
228 : {
2222 andres 229 CBC 29 : Oid indrelid = PG_GETARG_OID(0);
1835 andres 230 GIC 29 : bool heapallindexed = false;
1481 pg 231 CBC 29 : bool rootdescend = false;
1835 andres 232 ECB :
1481 pg 233 CBC 29 : if (PG_NARGS() >= 2)
1835 andres 234 GIC 24 : heapallindexed = PG_GETARG_BOOL(1);
1481 pg 235 CBC 29 : if (PG_NARGS() == 3)
236 22 : rootdescend = PG_GETARG_BOOL(2);
2222 andres 237 ECB :
1481 pg 238 CBC 29 : bt_index_check_internal(indrelid, true, heapallindexed, rootdescend);
239 :
2222 andres 240 19 : PG_RETURN_VOID();
241 : }
2222 andres 242 ECB :
243 : /*
244 : * Helper for bt_index_[parent_]check, coordinating the bulk of the work.
245 : */
246 : static void
1481 pg 247 GIC 2881 : bt_index_check_internal(Oid indrelid, bool parentcheck, bool heapallindexed,
248 : bool rootdescend)
2222 andres 249 ECB : {
250 : Oid heapid;
251 : Relation indrel;
252 : Relation heaprel;
253 : LOCKMODE lockmode;
254 : Oid save_userid;
255 : int save_sec_context;
256 : int save_nestlevel;
257 :
2222 andres 258 GIC 2881 : if (parentcheck)
259 29 : lockmode = ShareLock;
2222 andres 260 ECB : else
2222 andres 261 CBC 2852 : lockmode = AccessShareLock;
262 :
2222 andres 263 ECB : /*
264 : * We must lock table before index to avoid deadlocks. However, if the
265 : * passed indrelid isn't an index then IndexGetRelation() will fail.
266 : * Rather than emitting a not-very-helpful error message, postpone
267 : * complaining, expecting that the is-it-an-index test below will fail.
268 : *
269 : * In hot standby mode this will raise an error when parentcheck is true.
270 : */
2222 andres 271 GIC 2881 : heapid = IndexGetRelation(indrelid, true);
272 2881 : if (OidIsValid(heapid))
335 noah 273 ECB : {
1539 andres 274 CBC 2877 : heaprel = table_open(heapid, lockmode);
275 :
335 noah 276 ECB : /*
277 : * Switch to the table owner's userid, so that any index functions are
278 : * run as that user. Also lock down security-restricted operations
279 : * and arrange to make GUC variable changes local to this command.
280 : */
335 noah 281 GIC 2877 : GetUserIdAndSecContext(&save_userid, &save_sec_context);
282 2877 : SetUserIdAndSecContext(heaprel->rd_rel->relowner,
335 noah 283 ECB : save_sec_context | SECURITY_RESTRICTED_OPERATION);
335 noah 284 CBC 2877 : save_nestlevel = NewGUCNestLevel();
285 : }
2222 andres 286 ECB : else
287 : {
2222 andres 288 GIC 4 : heaprel = NULL;
289 : /* Set these just to suppress "uninitialized variable" warnings */
335 noah 290 CBC 4 : save_userid = InvalidOid;
335 noah 291 GIC 4 : save_sec_context = -1;
335 noah 292 CBC 4 : save_nestlevel = -1;
335 noah 293 ECB : }
2222 andres 294 :
295 : /*
296 : * Open the target index relations separately (like relation_openrv(), but
297 : * with heap relation locked first to prevent deadlocking). In hot
298 : * standby mode this will raise an error when parentcheck is true.
299 : *
300 : * There is no need for the usual indcheckxmin usability horizon test
301 : * here, even in the heapallindexed case, because index undergoing
302 : * verification only needs to have entries for a new transaction snapshot.
303 : * (If this is a parentcheck verification, there is no question about
304 : * committed or recently dead heap tuples lacking index entries due to
305 : * concurrent activity.)
306 : */
2222 andres 307 GIC 2881 : indrel = index_open(indrelid, lockmode);
308 :
2222 andres 309 ECB : /*
310 : * Since we did the IndexGetRelation call above without any lock, it's
311 : * barely possible that a race against an index drop/recreation could have
312 : * netted us the wrong table.
313 : */
2222 andres 314 GIC 2877 : if (heaprel == NULL || heapid != IndexGetRelation(indrelid, false))
2222 andres 315 UIC 0 : ereport(ERROR,
2222 andres 316 ECB : (errcode(ERRCODE_UNDEFINED_TABLE),
725 peter 317 EUB : errmsg("could not open parent table of index \"%s\"",
318 : RelationGetRelationName(indrel))));
319 :
320 : /* Relation suitable for checking as B-Tree? */
2222 andres 321 GIC 2877 : btree_index_checkable(indrel);
322 :
1336 pg 323 CBC 2876 : if (btree_index_mainfork_expected(indrel))
324 : {
1138 pg 325 ECB : bool heapkeyspace,
326 : allequalimage;
327 :
636 tgl 328 GIC 2876 : if (!smgrexists(RelationGetSmgr(indrel), MAIN_FORKNUM))
1336 pg 329 14 : ereport(ERROR,
1336 pg 330 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
331 : errmsg("index \"%s\" lacks a main relation fork",
332 : RelationGetRelationName(indrel))));
333 :
334 : /* Extract metadata from metapage, and sanitize it in passing */
8 andres 335 GNC 2862 : _bt_metaversion(indrel, heaprel, &heapkeyspace, &allequalimage);
976 pg 336 GIC 2862 : if (allequalimage && !heapkeyspace)
976 pg 337 LBC 0 : ereport(ERROR,
976 pg 338 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
976 pg 339 EUB : errmsg("index \"%s\" metapage has equalimage field set on unsupported nbtree version",
340 : RelationGetRelationName(indrel))));
976 pg 341 GIC 2862 : if (allequalimage && !_bt_allequalimage(indrel, false))
976 pg 342 UIC 0 : ereport(ERROR,
976 pg 343 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
976 pg 344 EUB : errmsg("index \"%s\" metapage incorrectly indicates that deduplication is safe",
345 : RelationGetRelationName(indrel))));
346 :
347 : /* Check index, possibly against table it is an index on */
1336 pg 348 GIC 2862 : bt_check_every_level(indrel, heaprel, heapkeyspace, parentcheck,
349 : heapallindexed, rootdescend);
1336 pg 350 ECB : }
351 :
352 : /* Roll back any GUC changes executed by index functions */
335 noah 353 GIC 2851 : AtEOXact_GUC(false, save_nestlevel);
354 :
335 noah 355 ECB : /* Restore userid and security context */
335 noah 356 GIC 2851 : SetUserIdAndSecContext(save_userid, save_sec_context);
357 :
2222 andres 358 ECB : /*
359 : * Release locks early. That's ok here because nothing in the called
360 : * routines will trigger shared cache invalidations to be sent, so we can
361 : * relax the usual pattern of only releasing locks after commit.
362 : */
2222 andres 363 GIC 2851 : index_close(indrel, lockmode);
364 2851 : if (heaprel)
1539 andres 365 CBC 2851 : table_close(heaprel, lockmode);
2222 366 2851 : }
2222 andres 367 ECB :
368 : /*
369 : * Basic checks about the suitability of a relation for checking as a B-Tree
370 : * index.
371 : *
372 : * NB: Intentionally not checking permissions, the function is normally not
373 : * callable by non-superusers. If granted, it's useful to be able to check a
374 : * whole cluster.
375 : */
376 : static inline void
2222 andres 377 GIC 2877 : btree_index_checkable(Relation rel)
378 : {
2222 andres 379 CBC 2877 : if (rel->rd_rel->relkind != RELKIND_INDEX ||
2222 andres 380 GIC 2877 : rel->rd_rel->relam != BTREE_AM_OID)
2222 andres 381 CBC 1 : ereport(ERROR,
2222 andres 382 ECB : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
383 : errmsg("only B-Tree indexes are supported as targets for verification"),
384 : errdetail("Relation \"%s\" is not a B-Tree index.",
385 : RelationGetRelationName(rel))));
386 :
2222 andres 387 GIC 2876 : if (RELATION_IS_OTHER_TEMP(rel))
2222 andres 388 UIC 0 : ereport(ERROR,
2222 andres 389 ECB : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2222 andres 390 EUB : errmsg("cannot access temporary tables of other sessions"),
391 : errdetail("Index \"%s\" is associated with temporary relation.",
392 : RelationGetRelationName(rel))));
393 :
1564 peter_e 394 GIC 2876 : if (!rel->rd_index->indisvalid)
2222 andres 395 UIC 0 : ereport(ERROR,
2222 andres 396 ECB : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2222 andres 397 EUB : errmsg("cannot check index \"%s\"",
398 : RelationGetRelationName(rel)),
399 : errdetail("Index is not valid.")));
2222 andres 400 GIC 2876 : }
401 :
1336 pg 402 ECB : /*
403 : * Check if B-Tree index relation should have a file for its main relation
404 : * fork. Verification uses this to skip unlogged indexes when in hot standby
405 : * mode, where there is simply nothing to verify. We behave as if the
406 : * relation is empty.
407 : *
408 : * NB: Caller should call btree_index_checkable() before calling here.
409 : */
410 : static inline bool
1336 pg 411 GIC 2876 : btree_index_mainfork_expected(Relation rel)
412 : {
1336 pg 413 CBC 2876 : if (rel->rd_rel->relpersistence != RELPERSISTENCE_UNLOGGED ||
1336 pg 414 UIC 0 : !RecoveryInProgress())
1336 pg 415 CBC 2876 : return true;
1336 pg 416 EUB :
545 pg 417 LBC 0 : ereport(DEBUG1,
418 : (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
1336 pg 419 EUB : errmsg("cannot verify unlogged index \"%s\" during recovery, skipping",
420 : RelationGetRelationName(rel))));
421 :
1336 pg 422 UIC 0 : return false;
423 : }
1336 pg 424 EUB :
425 : /*
426 : * Main entry point for B-Tree SQL-callable functions. Walks the B-Tree in
427 : * logical order, verifying invariants as it goes. Optionally, verification
428 : * checks if the heap relation contains any tuples that are not represented in
429 : * the index but should be.
430 : *
431 : * It is the caller's responsibility to acquire appropriate heavyweight lock on
432 : * the index relation, and advise us if extra checks are safe when a ShareLock
433 : * is held. (A lock of the same type must also have been acquired on the heap
434 : * relation.)
435 : *
436 : * A ShareLock is generally assumed to prevent any kind of physical
437 : * modification to the index structure, including modifications that VACUUM may
438 : * make. This does not include setting of the LP_DEAD bit by concurrent index
439 : * scans, although that is just metadata that is not able to directly affect
440 : * any check performed here. Any concurrent process that might act on the
441 : * LP_DEAD bit being set (recycle space) requires a heavyweight lock that
442 : * cannot be held while we hold a ShareLock. (Besides, even if that could
443 : * happen, the ad-hoc recycling when a page might otherwise split is performed
444 : * per-page, and requires an exclusive buffer lock, which wouldn't cause us
445 : * trouble. _bt_delitems_vacuum() may only delete leaf items, and so the extra
446 : * parent/child check cannot be affected.)
447 : */
448 : static void
1481 pg 449 GIC 2862 : bt_check_every_level(Relation rel, Relation heaprel, bool heapkeyspace,
450 : bool readonly, bool heapallindexed, bool rootdescend)
2222 andres 451 ECB : {
452 : BtreeCheckState *state;
453 : Page metapage;
454 : BTMetaPageData *metad;
455 : uint32 previouslevel;
456 : BtreeLevel current;
1835 andres 457 GIC 2862 : Snapshot snapshot = SnapshotAny;
458 :
1094 pg 459 CBC 2862 : if (!readonly)
1094 pg 460 GIC 2840 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\"",
1094 pg 461 ECB : RelationGetRelationName(rel));
462 : else
1094 pg 463 GIC 22 : elog(DEBUG1, "verifying consistency of tree structure for index \"%s\" with cross-level checks",
464 : RelationGetRelationName(rel));
1094 pg 465 ECB :
466 : /*
467 : * This assertion matches the one in index_getnext_tid(). See page
468 : * recycling/"visible to everyone" notes in nbtree README.
469 : */
970 andres 470 GIC 2862 : Assert(TransactionIdIsValid(RecentXmin));
471 :
2222 andres 472 ECB : /*
473 : * Initialize state for entire verification operation
474 : */
1810 teodor 475 GIC 2862 : state = palloc0(sizeof(BtreeCheckState));
2222 andres 476 2862 : state->rel = rel;
1835 andres 477 CBC 2862 : state->heaprel = heaprel;
1481 pg 478 2862 : state->heapkeyspace = heapkeyspace;
2222 andres 479 2862 : state->readonly = readonly;
1835 480 2862 : state->heapallindexed = heapallindexed;
1481 pg 481 2862 : state->rootdescend = rootdescend;
1835 andres 482 ECB :
1835 andres 483 CBC 2862 : if (state->heapallindexed)
484 : {
1359 pg 485 ECB : int64 total_pages;
486 : int64 total_elems;
487 : uint64 seed;
488 :
489 : /*
490 : * Size Bloom filter based on estimated number of tuples in index,
491 : * while conservatively assuming that each block must contain at least
492 : * MaxTIDsPerBTreePage / 3 "plain" tuples -- see
493 : * bt_posting_plain_tuple() for definition, and details of how posting
494 : * list tuples are handled.
495 : */
1359 pg 496 GIC 56 : total_pages = RelationGetNumberOfBlocks(rel);
1138 497 56 : total_elems = Max(total_pages * (MaxTIDsPerBTreePage / 3),
1359 pg 498 ECB : (int64) state->rel->rd_rel->reltuples);
497 tgl 499 : /* Generate a random seed to avoid repetition */
497 tgl 500 GIC 56 : seed = pg_prng_uint64(&pg_global_prng_state);
501 : /* Create Bloom filter to fingerprint index */
1835 andres 502 CBC 56 : state->filter = bloom_create(total_elems, maintenance_work_mem, seed);
1835 andres 503 GIC 56 : state->heaptuplespresent = 0;
1835 andres 504 ECB :
505 : /*
506 : * Register our own snapshot in !readonly case, rather than asking
507 : * table_index_build_scan() to do this for us later. This needs to
508 : * happen before index fingerprinting begins, so we can later be
509 : * certain that index fingerprinting should have reached all tuples
510 : * returned by table_index_build_scan().
511 : */
1835 andres 512 GIC 56 : if (!state->readonly)
513 : {
1835 andres 514 CBC 44 : snapshot = RegisterSnapshot(GetTransactionSnapshot());
515 :
1835 andres 516 ECB : /*
517 : * GetTransactionSnapshot() always acquires a new MVCC snapshot in
518 : * READ COMMITTED mode. A new snapshot is guaranteed to have all
519 : * the entries it requires in the index.
520 : *
521 : * We must defend against the possibility that an old xact
522 : * snapshot was returned at higher isolation levels when that
523 : * snapshot is not safe for index scans of the target index. This
524 : * is possible when the snapshot sees tuples that are before the
525 : * index's indcheckxmin horizon. Throwing an error here should be
526 : * very rare. It doesn't seem worth using a secondary snapshot to
527 : * avoid this.
528 : */
1835 andres 529 GIC 44 : if (IsolationUsesXactSnapshot() && rel->rd_index->indcheckxmin &&
1835 andres 530 UIC 0 : !TransactionIdPrecedes(HeapTupleHeaderGetXmin(rel->rd_indextuple->t_data),
1835 andres 531 ECB : snapshot->xmin))
1835 andres 532 UBC 0 : ereport(ERROR,
533 : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1835 andres 534 EUB : errmsg("index \"%s\" cannot be verified using transaction snapshot",
535 : RelationGetRelationName(rel))));
536 : }
537 : }
538 :
1481 pg 539 GIC 2862 : Assert(!state->rootdescend || state->readonly);
540 2862 : if (state->rootdescend && !state->heapkeyspace)
1481 pg 541 LBC 0 : ereport(ERROR,
1481 pg 542 ECB : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1481 pg 543 EUB : errmsg("cannot verify that tuples from index \"%s\" can each be found by an independent index search",
544 : RelationGetRelationName(rel)),
545 : errhint("Only B-Tree version 4 indexes support rootdescend verification.")));
546 :
547 : /* Create context for page */
2222 andres 548 GIC 2862 : state->targetcontext = AllocSetContextCreate(CurrentMemoryContext,
549 : "amcheck context",
1943 tgl 550 ECB : ALLOCSET_DEFAULT_SIZES);
2222 andres 551 GIC 2862 : state->checkstrategy = GetAccessStrategy(BAS_BULKREAD);
552 :
2222 andres 553 ECB : /* Get true root block from meta-page */
2222 andres 554 GIC 2862 : metapage = palloc_btree_page(state, BTREE_METAPAGE);
555 2862 : metad = BTPageGetMeta(metapage);
2222 andres 556 ECB :
557 : /*
558 : * Certain deletion patterns can result in "skinny" B-Tree indexes, where
559 : * the fast root and true root differ.
560 : *
561 : * Start from the true root, not the fast root, unlike conventional index
562 : * scans. This approach is more thorough, and removes the risk of
563 : * following a stale fast root from the meta page.
564 : */
2222 andres 565 GIC 2862 : if (metad->btm_fastroot != metad->btm_root)
566 11 : ereport(DEBUG1,
2222 andres 567 ECB : (errcode(ERRCODE_NO_DATA),
725 peter 568 : errmsg_internal("harmless fast root mismatch in index \"%s\"",
569 : RelationGetRelationName(rel)),
570 : errdetail_internal("Fast root block %u (level %u) differs from true root block %u (level %u).",
571 : metad->btm_fastroot, metad->btm_fastlevel,
572 : metad->btm_root, metad->btm_level)));
573 :
574 : /*
575 : * Starting at the root, verify every level. Move left to right, top to
576 : * bottom. Note that there may be no pages other than the meta page (meta
577 : * page can indicate that root is P_NONE when the index is totally empty).
578 : */
2222 andres 579 GIC 2862 : previouslevel = InvalidBtreeLevel;
580 2862 : current.level = metad->btm_level;
2222 andres 581 CBC 2862 : current.leftmost = metad->btm_root;
582 2862 : current.istruerootlevel = true;
583 4580 : while (current.leftmost != P_NONE)
2222 andres 584 ECB : {
585 : /*
586 : * Verify this level, and get left most page for next level down, if
587 : * not at leaf level
588 : */
2222 andres 589 GIC 1728 : current = bt_check_level_from_leftmost(state, current);
590 :
2222 andres 591 CBC 1718 : if (current.leftmost == InvalidBlockNumber)
2222 andres 592 UIC 0 : ereport(ERROR,
2222 andres 593 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 594 EUB : errmsg("index \"%s\" has no valid pages on level below %u or first level",
595 : RelationGetRelationName(rel), previouslevel)));
596 :
2222 andres 597 GIC 1718 : previouslevel = current.level;
598 : }
2222 andres 599 ECB :
600 : /*
601 : * * Check whether heap contains unindexed/malformed tuples *
602 : */
1835 andres 603 GIC 2852 : if (state->heapallindexed)
604 : {
1835 andres 605 CBC 55 : IndexInfo *indexinfo = BuildIndexInfo(state->rel);
606 : TableScanDesc scan;
1835 andres 607 ECB :
608 : /*
609 : * Create our own scan for table_index_build_scan(), rather than
610 : * getting it to do so for us. This is required so that we can
611 : * actually use the MVCC snapshot registered earlier in !readonly
612 : * case.
613 : *
614 : * Note that table_index_build_scan() calls heap_endscan() for us.
