Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * verify_heapam.c
4 : * Functions to check postgresql heap relations for corruption
5 : *
6 : * Copyright (c) 2016-2024, PostgreSQL Global Development Group
7 : *
8 : * contrib/amcheck/verify_heapam.c
9 : *-------------------------------------------------------------------------
10 : */
11 : #include "postgres.h"
12 :
13 : #include "access/detoast.h"
14 : #include "access/genam.h"
15 : #include "access/heapam.h"
16 : #include "access/heaptoast.h"
17 : #include "access/multixact.h"
18 : #include "access/toast_internals.h"
19 : #include "access/visibilitymap.h"
20 : #include "catalog/pg_am.h"
21 : #include "funcapi.h"
22 : #include "miscadmin.h"
23 : #include "storage/bufmgr.h"
24 : #include "storage/procarray.h"
25 : #include "utils/builtins.h"
26 : #include "utils/fmgroids.h"
27 :
28 604 : PG_FUNCTION_INFO_V1(verify_heapam);
29 :
30 : /* The number of columns in tuples returned by verify_heapam */
31 : #define HEAPCHECK_RELATION_COLS 4
32 :
33 : /* The largest valid toast va_rawsize */
34 : #define VARLENA_SIZE_LIMIT 0x3FFFFFFF
35 :
36 : /*
37 : * Despite the name, we use this for reporting problems with both XIDs and
38 : * MXIDs.
39 : */
40 : typedef enum XidBoundsViolation
41 : {
42 : XID_INVALID,
43 : XID_IN_FUTURE,
44 : XID_PRECEDES_CLUSTERMIN,
45 : XID_PRECEDES_RELMIN,
46 : XID_BOUNDS_OK,
47 : } XidBoundsViolation;
48 :
49 : typedef enum XidCommitStatus
50 : {
51 : XID_COMMITTED,
52 : XID_IS_CURRENT_XID,
53 : XID_IN_PROGRESS,
54 : XID_ABORTED,
55 : } XidCommitStatus;
56 :
57 : typedef enum SkipPages
58 : {
59 : SKIP_PAGES_ALL_FROZEN,
60 : SKIP_PAGES_ALL_VISIBLE,
61 : SKIP_PAGES_NONE,
62 : } SkipPages;
63 :
64 : /*
65 : * Struct holding information about a toasted attribute sufficient to both
66 : * check the toasted attribute and, if found to be corrupt, to report where it
67 : * was encountered in the main table.
68 : */
69 : typedef struct ToastedAttribute
70 : {
71 : struct varatt_external toast_pointer;
72 : BlockNumber blkno; /* block in main table */
73 : OffsetNumber offnum; /* offset in main table */
74 : AttrNumber attnum; /* attribute in main table */
75 : } ToastedAttribute;
76 :
77 : /*
78 : * Struct holding the running context information during
79 : * a lifetime of a verify_heapam execution.
80 : */
81 : typedef struct HeapCheckContext
82 : {
83 : /*
84 : * Cached copies of values from TransamVariables and computed values from
85 : * them.
86 : */
87 : FullTransactionId next_fxid; /* TransamVariables->nextXid */
88 : TransactionId next_xid; /* 32-bit version of next_fxid */
89 : TransactionId oldest_xid; /* TransamVariables->oldestXid */
90 : FullTransactionId oldest_fxid; /* 64-bit version of oldest_xid, computed
91 : * relative to next_fxid */
92 : TransactionId safe_xmin; /* this XID and newer ones can't become
93 : * all-visible while we're running */
94 :
95 : /*
96 : * Cached copy of value from MultiXactState
97 : */
98 : MultiXactId next_mxact; /* MultiXactState->nextMXact */
99 : MultiXactId oldest_mxact; /* MultiXactState->oldestMultiXactId */
100 :
101 : /*
102 : * Cached copies of the most recently checked xid and its status.
103 : */
104 : TransactionId cached_xid;
105 : XidCommitStatus cached_status;
106 :
107 : /* Values concerning the heap relation being checked */
108 : Relation rel;
109 : TransactionId relfrozenxid;
110 : FullTransactionId relfrozenfxid;
111 : TransactionId relminmxid;
112 : Relation toast_rel;
113 : Relation *toast_indexes;
114 : Relation valid_toast_index;
115 : int num_toast_indexes;
116 :
117 : /* Values for iterating over pages in the relation */
118 : BlockNumber blkno;
119 : BufferAccessStrategy bstrategy;
120 : Buffer buffer;
121 : Page page;
122 :
123 : /* Values for iterating over tuples within a page */
124 : OffsetNumber offnum;
125 : ItemId itemid;
126 : uint16 lp_len;
127 : uint16 lp_off;
128 : HeapTupleHeader tuphdr;
129 : int natts;
130 :
131 : /* Values for iterating over attributes within the tuple */
132 : uint32 offset; /* offset in tuple data */
133 : AttrNumber attnum;
134 :
135 : /* True if tuple's xmax makes it eligible for pruning */
136 : bool tuple_could_be_pruned;
137 :
138 : /*
139 : * List of ToastedAttribute structs for toasted attributes which are not
140 : * eligible for pruning and should be checked
141 : */
142 : List *toasted_attributes;
143 :
144 : /* Whether verify_heapam has yet encountered any corrupt tuples */
145 : bool is_corrupt;
146 :
147 : /* The descriptor and tuplestore for verify_heapam's result tuples */
148 : TupleDesc tupdesc;
149 : Tuplestorestate *tupstore;
150 : } HeapCheckContext;
151 :
152 : /* Internal implementation */
153 : static void check_tuple(HeapCheckContext *ctx,
154 : bool *xmin_commit_status_ok,
155 : XidCommitStatus *xmin_commit_status);
156 : static void check_toast_tuple(HeapTuple toasttup, HeapCheckContext *ctx,
157 : ToastedAttribute *ta, int32 *expected_chunk_seq,
158 : uint32 extsize);
159 :
160 : static bool check_tuple_attribute(HeapCheckContext *ctx);
161 : static void check_toasted_attribute(HeapCheckContext *ctx,
162 : ToastedAttribute *ta);
163 :
164 : static bool check_tuple_header(HeapCheckContext *ctx);
165 : static bool check_tuple_visibility(HeapCheckContext *ctx,
166 : bool *xmin_commit_status_ok,
167 : XidCommitStatus *xmin_commit_status);
168 :
169 : static void report_corruption(HeapCheckContext *ctx, char *msg);
170 : static void report_toast_corruption(HeapCheckContext *ctx,
171 : ToastedAttribute *ta, char *msg);
172 : static FullTransactionId FullTransactionIdFromXidAndCtx(TransactionId xid,
173 : const HeapCheckContext *ctx);
174 : static void update_cached_xid_range(HeapCheckContext *ctx);
175 : static void update_cached_mxid_range(HeapCheckContext *ctx);
176 : static XidBoundsViolation check_mxid_in_range(MultiXactId mxid,
177 : HeapCheckContext *ctx);
178 : static XidBoundsViolation check_mxid_valid_in_rel(MultiXactId mxid,
179 : HeapCheckContext *ctx);
180 : static XidBoundsViolation get_xid_status(TransactionId xid,
181 : HeapCheckContext *ctx,
182 : XidCommitStatus *status);
183 :
184 : /*
185 : * Scan and report corruption in heap pages, optionally reconciling toasted
186 : * attributes with entries in the associated toast table. Intended to be
187 : * called from SQL with the following parameters:
188 : *
189 : * relation:
190 : * The Oid of the heap relation to be checked.
191 : *
192 : * on_error_stop:
193 : * Whether to stop at the end of the first page for which errors are
194 : * detected. Note that multiple rows may be returned.
195 : *
196 : * check_toast:
197 : * Whether to check each toasted attribute against the toast table to
198 : * verify that it can be found there.
199 : *
200 : * skip:
201 : * What kinds of pages in the heap relation should be skipped. Valid
202 : * options are "all-visible", "all-frozen", and "none".
203 : *
204 : * Returns to the SQL caller a set of tuples, each containing the location
205 : * and a description of a corruption found in the heap.
206 : *
207 : * This code goes to some trouble to avoid crashing the server even if the
208 : * table pages are badly corrupted, but it's probably not perfect. If
209 : * check_toast is true, we'll use regular index lookups to try to fetch TOAST
210 : * tuples, which can certainly cause crashes if the right kind of corruption
211 : * exists in the toast table or index. No matter what parameters you pass,
212 : * we can't protect against crashes that might occur trying to look up the
213 : * commit status of transaction IDs (though we avoid trying to do such lookups
214 : * for transaction IDs that can't legally appear in the table).
215 : */
216 : Datum
217 6882 : verify_heapam(PG_FUNCTION_ARGS)
218 : {
219 6882 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
220 : HeapCheckContext ctx;
221 6882 : Buffer vmbuffer = InvalidBuffer;
222 : Oid relid;
223 : bool on_error_stop;
224 : bool check_toast;
225 6882 : SkipPages skip_option = SKIP_PAGES_NONE;
226 : BlockNumber first_block;
227 : BlockNumber last_block;
228 : BlockNumber nblocks;
229 : const char *skip;
230 :
231 : /* Check supplied arguments */
232 6882 : if (PG_ARGISNULL(0))
233 0 : ereport(ERROR,
234 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
235 : errmsg("relation cannot be null")));
236 6882 : relid = PG_GETARG_OID(0);
237 :
238 6882 : if (PG_ARGISNULL(1))
239 0 : ereport(ERROR,
240 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
241 : errmsg("on_error_stop cannot be null")));
242 6882 : on_error_stop = PG_GETARG_BOOL(1);
243 :
244 6882 : if (PG_ARGISNULL(2))
245 0 : ereport(ERROR,
246 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
247 : errmsg("check_toast cannot be null")));
248 6882 : check_toast = PG_GETARG_BOOL(2);
249 :
250 6882 : if (PG_ARGISNULL(3))
251 0 : ereport(ERROR,
252 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
253 : errmsg("skip cannot be null")));
254 6882 : skip = text_to_cstring(PG_GETARG_TEXT_PP(3));
255 6882 : if (pg_strcasecmp(skip, "all-visible") == 0)
256 170 : skip_option = SKIP_PAGES_ALL_VISIBLE;
257 6712 : else if (pg_strcasecmp(skip, "all-frozen") == 0)
258 174 : skip_option = SKIP_PAGES_ALL_FROZEN;
259 6538 : else if (pg_strcasecmp(skip, "none") == 0)
260 6536 : skip_option = SKIP_PAGES_NONE;
261 : else
262 2 : ereport(ERROR,
263 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
264 : errmsg("invalid skip option"),
265 : errhint("Valid skip options are \"all-visible\", \"all-frozen\", and \"none\".")));
266 :
267 6880 : memset(&ctx, 0, sizeof(HeapCheckContext));
268 6880 : ctx.cached_xid = InvalidTransactionId;
269 6880 : ctx.toasted_attributes = NIL;
270 :
271 : /*
272 : * Any xmin newer than the xmin of our snapshot can't become all-visible
273 : * while we're running.
274 : */
275 6880 : ctx.safe_xmin = GetTransactionSnapshot()->xmin;
276 :
277 : /*
278 : * If we report corruption when not examining some individual attribute,
279 : * we need attnum to be reported as NULL. Set that up before any
280 : * corruption reporting might happen.
