Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * verify_heapam.c
4 : * Functions to check postgresql heap relations for corruption
5 : *
6 : * Copyright (c) 2016-2026, 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/heaptoast.h"
16 : #include "access/multixact.h"
17 : #include "access/relation.h"
18 : #include "access/table.h"
19 : #include "access/toast_internals.h"
20 : #include "access/visibilitymap.h"
21 : #include "access/xact.h"
22 : #include "catalog/pg_am.h"
23 : #include "catalog/pg_class.h"
24 : #include "funcapi.h"
25 : #include "miscadmin.h"
26 : #include "storage/bufmgr.h"
27 : #include "storage/procarray.h"
28 : #include "storage/read_stream.h"
29 : #include "utils/builtins.h"
30 : #include "utils/fmgroids.h"
31 : #include "utils/rel.h"
32 :
33 0 : PG_FUNCTION_INFO_V1(verify_heapam);
34 :
35 : /* The number of columns in tuples returned by verify_heapam */
36 : #define HEAPCHECK_RELATION_COLS 4
37 :
38 : /* The largest valid toast va_rawsize */
39 : #define VARLENA_SIZE_LIMIT 0x3FFFFFFF
40 :
41 : /*
42 : * Despite the name, we use this for reporting problems with both XIDs and
43 : * MXIDs.
44 : */
45 : typedef enum XidBoundsViolation
46 : {
47 : XID_INVALID,
48 : XID_IN_FUTURE,
49 : XID_PRECEDES_CLUSTERMIN,
50 : XID_PRECEDES_RELMIN,
51 : XID_BOUNDS_OK,
52 : } XidBoundsViolation;
53 :
54 : typedef enum XidCommitStatus
55 : {
56 : XID_COMMITTED,
57 : XID_IS_CURRENT_XID,
58 : XID_IN_PROGRESS,
59 : XID_ABORTED,
60 : } XidCommitStatus;
61 :
62 : typedef enum SkipPages
63 : {
64 : SKIP_PAGES_ALL_FROZEN,
65 : SKIP_PAGES_ALL_VISIBLE,
66 : SKIP_PAGES_NONE,
67 : } SkipPages;
68 :
69 : /*
70 : * Struct holding information about a toasted attribute sufficient to both
71 : * check the toasted attribute and, if found to be corrupt, to report where it
72 : * was encountered in the main table.
73 : */
74 : typedef struct ToastedAttribute
75 : {
76 : struct varatt_external toast_pointer;
77 : BlockNumber blkno; /* block in main table */
78 : OffsetNumber offnum; /* offset in main table */
79 : AttrNumber attnum; /* attribute in main table */
80 : } ToastedAttribute;
81 :
82 : /*
83 : * Struct holding the running context information during
84 : * a lifetime of a verify_heapam execution.
85 : */
86 : typedef struct HeapCheckContext
87 : {
88 : /*
89 : * Cached copies of values from TransamVariables and computed values from
90 : * them.
91 : */
92 : FullTransactionId next_fxid; /* TransamVariables->nextXid */
93 : TransactionId next_xid; /* 32-bit version of next_fxid */
94 : TransactionId oldest_xid; /* TransamVariables->oldestXid */
95 : FullTransactionId oldest_fxid; /* 64-bit version of oldest_xid, computed
96 : * relative to next_fxid */
97 : TransactionId safe_xmin; /* this XID and newer ones can't become
98 : * all-visible while we're running */
99 :
100 : /*
101 : * Cached copy of value from MultiXactState
102 : */
103 : MultiXactId next_mxact; /* MultiXactState->nextMXact */
104 : MultiXactId oldest_mxact; /* MultiXactState->oldestMultiXactId */
105 :
106 : /*
107 : * Cached copies of the most recently checked xid and its status.
108 : */
109 : TransactionId cached_xid;
110 : XidCommitStatus cached_status;
111 :
112 : /* Values concerning the heap relation being checked */
113 : Relation rel;
114 : TransactionId relfrozenxid;
115 : FullTransactionId relfrozenfxid;
116 : TransactionId relminmxid;
117 : Relation toast_rel;
118 : Relation *toast_indexes;
119 : Relation valid_toast_index;
120 : int num_toast_indexes;
121 :
122 : /*
123 : * Values for iterating over pages in the relation. `blkno` is the most
124 : * recent block in the buffer yielded by the read stream API.