615 : */
1474 andres 616 GIC 55 : scan = table_beginscan_strat(state->heaprel, /* relation */
617 : snapshot, /* snapshot */
1474 andres 618 ECB : 0, /* number of keys */
619 : NULL, /* scan key */
620 : true, /* buffer access strategy OK */
621 : true); /* syncscan OK? */
622 :
623 : /*
624 : * Scan will behave as the first scan of a CREATE INDEX CONCURRENTLY
625 : * behaves in !readonly case.
626 : *
627 : * It's okay that we don't actually use the same lock strength for the
628 : * heap relation as any other ii_Concurrent caller would in !readonly
629 : * case. We have no reason to care about a concurrent VACUUM
630 : * operation, since there isn't going to be a second scan of the heap
631 : * that needs to be sure that there was no concurrent recycling of
632 : * TIDs.
633 : */
1835 andres 634 GIC 54 : indexinfo->ii_Concurrent = !state->readonly;
635 :
1835 andres 636 ECB : /*
637 : * Don't wait for uncommitted tuple xact commit/abort when index is a
638 : * unique index on a catalog (or an index used by an exclusion
639 : * constraint). This could otherwise happen in the readonly case.
640 : */
1835 andres 641 GIC 54 : indexinfo->ii_Unique = false;
642 54 : indexinfo->ii_ExclusionOps = NULL;
1835 andres 643 CBC 54 : indexinfo->ii_ExclusionProcs = NULL;
644 54 : indexinfo->ii_ExclusionStrats = NULL;
1835 andres 645 ECB :
1835 andres 646 CBC 54 : elog(DEBUG1, "verifying that tuples from index \"%s\" are present in \"%s\"",
647 : RelationGetRelationName(state->rel),
1835 andres 648 ECB : RelationGetRelationName(state->heaprel));
649 :
1468 alvherre 650 GIC 54 : table_index_build_scan(state->heaprel, state->rel, indexinfo, true, false,
651 : bt_tuple_present_callback, (void *) state, scan);
1835 andres 652 ECB :
1835 andres 653 GIC 54 : ereport(DEBUG1,
654 : (errmsg_internal("finished verifying presence of " INT64_FORMAT " tuples from table \"%s\" with bitset %.2f%% set",
1835 andres 655 ECB : state->heaptuplespresent, RelationGetRelationName(heaprel),
656 : 100.0 * bloom_prop_bits_set(state->filter))));
657 :
1835 andres 658 GIC 54 : if (snapshot != SnapshotAny)
659 44 : UnregisterSnapshot(snapshot);
1835 andres 660 ECB :
1835 andres 661 CBC 54 : bloom_free(state->filter);
662 : }
1835 andres 663 ECB :
664 : /* Be tidy: */
2222 andres 665 GIC 2851 : MemoryContextDelete(state->targetcontext);
666 2851 : }
2222 andres 667 ECB :
668 : /*
669 : * Given a left-most block at some level, move right, verifying each page
670 : * individually (with more verification across pages for "readonly"
671 : * callers). Caller should pass the true root page as the leftmost initially,
672 : * working their way down by passing what is returned for the last call here
673 : * until level 0 (leaf page level) was reached.
674 : *
675 : * Returns state for next call, if any. This includes left-most block number
676 : * one level lower that should be passed on next level/call, which is set to
677 : * P_NONE on last call here (when leaf level is verified). Level numbers
678 : * follow the nbtree convention: higher levels have higher numbers, because new
679 : * levels are added only due to a root page split. Note that prior to the
680 : * first root page split, the root is also a leaf page, so there is always a
681 : * level 0 (leaf level), and it's always the last level processed.
682 : *
683 : * Note on memory management: State's per-page context is reset here, between
684 : * each call to bt_target_page_check().
685 : */
686 : static BtreeLevel
2222 andres 687 GIC 1728 : bt_check_level_from_leftmost(BtreeCheckState *state, BtreeLevel level)
688 : {
2222 andres 689 ECB : /* State to establish early, concerning entire level */
690 : BTPageOpaque opaque;
691 : MemoryContext oldcontext;
692 : BtreeLevel nextleveldown;
693 :
694 : /* Variables for iterating across level using right links */
2222 andres 695 GIC 1728 : BlockNumber leftcurrent = P_NONE;
696 1728 : BlockNumber current = level.leftmost;
2222 andres 697 ECB :
698 : /* Initialize return state */
2222 andres 699 GIC 1728 : nextleveldown.leftmost = InvalidBlockNumber;
700 1728 : nextleveldown.level = InvalidBtreeLevel;
2222 andres 701 CBC 1728 : nextleveldown.istruerootlevel = false;
2222 andres 702 ECB :
703 : /* Use page-level context for duration of this call */
2222 andres 704 GIC 1728 : oldcontext = MemoryContextSwitchTo(state->targetcontext);
705 :
1094 pg 706 CBC 1728 : elog(DEBUG1, "verifying level %u%s", level.level,
707 : level.istruerootlevel ?
2222 andres 708 ECB : " (true root level)" : level.level == 0 ? " (leaf level)" : "");
709 :
1124 akorotkov 710 GIC 1728 : state->prevrightlink = InvalidBlockNumber;
711 1728 : state->previncompletesplit = false;
1124 akorotkov 712 ECB :
2222 andres 713 : do
714 : {
715 : /* Don't rely on CHECK_FOR_INTERRUPTS() calls at lower level */
2222 andres 716 GIC 6883 : CHECK_FOR_INTERRUPTS();
717 :
2222 andres 718 ECB : /* Initialize state for this iteration */
2222 andres 719 GIC 6883 : state->targetblock = current;
720 6883 : state->target = palloc_btree_page(state, state->targetblock);
2222 andres 721 CBC 6873 : state->targetlsn = PageGetLSN(state->target);
2222 andres 722 ECB :
373 michael 723 CBC 6873 : opaque = BTPageGetOpaque(state->target);
724 :
2222 andres 725 6873 : if (P_IGNORE(opaque))
726 : {
1810 teodor 727 ECB : /*
728 : * Since there cannot be a concurrent VACUUM operation in readonly
729 : * mode, and since a page has no links within other pages
730 : * (siblings and parent) once it is marked fully deleted, it
731 : * should be impossible to land on a fully deleted page in
732 : * readonly mode. See bt_child_check() for further details.
733 : *
734 : * The bt_child_check() P_ISDELETED() check is repeated here so
735 : * that pages that are only reachable through sibling links get
736 : * checked.
737 : */
1810 teodor 738 UIC 0 : if (state->readonly && P_ISDELETED(opaque))
739 0 : ereport(ERROR,
1810 teodor 740 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
741 : errmsg("downlink or sibling link points to deleted block in index \"%s\"",
742 : RelationGetRelationName(state->rel)),
743 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
744 : current, leftcurrent, opaque->btpo_prev)));
745 :
2222 andres 746 UIC 0 : if (P_RIGHTMOST(opaque))
747 0 : ereport(ERROR,
2222 andres 748 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
749 : errmsg("block %u fell off the end of index \"%s\"",
750 : current, RelationGetRelationName(state->rel))));
751 : else
2222 andres 752 UIC 0 : ereport(DEBUG1,
753 : (errcode(ERRCODE_NO_DATA),
754 pg 754 EUB : errmsg_internal("block %u of index \"%s\" concurrently deleted",
755 : current, RelationGetRelationName(state->rel))));
2222 andres 756 UIC 0 : goto nextpage;
757 : }
2222 andres 758 GBC 6873 : else if (nextleveldown.leftmost == InvalidBlockNumber)
759 : {
2222 andres 760 ECB : /*
761 : * A concurrent page split could make the caller supplied leftmost
762 : * block no longer contain the leftmost page, or no longer be the
763 : * true root, but where that isn't possible due to heavyweight
764 : * locking, check that the first valid page meets caller's
765 : * expectations.
766 : */
2222 andres 767 GIC 1718 : if (state->readonly)
768 : {
2222 andres 769 CBC 22 : if (!P_LEFTMOST(opaque))
2222 andres 770 UIC 0 : ereport(ERROR,
2222 andres 771 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2118 tgl 772 EUB : errmsg("block %u is not leftmost in index \"%s\"",
773 : current, RelationGetRelationName(state->rel))));
774 :
2222 andres 775 GIC 22 : if (level.istruerootlevel && !P_ISROOT(opaque))
2222 andres 776 UIC 0 : ereport(ERROR,
2222 andres 777 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2118 tgl 778 EUB : errmsg("block %u is not true root in index \"%s\"",
779 : current, RelationGetRelationName(state->rel))));
780 : }
781 :
782 : /*
783 : * Before beginning any non-trivial examination of level, prepare
784 : * state for next bt_check_level_from_leftmost() invocation for
785 : * the next level for the next level down (if any).
786 : *
787 : * There should be at least one non-ignorable page per level,
788 : * unless this is the leaf level, which is assumed by caller to be
789 : * final level.
790 : */
2222 andres 791 GIC 1718 : if (!P_ISLEAF(opaque))
792 : {
2222 andres 793 ECB : IndexTuple itup;
794 : ItemId itemid;
795 :
796 : /* Internal page -- downlink gets leftmost on next level */
1445 pg 797 GIC 358 : itemid = PageGetItemIdCareful(state, state->targetblock,
798 : state->target,
1445 pg 799 CBC 358 : P_FIRSTDATAKEY(opaque));
2222 andres 800 GIC 358 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1210 pg 801 CBC 358 : nextleveldown.leftmost = BTreeTupleGetDownLink(itup);
774 802 358 : nextleveldown.level = opaque->btpo_level - 1;
2222 andres 803 ECB : }
804 : else
805 : {
806 : /*
807 : * Leaf page -- final level caller must process.
808 : *
809 : * Note that this could also be the root page, if there has
810 : * been no root page split yet.
811 : */
2222 andres 812 GIC 1360 : nextleveldown.leftmost = P_NONE;
813 1360 : nextleveldown.level = InvalidBtreeLevel;
2222 andres 814 ECB : }
815 :
816 : /*
817 : * Finished setting up state for this call/level. Control will
818 : * never end up back here in any future loop iteration for this
819 : * level.
820 : */
821 : }
822 :
823 : /* Sibling links should be in mutual agreement */
974 pg 824 GIC 6873 : if (opaque->btpo_prev != leftcurrent)
974 pg 825 UIC 0 : bt_recheck_sibling_links(state, opaque->btpo_prev, leftcurrent);
2222 andres 826 ECB :
774 pg 827 EUB : /* Check level */
774 pg 828 GIC 6873 : if (level.level != opaque->btpo_level)
2222 andres 829 UIC 0 : ereport(ERROR,
2222 andres 830 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 831 EUB : errmsg("leftmost down link for level points to block in index \"%s\" whose level is not one level down",
832 : RelationGetRelationName(state->rel)),
833 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
834 : current, level.level, opaque->btpo_level)));
835 :
836 : /* Verify invariants for page */
2222 andres 837 GIC 6873 : bt_target_page_check(state);
838 :
2222 andres 839 CBC 6873 : nextpage:
840 :
2222 andres 841 ECB : /* Try to detect circular links */
2222 andres 842 GIC 6873 : if (current == leftcurrent || current == opaque->btpo_prev)
2222 andres 843 UIC 0 : ereport(ERROR,
2222 andres 844 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2118 tgl 845 EUB : errmsg("circular link chain found in block %u of index \"%s\"",
846 : current, RelationGetRelationName(state->rel))));
847 :
1124 akorotkov 848 GIC 6873 : leftcurrent = current;
849 6873 : current = opaque->btpo_next;
1124 akorotkov 850 ECB :
1124 akorotkov 851 CBC 6873 : if (state->lowkey)
852 : {
853 1586 : Assert(state->readonly);
1124 akorotkov 854 GIC 1586 : pfree(state->lowkey);
1124 akorotkov 855 CBC 1586 : state->lowkey = NULL;
1124 akorotkov 856 ECB : }
857 :
858 : /*
859 : * Copy current target high key as the low key of right sibling.
860 : * Allocate memory in upper level context, so it would be cleared
861 : * after reset of target context.
862 : *
863 : * We only need the low key in corner cases of checking child high
864 : * keys. We use high key only when incomplete split on the child level
865 : * falls to the boundary of pages on the target level. See
866 : * bt_child_highkey_check() for details. So, typically we won't end
867 : * up doing anything with low key, but it's simpler for general case
868 : * high key verification to always have it available.
869 : *
870 : * The correctness of managing low key in the case of concurrent
871 : * splits wasn't investigated yet. Thankfully we only need low key
872 : * for readonly verification and concurrent splits won't happen.
873 : */
1124 akorotkov 874 GIC 6873 : if (state->readonly && !P_RIGHTMOST(opaque))
875 : {
1124 akorotkov 876 ECB : IndexTuple itup;
877 : ItemId itemid;
878 :
1124 akorotkov 879 GIC 1586 : itemid = PageGetItemIdCareful(state, state->targetblock,
880 : state->target, P_HIKEY);
1124 akorotkov 881 CBC 1586 : itup = (IndexTuple) PageGetItem(state->target, itemid);
882 :
883 1586 : state->lowkey = MemoryContextAlloc(oldcontext, IndexTupleSize(itup));
1124 akorotkov 884 GIC 1586 : memcpy(state->lowkey, itup, IndexTupleSize(itup));
1124 akorotkov 885 ECB : }
2222 andres 886 :
887 : /* Free page and associated memory for this iteration */
2222 andres 888 GIC 6873 : MemoryContextReset(state->targetcontext);
889 : }
2222 andres 890 CBC 6873 : while (current != P_NONE);
891 :
1124 akorotkov 892 1718 : if (state->lowkey)
893 : {
1124 akorotkov 894 LBC 0 : Assert(state->readonly);
1124 akorotkov 895 UIC 0 : pfree(state->lowkey);
1124 akorotkov 896 UBC 0 : state->lowkey = NULL;
1124 akorotkov 897 EUB : }
898 :
899 : /* Don't change context for caller */
2222 andres 900 GIC 1718 : MemoryContextSwitchTo(oldcontext);
901 :
2222 andres 902 CBC 1718 : return nextleveldown;
903 : }
2222 andres 904 ECB :
905 : /*
906 : * Raise an error when target page's left link does not point back to the
907 : * previous target page, called leftcurrent here. The leftcurrent page's
908 : * right link was followed to get to the current target page, and we expect
909 : * mutual agreement among leftcurrent and the current target page. Make sure
910 : * that this condition has definitely been violated in the !readonly case,
911 : * where concurrent page splits are something that we need to deal with.
912 : *
913 : * Cross-page inconsistencies involving pages that don't agree about being
914 : * siblings are known to be a particularly good indicator of corruption
915 : * involving partial writes/lost updates. The bt_right_page_check_scankey
916 : * check also provides a way of detecting cross-page inconsistencies for
917 : * !readonly callers, but it can only detect sibling pages that have an
918 : * out-of-order keyspace, which can't catch many of the problems that we
919 : * expect to catch here.
920 : *
921 : * The classic example of the kind of inconsistency that we can only catch
922 : * with this check (when in !readonly mode) involves three sibling pages that
923 : * were affected by a faulty page split at some point in the past. The
924 : * effects of the split are reflected in the original page and its new right
925 : * sibling page, with a lack of any accompanying changes for the _original_
926 : * right sibling page. The original right sibling page's left link fails to
927 : * point to the new right sibling page (its left link still points to the
928 : * original page), even though the first phase of a page split is supposed to
929 : * work as a single atomic action. This subtle inconsistency will probably
930 : * only break backwards scans in practice.
931 : *
932 : * Note that this is the only place where amcheck will "couple" buffer locks
933 : * (and only for !readonly callers). In general we prefer to avoid more
934 : * thorough cross-page checks in !readonly mode, but it seems worth the
935 : * complexity here. Also, the performance overhead of performing lock
936 : * coupling here is negligible in practice. Control only reaches here with a
937 : * non-corrupt index when there is a concurrent page split at the instant
938 : * caller crossed over to target page from leftcurrent page.
939 : */
940 : static void
974 pg 941 UIC 0 : bt_recheck_sibling_links(BtreeCheckState *state,
942 : BlockNumber btpo_prev_from_target,
974 pg 943 EUB : BlockNumber leftcurrent)
944 : {
974 pg 945 UIC 0 : if (!state->readonly)
946 : {
974 pg 947 EUB : Buffer lbuf;
948 : Buffer newtargetbuf;
949 : Page page;
950 : BTPageOpaque opaque;
951 : BlockNumber newtargetblock;
952 :
953 : /* Couple locks in the usual order for nbtree: Left to right */
974 pg 954 UIC 0 : lbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM, leftcurrent,
955 : RBM_NORMAL, state->checkstrategy);
974 pg 956 UBC 0 : LockBuffer(lbuf, BT_READ);
974 pg 957 UIC 0 : _bt_checkpage(state->rel, lbuf);
974 pg 958 UBC 0 : page = BufferGetPage(lbuf);
373 michael 959 0 : opaque = BTPageGetOpaque(page);
974 pg 960 0 : if (P_ISDELETED(opaque))
974 pg 961 EUB : {
962 : /*
963 : * Cannot reason about concurrently deleted page -- the left link
964 : * in the page to the right is expected to point to some other
965 : * page to the left (not leftcurrent page).
966 : *
967 : * Note that we deliberately don't give up with a half-dead page.
968 : */
974 pg 969 UIC 0 : UnlockReleaseBuffer(lbuf);
970 0 : return;
974 pg 971 EUB : }
972 :
974 pg 973 UIC 0 : newtargetblock = opaque->btpo_next;
974 : /* Avoid self-deadlock when newtargetblock == leftcurrent */
974 pg 975 UBC 0 : if (newtargetblock != leftcurrent)
976 : {
977 0 : newtargetbuf = ReadBufferExtended(state->rel, MAIN_FORKNUM,
978 : newtargetblock, RBM_NORMAL,
974 pg 979 EUB : state->checkstrategy);
974 pg 980 UIC 0 : LockBuffer(newtargetbuf, BT_READ);
981 0 : _bt_checkpage(state->rel, newtargetbuf);
974 pg 982 UBC 0 : page = BufferGetPage(newtargetbuf);
373 michael 983 0 : opaque = BTPageGetOpaque(page);
974 pg 984 EUB : /* btpo_prev_from_target may have changed; update it */
974 pg 985 UBC 0 : btpo_prev_from_target = opaque->btpo_prev;
986 : }
974 pg 987 EUB : else
988 : {
989 : /*
990 : * leftcurrent right sibling points back to leftcurrent block.
991 : * Index is corrupt. Easiest way to handle this is to pretend
992 : * that we actually read from a distinct page that has an invalid
993 : * block number in its btpo_prev.
994 : */
974 pg 995 UIC 0 : newtargetbuf = InvalidBuffer;
996 0 : btpo_prev_from_target = InvalidBlockNumber;
974 pg 997 EUB : }
998 :
999 : /*
1000 : * No need to check P_ISDELETED here, since new target block cannot be
1001 : * marked deleted as long as we hold a lock on lbuf
1002 : */
974 pg 1003 UIC 0 : if (BufferIsValid(newtargetbuf))
1004 0 : UnlockReleaseBuffer(newtargetbuf);
974 pg 1005 UBC 0 : UnlockReleaseBuffer(lbuf);
974 pg 1006 EUB :
974 pg 1007 UBC 0 : if (btpo_prev_from_target == leftcurrent)
1008 : {
974 pg 1009 EUB : /* Report split in left sibling, not target (or new target) */
974 pg 1010 UIC 0 : ereport(DEBUG1,
1011 : (errcode(ERRCODE_INTERNAL_ERROR),
781 peter 1012 EUB : errmsg_internal("harmless concurrent page split detected in index \"%s\"",
1013 : RelationGetRelationName(state->rel)),
1014 : errdetail_internal("Block=%u new right sibling=%u original right sibling=%u.",
1015 : leftcurrent, newtargetblock,
1016 : state->targetblock)));
974 pg 1017 UIC 0 : return;
1018 : }
974 pg 1019 EUB :
1020 : /*
1021 : * Index is corrupt. Make sure that we report correct target page.