281 : */
282 6880 : ctx.attnum = -1;
283 :
284 : /* Construct the tuplestore and tuple descriptor */
285 6880 : InitMaterializedSRF(fcinfo, 0);
286 6880 : ctx.tupdesc = rsinfo->setDesc;
287 6880 : ctx.tupstore = rsinfo->setResult;
288 :
289 : /* Open relation, check relkind and access method */
290 6880 : ctx.rel = relation_open(relid, AccessShareLock);
291 :
292 : /*
293 : * Check that a relation's relkind and access method are both supported.
294 : */
295 6880 : if (!RELKIND_HAS_TABLE_AM(ctx.rel->rd_rel->relkind) &&
296 392 : ctx.rel->rd_rel->relkind != RELKIND_SEQUENCE)
297 8 : ereport(ERROR,
298 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
299 : errmsg("cannot check relation \"%s\"",
300 : RelationGetRelationName(ctx.rel)),
301 : errdetail_relkind_not_supported(ctx.rel->rd_rel->relkind)));
302 :
303 : /*
304 : * Sequences always use heap AM, but they don't show that in the catalogs.
305 : * Other relkinds might be using a different AM, so check.
306 : */
307 6872 : if (ctx.rel->rd_rel->relkind != RELKIND_SEQUENCE &&
308 6488 : ctx.rel->rd_rel->relam != HEAP_TABLE_AM_OID)
309 0 : ereport(ERROR,
310 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
311 : errmsg("only heap AM is supported")));
312 :
313 : /*
314 : * Early exit for unlogged relations during recovery. These will have no
315 : * relation fork, so there won't be anything to check. We behave as if
316 : * the relation is empty.
317 : */
318 6872 : if (ctx.rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
319 0 : RecoveryInProgress())
320 : {
321 0 : ereport(DEBUG1,
322 : (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
323 : errmsg("cannot verify unlogged relation \"%s\" during recovery, skipping",
324 : RelationGetRelationName(ctx.rel))));
325 0 : relation_close(ctx.rel, AccessShareLock);
326 0 : PG_RETURN_NULL();
327 : }
328 :
329 : /* Early exit if the relation is empty */
330 6872 : nblocks = RelationGetNumberOfBlocks(ctx.rel);
331 6838 : if (!nblocks)
332 : {
333 3882 : relation_close(ctx.rel, AccessShareLock);
334 3882 : PG_RETURN_NULL();
335 : }
336 :
337 2956 : ctx.bstrategy = GetAccessStrategy(BAS_BULKREAD);
338 2956 : ctx.buffer = InvalidBuffer;
339 2956 : ctx.page = NULL;
340 :
341 : /* Validate block numbers, or handle nulls. */
342 2956 : if (PG_ARGISNULL(4))
343 2708 : first_block = 0;
344 : else
345 : {
346 248 : int64 fb = PG_GETARG_INT64(4);
347 :
348 248 : if (fb < 0 || fb >= nblocks)
349 2 : ereport(ERROR,
350 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
351 : errmsg("starting block number must be between 0 and %u",
352 : nblocks - 1)));
353 246 : first_block = (BlockNumber) fb;
354 : }
355 2954 : if (PG_ARGISNULL(5))
356 2708 : last_block = nblocks - 1;
357 : else
358 : {
359 246 : int64 lb = PG_GETARG_INT64(5);
360 :
361 246 : if (lb < 0 || lb >= nblocks)
362 2 : ereport(ERROR,
363 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
364 : errmsg("ending block number must be between 0 and %u",
365 : nblocks - 1)));
366 244 : last_block = (BlockNumber) lb;
367 : }
368 :
369 : /* Optionally open the toast relation, if any. */
370 2952 : if (ctx.rel->rd_rel->reltoastrelid && check_toast)
371 1392 : {
372 : int offset;
373 :
374 : /* Main relation has associated toast relation */
375 1392 : ctx.toast_rel = table_open(ctx.rel->rd_rel->reltoastrelid,
376 : AccessShareLock);
377 1392 : offset = toast_open_indexes(ctx.toast_rel,
378 : AccessShareLock,
379 : &(ctx.toast_indexes),
380 : &(ctx.num_toast_indexes));
381 1392 : ctx.valid_toast_index = ctx.toast_indexes[offset];
382 : }
383 : else
384 : {
385 : /*
386 : * Main relation has no associated toast relation, or we're
387 : * intentionally skipping it.
388 : */
389 1560 : ctx.toast_rel = NULL;
390 1560 : ctx.toast_indexes = NULL;
391 1560 : ctx.num_toast_indexes = 0;
392 : }
393 :
394 2952 : update_cached_xid_range(&ctx);
395 2952 : update_cached_mxid_range(&ctx);
396 2952 : ctx.relfrozenxid = ctx.rel->rd_rel->relfrozenxid;
397 2952 : ctx.relfrozenfxid = FullTransactionIdFromXidAndCtx(ctx.relfrozenxid, &ctx);
398 2952 : ctx.relminmxid = ctx.rel->rd_rel->relminmxid;
399 :
400 2952 : if (TransactionIdIsNormal(ctx.relfrozenxid))
401 2568 : ctx.oldest_xid = ctx.relfrozenxid;
402 :
403 26394 : for (ctx.blkno = first_block; ctx.blkno <= last_block; ctx.blkno++)
404 : {
405 : OffsetNumber maxoff;
406 : OffsetNumber predecessor[MaxOffsetNumber];
407 : OffsetNumber successor[MaxOffsetNumber];
408 : bool lp_valid[MaxOffsetNumber];
409 : bool xmin_commit_status_ok[MaxOffsetNumber];
410 : XidCommitStatus xmin_commit_status[MaxOffsetNumber];
411 :
412 23448 : CHECK_FOR_INTERRUPTS();
413 :
414 23448 : memset(predecessor, 0, sizeof(OffsetNumber) * MaxOffsetNumber);
415 :
416 : /* Optionally skip over all-frozen or all-visible blocks */
417 23448 : if (skip_option != SKIP_PAGES_NONE)
418 : {
419 : int32 mapbits;
420 :
421 1472 : mapbits = (int32) visibilitymap_get_status(ctx.rel, ctx.blkno,
422 : &vmbuffer);
423 1472 : if (skip_option == SKIP_PAGES_ALL_FROZEN)
424 : {
425 768 : if ((mapbits & VISIBILITYMAP_ALL_FROZEN) != 0)
426 66 : continue;
427 : }
428 :
429 1408 : if (skip_option == SKIP_PAGES_ALL_VISIBLE)
430 : {
431 704 : if ((mapbits & VISIBILITYMAP_ALL_VISIBLE) != 0)
432 2 : continue;
433 : }
434 : }
435 :
436 : /* Read and lock the next page. */
437 23382 : ctx.buffer = ReadBufferExtended(ctx.rel, MAIN_FORKNUM, ctx.blkno,
438 : RBM_NORMAL, ctx.bstrategy);
439 23382 : LockBuffer(ctx.buffer, BUFFER_LOCK_SHARE);
440 23382 : ctx.page = BufferGetPage(ctx.buffer);
441 :
442 : /* Perform tuple checks */
443 23382 : maxoff = PageGetMaxOffsetNumber(ctx.page);
444 1111146 : for (ctx.offnum = FirstOffsetNumber; ctx.offnum <= maxoff;
445 1087764 : ctx.offnum = OffsetNumberNext(ctx.offnum))
446 : {
447 : BlockNumber nextblkno;
448 : OffsetNumber nextoffnum;
449 :
450 1087764 : successor[ctx.offnum] = InvalidOffsetNumber;
451 1087764 : lp_valid[ctx.offnum] = false;
452 1087764 : xmin_commit_status_ok[ctx.offnum] = false;
453 1087764 : ctx.itemid = PageGetItemId(ctx.page, ctx.offnum);
454 :
455 : /* Skip over unused/dead line pointers */
456 1087764 : if (!ItemIdIsUsed(ctx.itemid) || ItemIdIsDead(ctx.itemid))
457 10490 : continue;
458 :
459 : /*
460 : * If this line pointer has been redirected, check that it
461 : * redirects to a valid offset within the line pointer array
462 : */
463 1077274 : if (ItemIdIsRedirected(ctx.itemid))
464 : {
465 11288 : OffsetNumber rdoffnum = ItemIdGetRedirect(ctx.itemid);
466 : ItemId rditem;
467 :
468 11288 : if (rdoffnum < FirstOffsetNumber)
469 : {
470 12 : report_corruption(&ctx,
471 : psprintf("line pointer redirection to item at offset %u precedes minimum offset %u",
472 : (unsigned) rdoffnum,
473 : (unsigned) FirstOffsetNumber));
474 12 : continue;
475 : }
476 11276 : if (rdoffnum > maxoff)
477 : {
478 28 : report_corruption(&ctx,
479 : psprintf("line pointer redirection to item at offset %u exceeds maximum offset %u",
480 : (unsigned) rdoffnum,
481 : (unsigned) maxoff));
482 28 : continue;
483 : }
484 :
485 : /*
486 : * Since we've checked that this redirect points to a line
487 : * pointer between FirstOffsetNumber and maxoff, it should now
488 : * be safe to fetch the referenced line pointer. We expect it
489 : * to be LP_NORMAL; if not, that's corruption.
490 : */
491 11248 : rditem = PageGetItemId(ctx.page, rdoffnum);
492 11248 : if (!ItemIdIsUsed(rditem))
493 : {
494 0 : report_corruption(&ctx,
495 : psprintf("redirected line pointer points to an unused item at offset %u",
496 : (unsigned) rdoffnum));
497 0 : continue;
498 : }
499 11248 : else if (ItemIdIsDead(rditem))
500 : {
501 0 : report_corruption(&ctx,
502 : psprintf("redirected line pointer points to a dead item at offset %u",
503 : (unsigned) rdoffnum));
504 0 : continue;
505 : }
506 11248 : else if (ItemIdIsRedirected(rditem))
507 : {
508 2 : report_corruption(&ctx,
509 : psprintf("redirected line pointer points to another redirected line pointer at offset %u",
510 : (unsigned) rdoffnum));
511 2 : continue;
512 : }
513 :
514 : /*
515 : * Record the fact that this line pointer has passed basic
516 : * sanity checking, and also the offset number to which it
517 : * points.