125 : */
126 : BlockNumber blkno;
127 : BufferAccessStrategy bstrategy;
128 : Buffer buffer;
129 : Page page;
130 :
131 : /* Values for iterating over tuples within a page */
132 : OffsetNumber offnum;
133 : ItemId itemid;
134 : uint16 lp_len;
135 : uint16 lp_off;
136 : HeapTupleHeader tuphdr;
137 : int natts;
138 :
139 : /* Values for iterating over attributes within the tuple */
140 : uint32 offset; /* offset in tuple data */
141 : AttrNumber attnum;
142 :
143 : /* True if tuple's xmax makes it eligible for pruning */
144 : bool tuple_could_be_pruned;
145 :
146 : /*
147 : * List of ToastedAttribute structs for toasted attributes which are not
148 : * eligible for pruning and should be checked
149 : */
150 : List *toasted_attributes;
151 :
152 : /* Whether verify_heapam has yet encountered any corrupt tuples */
153 : bool is_corrupt;
154 :
155 : /* The descriptor and tuplestore for verify_heapam's result tuples */
156 : TupleDesc tupdesc;
157 : Tuplestorestate *tupstore;
158 : } HeapCheckContext;
159 :
160 : /*
161 : * The per-relation data provided to the read stream API for heap amcheck to
162 : * use in its callback for the SKIP_PAGES_ALL_FROZEN and
163 : * SKIP_PAGES_ALL_VISIBLE options.
164 : */
165 : typedef struct HeapCheckReadStreamData
166 : {
167 : /*
168 : * `range` is used by all SkipPages options. SKIP_PAGES_NONE uses the
169 : * default read stream callback, block_range_read_stream_cb(), which takes
170 : * a BlockRangeReadStreamPrivate as its callback_private_data. `range`
171 : * keeps track of the current block number across
172 : * read_stream_next_buffer() invocations.
173 : */
174 : BlockRangeReadStreamPrivate range;
175 : SkipPages skip_option;
176 : Relation rel;
177 : Buffer *vmbuffer;
178 : } HeapCheckReadStreamData;
179 :
180 :
181 : /* Internal implementation */
182 : static BlockNumber heapcheck_read_stream_next_unskippable(ReadStream *stream,
183 : void *callback_private_data,
184 : void *per_buffer_data);
185 :
186 : static void check_tuple(HeapCheckContext *ctx,
187 : bool *xmin_commit_status_ok,
188 : XidCommitStatus *xmin_commit_status);
189 : static void check_toast_tuple(HeapTuple toasttup, HeapCheckContext *ctx,
190 : ToastedAttribute *ta, int32 *expected_chunk_seq,
191 : uint32 extsize);
192 :
193 : static bool check_tuple_attribute(HeapCheckContext *ctx);
194 : static void check_toasted_attribute(HeapCheckContext *ctx,
195 : ToastedAttribute *ta);
196 :
197 : static bool check_tuple_header(HeapCheckContext *ctx);
198 : static bool check_tuple_visibility(HeapCheckContext *ctx,
199 : bool *xmin_commit_status_ok,
200 : XidCommitStatus *xmin_commit_status);
201 :
202 : static void report_corruption(HeapCheckContext *ctx, char *msg);
203 : static void report_toast_corruption(HeapCheckContext *ctx,
204 : ToastedAttribute *ta, char *msg);
205 : static FullTransactionId FullTransactionIdFromXidAndCtx(TransactionId xid,
206 : const HeapCheckContext *ctx);
207 : static void update_cached_xid_range(HeapCheckContext *ctx);
208 : static void update_cached_mxid_range(HeapCheckContext *ctx);
209 : static XidBoundsViolation check_mxid_in_range(MultiXactId mxid,
210 : HeapCheckContext *ctx);
211 : static XidBoundsViolation check_mxid_valid_in_rel(MultiXactId mxid,
212 : HeapCheckContext *ctx);
213 : static XidBoundsViolation get_xid_status(TransactionId xid,
214 : HeapCheckContext *ctx,
215 : XidCommitStatus *status);
216 :
217 : /*
218 : * Scan and report corruption in heap pages, optionally reconciling toasted
219 : * attributes with entries in the associated toast table. Intended to be
220 : * called from SQL with the following parameters:
221 : *
222 : * relation:
223 : * The Oid of the heap relation to be checked.