1022 : *
1023 : * This could have changed in cases where there was a concurrent page
1024 : * split, as well as index corruption (at least in theory). Note that
1025 : * btpo_prev_from_target was already updated above.
1026 : */
974 pg 1027 UIC 0 : state->targetblock = newtargetblock;
1028 : }
974 pg 1029 EUB :
974 pg 1030 UIC 0 : ereport(ERROR,
1031 : (errcode(ERRCODE_INDEX_CORRUPTED),
974 pg 1032 EUB : errmsg("left link/right link pair in index \"%s\" not in agreement",
1033 : RelationGetRelationName(state->rel)),
1034 : errdetail_internal("Block=%u left block=%u left link from block=%u.",
1035 : state->targetblock, leftcurrent,
1036 : btpo_prev_from_target)));
1037 : }
1038 :
1039 : /*
1040 : * Function performs the following checks on target page, or pages ancillary to
1041 : * target page:
1042 : *
1043 : * - That every "real" data item is less than or equal to the high key, which
1044 : * is an upper bound on the items on the page. Data items should be
1045 : * strictly less than the high key when the page is an internal page.
1046 : *
1047 : * - That within the page, every data item is strictly less than the item
1048 : * immediately to its right, if any (i.e., that the items are in order
1049 : * within the page, so that the binary searches performed by index scans are
1050 : * sane).
1051 : *
1052 : * - That the last data item stored on the page is strictly less than the
1053 : * first data item on the page to the right (when such a first item is
1054 : * available).
1055 : *
1056 : * - Various checks on the structure of tuples themselves. For example, check
1057 : * that non-pivot tuples have no truncated attributes.
1058 : *
1059 : * Furthermore, when state passed shows ShareLock held, function also checks:
1060 : *
1061 : * - That all child pages respect strict lower bound from parent's pivot
1062 : * tuple.
1063 : *
1064 : * - That downlink to block was encountered in parent where that's expected.
1065 : *
1066 : * - That high keys of child pages matches corresponding pivot keys in parent.
1067 : *
1068 : * This is also where heapallindexed callers use their Bloom filter to
1069 : * fingerprint IndexTuples for later table_index_build_scan() verification.
1070 : *
1071 : * Note: Memory allocated in this routine is expected to be released by caller
1072 : * resetting state->targetcontext.
1073 : */
1074 : static void
2222 andres 1075 GIC 6873 : bt_target_page_check(BtreeCheckState *state)
1076 : {
2222 andres 1077 ECB : OffsetNumber offset;
1078 : OffsetNumber max;
1079 : BTPageOpaque topaque;
1080 :
373 michael 1081 GIC 6873 : topaque = BTPageGetOpaque(state->target);
2222 andres 1082 6873 : max = PageGetMaxOffsetNumber(state->target);
2222 andres 1083 ECB :
2222 andres 1084 CBC 6873 : elog(DEBUG2, "verifying %u items on %s block %u", max,
1085 : P_ISLEAF(topaque) ? "leaf" : "internal", state->targetblock);
2222 andres 1086 ECB :
1087 : /*
1088 : * Check the number of attributes in high key. Note, rightmost page
1089 : * doesn't contain a high key, so nothing to check
1090 : */
1445 pg 1091 GIC 6873 : if (!P_RIGHTMOST(topaque))
1092 : {
1816 teodor 1093 ECB : ItemId itemid;
1094 : IndexTuple itup;
1095 :
1096 : /* Verify line pointer before checking tuple */
1445 pg 1097 GIC 5155 : itemid = PageGetItemIdCareful(state, state->targetblock,
1098 : state->target, P_HIKEY);
1445 pg 1099 CBC 5155 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1100 : P_HIKEY))
1445 pg 1101 ECB : {
1445 pg 1102 UIC 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1103 0 : ereport(ERROR,
1445 pg 1104 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
1105 : errmsg("wrong number of high key index tuple attributes in index \"%s\"",
1106 : RelationGetRelationName(state->rel)),
1107 : errdetail_internal("Index block=%u natts=%u block type=%s page lsn=%X/%X.",
1108 : state->targetblock,
1109 : BTreeTupleGetNAtts(itup, state->rel),
1110 : P_ISLEAF(topaque) ? "heap" : "index",
1111 : LSN_FORMAT_ARGS(state->targetlsn))));
1112 : }
1113 : }
1114 :
1115 : /*
1116 : * Loop over page items, starting from first non-highkey item, not high
1117 : * key (if any). Most tests are not performed for the "negative infinity"
1118 : * real item (if any).
1119 : */
2222 andres 1120 GIC 6873 : for (offset = P_FIRSTDATAKEY(topaque);
1121 1582525 : offset <= max;
2222 andres 1122 CBC 1575652 : offset = OffsetNumberNext(offset))
2222 andres 1123 ECB : {
1124 : ItemId itemid;
1125 : IndexTuple itup;
1126 : size_t tupsize;
1127 : BTScanInsert skey;
1128 : bool lowersizelimit;
1129 : ItemPointer scantid;
1130 :
2222 andres 1131 GIC 1575652 : CHECK_FOR_INTERRUPTS();
1132 :
1445 pg 1133 CBC 1575652 : itemid = PageGetItemIdCareful(state, state->targetblock,
1134 : state->target, offset);
1835 andres 1135 1575652 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1835 andres 1136 GIC 1575652 : tupsize = IndexTupleSize(itup);
1835 andres 1137 ECB :
1138 : /*
1139 : * lp_len should match the IndexTuple reported length exactly, since
1140 : * lp_len is completely redundant in indexes, and both sources of
1141 : * tuple length are MAXALIGN()'d. nbtree does not use lp_len all that
1142 : * frequently, and is surprisingly tolerant of corrupt lp_len fields.
1143 : */
1835 andres 1144 GIC 1575652 : if (tupsize != ItemIdGetLength(itemid))
1835 andres 1145 UIC 0 : ereport(ERROR,
1835 andres 1146 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1835 andres 1147 EUB : errmsg("index tuple size does not equal lp_len in index \"%s\"",
1148 : RelationGetRelationName(state->rel)),
1149 : errdetail_internal("Index tid=(%u,%u) tuple size=%zu lp_len=%u page lsn=%X/%X.",
1150 : state->targetblock, offset,
1151 : tupsize, ItemIdGetLength(itemid),
1152 : LSN_FORMAT_ARGS(state->targetlsn)),
1153 : errhint("This could be a torn page problem.")));
1154 :
1155 : /* Check the number of index tuple attributes */
1481 pg 1156 GIC 1575652 : if (!_bt_check_natts(state->rel, state->heapkeyspace, state->target,
1157 : offset))
1828 teodor 1158 ECB : {
1159 : ItemPointer tid;
1160 : char *itid,
1161 : *htid;
1162 :
1828 teodor 1163 UIC 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1138 pg 1164 0 : tid = BTreeTupleGetPointsToTID(itup);
1828 teodor 1165 UBC 0 : htid = psprintf("(%u,%u)",
1138 pg 1166 EUB : ItemPointerGetBlockNumberNoCheck(tid),
1138 pg 1167 UBC 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1168 :
1828 teodor 1169 0 : ereport(ERROR,
1170 : (errcode(ERRCODE_INDEX_CORRUPTED),
1816 teodor 1171 EUB : errmsg("wrong number of index tuple attributes in index \"%s\"",
1172 : RelationGetRelationName(state->rel)),
1173 : errdetail_internal("Index tid=%s natts=%u points to %s tid=%s page lsn=%X/%X.",
1174 : itid,
1175 : BTreeTupleGetNAtts(itup, state->rel),
1176 : P_ISLEAF(topaque) ? "heap" : "index",
1177 : htid,
1178 : LSN_FORMAT_ARGS(state->targetlsn))));
1179 : }
1180 :
1181 : /*
1182 : * Don't try to generate scankey using "negative infinity" item on
1183 : * internal pages. They are always truncated to zero attributes.
1184 : */
2222 andres 1185 GIC 1575652 : if (offset_is_negative_infinity(topaque, offset))
1186 : {
1124 akorotkov 1187 ECB : /*
1188 : * We don't call bt_child_check() for "negative infinity" items.
1189 : * But if we're performing downlink connectivity check, we do it
1190 : * for every item including "negative infinity" one.
1191 : */
1124 akorotkov 1192 GIC 360 : if (!P_ISLEAF(topaque) && state->readonly)
1193 : {
1124 akorotkov 1194 CBC 9 : bt_child_highkey_check(state,
1195 : offset,
1124 akorotkov 1196 ECB : NULL,
1197 : topaque->btpo_level);
1198 : }
2222 andres 1199 GIC 360 : continue;
1200 : }
2222 andres 1201 ECB :
1202 : /*
1203 : * Readonly callers may optionally verify that non-pivot tuples can
1204 : * each be found by an independent search that starts from the root.
1205 : * Note that we deliberately don't do individual searches for each
1206 : * TID, since the posting list itself is validated by other checks.
1207 : */
1481 pg 1208 GIC 1575292 : if (state->rootdescend && P_ISLEAF(topaque) &&
1209 200041 : !bt_rootdescend(state, itup))
1481 pg 1210 ECB : {
1138 pg 1211 LBC 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1212 : char *itid,
1481 pg 1213 EUB : *htid;
1214 :
1481 pg 1215 UIC 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1138 1216 0 : htid = psprintf("(%u,%u)", ItemPointerGetBlockNumber(tid),
1138 pg 1217 UBC 0 : ItemPointerGetOffsetNumber(tid));
1481 pg 1218 EUB :
1481 pg 1219 UBC 0 : ereport(ERROR,
1220 : (errcode(ERRCODE_INDEX_CORRUPTED),
1481 pg 1221 EUB : errmsg("could not find tuple using search from root page in index \"%s\"",
1222 : RelationGetRelationName(state->rel)),
1223 : errdetail_internal("Index tid=%s points to heap tid=%s page lsn=%X/%X.",
1224 : itid, htid,
1225 : LSN_FORMAT_ARGS(state->targetlsn))));
1226 : }
1227 :
1228 : /*
1229 : * If tuple is a posting list tuple, make sure posting list TIDs are
1230 : * in order
1231 : */
1138 pg 1232 GIC 1575292 : if (BTreeTupleIsPosting(itup))
1233 : {
1138 pg 1234 ECB : ItemPointerData last;
1235 : ItemPointer current;
1236 :
1138 pg 1237 GIC 7140 : ItemPointerCopy(BTreeTupleGetHeapTID(itup), &last);
1238 :
1138 pg 1239 CBC 45860 : for (int i = 1; i < BTreeTupleGetNPosting(itup); i++)
1240 : {
1138 pg 1241 ECB :
1138 pg 1242 GIC 38720 : current = BTreeTupleGetPostingN(itup, i);
1243 :
1138 pg 1244 CBC 38720 : if (ItemPointerCompare(current, &last) <= 0)
1245 : {
1138 pg 1246 LBC 0 : char *itid = psprintf("(%u,%u)", state->targetblock, offset);
1247 :
1138 pg 1248 UBC 0 : ereport(ERROR,
1249 : (errcode(ERRCODE_INDEX_CORRUPTED),
1138 pg 1250 EUB : errmsg_internal("posting list contains misplaced TID in index \"%s\"",
1251 : RelationGetRelationName(state->rel)),
1252 : errdetail_internal("Index tid=%s posting list offset=%d page lsn=%X/%X.",
1253 : itid, i,
1254 : LSN_FORMAT_ARGS(state->targetlsn))));
1255 : }
1256 :
1138 pg 1257 GIC 38720 : ItemPointerCopy(current, &last);
1258 : }
1138 pg 1259 ECB : }
1260 :
1261 : /* Build insertion scankey for current page offset */
8 andres 1262 GNC 1575292 : skey = bt_mkscankey_pivotsearch(state->rel, state->heaprel, itup);
1263 :
1481 pg 1264 ECB : /*
1265 : * Make sure tuple size does not exceed the relevant BTREE_VERSION
1266 : * specific limit.
1267 : *
1268 : * BTREE_VERSION 4 (which introduced heapkeyspace rules) requisitioned
1269 : * a small amount of space from BTMaxItemSize() in order to ensure
1270 : * that suffix truncation always has enough space to add an explicit
1271 : * heap TID back to a tuple -- we pessimistically assume that every
1272 : * newly inserted tuple will eventually need to have a heap TID
1273 : * appended during a future leaf page split, when the tuple becomes
1274 : * the basis of the new high key (pivot tuple) for the leaf page.
1275 : *
1276 : * Since the reclaimed space is reserved for that purpose, we must not
1277 : * enforce the slightly lower limit when the extra space has been used
1278 : * as intended. In other words, there is only a cross-version
1279 : * difference in the limit on tuple size within leaf pages.
1280 : *
1281 : * Still, we're particular about the details within BTREE_VERSION 4
1282 : * internal pages. Pivot tuples may only use the extra space for its
1283 : * designated purpose. Enforce the lower limit for pivot tuples when
1284 : * an explicit heap TID isn't actually present. (In all other cases
1285 : * suffix truncation is guaranteed to generate a pivot tuple that's no
1286 : * larger than the firstright tuple provided to it by its caller.)
1287 : */
1481 pg 1288 GIC 3150584 : lowersizelimit = skey->heapkeyspace &&
1289 1575292 : (P_ISLEAF(topaque) || BTreeTupleGetHeapTID(itup) == NULL);
1481 pg 1290 CBC 1575293 : if (tupsize > (lowersizelimit ? BTMaxItemSize(state->target) :
1291 1 : BTMaxItemSizeNoHeapTid(state->target)))
1481 pg 1292 ECB : {
1138 pg 1293 LBC 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1294 : char *itid,
1481 pg 1295 EUB : *htid;
1296 :
1481 pg 1297 UIC 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1298 0 : htid = psprintf("(%u,%u)",
1138 pg 1299 EUB : ItemPointerGetBlockNumberNoCheck(tid),
1138 pg 1300 UBC 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1301 :
1481 1302 0 : ereport(ERROR,
1303 : (errcode(ERRCODE_INDEX_CORRUPTED),
1481 pg 1304 EUB : errmsg("index row size %zu exceeds maximum for index \"%s\"",
1305 : tupsize, RelationGetRelationName(state->rel)),
1306 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%X.",
1307 : itid,
1308 : P_ISLEAF(topaque) ? "heap" : "index",
1309 : htid,
1310 : LSN_FORMAT_ARGS(state->targetlsn))));
1311 : }
1312 :
1313 : /* Fingerprint leaf page tuples (those that point to the heap) */
1835 andres 1314 GIC 1575292 : if (state->heapallindexed && P_ISLEAF(topaque) && !ItemIdIsDead(itemid))
1315 : {
1418 tgl 1316 ECB : IndexTuple norm;
1317 :
1138 pg 1318 GIC 401373 : if (BTreeTupleIsPosting(itup))
1319 : {
1138 pg 1320 ECB : /* Fingerprint all elements as distinct "plain" tuples */
1138 pg 1321 GIC 9454 : for (int i = 0; i < BTreeTupleGetNPosting(itup); i++)
1322 : {
1138 pg 1323 ECB : IndexTuple logtuple;
1324 :
1138 pg 1325 GIC 9369 : logtuple = bt_posting_plain_tuple(itup, i);
1326 9369 : norm = bt_normalize_tuple(state, logtuple);
1138 pg 1327 CBC 9369 : bloom_add_element(state->filter, (unsigned char *) norm,
1328 9369 : IndexTupleSize(norm));
1138 pg 1329 ECB : /* Be tidy */
1138 pg 1330 CBC 9369 : if (norm != logtuple)
1138 pg 1331 UIC 0 : pfree(norm);
1138 pg 1332 CBC 9369 : pfree(logtuple);
1138 pg 1333 EUB : }
1138 pg 1334 ECB : }
1335 : else
1336 : {
1138 pg 1337 GIC 401288 : norm = bt_normalize_tuple(state, itup);
1338 401288 : bloom_add_element(state->filter, (unsigned char *) norm,
1138 pg 1339 CBC 401288 : IndexTupleSize(norm));
1138 pg 1340 ECB : /* Be tidy */
1138 pg 1341 CBC 401288 : if (norm != itup)
1138 pg 1342 UIC 0 : pfree(norm);
1138 pg 1343 ECB : }
1523 pg 1344 EUB : }
1345 :
1346 : /*
1347 : * * High key check *
1348 : *
1349 : * If there is a high key (if this is not the rightmost page on its
1350 : * entire level), check that high key actually is upper bound on all
1351 : * page items. If this is a posting list tuple, we'll need to set
1352 : * scantid to be highest TID in posting list.
1353 : *
1354 : * We prefer to check all items against high key rather than checking
1355 : * just the last and trusting that the operator class obeys the
1356 : * transitive law (which implies that all previous items also
1357 : * respected the high key invariant if they pass the item order
1358 : * check).
1359 : *
1360 : * Ideally, we'd compare every item in the index against every other
1361 : * item in the index, and not trust opclass obedience of the
1362 : * transitive law to bridge the gap between children and their
1363 : * grandparents (as well as great-grandparents, and so on). We don't
1364 : * go to those lengths because that would be prohibitively expensive,
1365 : * and probably not markedly more effective in practice.
1366 : *
1367 : * On the leaf level, we check that the key is <= the highkey.
1368 : * However, on non-leaf levels we check that the key is < the highkey,
1369 : * because the high key is "just another separator" rather than a copy
1370 : * of some existing key item; we expect it to be unique among all keys
1371 : * on the same level. (Suffix truncation will sometimes produce a
1372 : * leaf highkey that is an untruncated copy of the lastleft item, but
1373 : * never any other item, which necessitates weakening the leaf level
1374 : * check to <=.)
1375 : *
1376 : * Full explanation for why a highkey is never truly a copy of another
1377 : * item from the same level on internal levels:
1378 : *
1379 : * While the new left page's high key is copied from the first offset
1380 : * on the right page during an internal page split, that's not the
1381 : * full story. In effect, internal pages are split in the middle of
1382 : * the firstright tuple, not between the would-be lastleft and
1383 : * firstright tuples: the firstright key ends up on the left side as
1384 : * left's new highkey, and the firstright downlink ends up on the
1385 : * right side as right's new "negative infinity" item. The negative
1386 : * infinity tuple is truncated to zero attributes, so we're only left
1387 : * with the downlink. In other words, the copying is just an
1388 : * implementation detail of splitting in the middle of a (pivot)
1389 : * tuple. (See also: "Notes About Data Representation" in the nbtree
1390 : * README.)