518 : */
519 11246 : lp_valid[ctx.offnum] = true;
520 11246 : successor[ctx.offnum] = rdoffnum;
521 11246 : continue;
522 : }
523 :
524 : /* Sanity-check the line pointer's offset and length values */
525 1065986 : ctx.lp_len = ItemIdGetLength(ctx.itemid);
526 1065986 : ctx.lp_off = ItemIdGetOffset(ctx.itemid);
527 :
528 1065986 : if (ctx.lp_off != MAXALIGN(ctx.lp_off))
529 : {
530 12 : report_corruption(&ctx,
531 : psprintf("line pointer to page offset %u is not maximally aligned",
532 12 : ctx.lp_off));
533 12 : continue;
534 : }
535 1065974 : if (ctx.lp_len < MAXALIGN(SizeofHeapTupleHeader))
536 : {
537 24 : report_corruption(&ctx,
538 : psprintf("line pointer length %u is less than the minimum tuple header size %u",
539 24 : ctx.lp_len,
540 : (unsigned) MAXALIGN(SizeofHeapTupleHeader)));
541 24 : continue;
542 : }
543 1065950 : if (ctx.lp_off + ctx.lp_len > BLCKSZ)
544 : {
545 28 : report_corruption(&ctx,
546 : psprintf("line pointer to page offset %u with length %u ends beyond maximum page offset %u",
547 28 : ctx.lp_off,
548 28 : ctx.lp_len,
549 : (unsigned) BLCKSZ));
550 28 : continue;
551 : }
552 :
553 : /* It should be safe to examine the tuple's header, at least */
554 1065922 : lp_valid[ctx.offnum] = true;
555 1065922 : ctx.tuphdr = (HeapTupleHeader) PageGetItem(ctx.page, ctx.itemid);
556 1065922 : ctx.natts = HeapTupleHeaderGetNatts(ctx.tuphdr);
557 :
558 : /* Ok, ready to check this next tuple */
559 1065922 : check_tuple(&ctx,
560 1065922 : &xmin_commit_status_ok[ctx.offnum],
561 1065922 : &xmin_commit_status[ctx.offnum]);
562 :
563 : /*
564 : * If the CTID field of this tuple seems to point to another tuple
565 : * on the same page, record that tuple as the successor of this
566 : * one.
567 : */
568 1065922 : nextblkno = ItemPointerGetBlockNumber(&(ctx.tuphdr)->t_ctid);
569 1065922 : nextoffnum = ItemPointerGetOffsetNumber(&(ctx.tuphdr)->t_ctid);
570 1065922 : if (nextblkno == ctx.blkno && nextoffnum != ctx.offnum &&
571 342 : nextoffnum >= FirstOffsetNumber && nextoffnum <= maxoff)
572 342 : successor[ctx.offnum] = nextoffnum;
573 : }
574 :
575 : /*
576 : * Update chain validation. Check each line pointer that's got a valid
577 : * successor against that successor.
578 : */
579 23382 : ctx.attnum = -1;
580 1111146 : for (ctx.offnum = FirstOffsetNumber; ctx.offnum <= maxoff;
581 1087764 : ctx.offnum = OffsetNumberNext(ctx.offnum))
582 : {
583 : ItemId curr_lp;
584 : ItemId next_lp;
585 : HeapTupleHeader curr_htup;
586 : HeapTupleHeader next_htup;
587 : TransactionId curr_xmin;
588 : TransactionId curr_xmax;
589 : TransactionId next_xmin;
590 1087764 : OffsetNumber nextoffnum = successor[ctx.offnum];
591 :
592 : /*
593 : * The current line pointer may not have a successor, either
594 : * because it's not valid or because it didn't point to anything.
595 : * In either case, we have to give up.
596 : *
597 : * If the current line pointer does point to something, it's
598 : * possible that the target line pointer isn't valid. We have to
599 : * give up in that case, too.
600 : */
601 1087764 : if (nextoffnum == InvalidOffsetNumber || !lp_valid[nextoffnum])
602 1076176 : continue;
603 :
604 : /* We have two valid line pointers that we can examine. */
605 11588 : curr_lp = PageGetItemId(ctx.page, ctx.offnum);
606 11588 : next_lp = PageGetItemId(ctx.page, nextoffnum);
607 :
608 : /* Handle the cases where the current line pointer is a redirect. */
609 11588 : if (ItemIdIsRedirected(curr_lp))
610 : {
611 : /*
612 : * We should not have set successor[ctx.offnum] to a value
613 : * other than InvalidOffsetNumber unless that line pointer is
614 : * LP_NORMAL.
615 : */
616 : Assert(ItemIdIsNormal(next_lp));
617 :
618 : /* Can only redirect to a HOT tuple. */
619 11246 : next_htup = (HeapTupleHeader) PageGetItem(ctx.page, next_lp);
620 11246 : if (!HeapTupleHeaderIsHeapOnly(next_htup))
621 : {
622 2 : report_corruption(&ctx,
623 : psprintf("redirected line pointer points to a non-heap-only tuple at offset %u",
624 : (unsigned) nextoffnum));
625 : }
626 :
627 : /* HOT chains should not intersect. */
628 11246 : if (predecessor[nextoffnum] != InvalidOffsetNumber)
629 : {
630 2 : report_corruption(&ctx,
631 : psprintf("redirect line pointer points to offset %u, but offset %u also points there",
632 2 : (unsigned) nextoffnum, (unsigned) predecessor[nextoffnum]));
633 2 : continue;
634 : }
635 :
636 : /*
637 : * This redirect and the tuple to which it points seem to be
638 : * part of an update chain.
639 : */
640 11244 : predecessor[nextoffnum] = ctx.offnum;
641 11244 : continue;
642 : }
643 :
644 : /*
645 : * If the next line pointer is a redirect, or if it's a tuple but
646 : * the XMAX of this tuple doesn't match the XMIN of the next
647 : * tuple, then the two aren't part of the same update chain and
648 : * there is nothing more to do.
649 : */
650 342 : if (ItemIdIsRedirected(next_lp))
651 0 : continue;
652 342 : curr_htup = (HeapTupleHeader) PageGetItem(ctx.page, curr_lp);
653 342 : curr_xmax = HeapTupleHeaderGetUpdateXid(curr_htup);
654 342 : next_htup = (HeapTupleHeader) PageGetItem(ctx.page, next_lp);
655 342 : next_xmin = HeapTupleHeaderGetXmin(next_htup);
656 342 : if (!TransactionIdIsValid(curr_xmax) ||
657 : !TransactionIdEquals(curr_xmax, next_xmin))
658 8 : continue;
659 :
660 : /* HOT chains should not intersect. */
661 334 : if (predecessor[nextoffnum] != InvalidOffsetNumber)
662 : {
663 2 : report_corruption(&ctx,
664 : psprintf("tuple points to new version at offset %u, but offset %u also points there",
665 2 : (unsigned) nextoffnum, (unsigned) predecessor[nextoffnum]));
666 2 : continue;
667 : }
668 :
669 : /*
670 : * This tuple and the tuple to which it points seem to be part of
671 : * an update chain.
672 : */
673 332 : predecessor[nextoffnum] = ctx.offnum;
674 :
675 : /*
676 : * If the current tuple is marked as HOT-updated, then the next
677 : * tuple should be marked as a heap-only tuple. Conversely, if the
678 : * current tuple isn't marked as HOT-updated, then the next tuple
679 : * shouldn't be marked as a heap-only tuple.
680 : *
681 : * NB: Can't use HeapTupleHeaderIsHotUpdated() as it checks if
682 : * hint bits indicate xmin/xmax aborted.
683 : */
684 332 : if (!(curr_htup->t_infomask2 & HEAP_HOT_UPDATED) &&
685 2 : HeapTupleHeaderIsHeapOnly(next_htup))
686 : {
687 2 : report_corruption(&ctx,
688 : psprintf("non-heap-only update produced a heap-only tuple at offset %u",
689 : (unsigned) nextoffnum));
690 : }
691 332 : if ((curr_htup->t_infomask2 & HEAP_HOT_UPDATED) &&
692 330 : !HeapTupleHeaderIsHeapOnly(next_htup))
693 : {
694 2 : report_corruption(&ctx,
695 : psprintf("heap-only update produced a non-heap only tuple at offset %u",
696 : (unsigned) nextoffnum));
697 : }
698 :
699 : /*
700 : * If the current tuple's xmin is still in progress but the
701 : * successor tuple's xmin is committed, that's corruption.
702 : *
703 : * NB: We recheck the commit status of the current tuple's xmin
704 : * here, because it might have committed after we checked it and
705 : * before we checked the commit status of the successor tuple's
706 : * xmin. This should be safe because the xmin itself can't have
707 : * changed, only its commit status.
708 : */
709 332 : curr_xmin = HeapTupleHeaderGetXmin(curr_htup);
710 332 : if (xmin_commit_status_ok[ctx.offnum] &&
711 332 : xmin_commit_status[ctx.offnum] == XID_IN_PROGRESS &&
712 2 : xmin_commit_status_ok[nextoffnum] &&
713 4 : xmin_commit_status[nextoffnum] == XID_COMMITTED &&
714 2 : TransactionIdIsInProgress(curr_xmin))
715 : {
716 2 : report_corruption(&ctx,
717 : psprintf("tuple with in-progress xmin %u was updated to produce a tuple at offset %u with committed xmin %u",
718 : (unsigned) curr_xmin,
719 2 : (unsigned) ctx.offnum,
720 : (unsigned) next_xmin));
721 : }
722 :
723 : /*
724 : * If the current tuple's xmin is aborted but the successor
725 : * tuple's xmin is in-progress or committed, that's corruption.
726 : */
727 332 : if (xmin_commit_status_ok[ctx.offnum] &&
728 332 : xmin_commit_status[ctx.offnum] == XID_ABORTED &&
729 4 : xmin_commit_status_ok[nextoffnum])
730 : {
731 4 : if (xmin_commit_status[nextoffnum] == XID_IN_PROGRESS)
732 2 : report_corruption(&ctx,
733 : psprintf("tuple with aborted xmin %u was updated to produce a tuple at offset %u with in-progress xmin %u",
734 : (unsigned) curr_xmin,
735 2 : (unsigned) ctx.offnum,
736 : (unsigned) next_xmin));
737 2 : else if (xmin_commit_status[nextoffnum] == XID_COMMITTED)
738 2 : report_corruption(&ctx,
739 : psprintf("tuple with aborted xmin %u was updated to produce a tuple at offset %u with committed xmin %u",
740 : (unsigned) curr_xmin,
741 2 : (unsigned) ctx.offnum,
742 : (unsigned) next_xmin));
743 : }
744 : }
745 :
746 : /*
747 : * An update chain can start either with a non-heap-only tuple or with
748 : * a redirect line pointer, but not with a heap-only tuple.
749 : *
750 : * (This check is in a separate loop because we need the predecessor
751 : * array to be fully populated before we can perform it.)
752 : */
753 23382 : for (ctx.offnum = FirstOffsetNumber;
754 1111146 : ctx.offnum <= maxoff;
755 1087764 : ctx.offnum = OffsetNumberNext(ctx.offnum))
756 : {
757 1087764 : if (xmin_commit_status_ok[ctx.offnum] &&
758 1065904 : (xmin_commit_status[ctx.offnum] == XID_COMMITTED ||
759 14 : xmin_commit_status[ctx.offnum] == XID_IN_PROGRESS) &&
760 1065894 : predecessor[ctx.offnum] == InvalidOffsetNumber)
761 : {
762 : ItemId curr_lp;
763 :
764 1054324 : curr_lp = PageGetItemId(ctx.page, ctx.offnum);
765 1054324 : if (!ItemIdIsRedirected(curr_lp))
766 : {
767 : HeapTupleHeader curr_htup;
768 :
769 : curr_htup = (HeapTupleHeader)
770 1054324 : PageGetItem(ctx.page, curr_lp);
771 1054324 : if (HeapTupleHeaderIsHeapOnly(curr_htup))
772 8 : report_corruption(&ctx,
773 : psprintf("tuple is root of chain but is marked as heap-only tuple"));
774 : }
775 : }
776 : }
777 :
778 : /* clean up */
779 23382 : UnlockReleaseBuffer(ctx.buffer);
780 :
781 : /*
782 : * Check any toast pointers from the page whose lock we just released
783 : */
784 23382 : if (ctx.toasted_attributes != NIL)
785 : {
786 : ListCell *cell;
787 :
788 25310 : foreach(cell, ctx.toasted_attributes)
789 23630 : check_toasted_attribute(&ctx, lfirst(cell));
790 1680 : list_free_deep(ctx.toasted_attributes);
791 1680 : ctx.toasted_attributes = NIL;
792 : }
793 :
794 23376 : if (on_error_stop && ctx.is_corrupt)
795 0 : break;
796 : }
797 :
798 2946 : if (vmbuffer != InvalidBuffer)
799 6 : ReleaseBuffer(vmbuffer);
800 :
801 : /* Close the associated toast table and indexes, if any. */
802 2946 : if (ctx.toast_indexes)
803 1386 : toast_close_indexes(ctx.toast_indexes, ctx.num_toast_indexes,
804 : AccessShareLock);
805 2946 : if (ctx.toast_rel)
806 1386 : table_close(ctx.toast_rel, AccessShareLock);
807 :
808 : /* Close the main relation */
809 2946 : relation_close(ctx.rel, AccessShareLock);
810 :
811 2946 : PG_RETURN_NULL();
812 : }
813 :
814 : /*
815 : * Shared internal implementation for report_corruption and
816 : * report_toast_corruption.