224 : *
225 : * on_error_stop:
226 : * Whether to stop at the end of the first page for which errors are
227 : * detected. Note that multiple rows may be returned.
228 : *
229 : * check_toast:
230 : * Whether to check each toasted attribute against the toast table to
231 : * verify that it can be found there.
232 : *
233 : * skip:
234 : * What kinds of pages in the heap relation should be skipped. Valid
235 : * options are "all-visible", "all-frozen", and "none".
236 : *
237 : * Returns to the SQL caller a set of tuples, each containing the location
238 : * and a description of a corruption found in the heap.
239 : *
240 : * This code goes to some trouble to avoid crashing the server even if the
241 : * table pages are badly corrupted, but it's probably not perfect. If
242 : * check_toast is true, we'll use regular index lookups to try to fetch TOAST
243 : * tuples, which can certainly cause crashes if the right kind of corruption
244 : * exists in the toast table or index. No matter what parameters you pass,
245 : * we can't protect against crashes that might occur trying to look up the
246 : * commit status of transaction IDs (though we avoid trying to do such lookups
247 : * for transaction IDs that can't legally appear in the table).
248 : */
249 : Datum
250 0 : verify_heapam(PG_FUNCTION_ARGS)
251 : {
252 0 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
253 0 : HeapCheckContext ctx;
254 0 : Buffer vmbuffer = InvalidBuffer;
255 0 : Oid relid;
256 0 : bool on_error_stop;
257 0 : bool check_toast;
258 0 : SkipPages skip_option = SKIP_PAGES_NONE;
259 0 : BlockNumber first_block;
260 0 : BlockNumber last_block;
261 0 : BlockNumber nblocks;
262 0 : const char *skip;
263 0 : ReadStream *stream;
264 0 : int stream_flags;
265 0 : ReadStreamBlockNumberCB stream_cb;
266 0 : void *stream_data;
267 0 : HeapCheckReadStreamData stream_skip_data;
268 :
269 : /* Check supplied arguments */
270 0 : if (PG_ARGISNULL(0))
271 0 : ereport(ERROR,
272 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
273 : errmsg("relation cannot be null")));
274 0 : relid = PG_GETARG_OID(0);
275 :
276 0 : if (PG_ARGISNULL(1))
277 0 : ereport(ERROR,
278 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
279 : errmsg("on_error_stop cannot be null")));
280 0 : on_error_stop = PG_GETARG_BOOL(1);
281 :
282 0 : if (PG_ARGISNULL(2))
283 0 : ereport(ERROR,
284 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
285 : errmsg("check_toast cannot be null")));
286 0 : check_toast = PG_GETARG_BOOL(2);
287 :
288 0 : if (PG_ARGISNULL(3))
289 0 : ereport(ERROR,
290 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
291 : errmsg("skip cannot be null")));
292 0 : skip = text_to_cstring(PG_GETARG_TEXT_PP(3));
293 0 : if (pg_strcasecmp(skip, "all-visible") == 0)
294 0 : skip_option = SKIP_PAGES_ALL_VISIBLE;
295 0 : else if (pg_strcasecmp(skip, "all-frozen") == 0)
296 0 : skip_option = SKIP_PAGES_ALL_FROZEN;
297 0 : else if (pg_strcasecmp(skip, "none") == 0)
298 0 : skip_option = SKIP_PAGES_NONE;
299 : else
300 0 : ereport(ERROR,
301 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
302 : errmsg("invalid skip option"),
303 : errhint("Valid skip options are \"all-visible\", \"all-frozen\", and \"none\".")));
304 :
305 0 : memset(&ctx, 0, sizeof(HeapCheckContext));
306 0 : ctx.cached_xid = InvalidTransactionId;
307 0 : ctx.toasted_attributes = NIL;
308 :
309 : /*
310 : * Any xmin newer than the xmin of our snapshot can't become all-visible
311 : * while we're running.