1391 : */
1138 pg 1392 GIC 1575292 : scantid = skey->scantid;
1393 1575292 : if (state->heapkeyspace && BTreeTupleIsPosting(itup))
1138 pg 1394 CBC 7140 : skey->scantid = BTreeTupleGetMaxHeapTID(itup);
1138 pg 1395 ECB :
2222 andres 1396 CBC 3026318 : if (!P_RIGHTMOST(topaque) &&
1481 pg 1397 GIC 1451026 : !(P_ISLEAF(topaque) ? invariant_leq_offset(state, skey, P_HIKEY) :
1481 pg 1398 CBC 566 : invariant_l_offset(state, skey, P_HIKEY)))
2222 andres 1399 ECB : {
1138 pg 1400 LBC 0 : ItemPointer tid = BTreeTupleGetPointsToTID(itup);
1401 : char *itid,
2222 andres 1402 EUB : *htid;
1403 :
2222 andres 1404 UIC 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1405 0 : htid = psprintf("(%u,%u)",
1138 pg 1406 EUB : ItemPointerGetBlockNumberNoCheck(tid),
1138 pg 1407 UBC 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1408 :
2222 andres 1409 0 : ereport(ERROR,
1410 : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 1411 EUB : errmsg("high key invariant violated for index \"%s\"",
1412 : RelationGetRelationName(state->rel)),
1413 : errdetail_internal("Index tid=%s points to %s tid=%s page lsn=%X/%X.",
1414 : itid,
1415 : P_ISLEAF(topaque) ? "heap" : "index",
1416 : htid,
1417 : LSN_FORMAT_ARGS(state->targetlsn))));
1418 : }
1419 : /* Reset, in case scantid was set to (itup) posting tuple's max TID */
1138 pg 1420 GIC 1575292 : skey->scantid = scantid;
1421 :
2222 andres 1422 ECB : /*
1423 : * * Item order check *
1424 : *
1425 : * Check that items are stored on page in logical order, by checking
1426 : * current item is strictly less than next item (if any).
1427 : */
2222 andres 1428 GIC 1575292 : if (OffsetNumberNext(offset) <= max &&
1481 pg 1429 1568421 : !invariant_l_offset(state, skey, OffsetNumberNext(offset)))
2222 andres 1430 ECB : {
1138 pg 1431 : ItemPointer tid;
1432 : char *itid,
1433 : *htid,
1434 : *nitid,
1435 : *nhtid;
1436 :
2222 andres 1437 UIC 0 : itid = psprintf("(%u,%u)", state->targetblock, offset);
1138 pg 1438 0 : tid = BTreeTupleGetPointsToTID(itup);
2222 andres 1439 UBC 0 : htid = psprintf("(%u,%u)",
1138 pg 1440 EUB : ItemPointerGetBlockNumberNoCheck(tid),
1138 pg 1441 UBC 0 : ItemPointerGetOffsetNumberNoCheck(tid));
2222 andres 1442 UIC 0 : nitid = psprintf("(%u,%u)", state->targetblock,
2222 andres 1443 UBC 0 : OffsetNumberNext(offset));
2222 andres 1444 EUB :
1445 : /* Reuse itup to get pointed-to heap location of second item */
1445 pg 1446 UIC 0 : itemid = PageGetItemIdCareful(state, state->targetblock,
1447 : state->target,
1445 pg 1448 UBC 0 : OffsetNumberNext(offset));
2222 andres 1449 UIC 0 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1138 pg 1450 UBC 0 : tid = BTreeTupleGetPointsToTID(itup);
2222 andres 1451 0 : nhtid = psprintf("(%u,%u)",
1138 pg 1452 EUB : ItemPointerGetBlockNumberNoCheck(tid),
1138 pg 1453 UBC 0 : ItemPointerGetOffsetNumberNoCheck(tid));
1454 :
2222 andres 1455 0 : ereport(ERROR,
1456 : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 1457 EUB : errmsg("item order invariant violated for index \"%s\"",
1458 : RelationGetRelationName(state->rel)),
1459 : errdetail_internal("Lower index tid=%s (points to %s tid=%s) "
1460 : "higher index tid=%s (points to %s tid=%s) "
1461 : "page lsn=%X/%X.",
1462 : itid,
1463 : P_ISLEAF(topaque) ? "heap" : "index",
1464 : htid,
1465 : nitid,
1466 : P_ISLEAF(topaque) ? "heap" : "index",
1467 : nhtid,
1468 : LSN_FORMAT_ARGS(state->targetlsn))));
1469 : }
1470 :
1471 : /*
1472 : * * Last item check *
1473 : *
1474 : * Check last item against next/right page's first data item's when
1475 : * last item on page is reached. This additional check will detect
1476 : * transposed pages iff the supposed right sibling page happens to
1477 : * belong before target in the key space. (Otherwise, a subsequent
1478 : * heap verification will probably detect the problem.)
1479 : *
1480 : * This check is similar to the item order check that will have
1481 : * already been performed for every other "real" item on target page
1482 : * when last item is checked. The difference is that the next item
1483 : * (the item that is compared to target's last item) needs to come
1484 : * from the next/sibling page. There may not be such an item
1485 : * available from sibling for various reasons, though (e.g., target is
1486 : * the rightmost page on level).
1487 : */
2222 andres 1488 GIC 1575292 : else if (offset == max)
1489 : {
1418 tgl 1490 ECB : BTScanInsert rightkey;
1491 :
1492 : /* Get item in next/right page */
2222 andres 1493 GIC 6871 : rightkey = bt_right_page_check_scankey(state);
1494 :
2222 andres 1495 CBC 6871 : if (rightkey &&
1481 pg 1496 GIC 5155 : !invariant_g_offset(state, rightkey, max))
2222 andres 1497 ECB : {
1498 : /*
1499 : * As explained at length in bt_right_page_check_scankey(),
1500 : * there is a known !readonly race that could account for
1501 : * apparent violation of invariant, which we must check for
1502 : * before actually proceeding with raising error. Our canary
1503 : * condition is that target page was deleted.
1504 : */
2222 andres 1505 UIC 0 : if (!state->readonly)
1506 : {
2222 andres 1507 EUB : /* Get fresh copy of target page */
2222 andres 1508 UIC 0 : state->target = palloc_btree_page(state, state->targetblock);
1509 : /* Note that we deliberately do not update target LSN */
373 michael 1510 UBC 0 : topaque = BTPageGetOpaque(state->target);
1511 :
2222 andres 1512 EUB : /*
1513 : * All !readonly checks now performed; just return
1514 : */
2222 andres 1515 UIC 0 : if (P_IGNORE(topaque))
1516 0 : return;
2222 andres 1517 EUB : }
1518 :
2222 andres 1519 UIC 0 : ereport(ERROR,
1520 : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 1521 EUB : errmsg("cross page item order invariant violated for index \"%s\"",
1522 : RelationGetRelationName(state->rel)),
1523 : errdetail_internal("Last item on page tid=(%u,%u) page lsn=%X/%X.",
1524 : state->targetblock, offset,
1525 : LSN_FORMAT_ARGS(state->targetlsn))));
1526 : }
1527 : }
1528 :
1529 : /*
1530 : * * Downlink check *
1531 : *
1532 : * Additional check of child items iff this is an internal page and
1533 : * caller holds a ShareLock. This happens for every downlink (item)
1534 : * in target excluding the negative-infinity downlink (again, this is
1535 : * because it has no useful value to compare).
1536 : */
2222 andres 1537 GIC 1575292 : if (!P_ISLEAF(topaque) && state->readonly)
1124 akorotkov 1538 1585 : bt_child_check(state, skey, offset);
2222 andres 1539 ECB : }
1810 teodor 1540 :
1541 : /*
1542 : * Special case bt_child_highkey_check() call
1543 : *
1544 : * We don't pass a real downlink, but we've to finish the level
1545 : * processing. If condition is satisfied, we've already processed all the
1546 : * downlinks from the target level. But there still might be pages to the
1547 : * right of the child page pointer to by our rightmost downlink. And they
1548 : * might have missing downlinks. This final call checks for them.
1549 : */
1124 akorotkov 1550 GIC 6873 : if (!P_ISLEAF(topaque) && P_RIGHTMOST(topaque) && state->readonly)
1551 : {
1124 akorotkov 1552 CBC 8 : bt_child_highkey_check(state, InvalidOffsetNumber,
1553 : NULL, topaque->btpo_level);
1124 akorotkov 1554 ECB : }
1555 : }
1556 :
1557 : /*
1558 : * Return a scankey for an item on page to right of current target (or the
1559 : * first non-ignorable page), sufficient to check ordering invariant on last
1560 : * item in current target page. Returned scankey relies on local memory
1561 : * allocated for the child page, which caller cannot pfree(). Caller's memory
1562 : * context should be reset between calls here.
1563 : *
1564 : * This is the first data item, and so all adjacent items are checked against
1565 : * their immediate sibling item (which may be on a sibling page, or even a
1566 : * "cousin" page at parent boundaries where target's rightlink points to page
1567 : * with different parent page). If no such valid item is available, return
1568 : * NULL instead.
1569 : *
1570 : * Note that !readonly callers must reverify that target page has not
1571 : * been concurrently deleted.
1572 : */
1573 : static BTScanInsert
2222 andres 1574 GIC 6871 : bt_right_page_check_scankey(BtreeCheckState *state)
1575 : {
2222 andres 1576 ECB : BTPageOpaque opaque;
1577 : ItemId rightitem;
1578 : IndexTuple firstitup;
1579 : BlockNumber targetnext;
1580 : Page rightpage;
1581 : OffsetNumber nline;
1582 :
1583 : /* Determine target's next block number */
373 michael 1584 GIC 6871 : opaque = BTPageGetOpaque(state->target);
1585 :
2222 andres 1586 ECB : /* If target is already rightmost, no right sibling; nothing to do here */
2222 andres 1587 GIC 6871 : if (P_RIGHTMOST(opaque))
1588 1716 : return NULL;
2222 andres 1589 ECB :
1590 : /*
1591 : * General notes on concurrent page splits and page deletion:
1592 : *
1593 : * Routines like _bt_search() don't require *any* page split interlock
1594 : * when descending the tree, including something very light like a buffer
1595 : * pin. That's why it's okay that we don't either. This avoidance of any
1596 : * need to "couple" buffer locks is the raison d' etre of the Lehman & Yao
1597 : * algorithm, in fact.
1598 : *
1599 : * That leaves deletion. A deleted page won't actually be recycled by
1600 : * VACUUM early enough for us to fail to at least follow its right link
1601 : * (or left link, or downlink) and find its sibling, because recycling
1602 : * does not occur until no possible index scan could land on the page.
1603 : * Index scans can follow links with nothing more than their snapshot as
1604 : * an interlock and be sure of at least that much. (See page
1605 : * recycling/"visible to everyone" notes in nbtree README.)
1606 : *
1607 : * Furthermore, it's okay if we follow a rightlink and find a half-dead or
1608 : * dead (ignorable) page one or more times. There will either be a
1609 : * further right link to follow that leads to a live page before too long
1610 : * (before passing by parent's rightmost child), or we will find the end
1611 : * of the entire level instead (possible when parent page is itself the
1612 : * rightmost on its level).
1613 : */
2222 andres 1614 GIC 5155 : targetnext = opaque->btpo_next;
1615 : for (;;)
2222 andres 1616 ECB : {
2222 andres 1617 GIC 5155 : CHECK_FOR_INTERRUPTS();
1618 :
2222 andres 1619 CBC 5155 : rightpage = palloc_btree_page(state, targetnext);
373 michael 1620 GIC 5155 : opaque = BTPageGetOpaque(rightpage);
2222 andres 1621 ECB :
2222 andres 1622 CBC 5155 : if (!P_IGNORE(opaque) || P_RIGHTMOST(opaque))
1623 : break;
2222 andres 1624 ECB :
1625 : /*
1626 : * We landed on a deleted or half-dead sibling page. Step right until
1627 : * we locate a live sibling page.
1628 : */
754 pg 1629 UIC 0 : ereport(DEBUG2,
1630 : (errcode(ERRCODE_NO_DATA),
754 pg 1631 EUB : errmsg_internal("level %u sibling page in block %u of index \"%s\" was found deleted or half dead",
1632 : opaque->btpo_level, targetnext, RelationGetRelationName(state->rel)),
1633 : errdetail_internal("Deleted page found when building scankey from right sibling.")));
1634 :
754 pg 1635 UIC 0 : targetnext = opaque->btpo_next;
1636 :
2222 andres 1637 EUB : /* Be slightly more pro-active in freeing this memory, just in case */
2222 andres 1638 UIC 0 : pfree(rightpage);
1639 : }
2222 andres 1640 EUB :
1641 : /*
1642 : * No ShareLock held case -- why it's safe to proceed.
1643 : *
1644 : * Problem:
1645 : *
1646 : * We must avoid false positive reports of corruption when caller treats
1647 : * item returned here as an upper bound on target's last item. In
1648 : * general, false positives are disallowed. Avoiding them here when
1649 : * caller is !readonly is subtle.
1650 : *
1651 : * A concurrent page deletion by VACUUM of the target page can result in
1652 : * the insertion of items on to this right sibling page that would
1653 : * previously have been inserted on our target page. There might have
1654 : * been insertions that followed the target's downlink after it was made
1655 : * to point to right sibling instead of target by page deletion's first
1656 : * phase. The inserters insert items that would belong on target page.
1657 : * This race is very tight, but it's possible. This is our only problem.
1658 : *
1659 : * Non-problems:
1660 : *
1661 : * We are not hindered by a concurrent page split of the target; we'll
1662 : * never land on the second half of the page anyway. A concurrent split
1663 : * of the right page will also not matter, because the first data item
1664 : * remains the same within the left half, which we'll reliably land on. If
1665 : * we had to skip over ignorable/deleted pages, it cannot matter because
1666 : * their key space has already been atomically merged with the first
1667 : * non-ignorable page we eventually find (doesn't matter whether the page
1668 : * we eventually find is a true sibling or a cousin of target, which we go
1669 : * into below).
1670 : *
1671 : * Solution:
1672 : *
1673 : * Caller knows that it should reverify that target is not ignorable
1674 : * (half-dead or deleted) when cross-page sibling item comparison appears
1675 : * to indicate corruption (invariant fails). This detects the single race
1676 : * condition that exists for caller. This is correct because the
1677 : * continued existence of target block as non-ignorable (not half-dead or
1678 : * deleted) implies that target page was not merged into from the right by
1679 : * deletion; the key space at or after target never moved left. Target's
1680 : * parent either has the same downlink to target as before, or a <
1681 : * downlink due to deletion at the left of target. Target either has the
1682 : * same highkey as before, or a highkey < before when there is a page
1683 : * split. (The rightmost concurrently-split-from-target-page page will
1684 : * still have the same highkey as target was originally found to have,
1685 : * which for our purposes is equivalent to target's highkey itself never
1686 : * changing, since we reliably skip over
1687 : * concurrently-split-from-target-page pages.)
1688 : *
1689 : * In simpler terms, we allow that the key space of the target may expand
1690 : * left (the key space can move left on the left side of target only), but
1691 : * the target key space cannot expand right and get ahead of us without
1692 : * our detecting it. The key space of the target cannot shrink, unless it
1693 : * shrinks to zero due to the deletion of the original page, our canary
1694 : * condition. (To be very precise, we're a bit stricter than that because
1695 : * it might just have been that the target page split and only the
1696 : * original target page was deleted. We can be more strict, just not more
1697 : * lax.)
1698 : *
1699 : * Top level tree walk caller moves on to next page (makes it the new
1700 : * target) following recovery from this race. (cf. The rationale for
1701 : * child/downlink verification needing a ShareLock within
1702 : * bt_child_check(), where page deletion is also the main source of
1703 : * trouble.)
1704 : *
1705 : * Note that it doesn't matter if right sibling page here is actually a
1706 : * cousin page, because in order for the key space to be readjusted in a
1707 : * way that causes us issues in next level up (guiding problematic
1708 : * concurrent insertions to the cousin from the grandparent rather than to
1709 : * the sibling from the parent), there'd have to be page deletion of
1710 : * target's parent page (affecting target's parent's downlink in target's
1711 : * grandparent page). Internal page deletion only occurs when there are
1712 : * no child pages (they were all fully deleted), and caller is checking
1713 : * that the target's parent has at least one non-deleted (so
1714 : * non-ignorable) child: the target page. (Note that the first phase of
1715 : * deletion atomically marks the page to be deleted half-dead/ignorable at
1716 : * the same time downlink in its parent is removed, so caller will
1717 : * definitely not fail to detect that this happened.)
1718 : *
1719 : * This trick is inspired by the method backward scans use for dealing
1720 : * with concurrent page splits; concurrent page deletion is a problem that
1721 : * similarly receives special consideration sometimes (it's possible that
1722 : * the backwards scan will re-read its "original" block after failing to
1723 : * find a right-link to it, having already moved in the opposite direction
1724 : * (right/"forwards") a few times to try to locate one). Just like us,
1725 : * that happens only to determine if there was a concurrent page deletion
1726 : * of a reference page, and just like us if there was a page deletion of
1727 : * that reference page it means we can move on from caring about the
1728 : * reference page. See the nbtree README for a full description of how
1729 : * that works.
1730 : */
2222 andres 1731 GIC 5155 : nline = PageGetMaxOffsetNumber(rightpage);
1732 :
2222 andres 1733 ECB : /*
1734 : * Get first data item, if any
1735 : */
2222 andres 1736 GIC 5155 : if (P_ISLEAF(opaque) && nline >= P_FIRSTDATAKEY(opaque))
1737 : {
2222 andres 1738 ECB : /* Return first data item (if any) */
1445 pg 1739 GIC 5153 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
1740 5153 : P_FIRSTDATAKEY(opaque));
2222 andres 1741 ECB : }
2222 andres 1742 CBC 4 : else if (!P_ISLEAF(opaque) &&
2222 andres 1743 GIC 2 : nline >= OffsetNumberNext(P_FIRSTDATAKEY(opaque)))
2222 andres 1744 ECB : {
1745 : /*
1746 : * Return first item after the internal page's "negative infinity"
1747 : * item
1748 : */
1445 pg 1749 GIC 2 : rightitem = PageGetItemIdCareful(state, targetnext, rightpage,
1750 2 : OffsetNumberNext(P_FIRSTDATAKEY(opaque)));
2222 andres 1751 ECB : }
1752 : else
1753 : {
1754 : /*
1755 : * No first item. Page is probably empty leaf page, but it's also
1756 : * possible that it's an internal page with only a negative infinity
1757 : * item.
1758 : */
754 pg 1759 UIC 0 : ereport(DEBUG2,
1760 : (errcode(ERRCODE_NO_DATA),
781 peter 1761 EUB : errmsg_internal("%s block %u of index \"%s\" has no first data item",
1762 : P_ISLEAF(opaque) ? "leaf" : "internal", targetnext,
1763 : RelationGetRelationName(state->rel))));
2222 andres 1764 UIC 0 : return NULL;
1765 : }
2222 andres 1766 EUB :
1767 : /*
1768 : * Return first real item scankey. Note that this relies on right page
1769 : * memory remaining allocated.
1770 : */
1481 pg 1771 GIC 5155 : firstitup = (IndexTuple) PageGetItem(rightpage, rightitem);
8 andres 1772 GNC 5155 : return bt_mkscankey_pivotsearch(state->rel, state->heaprel, firstitup);
2222 andres 1773 ECB : }
1774 :
1775 : /*
1776 : * Check if two tuples are binary identical except the block number. So,
1777 : * this function is capable to compare pivot keys on different levels.
1778 : */
1779 : static bool
935 pg 1780 GIC 1586 : bt_pivot_tuple_identical(bool heapkeyspace, IndexTuple itup1, IndexTuple itup2)
1781 : {
1124 akorotkov 1782 CBC 1586 : if (IndexTupleSize(itup1) != IndexTupleSize(itup2))
1124 akorotkov 1783 UIC 0 : return false;
1124 akorotkov 1784 ECB :
935 pg 1785 GBC 1586 : if (heapkeyspace)
1786 : {
935 pg 1787 ECB : /*
1788 : * Offset number will contain important information in heapkeyspace
1789 : * indexes: the number of attributes left in the pivot tuple following
1790 : * suffix truncation. Don't skip over it (compare it too).