817 : */
818 : static void
819 172 : report_corruption_internal(Tuplestorestate *tupstore, TupleDesc tupdesc,
820 : BlockNumber blkno, OffsetNumber offnum,
821 : AttrNumber attnum, char *msg)
822 : {
823 172 : Datum values[HEAPCHECK_RELATION_COLS] = {0};
824 172 : bool nulls[HEAPCHECK_RELATION_COLS] = {0};
825 : HeapTuple tuple;
826 :
827 172 : values[0] = Int64GetDatum(blkno);
828 172 : values[1] = Int32GetDatum(offnum);
829 172 : values[2] = Int32GetDatum(attnum);
830 172 : nulls[2] = (attnum < 0);
831 172 : values[3] = CStringGetTextDatum(msg);
832 :
833 : /*
834 : * In principle, there is nothing to prevent a scan over a large, highly
835 : * corrupted table from using work_mem worth of memory building up the
836 : * tuplestore. That's ok, but if we also leak the msg argument memory
837 : * until the end of the query, we could exceed work_mem by more than a
838 : * trivial amount. Therefore, free the msg argument each time we are
839 : * called rather than waiting for our current memory context to be freed.
840 : */
841 172 : pfree(msg);
842 :
843 172 : tuple = heap_form_tuple(tupdesc, values, nulls);
844 172 : tuplestore_puttuple(tupstore, tuple);
845 172 : }
846 :
847 : /*
848 : * Record a single corruption found in the main table. The values in ctx should
849 : * indicate the location of the corruption, and the msg argument should contain
850 : * a human-readable description of the corruption.
851 : *
852 : * The msg argument is pfree'd by this function.
853 : */
854 : static void
855 170 : report_corruption(HeapCheckContext *ctx, char *msg)
856 : {
857 170 : report_corruption_internal(ctx->tupstore, ctx->tupdesc, ctx->blkno,
858 170 : ctx->offnum, ctx->attnum, msg);
859 170 : ctx->is_corrupt = true;
860 170 : }
861 :
862 : /*
863 : * Record corruption found in the toast table. The values in ta should
864 : * indicate the location in the main table where the toast pointer was
865 : * encountered, and the msg argument should contain a human-readable
866 : * description of the toast table corruption.
867 : *
868 : * As above, the msg argument is pfree'd by this function.
869 : */
870 : static void
871 2 : report_toast_corruption(HeapCheckContext *ctx, ToastedAttribute *ta,
872 : char *msg)
873 : {
874 2 : report_corruption_internal(ctx->tupstore, ctx->tupdesc, ta->blkno,
875 2 : ta->offnum, ta->attnum, msg);
876 2 : ctx->is_corrupt = true;
877 2 : }
878 :
879 : /*
880 : * Check for tuple header corruption.
881 : *
882 : * Some kinds of corruption make it unsafe to check the tuple attributes, for
883 : * example when the line pointer refers to a range of bytes outside the page.
884 : * In such cases, we return false (not checkable) after recording appropriate
885 : * corruption messages.
886 : *
887 : * Some other kinds of tuple header corruption confuse the question of where
888 : * the tuple attributes begin, or how long the nulls bitmap is, etc., making it
889 : * unreasonable to attempt to check attributes, even if all candidate answers
890 : * to those questions would not result in reading past the end of the line
891 : * pointer or page. In such cases, like above, we record corruption messages
892 : * about the header and then return false.
893 : *
894 : * Other kinds of tuple header corruption do not bear on the question of
895 : * whether the tuple attributes can be checked, so we record corruption
896 : * messages for them but we do not return false merely because we detected
897 : * them.
898 : *
899 : * Returns whether the tuple is sufficiently sensible to undergo visibility and
900 : * attribute checks.
901 : */
902 : static bool
903 1065922 : check_tuple_header(HeapCheckContext *ctx)
904 : {
905 1065922 : HeapTupleHeader tuphdr = ctx->tuphdr;
906 1065922 : uint16 infomask = tuphdr->t_infomask;
907 1065922 : TransactionId curr_xmax = HeapTupleHeaderGetUpdateXid(tuphdr);
908 1065922 : bool result = true;
909 : unsigned expected_hoff;
910 :
911 1065922 : if (ctx->tuphdr->t_hoff > ctx->lp_len)
912 : {
913 2 : report_corruption(ctx,
914 : psprintf("data begins at offset %u beyond the tuple length %u",
915 2 : ctx->tuphdr->t_hoff, ctx->lp_len));
916 2 : result = false;
917 : }
918 :
919 1065922 : if ((ctx->tuphdr->t_infomask & HEAP_XMAX_COMMITTED) &&
920 262 : (ctx->tuphdr->t_infomask & HEAP_XMAX_IS_MULTI))
921 : {
922 4 : report_corruption(ctx,
923 : pstrdup("multixact should not be marked committed"));
924 :
925 : /*
926 : * This condition is clearly wrong, but it's not enough to justify
927 : * skipping further checks, because we don't rely on this to determine
928 : * whether the tuple is visible or to interpret other relevant header
929 : * fields.
930 : */
931 : }
932 :
933 1065922 : if (!TransactionIdIsValid(curr_xmax) &&
934 1065460 : HeapTupleHeaderIsHotUpdated(tuphdr))
935 : {
936 2 : report_corruption(ctx,
937 : psprintf("tuple has been HOT updated, but xmax is 0"));
938 :
939 : /*
940 : * As above, even though this shouldn't happen, it's not sufficient
941 : * justification for skipping further checks, we should still be able
942 : * to perform sensibly.
943 : */
944 : }
945 :
946 1065922 : if (HeapTupleHeaderIsHeapOnly(tuphdr) &&
947 11580 : ((tuphdr->t_infomask & HEAP_UPDATED) == 0))
948 : {
949 2 : report_corruption(ctx,
950 : psprintf("tuple is heap only, but not the result of an update"));
951 :
952 : /* Here again, we can still perform further checks. */
953 : }
954 :
955 1065922 : if (infomask & HEAP_HASNULL)
956 475662 : expected_hoff = MAXALIGN(SizeofHeapTupleHeader + BITMAPLEN(ctx->natts));
957 : else
958 590260 : expected_hoff = MAXALIGN(SizeofHeapTupleHeader);
959 1065922 : if (ctx->tuphdr->t_hoff != expected_hoff)
960 : {
961 10 : if ((infomask & HEAP_HASNULL) && ctx->natts == 1)
962 0 : report_corruption(ctx,
963 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (1 attribute, has nulls)",
964 0 : expected_hoff, ctx->tuphdr->t_hoff));
965 10 : else if ((infomask & HEAP_HASNULL))
966 2 : report_corruption(ctx,
967 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (%u attributes, has nulls)",
968 2 : expected_hoff, ctx->tuphdr->t_hoff, ctx->natts));
969 8 : else if (ctx->natts == 1)
970 0 : report_corruption(ctx,
971 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (1 attribute, no nulls)",
972 0 : expected_hoff, ctx->tuphdr->t_hoff));
973 : else
974 8 : report_corruption(ctx,
975 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (%u attributes, no nulls)",
976 8 : expected_hoff, ctx->tuphdr->t_hoff, ctx->natts));
977 10 : result = false;
978 : }
979 :
980 1065922 : return result;
981 : }
982 :
983 : /*
984 : * Checks tuple visibility so we know which further checks are safe to
985 : * perform.
986 : *
987 : * If a tuple could have been inserted by a transaction that also added a
988 : * column to the table, but which ultimately did not commit, or which has not
989 : * yet committed, then the table's current TupleDesc might differ from the one
990 : * used to construct this tuple, so we must not check it.
991 : *
992 : * As a special case, if our own transaction inserted the tuple, even if we
993 : * added a column to the table, our TupleDesc should match. We could check the
994 : * tuple, but choose not to do so.
995 : *
996 : * If a tuple has been updated or deleted, we can still read the old tuple for
997 : * corruption checking purposes, as long as we are careful about concurrent
998 : * vacuums. The main table tuple itself cannot be vacuumed away because we
999 : * hold a buffer lock on the page, but if the deleting transaction is older
1000 : * than our transaction snapshot's xmin, then vacuum could remove the toast at
1001 : * any time, so we must not try to follow TOAST pointers.
1002 : *
1003 : * If xmin or xmax values are older than can be checked against clog, or appear
1004 : * to be in the future (possibly due to wrap-around), then we cannot make a
1005 : * determination about the visibility of the tuple, so we skip further checks.
1006 : *
1007 : * Returns true if the tuple itself should be checked, false otherwise. Sets
1008 : * ctx->tuple_could_be_pruned if the tuple -- and thus also any associated
1009 : * TOAST tuples -- are eligible for pruning.
1010 : *
1011 : * Sets *xmin_commit_status_ok to true if the commit status of xmin is known
1012 : * and false otherwise. If it's set to true, then also set *xmin_commit_status
1013 : * to the actual commit status.
1014 : */
1015 : static bool
1016 1065912 : check_tuple_visibility(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
1017 : XidCommitStatus *xmin_commit_status)
1018 : {
1019 : TransactionId xmin;
1020 : TransactionId xvac;
1021 : TransactionId xmax;
1022 : XidCommitStatus xmin_status;
1023 : XidCommitStatus xvac_status;
1024 : XidCommitStatus xmax_status;
1025 1065912 : HeapTupleHeader tuphdr = ctx->tuphdr;
1026 :
1027 1065912 : ctx->tuple_could_be_pruned = true; /* have not yet proven otherwise */
1028 1065912 : *xmin_commit_status_ok = false; /* have not yet proven otherwise */
1029 :
1030 : /* If xmin is normal, it should be within valid range */
1031 1065912 : xmin = HeapTupleHeaderGetXmin(tuphdr);
1032 1065912 : switch (get_xid_status(xmin, ctx, &xmin_status))
1033 : {
1034 0 : case XID_INVALID:
1035 : /* Could be the result of a speculative insertion that aborted. */
1036 0 : return false;
1037 1065904 : case XID_BOUNDS_OK:
1038 1065904 : *xmin_commit_status_ok = true;
1039 1065904 : *xmin_commit_status = xmin_status;
1040 1065904 : break;
1041 2 : case XID_IN_FUTURE:
1042 2 : report_corruption(ctx,
1043 : psprintf("xmin %u equals or exceeds next valid transaction ID %u:%u",
1044 : xmin,
1045 2 : EpochFromFullTransactionId(ctx->next_fxid),
1046 2 : XidFromFullTransactionId(ctx->next_fxid)));
1047 2 : return false;
1048 4 : case XID_PRECEDES_CLUSTERMIN:
1049 4 : report_corruption(ctx,
1050 : psprintf("xmin %u precedes oldest valid transaction ID %u:%u",
1051 : xmin,
1052 4 : EpochFromFullTransactionId(ctx->oldest_fxid),
1053 4 : XidFromFullTransactionId(ctx->oldest_fxid)));
1054 4 : return false;
1055 2 : case XID_PRECEDES_RELMIN:
1056 2 : report_corruption(ctx,
1057 : psprintf("xmin %u precedes relation freeze threshold %u:%u",
1058 : xmin,
1059 2 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1060 2 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1061 2 : return false;
1062 : }
1063 :
1064 : /*
1065 : * Has inserting transaction committed?