312 : */
313 0 : ctx.safe_xmin = GetTransactionSnapshot()->xmin;
314 :
315 : /*
316 : * If we report corruption when not examining some individual attribute,
317 : * we need attnum to be reported as NULL. Set that up before any
318 : * corruption reporting might happen.
319 : */
320 0 : ctx.attnum = -1;
321 :
322 : /* Construct the tuplestore and tuple descriptor */
323 0 : InitMaterializedSRF(fcinfo, 0);
324 0 : ctx.tupdesc = rsinfo->setDesc;
325 0 : ctx.tupstore = rsinfo->setResult;
326 :
327 : /* Open relation, check relkind and access method */
328 0 : ctx.rel = relation_open(relid, AccessShareLock);
329 :
330 : /*
331 : * Check that a relation's relkind and access method are both supported.
332 : */
333 0 : if (!RELKIND_HAS_TABLE_AM(ctx.rel->rd_rel->relkind) &&
334 0 : ctx.rel->rd_rel->relkind != RELKIND_SEQUENCE)
335 0 : ereport(ERROR,
336 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
337 : errmsg("cannot check relation \"%s\"",
338 : RelationGetRelationName(ctx.rel)),
339 : errdetail_relkind_not_supported(ctx.rel->rd_rel->relkind)));
340 :
341 : /*
342 : * Sequences always use heap AM, but they don't show that in the catalogs.
343 : * Other relkinds might be using a different AM, so check.
344 : */
345 0 : if (ctx.rel->rd_rel->relkind != RELKIND_SEQUENCE &&
346 0 : ctx.rel->rd_rel->relam != HEAP_TABLE_AM_OID)
347 0 : ereport(ERROR,
348 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
349 : errmsg("only heap AM is supported")));
350 :
351 : /*
352 : * Early exit for unlogged relations during recovery. These will have no
353 : * relation fork, so there won't be anything to check. We behave as if
354 : * the relation is empty.
355 : */
356 0 : if (ctx.rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED &&
357 0 : RecoveryInProgress())
358 : {
359 0 : ereport(DEBUG1,
360 : (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
361 : errmsg("cannot verify unlogged relation \"%s\" during recovery, skipping",
362 : RelationGetRelationName(ctx.rel))));
363 0 : relation_close(ctx.rel, AccessShareLock);
364 0 : PG_RETURN_NULL();
365 0 : }
366 :
367 : /* Early exit if the relation is empty */
368 0 : nblocks = RelationGetNumberOfBlocks(ctx.rel);
369 0 : if (!nblocks)
370 : {
371 0 : relation_close(ctx.rel, AccessShareLock);
372 0 : PG_RETURN_NULL();
373 0 : }
374 :
375 0 : ctx.bstrategy = GetAccessStrategy(BAS_BULKREAD);
376 0 : ctx.buffer = InvalidBuffer;
377 0 : ctx.page = NULL;
378 :
379 : /* Validate block numbers, or handle nulls. */
380 0 : if (PG_ARGISNULL(4))
381 0 : first_block = 0;
382 : else
383 : {
384 0 : int64 fb = PG_GETARG_INT64(4);
385 :
386 0 : if (fb < 0 || fb >= nblocks)
387 0 : ereport(ERROR,
388 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
389 : errmsg("starting block number must be between 0 and %u",
390 : nblocks - 1)));
391 0 : first_block = (BlockNumber) fb;
392 0 : }
393 0 : if (PG_ARGISNULL(5))
394 0 : last_block = nblocks - 1;
395 : else
396 : {
397 0 : int64 lb = PG_GETARG_INT64(5);
398 :
399 0 : if (lb < 0 || lb >= nblocks)
400 0 : ereport(ERROR,
401 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
402 : errmsg("ending block number must be between 0 and %u",
403 : nblocks - 1)));
404 0 : last_block = (BlockNumber) lb;
405 0 : }
406 :
407 : /* Optionally open the toast relation, if any. */
408 0 : if (ctx.rel->rd_rel->reltoastrelid && check_toast)
409 : {
410 0 : int offset;
411 :
412 : /* Main relation has associated toast relation */
413 0 : ctx.toast_rel = table_open(ctx.rel->rd_rel->reltoastrelid,
414 : AccessShareLock);
415 0 : offset = toast_open_indexes(ctx.toast_rel,
416 : AccessShareLock,
417 0 : &(ctx.toast_indexes),
418 0 : &(ctx.num_toast_indexes));
419 0 : ctx.valid_toast_index = ctx.toast_indexes[offset];
420 0 : }
421 : else
422 : {
423 : /*
424 : * Main relation has no associated toast relation, or we're
425 : * intentionally skipping it.