1791 : */
935 pg 1792 GIC 1586 : if (memcmp(&itup1->t_tid.ip_posid, &itup2->t_tid.ip_posid,
1793 1586 : IndexTupleSize(itup1) -
935 pg 1794 ECB : offsetof(ItemPointerData, ip_posid)) != 0)
935 pg 1795 LBC 0 : return false;
1796 : }
935 pg 1797 EUB : else
1798 : {
1799 : /*
1800 : * Cannot rely on offset number field having consistent value across
1801 : * levels on pg_upgrade'd !heapkeyspace indexes. Compare contents of
1802 : * tuple starting from just after item pointer (i.e. after block
1803 : * number and offset number).
1804 : */
935 pg 1805 UIC 0 : if (memcmp(&itup1->t_info, &itup2->t_info,
1806 0 : IndexTupleSize(itup1) -
935 pg 1807 EUB : offsetof(IndexTupleData, t_info)) != 0)
935 pg 1808 UBC 0 : return false;
1809 : }
1124 akorotkov 1810 EUB :
1124 akorotkov 1811 GIC 1586 : return true;
1812 : }
1124 akorotkov 1813 ECB :
1814 : /*---
1815 : * Check high keys on the child level. Traverse rightlinks from previous
1816 : * downlink to the current one. Check that there are no intermediate pages
1817 : * with missing downlinks.
1818 : *
1819 : * If 'loaded_child' is given, it's assumed to be the page pointed to by the
1820 : * downlink referenced by 'downlinkoffnum' of the target page.
1821 : *
1822 : * Basically this function is called for each target downlink and checks two
1823 : * invariants:
1824 : *
1825 : * 1) You can reach the next child from previous one via rightlinks;
1826 : * 2) Each child high key have matching pivot key on target level.
1827 : *
1828 : * Consider the sample tree picture.
1829 : *
1830 : * 1
1831 : * / \
1832 : * 2 <-> 3
1833 : * / \ / \
1834 : * 4 <> 5 <> 6 <> 7 <> 8
1835 : *
1836 : * This function will be called for blocks 4, 5, 6 and 8. Consider what is
1837 : * happening for each function call.
1838 : *
1839 : * - The function call for block 4 initializes data structure and matches high
1840 : * key of block 4 to downlink's pivot key of block 2.
1841 : * - The high key of block 5 is matched to the high key of block 2.
1842 : * - The block 6 has an incomplete split flag set, so its high key isn't
1843 : * matched to anything.
1844 : * - The function call for block 8 checks that block 8 can be found while
1845 : * following rightlinks from block 6. The high key of block 7 will be
1846 : * matched to downlink's pivot key in block 3.
1847 : *
1848 : * There is also final call of this function, which checks that there is no
1849 : * missing downlinks for children to the right of the child referenced by
1850 : * rightmost downlink in target level.
1851 : */
1852 : static void
1124 akorotkov 1853 GIC 1602 : bt_child_highkey_check(BtreeCheckState *state,
1854 : OffsetNumber target_downlinkoffnum,
1124 akorotkov 1855 ECB : Page loaded_child,
1856 : uint32 target_level)
1857 : {
1124 akorotkov 1858 GIC 1602 : BlockNumber blkno = state->prevrightlink;
1859 : Page page;
1124 akorotkov 1860 ECB : BTPageOpaque opaque;
1124 akorotkov 1861 GIC 1602 : bool rightsplit = state->previncompletesplit;
1862 1602 : bool first = true;
1124 akorotkov 1863 ECB : ItemId itemid;
1864 : IndexTuple itup;
1865 : BlockNumber downlink;
1866 :
1124 akorotkov 1867 GIC 1602 : if (OffsetNumberIsValid(target_downlinkoffnum))
1868 : {
1124 akorotkov 1869 CBC 1594 : itemid = PageGetItemIdCareful(state, state->targetblock,
1870 : state->target, target_downlinkoffnum);
1871 1594 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1124 akorotkov 1872 GIC 1594 : downlink = BTreeTupleGetDownLink(itup);
1124 akorotkov 1873 ECB : }
1874 : else
1875 : {
1124 akorotkov 1876 GIC 8 : downlink = P_NONE;
1877 : }
1124 akorotkov 1878 ECB :
1879 : /*
1880 : * If no previous rightlink is memorized for current level just below
1881 : * target page's level, we are about to start from the leftmost page. We
1882 : * can't follow rightlinks from previous page, because there is no
1883 : * previous page. But we still can match high key.
1884 : *
1885 : * So we initialize variables for the loop above like there is previous
1886 : * page referencing current child. Also we imply previous page to not
1887 : * have incomplete split flag, that would make us require downlink for
1888 : * current child. That's correct, because leftmost page on the level
1889 : * should always have parent downlink.
1890 : */
1124 akorotkov 1891 GIC 1602 : if (!BlockNumberIsValid(blkno))
1892 : {
1124 akorotkov 1893 CBC 8 : blkno = downlink;
1124 akorotkov 1894 GIC 8 : rightsplit = false;
1124 akorotkov 1895 ECB : }
1896 :
1897 : /* Move to the right on the child level */
1898 : while (true)
1899 : {
1900 : /*
1901 : * Did we traverse the whole tree level and this is check for pages to
1902 : * the right of rightmost downlink?
1903 : */
1124 akorotkov 1904 GIC 1602 : if (blkno == P_NONE && downlink == P_NONE)
1905 : {
1124 akorotkov 1906 CBC 8 : state->prevrightlink = InvalidBlockNumber;
1124 akorotkov 1907 GIC 8 : state->previncompletesplit = false;
1124 akorotkov 1908 CBC 8 : return;
1124 akorotkov 1909 ECB : }
1910 :
1911 : /* Did we traverse the whole tree level and don't find next downlink? */
1124 akorotkov 1912 GIC 1594 : if (blkno == P_NONE)
1124 akorotkov 1913 UIC 0 : ereport(ERROR,
1124 akorotkov 1914 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1124 akorotkov 1915 EUB : errmsg("can't traverse from downlink %u to downlink %u of index \"%s\"",
1916 : state->prevrightlink, downlink,
1917 : RelationGetRelationName(state->rel))));
1918 :
1919 : /* Load page contents */
1124 akorotkov 1920 GIC 1594 : if (blkno == downlink && loaded_child)
1921 1585 : page = loaded_child;
1124 akorotkov 1922 ECB : else
1124 akorotkov 1923 CBC 9 : page = palloc_btree_page(state, blkno);
1924 :
373 michael 1925 1594 : opaque = BTPageGetOpaque(page);
1926 :
1124 akorotkov 1927 ECB : /* The first page we visit at the level should be leftmost */
1124 akorotkov 1928 GIC 1594 : if (first && !BlockNumberIsValid(state->prevrightlink) && !P_LEFTMOST(opaque))
1124 akorotkov 1929 UIC 0 : ereport(ERROR,
1124 akorotkov 1930 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1124 akorotkov 1931 EUB : errmsg("the first child of leftmost target page is not leftmost of its level in index \"%s\"",
1932 : RelationGetRelationName(state->rel)),
1933 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
1934 : state->targetblock, blkno,
1935 : LSN_FORMAT_ARGS(state->targetlsn))));
1936 :
1937 : /* Do level sanity check */
774 pg 1938 GIC 1594 : if ((!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque)) &&
1939 1594 : opaque->btpo_level != target_level - 1)
1124 akorotkov 1940 LBC 0 : ereport(ERROR,
1124 akorotkov 1941 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1124 akorotkov 1942 EUB : errmsg("block found while following rightlinks from child of index \"%s\" has invalid level",
1943 : RelationGetRelationName(state->rel)),
1944 : errdetail_internal("Block pointed to=%u expected level=%u level in pointed to block=%u.",
1945 : blkno, target_level - 1, opaque->btpo_level)));
1946 :
1947 : /* Try to detect circular links */
1124 akorotkov 1948 GIC 1594 : if ((!first && blkno == state->prevrightlink) || blkno == opaque->btpo_prev)
1124 akorotkov 1949 UIC 0 : ereport(ERROR,
1124 akorotkov 1950 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1124 akorotkov 1951 EUB : errmsg("circular link chain found in block %u of index \"%s\"",
1952 : blkno, RelationGetRelationName(state->rel))));
1953 :
1124 akorotkov 1954 GIC 1594 : if (blkno != downlink && !P_IGNORE(opaque))
1955 : {
1124 akorotkov 1956 ECB : /* blkno probably has missing parent downlink */
1124 akorotkov 1957 UIC 0 : bt_downlink_missing_check(state, rightsplit, blkno, page);
1958 : }
1124 akorotkov 1959 EUB :
1124 akorotkov 1960 GIC 1594 : rightsplit = P_INCOMPLETE_SPLIT(opaque);
1961 :
1124 akorotkov 1962 ECB : /*
1963 : * If we visit page with high key, check that it is equal to the
1964 : * target key next to corresponding downlink.
1965 : */
1124 akorotkov 1966 GIC 1594 : if (!rightsplit && !P_RIGHTMOST(opaque))
1967 : {
1124 akorotkov 1968 ECB : BTPageOpaque topaque;
1969 : IndexTuple highkey;
1970 : OffsetNumber pivotkey_offset;
1971 :
1972 : /* Get high key */
1124 akorotkov 1973 GIC 1586 : itemid = PageGetItemIdCareful(state, blkno, page, P_HIKEY);
1974 1586 : highkey = (IndexTuple) PageGetItem(page, itemid);
1124 akorotkov 1975 ECB :
1976 : /*
1977 : * There might be two situations when we examine high key. If
1978 : * current child page is referenced by given target downlink, we
1979 : * should look to the next offset number for matching key from
1980 : * target page.
1981 : *
1982 : * Alternatively, we're following rightlinks somewhere in the
1983 : * middle between page referenced by previous target's downlink
1984 : * and the page referenced by current target's downlink. If
1985 : * current child page hasn't incomplete split flag set, then its
1986 : * high key should match to the target's key of current offset
1987 : * number. This happens when a previous call here (to
1988 : * bt_child_highkey_check()) found an incomplete split, and we
1989 : * reach a right sibling page without a downlink -- the right
1990 : * sibling page's high key still needs to be matched to a
1991 : * separator key on the parent/target level.
1992 : *
1993 : * Don't apply OffsetNumberNext() to target_downlinkoffnum when we
1994 : * already had to step right on the child level. Our traversal of
1995 : * the child level must try to move in perfect lockstep behind (to
1996 : * the left of) the target/parent level traversal.
1997 : */
1124 akorotkov 1998 GIC 1586 : if (blkno == downlink)
1999 1586 : pivotkey_offset = OffsetNumberNext(target_downlinkoffnum);
1124 akorotkov 2000 ECB : else
1124 akorotkov 2001 LBC 0 : pivotkey_offset = target_downlinkoffnum;
2002 :
373 michael 2003 GBC 1586 : topaque = BTPageGetOpaque(state->target);
2004 :
1124 akorotkov 2005 CBC 1586 : if (!offset_is_negative_infinity(topaque, pivotkey_offset))
2006 : {
1124 akorotkov 2007 ECB : /*
2008 : * If we're looking for the next pivot tuple in target page,
2009 : * but there is no more pivot tuples, then we should match to
2010 : * high key instead.
2011 : */
1124 akorotkov 2012 GIC 1586 : if (pivotkey_offset > PageGetMaxOffsetNumber(state->target))
2013 : {
1124 akorotkov 2014 CBC 1 : if (P_RIGHTMOST(topaque))
1124 akorotkov 2015 UIC 0 : ereport(ERROR,
1124 akorotkov 2016 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1124 akorotkov 2017 EUB : errmsg("child high key is greater than rightmost pivot key on target level in index \"%s\"",
2018 : RelationGetRelationName(state->rel)),
2019 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2020 : state->targetblock, blkno,
2021 : LSN_FORMAT_ARGS(state->targetlsn))));
1124 akorotkov 2022 GIC 1 : pivotkey_offset = P_HIKEY;
2023 : }
1124 akorotkov 2024 CBC 1586 : itemid = PageGetItemIdCareful(state, state->targetblock,
2025 : state->target, pivotkey_offset);
2026 1586 : itup = (IndexTuple) PageGetItem(state->target, itemid);
2027 : }
1124 akorotkov 2028 ECB : else
2029 : {
2030 : /*
2031 : * We cannot try to match child's high key to a negative
2032 : * infinity key in target, since there is nothing to compare.
2033 : * However, it's still possible to match child's high key
2034 : * outside of target page. The reason why we're are is that
2035 : * bt_child_highkey_check() was previously called for the
2036 : * cousin page of 'loaded_child', which is incomplete split.
2037 : * So, now we traverse to the right of that cousin page and
2038 : * current child level page under consideration still belongs
2039 : * to the subtree of target's left sibling. Thus, we need to
2040 : * match child's high key to it's left uncle page high key.
2041 : * Thankfully we saved it, it's called a "low key" of target
2042 : * page.
2043 : */
1124 akorotkov 2044 UIC 0 : if (!state->lowkey)
2045 0 : ereport(ERROR,
1124 akorotkov 2046 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
2047 : errmsg("can't find left sibling high key in index \"%s\"",
2048 : RelationGetRelationName(state->rel)),
2049 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2050 : state->targetblock, blkno,
2051 : LSN_FORMAT_ARGS(state->targetlsn))));
1124 akorotkov 2052 UIC 0 : itup = state->lowkey;
2053 : }
1124 akorotkov 2054 EUB :
935 pg 2055 GIC 1586 : if (!bt_pivot_tuple_identical(state->heapkeyspace, highkey, itup))
2056 : {
1124 akorotkov 2057 LBC 0 : ereport(ERROR,
2058 : (errcode(ERRCODE_INDEX_CORRUPTED),
1124 akorotkov 2059 EUB : errmsg("mismatch between parent key and child high key in index \"%s\"",
2060 : RelationGetRelationName(state->rel)),
2061 : errdetail_internal("Target block=%u child block=%u target page lsn=%X/%X.",
2062 : state->targetblock, blkno,
2063 : LSN_FORMAT_ARGS(state->targetlsn))));
2064 : }
2065 : }
2066 :
2067 : /* Exit if we already found next downlink */
1124 akorotkov 2068 GIC 1594 : if (blkno == downlink)
2069 : {
1124 akorotkov 2070 CBC 1594 : state->prevrightlink = opaque->btpo_next;
1124 akorotkov 2071 GIC 1594 : state->previncompletesplit = rightsplit;
1124 akorotkov 2072 CBC 1594 : return;
1124 akorotkov 2073 ECB : }
2074 :
2075 : /* Traverse to the next page using rightlink */
1124 akorotkov 2076 UIC 0 : blkno = opaque->btpo_next;
2077 :
1124 akorotkov 2078 EUB : /* Free page contents if it's allocated by us */
1124 akorotkov 2079 UIC 0 : if (page != loaded_child)
2080 0 : pfree(page);
1124 akorotkov 2081 UBC 0 : first = false;
1124 akorotkov 2082 EUB : }
2083 : }
2084 :
2085 : /*
2086 : * Checks one of target's downlink against its child page.
2087 : *
2088 : * Conceptually, the target page continues to be what is checked here. The
2089 : * target block is still blamed in the event of finding an invariant violation.
2090 : * The downlink insertion into the target is probably where any problem raised
2091 : * here arises, and there is no such thing as a parent link, so doing the
2092 : * verification this way around is much more practical.
2093 : *
2094 : * This function visits child page and it's sequentially called for each
2095 : * downlink of target page. Assuming this we also check downlink connectivity
2096 : * here in order to save child page visits.
2097 : */
2098 : static void
1124 akorotkov 2099 GIC 1585 : bt_child_check(BtreeCheckState *state, BTScanInsert targetkey,
2100 : OffsetNumber downlinkoffnum)
2222 andres 2101 ECB : {
2102 : ItemId itemid;
2103 : IndexTuple itup;
2104 : BlockNumber childblock;
2105 : OffsetNumber offset;
2106 : OffsetNumber maxoffset;
2107 : Page child;
2108 : BTPageOpaque copaque;
2109 : BTPageOpaque topaque;
2110 :
1124 akorotkov 2111 GIC 1585 : itemid = PageGetItemIdCareful(state, state->targetblock,
2112 : state->target, downlinkoffnum);
1124 akorotkov 2113 CBC 1585 : itup = (IndexTuple) PageGetItem(state->target, itemid);
1124 akorotkov 2114 GIC 1585 : childblock = BTreeTupleGetDownLink(itup);
2222 andres 2115 ECB :
2116 : /*
2117 : * Caller must have ShareLock on target relation, because of
2118 : * considerations around page deletion by VACUUM.
2119 : *
2120 : * NB: In general, page deletion deletes the right sibling's downlink, not
2121 : * the downlink of the page being deleted; the deleted page's downlink is
2122 : * reused for its sibling. The key space is thereby consolidated between
2123 : * the deleted page and its right sibling. (We cannot delete a parent
2124 : * page's rightmost child unless it is the last child page, and we intend
2125 : * to also delete the parent itself.)
2126 : *
2127 : * If this verification happened without a ShareLock, the following race
2128 : * condition could cause false positives:
2129 : *
2130 : * In general, concurrent page deletion might occur, including deletion of
2131 : * the left sibling of the child page that is examined here. If such a
2132 : * page deletion were to occur, closely followed by an insertion into the
2133 : * newly expanded key space of the child, a window for the false positive
2134 : * opens up: the stale parent/target downlink originally followed to get
2135 : * to the child legitimately ceases to be a lower bound on all items in
2136 : * the page, since the key space was concurrently expanded "left".
2137 : * (Insertion followed the "new" downlink for the child, not our now-stale
2138 : * downlink, which was concurrently physically removed in target/parent as
2139 : * part of deletion's first phase.)
2140 : *
2141 : * While we use various techniques elsewhere to perform cross-page
2142 : * verification for !readonly callers, a similar trick seems difficult
2143 : * here. The tricks used by bt_recheck_sibling_links and by
2144 : * bt_right_page_check_scankey both involve verification of a same-level,
2145 : * cross-sibling invariant. Cross-level invariants are far more squishy,
2146 : * though. The nbtree REDO routines do not actually couple buffer locks
2147 : * across levels during page splits, so making any cross-level check work
2148 : * reliably in !readonly mode may be impossible.
2149 : */
2222 andres 2150 GIC 1585 : Assert(state->readonly);
2151 :
2222 andres 2152 ECB : /*
2153 : * Verify child page has the downlink key from target page (its parent) as
2154 : * a lower bound; downlink must be strictly less than all keys on the
2155 : * page.
2156 : *
2157 : * Check all items, rather than checking just the first and trusting that
2158 : * the operator class obeys the transitive law.
2159 : */
373 michael 2160 GIC 1585 : topaque = BTPageGetOpaque(state->target);
2222 andres 2161 1585 : child = palloc_btree_page(state, childblock);
373 michael 2162 CBC 1585 : copaque = BTPageGetOpaque(child);
2222 andres 2163 1585 : maxoffset = PageGetMaxOffsetNumber(child);
2222 andres 2164 ECB :
1124 akorotkov 2165 : /*
2166 : * Since we've already loaded the child block, combine this check with
2167 : * check for downlink connectivity.
2168 : */
1124 akorotkov 2169 GIC 1585 : bt_child_highkey_check(state, downlinkoffnum,
2170 : child, topaque->btpo_level);
1124 akorotkov 2171 ECB :
2172 : /*
2173 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2174 : * and since a page has no links within other pages (siblings and parent)
2175 : * once it is marked fully deleted, it should be impossible to land on a
2176 : * fully deleted page.
2177 : *
2178 : * It does not quite make sense to enforce that the page cannot even be
2179 : * half-dead, despite the fact the downlink is modified at the same stage
2180 : * that the child leaf page is marked half-dead. That's incorrect because
2181 : * there may occasionally be multiple downlinks from a chain of pages
2182 : * undergoing deletion, where multiple successive calls are made to
2183 : * _bt_unlink_halfdead_page() by VACUUM before it can finally safely mark
2184 : * the leaf page as fully dead. While _bt_mark_page_halfdead() usually
2185 : * removes the downlink to the leaf page that is marked half-dead, that's
2186 : * not guaranteed, so it's possible we'll land on a half-dead page with a
2187 : * downlink due to an interrupted multi-level page deletion.