1066 : */
1067 1065904 : if (!HeapTupleHeaderXminCommitted(tuphdr))
1068 : {
1069 36906 : if (HeapTupleHeaderXminInvalid(tuphdr))
1070 0 : return false; /* inserter aborted, don't check */
1071 : /* Used by pre-9.0 binary upgrades */
1072 36906 : else if (tuphdr->t_infomask & HEAP_MOVED_OFF)
1073 : {
1074 0 : xvac = HeapTupleHeaderGetXvac(tuphdr);
1075 :
1076 0 : switch (get_xid_status(xvac, ctx, &xvac_status))
1077 : {
1078 0 : case XID_INVALID:
1079 0 : report_corruption(ctx,
1080 : pstrdup("old-style VACUUM FULL transaction ID for moved off tuple is invalid"));
1081 0 : return false;
1082 0 : case XID_IN_FUTURE:
1083 0 : report_corruption(ctx,
1084 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple equals or exceeds next valid transaction ID %u:%u",
1085 : xvac,
1086 0 : EpochFromFullTransactionId(ctx->next_fxid),
1087 0 : XidFromFullTransactionId(ctx->next_fxid)));
1088 0 : return false;
1089 0 : case XID_PRECEDES_RELMIN:
1090 0 : report_corruption(ctx,
1091 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple precedes relation freeze threshold %u:%u",
1092 : xvac,
1093 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1094 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1095 0 : return false;
1096 0 : case XID_PRECEDES_CLUSTERMIN:
1097 0 : report_corruption(ctx,
1098 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple precedes oldest valid transaction ID %u:%u",
1099 : xvac,
1100 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1101 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1102 0 : return false;
1103 0 : case XID_BOUNDS_OK:
1104 0 : break;
1105 : }
1106 :
1107 0 : switch (xvac_status)
1108 : {
1109 0 : case XID_IS_CURRENT_XID:
1110 0 : report_corruption(ctx,
1111 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple matches our current transaction ID",
1112 : xvac));
1113 0 : return false;
1114 0 : case XID_IN_PROGRESS:
1115 0 : report_corruption(ctx,
1116 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple appears to be in progress",
1117 : xvac));
1118 0 : return false;
1119 :
1120 0 : case XID_COMMITTED:
1121 :
1122 : /*
1123 : * The tuple is dead, because the xvac transaction moved
1124 : * it off and committed. It's checkable, but also
1125 : * prunable.
1126 : */
1127 0 : return true;
1128 :
1129 0 : case XID_ABORTED:
1130 :
1131 : /*
1132 : * The original xmin must have committed, because the xvac
1133 : * transaction tried to move it later. Since xvac is
1134 : * aborted, whether it's still alive now depends on the
1135 : * status of xmax.
1136 : */
1137 0 : break;
1138 : }
1139 0 : }
1140 : /* Used by pre-9.0 binary upgrades */
1141 36906 : else if (tuphdr->t_infomask & HEAP_MOVED_IN)
1142 : {
1143 0 : xvac = HeapTupleHeaderGetXvac(tuphdr);
1144 :
1145 0 : switch (get_xid_status(xvac, ctx, &xvac_status))
1146 : {
1147 0 : case XID_INVALID:
1148 0 : report_corruption(ctx,
1149 : pstrdup("old-style VACUUM FULL transaction ID for moved in tuple is invalid"));
1150 0 : return false;
1151 0 : case XID_IN_FUTURE:
1152 0 : report_corruption(ctx,
1153 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple equals or exceeds next valid transaction ID %u:%u",
1154 : xvac,
1155 0 : EpochFromFullTransactionId(ctx->next_fxid),
1156 0 : XidFromFullTransactionId(ctx->next_fxid)));
1157 0 : return false;
1158 0 : case XID_PRECEDES_RELMIN:
1159 0 : report_corruption(ctx,
1160 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple precedes relation freeze threshold %u:%u",
1161 : xvac,
1162 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1163 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1164 0 : return false;
1165 0 : case XID_PRECEDES_CLUSTERMIN:
1166 0 : report_corruption(ctx,
1167 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple precedes oldest valid transaction ID %u:%u",
1168 : xvac,
1169 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1170 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1171 0 : return false;
1172 0 : case XID_BOUNDS_OK:
1173 0 : break;
1174 : }
1175 :
1176 0 : switch (xvac_status)
1177 : {
1178 0 : case XID_IS_CURRENT_XID:
1179 0 : report_corruption(ctx,
1180 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple matches our current transaction ID",
1181 : xvac));
1182 0 : return false;
1183 0 : case XID_IN_PROGRESS:
1184 0 : report_corruption(ctx,
1185 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple appears to be in progress",
1186 : xvac));
1187 0 : return false;
1188 :
1189 0 : case XID_COMMITTED:
1190 :
1191 : /*
1192 : * The original xmin must have committed, because the xvac
1193 : * transaction moved it later. Whether it's still alive
1194 : * now depends on the status of xmax.
1195 : */
1196 0 : break;
1197 :
1198 0 : case XID_ABORTED:
1199 :
1200 : /*
1201 : * The tuple is dead, because the xvac transaction moved
1202 : * it off and committed. It's checkable, but also
1203 : * prunable.
1204 : */
1205 0 : return true;
1206 : }
1207 0 : }
1208 36906 : else if (xmin_status != XID_COMMITTED)
1209 : {
1210 : /*
1211 : * Inserting transaction is not in progress, and not committed, so
1212 : * it might have changed the TupleDesc in ways we don't know
1213 : * about. Thus, don't try to check the tuple structure.
1214 : *
1215 : * If xmin_status happens to be XID_IS_CURRENT_XID, then in theory
1216 : * any such DDL changes ought to be visible to us, so perhaps we
1217 : * could check anyway in that case. But, for now, let's be
1218 : * conservative and treat this like any other uncommitted insert.
1219 : */
1220 14 : return false;
1221 : }
1222 : }
1223 :
1224 : /*
1225 : * Okay, the inserter committed, so it was good at some point. Now what
1226 : * about the deleting transaction?
1227 : */
1228 :
1229 1065890 : if (tuphdr->t_infomask & HEAP_XMAX_IS_MULTI)
1230 : {
1231 : /*
1232 : * xmax is a multixact, so sanity-check the MXID. Note that we do this
1233 : * prior to checking for HEAP_XMAX_INVALID or
1234 : * HEAP_XMAX_IS_LOCKED_ONLY. This might therefore complain about
1235 : * things that wouldn't actually be a problem during a normal scan,
1236 : * but eventually we're going to have to freeze, and that process will
1237 : * ignore hint bits.
1238 : *
1239 : * Even if the MXID is out of range, we still know that the original
1240 : * insert committed, so we can check the tuple itself. However, we
1241 : * can't rule out the possibility that this tuple is dead, so don't
1242 : * clear ctx->tuple_could_be_pruned. Possibly we should go ahead and
1243 : * clear that flag anyway if HEAP_XMAX_INVALID is set or if
1244 : * HEAP_XMAX_IS_LOCKED_ONLY is true, but for now we err on the side of
1245 : * avoiding possibly-bogus complaints about missing TOAST entries.
1246 : */
1247 116 : xmax = HeapTupleHeaderGetRawXmax(tuphdr);
1248 116 : switch (check_mxid_valid_in_rel(xmax, ctx))
1249 : {
1250 0 : case XID_INVALID:
1251 0 : report_corruption(ctx,
1252 : pstrdup("multitransaction ID is invalid"));
1253 0 : return true;
1254 2 : case XID_PRECEDES_RELMIN:
1255 2 : report_corruption(ctx,
1256 : psprintf("multitransaction ID %u precedes relation minimum multitransaction ID threshold %u",
1257 : xmax, ctx->relminmxid));
1258 2 : return true;
1259 0 : case XID_PRECEDES_CLUSTERMIN:
1260 0 : report_corruption(ctx,
1261 : psprintf("multitransaction ID %u precedes oldest valid multitransaction ID threshold %u",
1262 : xmax, ctx->oldest_mxact));
1263 0 : return true;
1264 2 : case XID_IN_FUTURE:
1265 2 : report_corruption(ctx,
1266 : psprintf("multitransaction ID %u equals or exceeds next valid multitransaction ID %u",
1267 : xmax,
1268 : ctx->next_mxact));
1269 2 : return true;
1270 112 : case XID_BOUNDS_OK:
1271 112 : break;
1272 : }
1273 1065774 : }
1274 :
1275 1065886 : if (tuphdr->t_infomask & HEAP_XMAX_INVALID)
1276 : {
1277 : /*
1278 : * This tuple is live. A concurrently running transaction could
1279 : * delete it before we get around to checking the toast, but any such
1280 : * running transaction is surely not less than our safe_xmin, so the
1281 : * toast cannot be vacuumed out from under us.
1282 : */
1283 1065434 : ctx->tuple_could_be_pruned = false;
1284 1065434 : return true;
1285 : }
1286 :
1287 452 : if (HEAP_XMAX_IS_LOCKED_ONLY(tuphdr->t_infomask))
1288 : {
1289 : /*
1290 : * "Deleting" xact really only locked it, so the tuple is live in any
1291 : * case. As above, a concurrently running transaction could delete
1292 : * it, but it cannot be vacuumed out from under us.
1293 : */
1294 56 : ctx->tuple_could_be_pruned = false;
1295 56 : return true;
1296 : }
1297 :
1298 396 : if (tuphdr->t_infomask & HEAP_XMAX_IS_MULTI)
1299 : {
1300 : /*
1301 : * We already checked above that this multixact is within limits for
1302 : * this table. Now check the update xid from this multixact.