426 : */
427 0 : ctx.toast_rel = NULL;
428 0 : ctx.toast_indexes = NULL;
429 0 : ctx.num_toast_indexes = 0;
430 : }
431 :
432 0 : update_cached_xid_range(&ctx);
433 0 : update_cached_mxid_range(&ctx);
434 0 : ctx.relfrozenxid = ctx.rel->rd_rel->relfrozenxid;
435 0 : ctx.relfrozenfxid = FullTransactionIdFromXidAndCtx(ctx.relfrozenxid, &ctx);
436 0 : ctx.relminmxid = ctx.rel->rd_rel->relminmxid;
437 :
438 0 : if (TransactionIdIsNormal(ctx.relfrozenxid))
439 0 : ctx.oldest_xid = ctx.relfrozenxid;
440 :
441 : /* Now that `ctx` is set up, set up the read stream */
442 0 : stream_skip_data.range.current_blocknum = first_block;
443 0 : stream_skip_data.range.last_exclusive = last_block + 1;
444 0 : stream_skip_data.skip_option = skip_option;
445 0 : stream_skip_data.rel = ctx.rel;
446 0 : stream_skip_data.vmbuffer = &vmbuffer;
447 :
448 0 : if (skip_option == SKIP_PAGES_NONE)
449 : {
450 : /*
451 : * It is safe to use batchmode as block_range_read_stream_cb takes no
452 : * locks.
453 : */
454 0 : stream_cb = block_range_read_stream_cb;
455 0 : stream_flags = READ_STREAM_SEQUENTIAL |
456 : READ_STREAM_FULL |
457 : READ_STREAM_USE_BATCHING;
458 0 : stream_data = &stream_skip_data.range;
459 0 : }
460 : else
461 : {
462 : /*
463 : * It would not be safe to naively use batchmode, as
464 : * heapcheck_read_stream_next_unskippable takes locks. It shouldn't be
465 : * too hard to convert though.
466 : */
467 0 : stream_cb = heapcheck_read_stream_next_unskippable;
468 0 : stream_flags = READ_STREAM_DEFAULT;
469 0 : stream_data = &stream_skip_data;
470 : }
471 :
472 0 : stream = read_stream_begin_relation(stream_flags,
473 0 : ctx.bstrategy,
474 0 : ctx.rel,
475 : MAIN_FORKNUM,
476 0 : stream_cb,
477 0 : stream_data,
478 : 0);
479 :
480 0 : while ((ctx.buffer = read_stream_next_buffer(stream, NULL)) != InvalidBuffer)
481 : {
482 0 : OffsetNumber maxoff;
483 0 : OffsetNumber predecessor[MaxOffsetNumber];
484 0 : OffsetNumber successor[MaxOffsetNumber];
485 0 : bool lp_valid[MaxOffsetNumber];
486 0 : bool xmin_commit_status_ok[MaxOffsetNumber];
487 0 : XidCommitStatus xmin_commit_status[MaxOffsetNumber];
488 :
489 0 : CHECK_FOR_INTERRUPTS();
490 :
491 0 : memset(predecessor, 0, sizeof(OffsetNumber) * MaxOffsetNumber);
492 :
493 : /* Lock the next page. */
494 0 : Assert(BufferIsValid(ctx.buffer));
495 0 : LockBuffer(ctx.buffer, BUFFER_LOCK_SHARE);
496 :
497 0 : ctx.blkno = BufferGetBlockNumber(ctx.buffer);
498 0 : ctx.page = BufferGetPage(ctx.buffer);
499 :
500 : /* Perform tuple checks */
501 0 : maxoff = PageGetMaxOffsetNumber(ctx.page);
502 0 : for (ctx.offnum = FirstOffsetNumber; ctx.offnum <= maxoff;
503 0 : ctx.offnum = OffsetNumberNext(ctx.offnum))
504 : {
505 0 : BlockNumber nextblkno;
506 0 : OffsetNumber nextoffnum;
507 :
508 0 : successor[ctx.offnum] = InvalidOffsetNumber;
509 0 : lp_valid[ctx.offnum] = false;
510 0 : xmin_commit_status_ok[ctx.offnum] = false;
511 0 : ctx.itemid = PageGetItemId(ctx.page, ctx.offnum);