2188 : *
2189 : * We go ahead with our checks if the child page is half-dead. It's safe
2190 : * to do so because we do not test the child's high key, so it does not
2191 : * matter that the original high key will have been replaced by a dummy
2192 : * truncated high key within _bt_mark_page_halfdead(). All other page
2193 : * items are left intact on a half-dead page, so there is still something
2194 : * to test.
2195 : */
1810 teodor 2196 GIC 1585 : if (P_ISDELETED(copaque))
1810 teodor 2197 UIC 0 : ereport(ERROR,
1810 teodor 2198 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1810 teodor 2199 EUB : errmsg("downlink to deleted page found in index \"%s\"",
2200 : RelationGetRelationName(state->rel)),
2201 : errdetail_internal("Parent block=%u child block=%u parent page lsn=%X/%X.",
2202 : state->targetblock, childblock,
2203 : LSN_FORMAT_ARGS(state->targetlsn))));
2204 :
2222 andres 2205 GIC 1585 : for (offset = P_FIRSTDATAKEY(copaque);
2206 499804 : offset <= maxoffset;
2222 andres 2207 CBC 498219 : offset = OffsetNumberNext(offset))
2222 andres 2208 ECB : {
2209 : /*
2210 : * Skip comparison of target page key against "negative infinity"
2211 : * item, if any. Checking it would indicate that it's not a strict
2212 : * lower bound, but that's only because of the hard-coding for
2213 : * negative infinity items within _bt_compare().
2214 : *
2215 : * If nbtree didn't truncate negative infinity tuples during internal
2216 : * page splits then we'd expect child's negative infinity key to be
2217 : * equal to the scankey/downlink from target/parent (it would be a
2218 : * "low key" in this hypothetical scenario, and so it would still need
2219 : * to be treated as a special case here).
2220 : *
2221 : * Negative infinity items can be thought of as a strict lower bound
2222 : * that works transitively, with the last non-negative-infinity pivot
2223 : * followed during a descent from the root as its "true" strict lower
2224 : * bound. Only a small number of negative infinity items are truly
2225 : * negative infinity; those that are the first items of leftmost
2226 : * internal pages. In more general terms, a negative infinity item is
2227 : * only negative infinity with respect to the subtree that the page is
2228 : * at the root of.
2229 : *
2230 : * See also: bt_rootdescend(), which can even detect transitive
2231 : * inconsistencies on cousin leaf pages.
2232 : */
2222 andres 2233 GIC 498219 : if (offset_is_negative_infinity(copaque, offset))
2234 1 : continue;
2222 andres 2235 ECB :
1445 pg 2236 CBC 498218 : if (!invariant_l_nontarget_offset(state, targetkey, childblock, child,
2237 : offset))
2222 andres 2238 LBC 0 : ereport(ERROR,
2239 : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 2240 EUB : errmsg("down-link lower bound invariant violated for index \"%s\"",
2241 : RelationGetRelationName(state->rel)),
2242 : errdetail_internal("Parent block=%u child index tid=(%u,%u) parent page lsn=%X/%X.",
2243 : state->targetblock, childblock, offset,
2244 : LSN_FORMAT_ARGS(state->targetlsn))));
2245 : }
2246 :
2222 andres 2247 GIC 1585 : pfree(child);
2248 1585 : }
2222 andres 2249 ECB :
1810 teodor 2250 : /*
2251 : * Checks if page is missing a downlink that it should have.
2252 : *
2253 : * A page that lacks a downlink/parent may indicate corruption. However, we
2254 : * must account for the fact that a missing downlink can occasionally be
2255 : * encountered in a non-corrupt index. This can be due to an interrupted page
2256 : * split, or an interrupted multi-level page deletion (i.e. there was a hard
2257 : * crash or an error during a page split, or while VACUUM was deleting a
2258 : * multi-level chain of pages).
2259 : *
2260 : * Note that this can only be called in readonly mode, so there is no need to
2261 : * be concerned about concurrent page splits or page deletions.
2262 : */
2263 : static void
1124 akorotkov 2264 UIC 0 : bt_downlink_missing_check(BtreeCheckState *state, bool rightsplit,
2265 : BlockNumber blkno, Page page)
1810 teodor 2266 EUB : {
373 michael 2267 UIC 0 : BTPageOpaque opaque = BTPageGetOpaque(page);
2268 : ItemId itemid;
1809 tgl 2269 EUB : IndexTuple itup;
2270 : Page child;
2271 : BTPageOpaque copaque;
2272 : uint32 level;
2273 : BlockNumber childblk;
2274 : XLogRecPtr pagelsn;
2275 :
1124 akorotkov 2276 UIC 0 : Assert(state->readonly);
2277 0 : Assert(!P_IGNORE(opaque));
1810 teodor 2278 EUB :
2279 : /* No next level up with downlinks to fingerprint from the true root */
1124 akorotkov 2280 UIC 0 : if (P_ISROOT(opaque))
1810 teodor 2281 0 : return;
1810 teodor 2282 EUB :
1124 akorotkov 2283 UBC 0 : pagelsn = PageGetLSN(page);
2284 :
1810 teodor 2285 EUB : /*
2286 : * Incomplete (interrupted) page splits can account for the lack of a
2287 : * downlink. Some inserting transaction should eventually complete the
2288 : * page split in passing, when it notices that the left sibling page is
2289 : * P_INCOMPLETE_SPLIT().
2290 : *
2291 : * In general, VACUUM is not prepared for there to be no downlink to a
2292 : * page that it deletes. This is the main reason why the lack of a
2293 : * downlink can be reported as corruption here. It's not obvious that an
2294 : * invalid missing downlink can result in wrong answers to queries,
2295 : * though, since index scans that land on the child may end up
2296 : * consistently moving right. The handling of concurrent page splits (and
2297 : * page deletions) within _bt_moveright() cannot distinguish
2298 : * inconsistencies that last for a moment from inconsistencies that are
2299 : * permanent and irrecoverable.
2300 : *
2301 : * VACUUM isn't even prepared to delete pages that have no downlink due to
2302 : * an incomplete page split, but it can detect and reason about that case
2303 : * by design, so it shouldn't be taken to indicate corruption. See
2304 : * _bt_pagedel() for full details.
2305 : */
1124 akorotkov 2306 UIC 0 : if (rightsplit)
2307 : {
1810 teodor 2308 UBC 0 : ereport(DEBUG1,
2309 : (errcode(ERRCODE_NO_DATA),
725 peter 2310 EUB : errmsg_internal("harmless interrupted page split detected in index \"%s\"",
2311 : RelationGetRelationName(state->rel)),
2312 : errdetail_internal("Block=%u level=%u left sibling=%u page lsn=%X/%X.",
2313 : blkno, opaque->btpo_level,
2314 : opaque->btpo_prev,
2315 : LSN_FORMAT_ARGS(pagelsn))));
1810 teodor 2316 UIC 0 : return;
2317 : }
1810 teodor 2318 EUB :
2319 : /*
2320 : * Page under check is probably the "top parent" of a multi-level page
2321 : * deletion. We'll need to descend the subtree to make sure that
2322 : * descendant pages are consistent with that, though.
2323 : *
2324 : * If the page (which must be non-ignorable) is a leaf page, then clearly
2325 : * it can't be the top parent. The lack of a downlink is probably a
2326 : * symptom of a broad problem that could just as easily cause
2327 : * inconsistencies anywhere else.
2328 : */
1124 akorotkov 2329 UIC 0 : if (P_ISLEAF(opaque))
1810 teodor 2330 0 : ereport(ERROR,
1810 teodor 2331 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
2332 : errmsg("leaf index block lacks downlink in index \"%s\"",
2333 : RelationGetRelationName(state->rel)),
2334 : errdetail_internal("Block=%u page lsn=%X/%X.",
2335 : blkno,
2336 : LSN_FORMAT_ARGS(pagelsn))));
2337 :
2338 : /* Descend from the given page, which is an internal page */
1810 teodor 2339 UIC 0 : elog(DEBUG1, "checking for interrupted multi-level deletion due to missing downlink in index \"%s\"",
2340 : RelationGetRelationName(state->rel));
1810 teodor 2341 EUB :
774 pg 2342 UIC 0 : level = opaque->btpo_level;
1124 akorotkov 2343 0 : itemid = PageGetItemIdCareful(state, blkno, page, P_FIRSTDATAKEY(opaque));
1124 akorotkov 2344 UBC 0 : itup = (IndexTuple) PageGetItem(page, itemid);
1210 pg 2345 0 : childblk = BTreeTupleGetDownLink(itup);
1810 teodor 2346 EUB : for (;;)
2347 : {
1810 teodor 2348 UIC 0 : CHECK_FOR_INTERRUPTS();
2349 :
1810 teodor 2350 UBC 0 : child = palloc_btree_page(state, childblk);
373 michael 2351 UIC 0 : copaque = BTPageGetOpaque(child);
1810 teodor 2352 EUB :
1810 teodor 2353 UBC 0 : if (P_ISLEAF(copaque))
1810 teodor 2354 UIC 0 : break;
1810 teodor 2355 EUB :
2356 : /* Do an extra sanity check in passing on internal pages */
774 pg 2357 UIC 0 : if (copaque->btpo_level != level - 1)
1810 teodor 2358 0 : ereport(ERROR,
1810 teodor 2359 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
2360 : errmsg_internal("downlink points to block in index \"%s\" whose level is not one level down",
2361 : RelationGetRelationName(state->rel)),
2362 : errdetail_internal("Top parent/under check block=%u block pointed to=%u expected level=%u level in pointed to block=%u.",
2363 : blkno, childblk,
2364 : level - 1, copaque->btpo_level)));
2365 :
774 pg 2366 UIC 0 : level = copaque->btpo_level;
1445 2367 0 : itemid = PageGetItemIdCareful(state, childblk, child,
1445 pg 2368 UBC 0 : P_FIRSTDATAKEY(copaque));
1810 teodor 2369 0 : itup = (IndexTuple) PageGetItem(child, itemid);
1210 pg 2370 0 : childblk = BTreeTupleGetDownLink(itup);
1810 teodor 2371 EUB : /* Be slightly more pro-active in freeing this memory, just in case */
1810 teodor 2372 UBC 0 : pfree(child);
2373 : }
1810 teodor 2374 EUB :
2375 : /*
2376 : * Since there cannot be a concurrent VACUUM operation in readonly mode,
2377 : * and since a page has no links within other pages (siblings and parent)
2378 : * once it is marked fully deleted, it should be impossible to land on a
2379 : * fully deleted page. See bt_child_check() for further details.
2380 : *
2381 : * The bt_child_check() P_ISDELETED() check is repeated here because
2382 : * bt_child_check() does not visit pages reachable through negative
2383 : * infinity items. Besides, bt_child_check() is unwilling to descend
2384 : * multiple levels. (The similar bt_child_check() P_ISDELETED() check
2385 : * within bt_check_level_from_leftmost() won't reach the page either,
2386 : * since the leaf's live siblings should have their sibling links updated
2387 : * to bypass the deletion target page when it is marked fully dead.)
2388 : *
2389 : * If this error is raised, it might be due to a previous multi-level page
2390 : * deletion that failed to realize that it wasn't yet safe to mark the
2391 : * leaf page as fully dead. A "dangling downlink" will still remain when
2392 : * this happens. The fact that the dangling downlink's page (the leaf's
2393 : * parent/ancestor page) lacked a downlink is incidental.
2394 : */
1810 teodor 2395 UIC 0 : if (P_ISDELETED(copaque))
2396 0 : ereport(ERROR,
1810 teodor 2397 EUB : (errcode(ERRCODE_INDEX_CORRUPTED),
2398 : errmsg_internal("downlink to deleted leaf page found in index \"%s\"",
2399 : RelationGetRelationName(state->rel)),
2400 : errdetail_internal("Top parent/target block=%u leaf block=%u top parent/under check lsn=%X/%X.",
2401 : blkno, childblk,
2402 : LSN_FORMAT_ARGS(pagelsn))));
2403 :
2404 : /*
2405 : * Iff leaf page is half-dead, its high key top parent link should point
2406 : * to what VACUUM considered to be the top parent page at the instant it
2407 : * was interrupted. Provided the high key link actually points to the
2408 : * page under check, the missing downlink we detected is consistent with
2409 : * there having been an interrupted multi-level page deletion. This means
2410 : * that the subtree with the page under check at its root (a page deletion
2411 : * chain) is in a consistent state, enabling VACUUM to resume deleting the
2412 : * entire chain the next time it encounters the half-dead leaf page.
2413 : */
1810 teodor 2414 UIC 0 : if (P_ISHALFDEAD(copaque) && !P_RIGHTMOST(copaque))
2415 : {
1445 pg 2416 UBC 0 : itemid = PageGetItemIdCareful(state, childblk, child, P_HIKEY);
1810 teodor 2417 UIC 0 : itup = (IndexTuple) PageGetItem(child, itemid);
1124 akorotkov 2418 UBC 0 : if (BTreeTupleGetTopParent(itup) == blkno)
1810 teodor 2419 0 : return;
1810 teodor 2420 EUB : }
2421 :
1810 teodor 2422 UIC 0 : ereport(ERROR,
2423 : (errcode(ERRCODE_INDEX_CORRUPTED),
1810 teodor 2424 EUB : errmsg("internal index block lacks downlink in index \"%s\"",
2425 : RelationGetRelationName(state->rel)),
2426 : errdetail_internal("Block=%u level=%u page lsn=%X/%X.",
2427 : blkno, opaque->btpo_level,
2428 : LSN_FORMAT_ARGS(pagelsn))));
2429 : }
2430 :
2431 : /*
2432 : * Per-tuple callback from table_index_build_scan, used to determine if index has
2433 : * all the entries that definitely should have been observed in leaf pages of
2434 : * the target index (that is, all IndexTuples that were fingerprinted by our
2435 : * Bloom filter). All heapallindexed checks occur here.
2436 : *
2437 : * The redundancy between an index and the table it indexes provides a good
2438 : * opportunity to detect corruption, especially corruption within the table.
2439 : * The high level principle behind the verification performed here is that any
2440 : * IndexTuple that should be in an index following a fresh CREATE INDEX (based
2441 : * on the same index definition) should also have been in the original,
2442 : * existing index, which should have used exactly the same representation
2443 : *
2444 : * Since the overall structure of the index has already been verified, the most
2445 : * likely explanation for error here is a corrupt heap page (could be logical
2446 : * or physical corruption). Index corruption may still be detected here,
2447 : * though. Only readonly callers will have verified that left links and right
2448 : * links are in agreement, and so it's possible that a leaf page transposition
2449 : * within index is actually the source of corruption detected here (for
2450 : * !readonly callers). The checks performed only for readonly callers might
2451 : * more accurately frame the problem as a cross-page invariant issue (this
2452 : * could even be due to recovery not replaying all WAL records). The !readonly
2453 : * ERROR message raised here includes a HINT about retrying with readonly
2454 : * verification, just in case it's a cross-page invariant issue, though that
2455 : * isn't particularly likely.
2456 : *
2457 : * table_index_build_scan() expects to be able to find the root tuple when a
2458 : * heap-only tuple (the live tuple at the end of some HOT chain) needs to be
2459 : * indexed, in order to replace the actual tuple's TID with the root tuple's
2460 : * TID (which is what we're actually passed back here). The index build heap
2461 : * scan code will raise an error when a tuple that claims to be the root of the
2462 : * heap-only tuple's HOT chain cannot be located. This catches cases where the
2463 : * original root item offset/root tuple for a HOT chain indicates (for whatever
2464 : * reason) that the entire HOT chain is dead, despite the fact that the latest
2465 : * heap-only tuple should be indexed. When this happens, sequential scans may
2466 : * always give correct answers, and all indexes may be considered structurally
2467 : * consistent (i.e. the nbtree structural checks would not detect corruption).
2468 : * It may be the case that only index scans give wrong answers, and yet heap or
2469 : * SLRU corruption is the real culprit. (While it's true that LP_DEAD bit
2470 : * setting will probably also leave the index in a corrupt state before too
2471 : * long, the problem is nonetheless that there is heap corruption.)
2472 : *
2473 : * Heap-only tuple handling within table_index_build_scan() works in a way that
2474 : * helps us to detect index tuples that contain the wrong values (values that
2475 : * don't match the latest tuple in the HOT chain). This can happen when there
2476 : * is no superseding index tuple due to a faulty assessment of HOT safety,
2477 : * perhaps during the original CREATE INDEX. Because the latest tuple's
2478 : * contents are used with the root TID, an error will be raised when a tuple
2479 : * with the same TID but non-matching attribute values is passed back to us.
2480 : * Faulty assessment of HOT-safety was behind at least two distinct CREATE
2481 : * INDEX CONCURRENTLY bugs that made it into stable releases, one of which was
2482 : * undetected for many years. In short, the same principle that allows a
2483 : * REINDEX to repair corruption when there was an (undetected) broken HOT chain
2484 : * also allows us to detect the corruption in many cases.
2485 : */
2486 : static void
1248 andres 2487 GIC 410615 : bt_tuple_present_callback(Relation index, ItemPointer tid, Datum *values,
2488 : bool *isnull, bool tupleIsAlive, void *checkstate)
1835 andres 2489 ECB : {
1835 andres 2490 GIC 410615 : BtreeCheckState *state = (BtreeCheckState *) checkstate;
2491 : IndexTuple itup,
1418 tgl 2492 ECB : norm;
2493 :
1835 andres 2494 GIC 410615 : Assert(state->heapallindexed);
2495 :
1523 pg 2496 ECB : /* Generate a normalized index tuple for fingerprinting */
1835 andres 2497 GIC 410615 : itup = index_form_tuple(RelationGetDescr(index), values, isnull);
1248 2498 410615 : itup->t_tid = *tid;
1523 pg 2499 CBC 410615 : norm = bt_normalize_tuple(state, itup);
1835 andres 2500 ECB :
2501 : /* Probe Bloom filter -- tuple should be present */
1523 pg 2502 GIC 410615 : if (bloom_lacks_element(state->filter, (unsigned char *) norm,
2503 410615 : IndexTupleSize(norm)))
1835 andres 2504 LBC 0 : ereport(ERROR,
1835 andres 2505 ECB : (errcode(ERRCODE_DATA_CORRUPTED),
1835 andres 2506 EUB : errmsg("heap tuple (%u,%u) from table \"%s\" lacks matching index tuple within index \"%s\"",
2507 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2508 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2509 : RelationGetRelationName(state->heaprel),
2510 : RelationGetRelationName(state->rel)),
2511 : !state->readonly
2512 : ? errhint("Retrying verification using the function bt_index_parent_check() might provide a more specific error.")
2513 : : 0));
2514 :
1835 andres 2515 GIC 410615 : state->heaptuplespresent++;
2516 410615 : pfree(itup);
1523 pg 2517 ECB : /* Cannot leak memory here */
1523 pg 2518 CBC 410615 : if (norm != itup)
1523 pg 2519 GIC 1 : pfree(norm);
1523 pg 2520 CBC 410615 : }
1523 pg 2521 ECB :
2522 : /*
2523 : * Normalize an index tuple for fingerprinting.
2524 : *
2525 : * In general, index tuple formation is assumed to be deterministic by
2526 : * heapallindexed verification, and IndexTuples are assumed immutable. While
2527 : * the LP_DEAD bit is mutable in leaf pages, that's ItemId metadata, which is
2528 : * not fingerprinted. Normalization is required to compensate for corner
2529 : * cases where the determinism assumption doesn't quite work.