1303 : */
1304 56 : xmax = HeapTupleGetUpdateXid(tuphdr);
1305 56 : switch (get_xid_status(xmax, ctx, &xmax_status))
1306 : {
1307 0 : case XID_INVALID:
1308 : /* not LOCKED_ONLY, so it has to have an xmax */
1309 0 : report_corruption(ctx,
1310 : pstrdup("update xid is invalid"));
1311 0 : return true;
1312 0 : case XID_IN_FUTURE:
1313 0 : report_corruption(ctx,
1314 : psprintf("update xid %u equals or exceeds next valid transaction ID %u:%u",
1315 : xmax,
1316 0 : EpochFromFullTransactionId(ctx->next_fxid),
1317 0 : XidFromFullTransactionId(ctx->next_fxid)));
1318 0 : return true;
1319 0 : case XID_PRECEDES_RELMIN:
1320 0 : report_corruption(ctx,
1321 : psprintf("update xid %u precedes relation freeze threshold %u:%u",
1322 : xmax,
1323 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1324 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1325 0 : return true;
1326 0 : case XID_PRECEDES_CLUSTERMIN:
1327 0 : report_corruption(ctx,
1328 : psprintf("update xid %u precedes oldest valid transaction ID %u:%u",
1329 : xmax,
1330 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1331 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1332 0 : return true;
1333 56 : case XID_BOUNDS_OK:
1334 56 : break;
1335 : }
1336 :
1337 56 : switch (xmax_status)
1338 : {
1339 0 : case XID_IS_CURRENT_XID:
1340 : case XID_IN_PROGRESS:
1341 :
1342 : /*
1343 : * The delete is in progress, so it cannot be visible to our
1344 : * snapshot.
1345 : */
1346 0 : ctx->tuple_could_be_pruned = false;
1347 0 : break;
1348 56 : case XID_COMMITTED:
1349 :
1350 : /*
1351 : * The delete committed. Whether the toast can be vacuumed
1352 : * away depends on how old the deleting transaction is.
1353 : */
1354 56 : ctx->tuple_could_be_pruned = TransactionIdPrecedes(xmax,
1355 : ctx->safe_xmin);
1356 56 : break;
1357 0 : case XID_ABORTED:
1358 :
1359 : /*
1360 : * The delete aborted or crashed. The tuple is still live.
1361 : */
1362 0 : ctx->tuple_could_be_pruned = false;
1363 0 : break;
1364 : }
1365 :
1366 : /* Tuple itself is checkable even if it's dead. */
1367 56 : return true;
1368 : }
1369 :
1370 : /* xmax is an XID, not a MXID. Sanity check it. */
1371 340 : xmax = HeapTupleHeaderGetRawXmax(tuphdr);
1372 340 : switch (get_xid_status(xmax, ctx, &xmax_status))
1373 : {
1374 2 : case XID_INVALID:
1375 2 : ctx->tuple_could_be_pruned = false;
1376 2 : return true;
1377 0 : case XID_IN_FUTURE:
1378 0 : report_corruption(ctx,
1379 : psprintf("xmax %u equals or exceeds next valid transaction ID %u:%u",
1380 : xmax,
1381 0 : EpochFromFullTransactionId(ctx->next_fxid),
1382 0 : XidFromFullTransactionId(ctx->next_fxid)));
1383 0 : return false; /* corrupt */
1384 0 : case XID_PRECEDES_RELMIN:
1385 0 : report_corruption(ctx,
1386 : psprintf("xmax %u precedes relation freeze threshold %u:%u",
1387 : xmax,
1388 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1389 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1390 0 : return false; /* corrupt */
1391 2 : case XID_PRECEDES_CLUSTERMIN:
1392 2 : report_corruption(ctx,
1393 : psprintf("xmax %u precedes oldest valid transaction ID %u:%u",
1394 : xmax,
1395 2 : EpochFromFullTransactionId(ctx->oldest_fxid),
1396 2 : XidFromFullTransactionId(ctx->oldest_fxid)));
1397 2 : return false; /* corrupt */
1398 336 : case XID_BOUNDS_OK:
1399 336 : break;
1400 : }
1401 :
1402 : /*
1403 : * Whether the toast can be vacuumed away depends on how old the deleting
1404 : * transaction is.
1405 : */
1406 336 : switch (xmax_status)
1407 : {
1408 0 : case XID_IS_CURRENT_XID:
1409 : case XID_IN_PROGRESS:
1410 :
1411 : /*
1412 : * The delete is in progress, so it cannot be visible to our
1413 : * snapshot.
1414 : */
1415 0 : ctx->tuple_could_be_pruned = false;
1416 0 : break;
1417 :
1418 330 : case XID_COMMITTED:
1419 :
1420 : /*
1421 : * The delete committed. Whether the toast can be vacuumed away
1422 : * depends on how old the deleting transaction is.
1423 : */
1424 330 : ctx->tuple_could_be_pruned = TransactionIdPrecedes(xmax,
1425 : ctx->safe_xmin);
1426 330 : break;
1427 :
1428 6 : case XID_ABORTED:
1429 :
1430 : /*
1431 : * The delete aborted or crashed. The tuple is still live.
1432 : */
1433 6 : ctx->tuple_could_be_pruned = false;
1434 6 : break;
1435 : }
1436 :
1437 : /* Tuple itself is checkable even if it's dead. */
1438 336 : return true;
1439 : }
1440 :
1441 :
1442 : /*
1443 : * Check the current toast tuple against the state tracked in ctx, recording
1444 : * any corruption found in ctx->tupstore.
1445 : *
1446 : * This is not equivalent to running verify_heapam on the toast table itself,
1447 : * and is not hardened against corruption of the toast table. Rather, when
1448 : * validating a toasted attribute in the main table, the sequence of toast
1449 : * tuples that store the toasted value are retrieved and checked in order, with
1450 : * each toast tuple being checked against where we are in the sequence, as well
1451 : * as each toast tuple having its varlena structure sanity checked.
1452 : *
1453 : * On entry, *expected_chunk_seq should be the chunk_seq value that we expect
1454 : * to find in toasttup. On exit, it will be updated to the value the next call
1455 : * to this function should expect to see.
1456 : */
1457 : static void
1458 83138 : check_toast_tuple(HeapTuple toasttup, HeapCheckContext *ctx,
1459 : ToastedAttribute *ta, int32 *expected_chunk_seq,
1460 : uint32 extsize)
1461 : {
1462 : int32 chunk_seq;
1463 83138 : int32 last_chunk_seq = (extsize - 1) / TOAST_MAX_CHUNK_SIZE;
1464 : Pointer chunk;
1465 : bool isnull;
1466 : int32 chunksize;
1467 : int32 expected_size;
1468 :
1469 : /* Sanity-check the sequence number. */
1470 83138 : chunk_seq = DatumGetInt32(fastgetattr(toasttup, 2,
1471 83138 : ctx->toast_rel->rd_att, &isnull));
1472 83138 : if (isnull)
1473 : {
1474 0 : report_toast_corruption(ctx, ta,
1475 : psprintf("toast value %u has toast chunk with null sequence number",
1476 : ta->toast_pointer.va_valueid));
1477 0 : return;
1478 : }
1479 83138 : if (chunk_seq != *expected_chunk_seq)
1480 : {
1481 : /* Either the TOAST index is corrupt, or we don't have all chunks. */
1482 0 : report_toast_corruption(ctx, ta,
1483 : psprintf("toast value %u index scan returned chunk %d when expecting chunk %d",
1484 : ta->toast_pointer.va_valueid,
1485 : chunk_seq, *expected_chunk_seq));
1486 : }
1487 83138 : *expected_chunk_seq = chunk_seq + 1;
1488 :
1489 : /* Sanity-check the chunk data. */
1490 83138 : chunk = DatumGetPointer(fastgetattr(toasttup, 3,
1491 83138 : ctx->toast_rel->rd_att, &isnull));
1492 83138 : if (isnull)
1493 : {
1494 0 : report_toast_corruption(ctx, ta,
1495 : psprintf("toast value %u chunk %d has null data",
1496 : ta->toast_pointer.va_valueid,
1497 : chunk_seq));
1498 0 : return;
1499 : }
1500 83138 : if (!VARATT_IS_EXTENDED(chunk))
1501 83138 : chunksize = VARSIZE(chunk) - VARHDRSZ;
1502 0 : else if (VARATT_IS_SHORT(chunk))
1503 : {
1504 : /*
1505 : * could happen due to heap_form_tuple doing its thing
1506 : */
1507 0 : chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
1508 : }
1509 : else
1510 : {
1511 : /* should never happen */
1512 0 : uint32 header = ((varattrib_4b *) chunk)->va_4byte.va_header;
1513 :
1514 0 : report_toast_corruption(ctx, ta,
1515 : psprintf("toast value %u chunk %d has invalid varlena header %0x",
1516 : ta->toast_pointer.va_valueid,
1517 : chunk_seq, header));
1518 0 : return;
1519 : }
1520 :
1521 : /*
1522 : * Some checks on the data we've found
1523 : */
1524 83138 : if (chunk_seq > last_chunk_seq)
1525 : {
1526 0 : report_toast_corruption(ctx, ta,
1527 : psprintf("toast value %u chunk %d follows last expected chunk %d",
1528 : ta->toast_pointer.va_valueid,
1529 : chunk_seq, last_chunk_seq));
1530 0 : return;
1531 : }
1532 :
1533 83138 : expected_size = chunk_seq < last_chunk_seq ? TOAST_MAX_CHUNK_SIZE
1534 23622 : : extsize - (last_chunk_seq * TOAST_MAX_CHUNK_SIZE);
1535 :
1536 83138 : if (chunksize != expected_size)
1537 0 : report_toast_corruption(ctx, ta,
1538 : psprintf("toast value %u chunk %d has size %u, but expected size %u",
1539 : ta->toast_pointer.va_valueid,
1540 : chunk_seq, chunksize, expected_size));
1541 : }
1542 :
1543 : /*
1544 : * Check the current attribute as tracked in ctx, recording any corruption
1545 : * found in ctx->tupstore.
1546 : *
1547 : * This function follows the logic performed by heap_deform_tuple(), and in the
1548 : * case of a toasted value, optionally stores the toast pointer so later it can
1549 : * be checked following the logic of detoast_external_attr(), checking for any
1550 : * conditions that would result in either of those functions Asserting or
1551 : * crashing the backend. The checks performed by Asserts present in those two
1552 : * functions are also performed here and in check_toasted_attribute. In cases
1553 : * where those two functions are a bit cavalier in their assumptions about data
1554 : * being correct, we perform additional checks not present in either of those
1555 : * two functions. Where some condition is checked in both of those functions,
1556 : * we perform it here twice, as we parallel the logical flow of those two
1557 : * functions. The presence of duplicate checks seems a reasonable price to pay
1558 : * for keeping this code tightly coupled with the code it protects.
1559 : *
1560 : * Returns true if the tuple attribute is sane enough for processing to
1561 : * continue on to the next attribute, false otherwise.
1562 : */
1563 : static bool
1564 15397750 : check_tuple_attribute(HeapCheckContext *ctx)
1565 : {
1566 : Datum attdatum;
1567 : struct varlena *attr;
1568 : char *tp; /* pointer to the tuple data */
1569 : uint16 infomask;
1570 : Form_pg_attribute thisatt;
1571 : struct varatt_external toast_pointer;
1572 :
1573 15397750 : infomask = ctx->tuphdr->t_infomask;
1574 15397750 : thisatt = TupleDescAttr(RelationGetDescr(ctx->rel), ctx->attnum);
1575 :
1576 15397750 : tp = (char *) ctx->tuphdr + ctx->tuphdr->t_hoff;
1577 :
1578 15397750 : if (ctx->tuphdr->t_hoff + ctx->offset > ctx->lp_len)
1579 : {
1580 0 : report_corruption(ctx,
1581 : psprintf("attribute with length %u starts at offset %u beyond total tuple length %u",
1582 0 : thisatt->attlen,
1583 0 : ctx->tuphdr->t_hoff + ctx->offset,
1584 0 : ctx->lp_len));
1585 0 : return false;
1586 : }
1587 :
1588 : /* Skip null values */
1589 15397750 : if (infomask & HEAP_HASNULL && att_isnull(ctx->attnum, ctx->tuphdr->t_bits))
1590 2601596 : return true;
1591 :
1592 : /* Skip non-varlena values, but update offset first */
1593 12796154 : if (thisatt->attlen != -1)
1594 : {
1595 11749452 : ctx->offset = att_align_nominal(ctx->offset, thisatt->attalign);
1596 11749452 : ctx->offset = att_addlength_pointer(ctx->offset, thisatt->attlen,
1597 : tp + ctx->offset);
1598 11749452 : if (ctx->tuphdr->t_hoff + ctx->offset > ctx->lp_len)
1599 : {
1600 0 : report_corruption(ctx,
1601 : psprintf("attribute with length %u ends at offset %u beyond total tuple length %u",
1602 0 : thisatt->attlen,
1603 0 : ctx->tuphdr->t_hoff + ctx->offset,
1604 0 : ctx->lp_len));
1605 0 : return false;
1606 : }
1607 11749452 : return true;
1608 : }
1609 :
1610 : /* Ok, we're looking at a varlena attribute. */
1611 1046702 : ctx->offset = att_align_pointer(ctx->offset, thisatt->attalign, -1,
1612 : tp + ctx->offset);
1613 :
1614 : /* Get the (possibly corrupt) varlena datum */
1615 1046702 : attdatum = fetchatt(thisatt, tp + ctx->offset);
1616 :
1617 : /*
1618 : * We have the datum, but we cannot decode it carelessly, as it may still
1619 : * be corrupt.
1620 : */
1621 :
1622 : /*
1623 : * Check that VARTAG_SIZE won't hit an Assert on a corrupt va_tag before
1624 : * risking a call into att_addlength_pointer
1625 : */
1626 1046702 : if (VARATT_IS_EXTERNAL(tp + ctx->offset))
1627 : {
1628 52504 : uint8 va_tag = VARTAG_EXTERNAL(tp + ctx->offset);
1629 :
1630 52504 : if (va_tag != VARTAG_ONDISK)
1631 : {
1632 0 : report_corruption(ctx,
1633 : psprintf("toasted attribute has unexpected TOAST tag %u",
1634 : va_tag));
1635 : /* We can't know where the next attribute begins */
1636 0 : return false;
1637 : }
1638 : }
1639 :
1640 : /* Ok, should be safe now */
1641 1046702 : ctx->offset = att_addlength_pointer(ctx->offset, thisatt->attlen,
1642 : tp + ctx->offset);
1643 :
1644 1046702 : if (ctx->tuphdr->t_hoff + ctx->offset > ctx->lp_len)
1645 : {
1646 2 : report_corruption(ctx,
1647 : psprintf("attribute with length %u ends at offset %u beyond total tuple length %u",
1648 2 : thisatt->attlen,
1649 2 : ctx->tuphdr->t_hoff + ctx->offset,
1650 2 : ctx->lp_len));
1651 :
1652 2 : return false;
1653 : }
1654 :
1655 : /*
1656 : * heap_deform_tuple would be done with this attribute at this point,
1657 : * having stored it in values[], and would continue to the next attribute.
1658 : * We go further, because we need to check if the toast datum is corrupt.
1659 : */
1660 :
1661 1046700 : attr = (struct varlena *) DatumGetPointer(attdatum);
1662 :
1663 : /*
1664 : * Now we follow the logic of detoast_external_attr(), with the same
1665 : * caveats about being paranoid about corruption.
1666 : */
1667 :
1668 : /* Skip values that are not external */
1669 1046700 : if (!VARATT_IS_EXTERNAL(attr))
1670 994196 : return true;
1671 :
1672 : /* It is external, and we're looking at a page on disk */
1673 :
1674 : /*
1675 : * Must copy attr into toast_pointer for alignment considerations
1676 : */
1677 52504 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
1678 :
1679 : /* Toasted attributes too large to be untoasted should never be stored */
1680 52504 : if (toast_pointer.va_rawsize > VARLENA_SIZE_LIMIT)
1681 0 : report_corruption(ctx,
1682 : psprintf("toast value %u rawsize %d exceeds limit %d",
1683 : toast_pointer.va_valueid,
1684 : toast_pointer.va_rawsize,
1685 : VARLENA_SIZE_LIMIT));
1686 :
1687 52504 : if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
1688 : {
1689 : ToastCompressionId cmid;
1690 4048 : bool valid = false;
1691 :
1692 : /* Compressed attributes should have a valid compression method */
1693 4048 : cmid = TOAST_COMPRESS_METHOD(&toast_pointer);
1694 4048 : switch (cmid)
1695 : {
1696 : /* List of all valid compression method IDs */
1697 4048 : case TOAST_PGLZ_COMPRESSION_ID:
1698 : case TOAST_LZ4_COMPRESSION_ID:
1699 4048 : valid = true;
1700 4048 : break;
1701 :
1702 : /* Recognized but invalid compression method ID */
1703 0 : case TOAST_INVALID_COMPRESSION_ID:
1704 0 : break;
1705 :
1706 : /* Intentionally no default here */
1707 : }
1708 4048 : if (!valid)
1709 0 : report_corruption(ctx,
1710 : psprintf("toast value %u has invalid compression method id %d",
1711 : toast_pointer.va_valueid, cmid));
1712 : }
1713 :
1714 : /* The tuple header better claim to contain toasted values */
1715 52504 : if (!(infomask & HEAP_HASEXTERNAL))
1716 : {
1717 0 : report_corruption(ctx,
1718 : psprintf("toast value %u is external but tuple header flag HEAP_HASEXTERNAL not set",
1719 : toast_pointer.va_valueid));
1720 0 : return true;
1721 : }
1722 :
1723 : /* The relation better have a toast table */
1724 52504 : if (!ctx->rel->rd_rel->reltoastrelid)
1725 : {
1726 0 : report_corruption(ctx,
1727 : psprintf("toast value %u is external but relation has no toast relation",
1728 : toast_pointer.va_valueid));
1729 0 : return true;
1730 : }
1731 :
1732 : /* If we were told to skip toast checking, then we're done. */
1733 52504 : if (ctx->toast_rel == NULL)
1734 28858 : return true;
1735 :
1736 : /*
1737 : * If this tuple is eligible to be pruned, we cannot check the toast.
1738 : * Otherwise, we push a copy of the toast tuple so we can check it after
1739 : * releasing the main table buffer lock.
1740 : */
1741 23646 : if (!ctx->tuple_could_be_pruned)
1742 : {
1743 : ToastedAttribute *ta;
1744 :
1745 23642 : ta = (ToastedAttribute *) palloc0(sizeof(ToastedAttribute));
1746 :
1747 23642 : VARATT_EXTERNAL_GET_POINTER(ta->toast_pointer, attr);
1748 23642 : ta->blkno = ctx->blkno;
1749 23642 : ta->offnum = ctx->offnum;
1750 23642 : ta->attnum = ctx->attnum;
1751 23642 : ctx->toasted_attributes = lappend(ctx->toasted_attributes, ta);
1752 : }
1753 :
1754 23646 : return true;
1755 : }
1756 :
1757 : /*
1758 : * For each attribute collected in ctx->toasted_attributes, look up the value
1759 : * in the toast table and perform checks on it. This function should only be
1760 : * called on toast pointers which cannot be vacuumed away during our
1761 : * processing.
1762 : */
1763 : static void
1764 23630 : check_toasted_attribute(HeapCheckContext *ctx, ToastedAttribute *ta)
1765 : {
1766 : SnapshotData SnapshotToast;
1767 : ScanKeyData toastkey;
1768 : SysScanDesc toastscan;
1769 : bool found_toasttup;
1770 : HeapTuple toasttup;
1771 : uint32 extsize;
1772 23630 : int32 expected_chunk_seq = 0;
1773 : int32 last_chunk_seq;
1774 :
1775 23630 : extsize = VARATT_EXTERNAL_GET_EXTSIZE(ta->toast_pointer);
1776 23630 : last_chunk_seq = (extsize - 1) / TOAST_MAX_CHUNK_SIZE;
1777 :
1778 : /*
1779 : * Setup a scan key to find chunks in toast table with matching va_valueid
1780 : */
1781 23630 : ScanKeyInit(&toastkey,
1782 : (AttrNumber) 1,
1783 : BTEqualStrategyNumber, F_OIDEQ,
1784 : ObjectIdGetDatum(ta->toast_pointer.va_valueid));
1785 :
1786 : /*
1787 : * Check if any chunks for this toasted object exist in the toast table,
1788 : * accessible via the index.
1789 : */
1790 23630 : init_toast_snapshot(&SnapshotToast);
1791 23630 : toastscan = systable_beginscan_ordered(ctx->toast_rel,
1792 : ctx->valid_toast_index,
1793 : &SnapshotToast, 1,
1794 : &toastkey);
1795 23630 : found_toasttup = false;
1796 130392 : while ((toasttup =
1797 106768 : systable_getnext_ordered(toastscan,
1798 : ForwardScanDirection)) != NULL)
1799 : {
1800 83138 : found_toasttup = true;
1801 83138 : check_toast_tuple(toasttup, ctx, ta, &expected_chunk_seq, extsize);
1802 : }
1803 23624 : systable_endscan_ordered(toastscan);
1804 :
1805 23624 : if (!found_toasttup)
1806 2 : report_toast_corruption(ctx, ta,
1807 : psprintf("toast value %u not found in toast table",
1808 : ta->toast_pointer.va_valueid));
1809 23622 : else if (expected_chunk_seq <= last_chunk_seq)
1810 0 : report_toast_corruption(ctx, ta,
1811 : psprintf("toast value %u was expected to end at chunk %d, but ended while expecting chunk %d",
1812 : ta->toast_pointer.va_valueid,
1813 : last_chunk_seq, expected_chunk_seq));
1814 23624 : }
1815 :
1816 : /*
1817 : * Check the current tuple as tracked in ctx, recording any corruption found in
1818 : * ctx->tupstore.
1819 : *
1820 : * We return some information about the status of xmin to aid in validating
1821 : * update chains.
1822 : */
1823 : static void
1824 1065922 : check_tuple(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
1825 : XidCommitStatus *xmin_commit_status)
1826 : {
1827 : /*
1828 : * Check various forms of tuple header corruption, and if the header is
1829 : * too corrupt, do not continue with other checks.
1830 : */
1831 1065922 : if (!check_tuple_header(ctx))
1832 10 : return;
1833 :
1834 : /*
1835 : * Check tuple visibility. If the inserting transaction aborted, we
1836 : * cannot assume our relation description matches the tuple structure, and
1837 : * therefore cannot check it.
1838 : */
1839 1065912 : if (!check_tuple_visibility(ctx, xmin_commit_status_ok,
1840 : xmin_commit_status))
1841 24 : return;
1842 :
1843 : /*
1844 : * The tuple is visible, so it must be compatible with the current version
1845 : * of the relation descriptor. It might have fewer columns than are
1846 : * present in the relation descriptor, but it cannot have more.