2530 : *
2531 : * There is currently one such case: index_form_tuple() does not try to hide
2532 : * the source TOAST state of input datums. The executor applies TOAST
2533 : * compression for heap tuples based on different criteria to the compression
2534 : * applied within btinsert()'s call to index_form_tuple(): it sometimes
2535 : * compresses more aggressively, resulting in compressed heap tuple datums but
2536 : * uncompressed corresponding index tuple datums. A subsequent heapallindexed
2537 : * verification will get a logically equivalent though bitwise unequal tuple
2538 : * from index_form_tuple(). False positive heapallindexed corruption reports
2539 : * could occur without normalizing away the inconsistency.
2540 : *
2541 : * Returned tuple is often caller's own original tuple. Otherwise, it is a
2542 : * new representation of caller's original index tuple, palloc()'d in caller's
2543 : * memory context.
2544 : *
2545 : * Note: This routine is not concerned with distinctions about the
2546 : * representation of tuples beyond those that might break heapallindexed
2547 : * verification. In particular, it won't try to normalize opclass-equal
2548 : * datums with potentially distinct representations (e.g., btree/numeric_ops
2549 : * index datums will not get their display scale normalized-away here).
2550 : * Caller does normalization for non-pivot tuples that have a posting list,
2551 : * since dummy CREATE INDEX callback code generates new tuples with the same
2552 : * normalized representation.
2553 : */
2554 : static IndexTuple
1523 pg 2555 GIC 821272 : bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
2556 : {
1523 pg 2557 CBC 821272 : TupleDesc tupleDescriptor = RelationGetDescr(state->rel);
2558 : Datum normalized[INDEX_MAX_KEYS];
1523 pg 2559 ECB : bool isnull[INDEX_MAX_KEYS];
2560 : bool toast_free[INDEX_MAX_KEYS];
1523 pg 2561 GIC 821272 : bool formnewtup = false;
2562 : IndexTuple reformed;
1523 pg 2563 ECB : int i;
2564 :
2565 : /* Caller should only pass "logical" non-pivot tuples here */
1138 pg 2566 GIC 821272 : Assert(!BTreeTupleIsPosting(itup) && !BTreeTupleIsPivot(itup));
2567 :
1523 pg 2568 ECB : /* Easy case: It's immediately clear that tuple has no varlena datums */
1523 pg 2569 GIC 821272 : if (!IndexTupleHasVarwidths(itup))
2570 821270 : return itup;
1523 pg 2571 ECB :
1523 pg 2572 CBC 4 : for (i = 0; i < tupleDescriptor->natts; i++)
2573 : {
1418 tgl 2574 ECB : Form_pg_attribute att;
2575 :
1523 pg 2576 GIC 2 : att = TupleDescAttr(tupleDescriptor, i);
2577 :
1523 pg 2578 ECB : /* Assume untoasted/already normalized datum initially */
1523 pg 2579 GIC 2 : toast_free[i] = false;
2580 2 : normalized[i] = index_getattr(itup, att->attnum,
1523 pg 2581 ECB : tupleDescriptor,
2582 : &isnull[i]);
1523 pg 2583 GIC 2 : if (att->attbyval || att->attlen != -1 || isnull[i])
1523 pg 2584 UIC 0 : continue;
1523 pg 2585 ECB :
1523 pg 2586 EUB : /*
2587 : * Callers always pass a tuple that could safely be inserted into the
2588 : * index without further processing, so an external varlena header
2589 : * should never be encountered here
2590 : */
1523 pg 2591 GIC 2 : if (VARATT_IS_EXTERNAL(DatumGetPointer(normalized[i])))
1523 pg 2592 UIC 0 : ereport(ERROR,
1523 pg 2593 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1523 pg 2594 EUB : errmsg("external varlena datum in tuple that references heap row (%u,%u) in index \"%s\"",
2595 : ItemPointerGetBlockNumber(&(itup->t_tid)),
2596 : ItemPointerGetOffsetNumber(&(itup->t_tid)),
2597 : RelationGetRelationName(state->rel))));
1523 pg 2598 GIC 2 : else if (VARATT_IS_COMPRESSED(DatumGetPointer(normalized[i])))
2599 : {
1523 pg 2600 CBC 1 : formnewtup = true;
1523 pg 2601 GIC 1 : normalized[i] = PointerGetDatum(PG_DETOAST_DATUM(normalized[i]));
1523 pg 2602 CBC 1 : toast_free[i] = true;
1523 pg 2603 ECB : }
2604 : }
2605 :
2606 : /* Easier case: Tuple has varlena datums, none of which are compressed */
1523 pg 2607 GIC 2 : if (!formnewtup)
2608 1 : return itup;
1523 pg 2609 ECB :
2610 : /*
2611 : * Hard case: Tuple had compressed varlena datums that necessitate
2612 : * creating normalized version of the tuple from uncompressed input datums
2613 : * (normalized input datums). This is rather naive, but shouldn't be
2614 : * necessary too often.
2615 : *
2616 : * Note that we rely on deterministic index_form_tuple() TOAST compression
2617 : * of normalized input.
2618 : */
1523 pg 2619 GIC 1 : reformed = index_form_tuple(tupleDescriptor, normalized, isnull);
2620 1 : reformed->t_tid = itup->t_tid;
1523 pg 2621 ECB :
2622 : /* Cannot leak memory here */
1523 pg 2623 GIC 2 : for (i = 0; i < tupleDescriptor->natts; i++)
2624 1 : if (toast_free[i])
1523 pg 2625 CBC 1 : pfree(DatumGetPointer(normalized[i]));
1523 pg 2626 ECB :
1523 pg 2627 CBC 1 : return reformed;
2628 : }
1835 andres 2629 ECB :
2630 : /*
2631 : * Produce palloc()'d "plain" tuple for nth posting list entry/TID.
2632 : *
2633 : * In general, deduplication is not supposed to change the logical contents of
2634 : * an index. Multiple index tuples are merged together into one equivalent
2635 : * posting list index tuple when convenient.
2636 : *
2637 : * heapallindexed verification must normalize-away this variation in
2638 : * representation by converting posting list tuples into two or more "plain"
2639 : * tuples. Each tuple must be fingerprinted separately -- there must be one
2640 : * tuple for each corresponding Bloom filter probe during the heap scan.
2641 : *
2642 : * Note: Caller still needs to call bt_normalize_tuple() with returned tuple.
2643 : */
2644 : static inline IndexTuple
1138 pg 2645 GIC 9369 : bt_posting_plain_tuple(IndexTuple itup, int n)
2646 : {
1138 pg 2647 CBC 9369 : Assert(BTreeTupleIsPosting(itup));
2648 :
1138 pg 2649 ECB : /* Returns non-posting-list tuple */
1138 pg 2650 GIC 9369 : return _bt_form_posting(itup, BTreeTupleGetPostingN(itup, n), 1);
2651 : }
1138 pg 2652 ECB :
2653 : /*
2654 : * Search for itup in index, starting from fast root page. itup must be a
2655 : * non-pivot tuple. This is only supported with heapkeyspace indexes, since
2656 : * we rely on having fully unique keys to find a match with only a single
2657 : * visit to a leaf page, barring an interrupted page split, where we may have
2658 : * to move right. (A concurrent page split is impossible because caller must
2659 : * be readonly caller.)
2660 : *
2661 : * This routine can detect very subtle transitive consistency issues across
2662 : * more than one level of the tree. Leaf pages all have a high key (even the
2663 : * rightmost page has a conceptual positive infinity high key), but not a low
2664 : * key. Their downlink in parent is a lower bound, which along with the high
2665 : * key is almost enough to detect every possible inconsistency. A downlink
2666 : * separator key value won't always be available from parent, though, because
2667 : * the first items of internal pages are negative infinity items, truncated
2668 : * down to zero attributes during internal page splits. While it's true that
2669 : * bt_child_check() and the high key check can detect most imaginable key
2670 : * space problems, there are remaining problems it won't detect with non-pivot
2671 : * tuples in cousin leaf pages. Starting a search from the root for every
2672 : * existing leaf tuple detects small inconsistencies in upper levels of the
2673 : * tree that cannot be detected any other way. (Besides all this, this is
2674 : * probably also useful as a direct test of the code used by index scans
2675 : * themselves.)
2676 : */
2677 : static bool
1481 pg 2678 GIC 200041 : bt_rootdescend(BtreeCheckState *state, IndexTuple itup)
2679 : {
1481 pg 2680 ECB : BTScanInsert key;
2681 : BTStack stack;
2682 : Buffer lbuf;
2683 : bool exists;
2684 :
8 andres 2685 GNC 200041 : key = _bt_mkscankey(state->rel, state->heaprel, itup);
1481 pg 2686 GIC 200041 : Assert(key->heapkeyspace && key->scantid != NULL);
1481 pg 2687 ECB :
2688 : /*
2689 : * Search from root.
2690 : *
2691 : * Ideally, we would arrange to only move right within _bt_search() when
2692 : * an interrupted page split is detected (i.e. when the incomplete split
2693 : * bit is found to be set), but for now we accept the possibility that
2694 : * that could conceal an inconsistency.
2695 : */
1481 pg 2696 GIC 200041 : Assert(state->readonly && state->rootdescend);
2697 200041 : exists = false;
8 andres 2698 GNC 200041 : stack = _bt_search(state->rel, state->heaprel, key, &lbuf, BT_READ, NULL);
1481 pg 2699 ECB :
1481 pg 2700 CBC 200041 : if (BufferIsValid(lbuf))
2701 : {
1481 pg 2702 ECB : BTInsertStateData insertstate;
2703 : OffsetNumber offnum;
2704 : Page page;
2705 :
1481 pg 2706 GIC 200041 : insertstate.itup = itup;
2707 200041 : insertstate.itemsz = MAXALIGN(IndexTupleSize(itup));
1481 pg 2708 CBC 200041 : insertstate.itup_key = key;
1138 2709 200041 : insertstate.postingoff = 0;
1481 2710 200041 : insertstate.bounds_valid = false;
2711 200041 : insertstate.buf = lbuf;
1481 pg 2712 ECB :
2713 : /* Get matching tuple on leaf page */
1481 pg 2714 GIC 200041 : offnum = _bt_binsrch_insert(state->rel, &insertstate);
2715 : /* Compare first >= matching item on leaf page, if any */
1481 pg 2716 CBC 200041 : page = BufferGetPage(lbuf);
2717 : /* Should match on first heap TID when tuple has a posting list */
2718 200041 : if (offnum <= PageGetMaxOffsetNumber(page) &&
1138 pg 2719 GIC 400082 : insertstate.postingoff <= 0 &&
1481 pg 2720 CBC 200041 : _bt_compare(state->rel, key, page, offnum) == 0)
2721 200041 : exists = true;
2722 200041 : _bt_relbuf(state->rel, lbuf);
1481 pg 2723 ECB : }
2724 :
1481 pg 2725 GIC 200041 : _bt_freestack(stack);
2726 200041 : pfree(key);
1481 pg 2727 ECB :
1481 pg 2728 CBC 200041 : return exists;
2729 : }
1481 pg 2730 ECB :
2731 : /*
2732 : * Is particular offset within page (whose special state is passed by caller)
2733 : * the page negative-infinity item?
2734 : *
2735 : * As noted in comments above _bt_compare(), there is special handling of the
2736 : * first data item as a "negative infinity" item. The hard-coding within
2737 : * _bt_compare() makes comparing this item for the purposes of verification
2738 : * pointless at best, since the IndexTuple only contains a valid TID (a
2739 : * reference TID to child page).
2740 : */
2741 : static inline bool
2222 andres 2742 GIC 2075457 : offset_is_negative_infinity(BTPageOpaque opaque, OffsetNumber offset)
2743 : {
2222 andres 2744 ECB : /*
2745 : * For internal pages only, the first item after high key, if any, is
2746 : * negative infinity item. Internal pages always have a negative infinity
2747 : * item, whereas leaf pages never have one. This implies that negative
2748 : * infinity item is either first or second line item, or there is none
2749 : * within page.
2750 : *
2751 : * Negative infinity items are a special case among pivot tuples. They
2752 : * always have zero attributes, while all other pivot tuples always have
2753 : * nkeyatts attributes.
2754 : *
2755 : * Right-most pages don't have a high key, but could be said to
2756 : * conceptually have a "positive infinity" high key. Thus, there is a
2757 : * symmetry between down link items in parent pages, and high keys in
2758 : * children. Together, they represent the part of the key space that
2759 : * belongs to each page in the index. For example, all children of the
2760 : * root page will have negative infinity as a lower bound from root
2761 : * negative infinity downlink, and positive infinity as an upper bound
2762 : * (implicitly, from "imaginary" positive infinity high key in root).
2763 : */
2222 andres 2764 GIC 2075457 : return !P_ISLEAF(opaque) && offset == P_FIRSTDATAKEY(opaque);
2765 : }
2222 andres 2766 ECB :
2767 : /*
2768 : * Does the invariant hold that the key is strictly less than a given upper
2769 : * bound offset item?
2770 : *
2771 : * Verifies line pointer on behalf of caller.
2772 : *
2773 : * If this function returns false, convention is that caller throws error due
2774 : * to corruption.
2775 : */
2776 : static inline bool
1481 pg 2777 GIC 1568987 : invariant_l_offset(BtreeCheckState *state, BTScanInsert key,
2778 : OffsetNumber upperbound)
1481 pg 2779 ECB : {
2780 : ItemId itemid;
2781 : int32 cmp;
2782 :
1481 pg 2783 GIC 1568987 : Assert(key->pivotsearch);
2784 :
1445 pg 2785 ECB : /* Verify line pointer before checking tuple */
1445 pg 2786 GIC 1568987 : itemid = PageGetItemIdCareful(state, state->targetblock, state->target,
2787 : upperbound);
1481 pg 2788 ECB : /* pg_upgrade'd indexes may legally have equal sibling tuples */
1481 pg 2789 GIC 1568987 : if (!key->heapkeyspace)
1481 pg 2790 UIC 0 : return invariant_leq_offset(state, key, upperbound);
1481 pg 2791 ECB :
1481 pg 2792 GBC 1568987 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
2793 :
1481 pg 2794 ECB : /*
2795 : * _bt_compare() is capable of determining that a scankey with a
2796 : * filled-out attribute is greater than pivot tuples where the comparison
2797 : * is resolved at a truncated attribute (value of attribute in pivot is
2798 : * minus infinity). However, it is not capable of determining that a
2799 : * scankey is _less than_ a tuple on the basis of a comparison resolved at
2800 : * _scankey_ minus infinity attribute. Complete an extra step to simulate
2801 : * having minus infinity values for omitted scankey attribute(s).
2802 : */
1481 pg 2803 GIC 1568987 : if (cmp == 0)
2804 : {
1481 pg 2805 ECB : BTPageOpaque topaque;
2806 : IndexTuple ritup;
2807 : int uppnkeyatts;
2808 : ItemPointer rheaptid;
2809 : bool nonpivot;
2810 :
1481 pg 2811 UIC 0 : ritup = (IndexTuple) PageGetItem(state->target, itemid);
373 michael 2812 0 : topaque = BTPageGetOpaque(state->target);
1481 pg 2813 UBC 0 : nonpivot = P_ISLEAF(topaque) && upperbound >= P_FIRSTDATAKEY(topaque);
1481 pg 2814 EUB :
2815 : /* Get number of keys + heap TID for item to the right */
1481 pg 2816 UIC 0 : uppnkeyatts = BTreeTupleGetNKeyAtts(ritup, state->rel);
2817 0 : rheaptid = BTreeTupleGetHeapTIDCareful(state, ritup, nonpivot);
1481 pg 2818 EUB :
2819 : /* Heap TID is tiebreaker key attribute */
1481 pg 2820 UIC 0 : if (key->keysz == uppnkeyatts)
2821 0 : return key->scantid == NULL && rheaptid != NULL;
1481 pg 2822 EUB :
1481 pg 2823 UBC 0 : return key->keysz < uppnkeyatts;
2824 : }
1481 pg 2825 EUB :
1481 pg 2826 GIC 1568987 : return cmp < 0;
2827 : }
1481 pg 2828 ECB :
2829 : /*
2830 : * Does the invariant hold that the key is less than or equal to a given upper
2831 : * bound offset item?
2832 : *
2833 : * Caller should have verified that upperbound's line pointer is consistent
2834 : * using PageGetItemIdCareful() call.
2835 : *
2836 : * If this function returns false, convention is that caller throws error due
2837 : * to corruption.
2838 : */
2839 : static inline bool
1481 pg 2840 GIC 1450460 : invariant_leq_offset(BtreeCheckState *state, BTScanInsert key,
2841 : OffsetNumber upperbound)
2222 andres 2842 ECB : {
2843 : int32 cmp;
2844 :
1481 pg 2845 GIC 1450460 : Assert(key->pivotsearch);
2846 :
1481 pg 2847 CBC 1450460 : cmp = _bt_compare(state->rel, key, state->target, upperbound);
2848 :
2222 andres 2849 1450460 : return cmp <= 0;
2850 : }
2222 andres 2851 ECB :
2852 : /*
2853 : * Does the invariant hold that the key is strictly greater than a given lower
2854 : * bound offset item?
2855 : *
2856 : * Caller should have verified that lowerbound's line pointer is consistent
2857 : * using PageGetItemIdCareful() call.
2858 : *
2859 : * If this function returns false, convention is that caller throws error due
2860 : * to corruption.
2861 : */
2862 : static inline bool
1481 pg 2863 GIC 5155 : invariant_g_offset(BtreeCheckState *state, BTScanInsert key,
2864 : OffsetNumber lowerbound)
2222 andres 2865 ECB : {
2866 : int32 cmp;
2867 :
1481 pg 2868 GIC 5155 : Assert(key->pivotsearch);
2869 :
1481 pg 2870 CBC 5155 : cmp = _bt_compare(state->rel, key, state->target, lowerbound);
2871 :
1481 pg 2872 ECB : /* pg_upgrade'd indexes may legally have equal sibling tuples */
1481 pg 2873 GIC 5155 : if (!key->heapkeyspace)
1481 pg 2874 UIC 0 : return cmp >= 0;
1481 pg 2875 ECB :
1481 pg 2876 EUB : /*
2877 : * No need to consider the possibility that scankey has attributes that we
2878 : * need to force to be interpreted as negative infinity. _bt_compare() is
2879 : * able to determine that scankey is greater than negative infinity. The
2880 : * distinction between "==" and "<" isn't interesting here, since
2881 : * corruption is indicated either way.
2882 : */
1481 pg 2883 GIC 5155 : return cmp > 0;
2884 : }
2222 andres 2885 ECB :
2886 : /*
2887 : * Does the invariant hold that the key is strictly less than a given upper
2888 : * bound offset item, with the offset relating to a caller-supplied page that
2889 : * is not the current target page?
2890 : *
2891 : * Caller's non-target page is a child page of the target, checked as part of
2892 : * checking a property of the target page (i.e. the key comes from the
2893 : * target). Verifies line pointer on behalf of caller.
2894 : *
2895 : * If this function returns false, convention is that caller throws error due
2896 : * to corruption.