1847 : */
1848 1065888 : if (RelationGetDescr(ctx->rel)->natts < ctx->natts)
1849 : {
1850 4 : report_corruption(ctx,
1851 : psprintf("number of attributes %u exceeds maximum expected for table %u",
1852 : ctx->natts,
1853 4 : RelationGetDescr(ctx->rel)->natts));
1854 4 : return;
1855 : }
1856 :
1857 : /*
1858 : * Check each attribute unless we hit corruption that confuses what to do
1859 : * next, at which point we abort further attribute checks for this tuple.
1860 : * Note that we don't abort for all types of corruption, only for those
1861 : * types where we don't know how to continue. We also don't abort the
1862 : * checking of toasted attributes collected from the tuple prior to
1863 : * aborting. Those will still be checked later along with other toasted
1864 : * attributes collected from the page.
1865 : */
1866 1065884 : ctx->offset = 0;
1867 16463632 : for (ctx->attnum = 0; ctx->attnum < ctx->natts; ctx->attnum++)
1868 15397750 : if (!check_tuple_attribute(ctx))
1869 2 : break; /* cannot continue */
1870 :
1871 : /* revert attnum to -1 until we again examine individual attributes */
1872 1065884 : ctx->attnum = -1;
1873 : }
1874 :
1875 : /*
1876 : * Convert a TransactionId into a FullTransactionId using our cached values of
1877 : * the valid transaction ID range. It is the caller's responsibility to have
1878 : * already updated the cached values, if necessary.
1879 : */
1880 : static FullTransactionId
1881 147530 : FullTransactionIdFromXidAndCtx(TransactionId xid, const HeapCheckContext *ctx)
1882 : {
1883 : uint64 nextfxid_i;
1884 : int32 diff;
1885 : FullTransactionId fxid;
1886 :
1887 : Assert(TransactionIdIsNormal(ctx->next_xid));
1888 : Assert(FullTransactionIdIsNormal(ctx->next_fxid));
1889 : Assert(XidFromFullTransactionId(ctx->next_fxid) == ctx->next_xid);
1890 :
1891 147530 : if (!TransactionIdIsNormal(xid))
1892 384 : return FullTransactionIdFromEpochAndXid(0, xid);
1893 :
1894 147146 : nextfxid_i = U64FromFullTransactionId(ctx->next_fxid);
1895 :
1896 : /* compute the 32bit modulo difference */
1897 147146 : diff = (int32) (ctx->next_xid - xid);
1898 :
1899 : /*
1900 : * In cases of corruption we might see a 32bit xid that is before epoch 0.
1901 : * We can't represent that as a 64bit xid, due to 64bit xids being
1902 : * unsigned integers, without the modulo arithmetic of 32bit xid. There's
1903 : * no really nice way to deal with that, but it works ok enough to use
1904 : * FirstNormalFullTransactionId in that case, as a freshly initdb'd
1905 : * cluster already has a newer horizon.
1906 : */
1907 147146 : if (diff > 0 && (nextfxid_i - FirstNormalTransactionId) < (int64) diff)
1908 : {
1909 : Assert(EpochFromFullTransactionId(ctx->next_fxid) == 0);
1910 8 : fxid = FirstNormalFullTransactionId;
1911 : }
1912 : else
1913 147138 : fxid = FullTransactionIdFromU64(nextfxid_i - diff);
1914 :
1915 : Assert(FullTransactionIdIsNormal(fxid));
1916 147146 : return fxid;
1917 : }
1918 :
1919 : /*
1920 : * Update our cached range of valid transaction IDs.
1921 : */
1922 : static void
1923 2960 : update_cached_xid_range(HeapCheckContext *ctx)
1924 : {
1925 : /* Make cached copies */
1926 2960 : LWLockAcquire(XidGenLock, LW_SHARED);
1927 2960 : ctx->next_fxid = TransamVariables->nextXid;
1928 2960 : ctx->oldest_xid = TransamVariables->oldestXid;
1929 2960 : LWLockRelease(XidGenLock);
1930 :
1931 : /* And compute alternate versions of the same */
1932 2960 : ctx->next_xid = XidFromFullTransactionId(ctx->next_fxid);
1933 2960 : ctx->oldest_fxid = FullTransactionIdFromXidAndCtx(ctx->oldest_xid, ctx);
1934 2960 : }
1935 :
1936 : /*
1937 : * Update our cached range of valid multitransaction IDs.
1938 : */
1939 : static void
1940 2956 : update_cached_mxid_range(HeapCheckContext *ctx)
1941 : {
1942 2956 : ReadMultiXactIdRange(&ctx->oldest_mxact, &ctx->next_mxact);
1943 2956 : }
1944 :
1945 : /*
1946 : * Return whether the given FullTransactionId is within our cached valid
1947 : * transaction ID range.
1948 : */
1949 : static inline bool
1950 122852 : fxid_in_cached_range(FullTransactionId fxid, const HeapCheckContext *ctx)
1951 : {
1952 245698 : return (FullTransactionIdPrecedesOrEquals(ctx->oldest_fxid, fxid) &&
1953 122846 : FullTransactionIdPrecedes(fxid, ctx->next_fxid));
1954 : }
1955 :
1956 : /*
1957 : * Checks whether a multitransaction ID is in the cached valid range, returning
1958 : * the nature of the range violation, if any.
1959 : */
1960 : static XidBoundsViolation
1961 120 : check_mxid_in_range(MultiXactId mxid, HeapCheckContext *ctx)
1962 : {
1963 120 : if (!TransactionIdIsValid(mxid))
1964 0 : return XID_INVALID;
1965 120 : if (MultiXactIdPrecedes(mxid, ctx->relminmxid))
1966 4 : return XID_PRECEDES_RELMIN;
1967 116 : if (MultiXactIdPrecedes(mxid, ctx->oldest_mxact))
1968 0 : return XID_PRECEDES_CLUSTERMIN;
1969 116 : if (MultiXactIdPrecedesOrEquals(ctx->next_mxact, mxid))
1970 4 : return XID_IN_FUTURE;
1971 112 : return XID_BOUNDS_OK;
1972 : }
1973 :
1974 : /*
1975 : * Checks whether the given mxid is valid to appear in the heap being checked,
1976 : * returning the nature of the range violation, if any.
1977 : *
1978 : * This function attempts to return quickly by caching the known valid mxid
1979 : * range in ctx. Callers should already have performed the initial setup of
1980 : * the cache prior to the first call to this function.
1981 : */
1982 : static XidBoundsViolation
1983 116 : check_mxid_valid_in_rel(MultiXactId mxid, HeapCheckContext *ctx)
1984 : {
1985 : XidBoundsViolation result;
1986 :
1987 116 : result = check_mxid_in_range(mxid, ctx);
1988 116 : if (result == XID_BOUNDS_OK)
1989 112 : return XID_BOUNDS_OK;
1990 :
1991 : /* The range may have advanced. Recheck. */
1992 4 : update_cached_mxid_range(ctx);
1993 4 : return check_mxid_in_range(mxid, ctx);
1994 : }
1995 :
1996 : /*
1997 : * Checks whether the given transaction ID is (or was recently) valid to appear
1998 : * in the heap being checked, or whether it is too old or too new to appear in
1999 : * the relation, returning information about the nature of the bounds violation.
2000 : *
2001 : * We cache the range of valid transaction IDs. If xid is in that range, we
2002 : * conclude that it is valid, even though concurrent changes to the table might
2003 : * invalidate it under certain corrupt conditions. (For example, if the table
2004 : * contains corrupt all-frozen bits, a concurrent vacuum might skip the page(s)
2005 : * containing the xid and then truncate clog and advance the relfrozenxid
2006 : * beyond xid.) Reporting the xid as valid under such conditions seems
2007 : * acceptable, since if we had checked it earlier in our scan it would have
2008 : * truly been valid at that time.
2009 : *
2010 : * If the status argument is not NULL, and if and only if the transaction ID
2011 : * appears to be valid in this relation, the status argument will be set with
2012 : * the commit status of the transaction ID.
2013 : */
2014 : static XidBoundsViolation
2015 1066308 : get_xid_status(TransactionId xid, HeapCheckContext *ctx,
2016 : XidCommitStatus *status)
2017 : {
2018 : FullTransactionId fxid;
2019 : FullTransactionId clog_horizon;
2020 :
2021 : /* Quick check for special xids */
2022 1066308 : if (!TransactionIdIsValid(xid))
2023 2 : return XID_INVALID;
2024 1066306 : else if (xid == BootstrapTransactionId || xid == FrozenTransactionId)
2025 : {
2026 943454 : if (status != NULL)
2027 943454 : *status = XID_COMMITTED;
2028 943454 : return XID_BOUNDS_OK;
2029 : }
2030 :
2031 : /* Check if the xid is within bounds */
2032 122852 : fxid = FullTransactionIdFromXidAndCtx(xid, ctx);
2033 122852 : if (!fxid_in_cached_range(fxid, ctx))
2034 : {
2035 : /*
2036 : * We may have been checking against stale values. Update the cached
2037 : * range to be sure, and since we relied on the cached range when we
2038 : * performed the full xid conversion, reconvert.
2039 : */
2040 8 : update_cached_xid_range(ctx);
2041 8 : fxid = FullTransactionIdFromXidAndCtx(xid, ctx);
2042 : }
2043 :
2044 122852 : if (FullTransactionIdPrecedesOrEquals(ctx->next_fxid, fxid))
2045 2 : return XID_IN_FUTURE;
2046 122850 : if (FullTransactionIdPrecedes(fxid, ctx->oldest_fxid))
2047 6 : return XID_PRECEDES_CLUSTERMIN;
2048 122844 : if (FullTransactionIdPrecedes(fxid, ctx->relfrozenfxid))
2049 2 : return XID_PRECEDES_RELMIN;
2050 :
2051 : /* Early return if the caller does not request clog checking */
2052 122842 : if (status == NULL)
2053 0 : return XID_BOUNDS_OK;
2054 :
2055 : /* Early return if we just checked this xid in a prior call */
2056 122842 : if (xid == ctx->cached_xid)
2057 : {
2058 104084 : *status = ctx->cached_status;
2059 104084 : return XID_BOUNDS_OK;
2060 : }
2061 :
2062 18758 : *status = XID_COMMITTED;
2063 18758 : LWLockAcquire(XactTruncationLock, LW_SHARED);
2064 : clog_horizon =
2065 18758 : FullTransactionIdFromXidAndCtx(TransamVariables->oldestClogXid,
2066 : ctx);
2067 18758 : if (FullTransactionIdPrecedesOrEquals(clog_horizon, fxid))
2068 : {
2069 18758 : if (TransactionIdIsCurrentTransactionId(xid))
2070 0 : *status = XID_IS_CURRENT_XID;
2071 18758 : else if (TransactionIdIsInProgress(xid))
2072 4 : *status = XID_IN_PROGRESS;
2073 18754 : else if (TransactionIdDidCommit(xid))
2074 18740 : *status = XID_COMMITTED;
2075 : else
2076 14 : *status = XID_ABORTED;
2077 : }
2078 18758 : LWLockRelease(XactTruncationLock);
2079 18758 : ctx->cached_xid = xid;
2080 18758 : ctx->cached_status = *status;
2081 18758 : return XID_BOUNDS_OK;
2082 : }
|