2897 : */
2898 : static inline bool
1481 pg 2899 GIC 498218 : invariant_l_nontarget_offset(BtreeCheckState *state, BTScanInsert key,
2900 : BlockNumber nontargetblock, Page nontarget,
1445 pg 2901 ECB : OffsetNumber upperbound)
2902 : {
2903 : ItemId itemid;
2904 : int32 cmp;
2905 :
1481 pg 2906 GIC 498218 : Assert(key->pivotsearch);
2907 :
1445 pg 2908 ECB : /* Verify line pointer before checking tuple */
1445 pg 2909 GIC 498218 : itemid = PageGetItemIdCareful(state, nontargetblock, nontarget,
2910 : upperbound);
1481 pg 2911 CBC 498218 : cmp = _bt_compare(state->rel, key, nontarget, upperbound);
2912 :
1481 pg 2913 ECB : /* pg_upgrade'd indexes may legally have equal sibling tuples */
1481 pg 2914 GIC 498218 : if (!key->heapkeyspace)
1481 pg 2915 UIC 0 : return cmp <= 0;
1481 pg 2916 ECB :
1481 pg 2917 EUB : /* See invariant_l_offset() for an explanation of this extra step */
1481 pg 2918 GIC 498218 : if (cmp == 0)
2919 : {
1481 pg 2920 ECB : IndexTuple child;
2921 : int uppnkeyatts;
2922 : ItemPointer childheaptid;
2923 : BTPageOpaque copaque;
2924 : bool nonpivot;
2925 :
1481 pg 2926 GIC 1584 : child = (IndexTuple) PageGetItem(nontarget, itemid);
373 michael 2927 1584 : copaque = BTPageGetOpaque(nontarget);
1481 pg 2928 CBC 1584 : nonpivot = P_ISLEAF(copaque) && upperbound >= P_FIRSTDATAKEY(copaque);
1481 pg 2929 ECB :
2930 : /* Get number of keys + heap TID for child/non-target item */
1481 pg 2931 GIC 1584 : uppnkeyatts = BTreeTupleGetNKeyAtts(child, state->rel);
2932 1584 : childheaptid = BTreeTupleGetHeapTIDCareful(state, child, nonpivot);
1481 pg 2933 ECB :
2934 : /* Heap TID is tiebreaker key attribute */
1481 pg 2935 GIC 1584 : if (key->keysz == uppnkeyatts)
2936 1584 : return key->scantid == NULL && childheaptid != NULL;
1481 pg 2937 ECB :
1481 pg 2938 LBC 0 : return key->keysz < uppnkeyatts;
2939 : }
1481 pg 2940 EUB :
1481 pg 2941 GIC 496634 : return cmp < 0;
2942 : }
2222 andres 2943 ECB :
2944 : /*
2945 : * Given a block number of a B-Tree page, return page in palloc()'d memory.
2946 : * While at it, perform some basic checks of the page.
2947 : *
2948 : * There is never an attempt to get a consistent view of multiple pages using
2949 : * multiple concurrent buffer locks; in general, we only acquire a single pin
2950 : * and buffer lock at a time, which is often all that the nbtree code requires.
2951 : * (Actually, bt_recheck_sibling_links couples buffer locks, which is the only
2952 : * exception to this general rule.)
2953 : *
2954 : * Operating on a copy of the page is useful because it prevents control
2955 : * getting stuck in an uninterruptible state when an underlying operator class
2956 : * misbehaves.
2957 : */
2958 : static Page
2222 andres 2959 GIC 16494 : palloc_btree_page(BtreeCheckState *state, BlockNumber blocknum)
2960 : {
2222 andres 2961 ECB : Buffer buffer;
2962 : Page page;
2963 : BTPageOpaque opaque;
2964 : OffsetNumber maxoffset;
2965 :
2222 andres 2966 GIC 16494 : page = palloc(BLCKSZ);
2967 :
2222 andres 2968 ECB : /*
2969 : * We copy the page into local storage to avoid holding pin on the buffer
2970 : * longer than we must.
2971 : */
2222 andres 2972 GIC 16494 : buffer = ReadBufferExtended(state->rel, MAIN_FORKNUM, blocknum, RBM_NORMAL,
2973 : state->checkstrategy);
2222 andres 2974 CBC 16484 : LockBuffer(buffer, BT_READ);
2975 :
2222 andres 2976 ECB : /*
2977 : * Perform the same basic sanity checking that nbtree itself performs for
2978 : * every page:
2979 : */
2222 andres 2980 GIC 16484 : _bt_checkpage(state->rel, buffer);
2981 :
2222 andres 2982 ECB : /* Only use copy of page in palloc()'d memory */
2222 andres 2983 GIC 16484 : memcpy(page, BufferGetPage(buffer), BLCKSZ);
2984 16484 : UnlockReleaseBuffer(buffer);
2222 andres 2985 ECB :
373 michael 2986 CBC 16484 : opaque = BTPageGetOpaque(page);
2987 :
2029 tgl 2988 16484 : if (P_ISMETA(opaque) && blocknum != BTREE_METAPAGE)
2222 andres 2989 UIC 0 : ereport(ERROR,
2222 andres 2990 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 2991 EUB : errmsg("invalid meta page found at block %u in index \"%s\"",
2992 : blocknum, RelationGetRelationName(state->rel))));
2993 :
2994 : /* Check page from block that ought to be meta page */
2222 andres 2995 GIC 16484 : if (blocknum == BTREE_METAPAGE)
2996 : {
2222 andres 2997 CBC 2862 : BTMetaPageData *metad = BTPageGetMeta(page);
2998 :
2029 tgl 2999 2862 : if (!P_ISMETA(opaque) ||
2222 andres 3000 GIC 2862 : metad->btm_magic != BTREE_MAGIC)
2222 andres 3001 LBC 0 : ereport(ERROR,
2222 andres 3002 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
2222 andres 3003 EUB : errmsg("index \"%s\" meta page is corrupt",
3004 : RelationGetRelationName(state->rel))));
3005 :
1831 teodor 3006 GIC 2862 : if (metad->btm_version < BTREE_MIN_VERSION ||
3007 2862 : metad->btm_version > BTREE_VERSION)
2222 andres 3008 LBC 0 : ereport(ERROR,
2222 andres 3009 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1831 teodor 3010 EUB : errmsg("version mismatch in index \"%s\": file version %d, "
3011 : "current version %d, minimum supported version %d",
3012 : RelationGetRelationName(state->rel),
3013 : metad->btm_version, BTREE_VERSION,
3014 : BTREE_MIN_VERSION)));
3015 :
3016 : /* Finished with metapage checks */
1810 teodor 3017 GIC 2862 : return page;
3018 : }
2222 andres 3019 ECB :
3020 : /*
3021 : * Deleted pages that still use the old 32-bit XID representation have no
3022 : * sane "level" field because they type pun the field, but all other pages
3023 : * (including pages deleted on Postgres 14+) have a valid value.
3024 : */
774 pg 3025 GIC 13622 : if (!P_ISDELETED(opaque) || P_HAS_FULLXID(opaque))
3026 : {
774 pg 3027 ECB : /* Okay, no reason not to trust btpo_level field from page */
3028 :
774 pg 3029 GIC 13622 : if (P_ISLEAF(opaque) && opaque->btpo_level != 0)
774 pg 3030 UIC 0 : ereport(ERROR,
774 pg 3031 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
774 pg 3032 EUB : errmsg_internal("invalid leaf page level %u for block %u in index \"%s\"",
3033 : opaque->btpo_level, blocknum,
3034 : RelationGetRelationName(state->rel))));
3035 :
774 pg 3036 GIC 13622 : if (!P_ISLEAF(opaque) && opaque->btpo_level == 0)
774 pg 3037 UIC 0 : ereport(ERROR,
774 pg 3038 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
774 pg 3039 EUB : errmsg_internal("invalid internal page level 0 for block %u in index \"%s\"",
3040 : blocknum,
3041 : RelationGetRelationName(state->rel))));
3042 : }
3043 :
3044 : /*
3045 : * Sanity checks for number of items on page.
3046 : *
3047 : * As noted at the beginning of _bt_binsrch(), an internal page must have
3048 : * children, since there must always be a negative infinity downlink
3049 : * (there may also be a highkey). In the case of non-rightmost leaf
3050 : * pages, there must be at least a highkey. The exceptions are deleted
3051 : * pages, which contain no items.
3052 : *
3053 : * This is correct when pages are half-dead, since internal pages are
3054 : * never half-dead, and leaf pages must have a high key when half-dead
3055 : * (the rightmost page can never be deleted). It's also correct with
3056 : * fully deleted pages: _bt_unlink_halfdead_page() doesn't change anything
3057 : * about the target page other than setting the page as fully dead, and
3058 : * setting its xact field. In particular, it doesn't change the sibling
3059 : * links in the deletion target itself, since they're required when index
3060 : * scans land on the deletion target, and then need to move right (or need
3061 : * to move left, in the case of backward index scans).
3062 : */
1810 teodor 3063 GIC 13622 : maxoffset = PageGetMaxOffsetNumber(page);
3064 13622 : if (maxoffset > MaxIndexTuplesPerPage)
1810 teodor 3065 LBC 0 : ereport(ERROR,
1810 teodor 3066 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1810 teodor 3067 EUB : errmsg("Number of items on block %u of index \"%s\" exceeds MaxIndexTuplesPerPage (%u)",
3068 : blocknum, RelationGetRelationName(state->rel),
3069 : MaxIndexTuplesPerPage)));
3070 :
1060 akorotkov 3071 GIC 13622 : if (!P_ISLEAF(opaque) && !P_ISDELETED(opaque) && maxoffset < P_FIRSTDATAKEY(opaque))
1810 teodor 3072 UIC 0 : ereport(ERROR,
1810 teodor 3073 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1810 teodor 3074 EUB : errmsg("internal block %u in index \"%s\" lacks high key and/or at least one downlink",
3075 : blocknum, RelationGetRelationName(state->rel))));
3076 :
1060 akorotkov 3077 GIC 13622 : if (P_ISLEAF(opaque) && !P_ISDELETED(opaque) && !P_RIGHTMOST(opaque) && maxoffset < P_HIKEY)
1810 teodor 3078 UIC 0 : ereport(ERROR,
1810 teodor 3079 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1810 teodor 3080 EUB : errmsg("non-rightmost leaf block %u in index \"%s\" lacks high key item",
3081 : blocknum, RelationGetRelationName(state->rel))));
3082 :
3083 : /*
3084 : * In general, internal pages are never marked half-dead, except on
3085 : * versions of Postgres prior to 9.4, where it can be valid transient
3086 : * state. This state is nonetheless treated as corruption by VACUUM on
3087 : * from version 9.4 on, so do the same here. See _bt_pagedel() for full
3088 : * details.
3089 : */
1810 teodor 3090 GIC 13622 : if (!P_ISLEAF(opaque) && P_ISHALFDEAD(opaque))
1810 teodor 3091 UIC 0 : ereport(ERROR,
1810 teodor 3092 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1810 teodor 3093 EUB : errmsg("internal page block %u in index \"%s\" is half-dead",
3094 : blocknum, RelationGetRelationName(state->rel)),
3095 : errhint("This can be caused by an interrupted VACUUM in version 9.3 or older, before upgrade. Please REINDEX it.")));
3096 :
3097 : /*
3098 : * Check that internal pages have no garbage items, and that no page has
3099 : * an invalid combination of deletion-related page level flags
3100 : */
2222 andres 3101 GIC 13622 : if (!P_ISLEAF(opaque) && P_HAS_GARBAGE(opaque))
2222 andres 3102 UIC 0 : ereport(ERROR,
2222 andres 3103 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
774 pg 3104 EUB : errmsg_internal("internal page block %u in index \"%s\" has garbage items",
3105 : blocknum, RelationGetRelationName(state->rel))));
3106 :
774 pg 3107 GIC 13622 : if (P_HAS_FULLXID(opaque) && !P_ISDELETED(opaque))
774 pg 3108 UIC 0 : ereport(ERROR,
774 pg 3109 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
774 pg 3110 EUB : errmsg_internal("full transaction id page flag appears in non-deleted block %u in index \"%s\"",
3111 : blocknum, RelationGetRelationName(state->rel))));
3112 :
774 pg 3113 GIC 13622 : if (P_ISDELETED(opaque) && P_ISHALFDEAD(opaque))
774 pg 3114 UIC 0 : ereport(ERROR,
774 pg 3115 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
774 pg 3116 EUB : errmsg_internal("deleted page block %u in index \"%s\" is half-dead",
3117 : blocknum, RelationGetRelationName(state->rel))));
3118 :
2222 andres 3119 GIC 13622 : return page;
3120 : }
1481 pg 3121 ECB :
3122 : /*
3123 : * _bt_mkscankey() wrapper that automatically prevents insertion scankey from
3124 : * being considered greater than the pivot tuple that its values originated
3125 : * from (or some other identical pivot tuple) in the common case where there
3126 : * are truncated/minus infinity attributes. Without this extra step, there
3127 : * are forms of corruption that amcheck could theoretically fail to report.
3128 : *
3129 : * For example, invariant_g_offset() might miss a cross-page invariant failure
3130 : * on an internal level if the scankey built from the first item on the
3131 : * target's right sibling page happened to be equal to (not greater than) the
3132 : * last item on target page. The !pivotsearch tiebreaker in _bt_compare()
3133 : * might otherwise cause amcheck to assume (rather than actually verify) that
3134 : * the scankey is greater.
3135 : */
3136 : static inline BTScanInsert
8 andres 3137 GNC 1580447 : bt_mkscankey_pivotsearch(Relation rel, Relation heaprel, IndexTuple itup)
3138 : {
1481 pg 3139 ECB : BTScanInsert skey;
3140 :
8 andres 3141 GNC 1580447 : skey = _bt_mkscankey(rel, heaprel, itup);
1481 pg 3142 GIC 1580447 : skey->pivotsearch = true;
1481 pg 3143 ECB :
1481 pg 3144 CBC 1580447 : return skey;
3145 : }
1481 pg 3146 ECB :
3147 : /*
3148 : * PageGetItemId() wrapper that validates returned line pointer.
3149 : *
3150 : * Buffer page/page item access macros generally trust that line pointers are
3151 : * not corrupt, which might cause problems for verification itself. For
3152 : * example, there is no bounds checking in PageGetItem(). Passing it a
3153 : * corrupt line pointer can cause it to return a tuple/pointer that is unsafe
3154 : * to dereference.
3155 : *
3156 : * Validating line pointers before tuples avoids undefined behavior and
3157 : * assertion failures with corrupt indexes, making the verification process
3158 : * more robust and predictable.
3159 : */
3160 : static ItemId
1445 pg 3161 GIC 3661462 : PageGetItemIdCareful(BtreeCheckState *state, BlockNumber block, Page page,
3162 : OffsetNumber offset)
1445 pg 3163 ECB : {
1445 pg 3164 GIC 3661462 : ItemId itemid = PageGetItemId(page, offset);
3165 :
1445 pg 3166 CBC 3661462 : if (ItemIdGetOffset(itemid) + ItemIdGetLength(itemid) >
3167 : BLCKSZ - MAXALIGN(sizeof(BTPageOpaqueData)))
1445 pg 3168 LBC 0 : ereport(ERROR,
3169 : (errcode(ERRCODE_INDEX_CORRUPTED),
1445 pg 3170 EUB : errmsg("line pointer points past end of tuple space in index \"%s\"",
3171 : RelationGetRelationName(state->rel)),
3172 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3173 : block, offset, ItemIdGetOffset(itemid),
3174 : ItemIdGetLength(itemid),
3175 : ItemIdGetFlags(itemid))));
3176 :
3177 : /*
3178 : * Verify that line pointer isn't LP_REDIRECT or LP_UNUSED, since nbtree
3179 : * never uses either. Verify that line pointer has storage, too, since
3180 : * even LP_DEAD items should within nbtree.
3181 : */
1445 pg 3182 GIC 3661462 : if (ItemIdIsRedirected(itemid) || !ItemIdIsUsed(itemid) ||
3183 3661462 : ItemIdGetLength(itemid) == 0)
1445 pg 3184 LBC 0 : ereport(ERROR,
1445 pg 3185 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1445 pg 3186 EUB : errmsg("invalid line pointer storage in index \"%s\"",
3187 : RelationGetRelationName(state->rel)),
3188 : errdetail_internal("Index tid=(%u,%u) lp_off=%u, lp_len=%u lp_flags=%u.",
3189 : block, offset, ItemIdGetOffset(itemid),
3190 : ItemIdGetLength(itemid),
3191 : ItemIdGetFlags(itemid))));
3192 :
1445 pg 3193 GIC 3661462 : return itemid;
3194 : }
1445 pg 3195 ECB :
3196 : /*
3197 : * BTreeTupleGetHeapTID() wrapper that enforces that a heap TID is present in
3198 : * cases where that is mandatory (i.e. for non-pivot tuples)
3199 : */
3200 : static inline ItemPointer
1481 pg 3201 GIC 1584 : BTreeTupleGetHeapTIDCareful(BtreeCheckState *state, IndexTuple itup,
3202 : bool nonpivot)
1481 pg 3203 ECB : {
3204 : ItemPointer htid;
3205 :
3206 : /*
3207 : * Caller determines whether this is supposed to be a pivot or non-pivot
3208 : * tuple using page type and item offset number. Verify that tuple
3209 : * metadata agrees with this.
3210 : */
1138 pg 3211 GIC 1584 : Assert(state->heapkeyspace);
3212 1584 : if (BTreeTupleIsPivot(itup) && nonpivot)
1138 pg 3213 LBC 0 : ereport(ERROR,
1138 pg 3214 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1138 pg 3215 EUB : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected pivot tuple",
3216 : state->targetblock,
3217 : RelationGetRelationName(state->rel))));
3218 :
1138 pg 3219 GIC 1584 : if (!BTreeTupleIsPivot(itup) && !nonpivot)
1138 pg 3220 UIC 0 : ereport(ERROR,
1138 pg 3221 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1138 pg 3222 EUB : errmsg_internal("block %u or its right sibling block or child block in index \"%s\" has unexpected non-pivot tuple",
3223 : state->targetblock,
3224 : RelationGetRelationName(state->rel))));
3225 :
1138 pg 3226 GIC 1584 : htid = BTreeTupleGetHeapTID(itup);
3227 1584 : if (!ItemPointerIsValid(htid) && nonpivot)
1481 pg 3228 LBC 0 : ereport(ERROR,
1481 pg 3229 ECB : (errcode(ERRCODE_INDEX_CORRUPTED),
1481 pg 3230 EUB : errmsg("block %u or its right sibling block or child block in index \"%s\" contains non-pivot tuple that lacks a heap TID",
3231 : state->targetblock,
3232 : RelationGetRelationName(state->rel))));
3233 :
1138 pg 3234 GIC 1584 : return htid;
3235 : }
1138 pg 3236 ECB :
3237 : /*
3238 : * Return the "pointed to" TID for itup, which is used to generate a
3239 : * descriptive error message. itup must be a "data item" tuple (it wouldn't
3240 : * make much sense to call here with a high key tuple, since there won't be a
3241 : * valid downlink/block number to display).
3242 : *
3243 : * Returns either a heap TID (which will be the first heap TID in posting list
3244 : * if itup is posting list tuple), or a TID that contains downlink block
3245 : * number, plus some encoded metadata (e.g., the number of attributes present
3246 : * in itup).
3247 : */
3248 : static inline ItemPointer
1138 pg 3249 UIC 0 : BTreeTupleGetPointsToTID(IndexTuple itup)
3250 : {
1138 pg 3251 EUB : /*
3252 : * Rely on the assumption that !heapkeyspace internal page data items will
3253 : * correctly return TID with downlink here -- BTreeTupleGetHeapTID() won't
3254 : * recognize it as a pivot tuple, but everything still works out because
3255 : * the t_tid field is still returned
3256 : */
1138 pg 3257 UIC 0 : if (!BTreeTupleIsPivot(itup))
3258 0 : return BTreeTupleGetHeapTID(itup);
1481 pg 3259 EUB :
1138 3260 : /* Pivot tuple returns TID with downlink block (heapkeyspace variant) */
1138 pg 3261 UIC 0 : return &itup->t_tid;
3262 : }
|