512 :
513 : /* Skip over unused/dead line pointers */
514 0 : if (!ItemIdIsUsed(ctx.itemid) || ItemIdIsDead(ctx.itemid))
515 0 : continue;
516 :
517 : /*
518 : * If this line pointer has been redirected, check that it
519 : * redirects to a valid offset within the line pointer array
520 : */
521 0 : if (ItemIdIsRedirected(ctx.itemid))
522 : {
523 0 : OffsetNumber rdoffnum = ItemIdGetRedirect(ctx.itemid);
524 0 : ItemId rditem;
525 :
526 0 : if (rdoffnum < FirstOffsetNumber)
527 : {
528 0 : report_corruption(&ctx,
529 0 : psprintf("line pointer redirection to item at offset %d precedes minimum offset %d",
530 0 : rdoffnum,
531 : FirstOffsetNumber));
532 0 : continue;
533 : }
534 0 : if (rdoffnum > maxoff)
535 : {
536 0 : report_corruption(&ctx,
537 0 : psprintf("line pointer redirection to item at offset %d exceeds maximum offset %d",
538 0 : rdoffnum,
539 0 : maxoff));
540 0 : continue;
541 : }
542 :
543 : /*
544 : * Since we've checked that this redirect points to a line
545 : * pointer between FirstOffsetNumber and maxoff, it should now
546 : * be safe to fetch the referenced line pointer. We expect it
547 : * to be LP_NORMAL; if not, that's corruption.
548 : */
549 0 : rditem = PageGetItemId(ctx.page, rdoffnum);
550 0 : if (!ItemIdIsUsed(rditem))
551 : {
552 0 : report_corruption(&ctx,
553 0 : psprintf("redirected line pointer points to an unused item at offset %d",
554 0 : rdoffnum));
555 0 : continue;
556 : }
557 0 : else if (ItemIdIsDead(rditem))
558 : {
559 0 : report_corruption(&ctx,
560 0 : psprintf("redirected line pointer points to a dead item at offset %d",
561 0 : rdoffnum));
562 0 : continue;
563 : }
564 0 : else if (ItemIdIsRedirected(rditem))
565 : {
566 0 : report_corruption(&ctx,
567 0 : psprintf("redirected line pointer points to another redirected line pointer at offset %d",
568 0 : rdoffnum));
569 0 : continue;
570 : }
571 :
572 : /*
573 : * Record the fact that this line pointer has passed basic
574 : * sanity checking, and also the offset number to which it
575 : * points.
576 : */
577 0 : lp_valid[ctx.offnum] = true;
578 0 : successor[ctx.offnum] = rdoffnum;
579 0 : continue;
580 0 : }
581 :
582 : /* Sanity-check the line pointer's offset and length values */
583 0 : ctx.lp_len = ItemIdGetLength(ctx.itemid);
584 0 : ctx.lp_off = ItemIdGetOffset(ctx.itemid);
585 :
586 0 : if (ctx.lp_off != MAXALIGN(ctx.lp_off))
587 : {
588 0 : report_corruption(&ctx,
589 0 : psprintf("line pointer to page offset %u is not maximally aligned",
590 0 : ctx.lp_off));
591 0 : continue;
592 : }
593 0 : if (ctx.lp_len < MAXALIGN(SizeofHeapTupleHeader))
594 : {
595 0 : report_corruption(&ctx,
596 0 : psprintf("line pointer length %u is less than the minimum tuple header size %u",
597 0 : ctx.lp_len,
598 : (unsigned) MAXALIGN(SizeofHeapTupleHeader)));
599 0 : continue;
600 : }
601 0 : if (ctx.lp_off + ctx.lp_len > BLCKSZ)
602 : {
603 0 : report_corruption(&ctx,
604 0 : psprintf("line pointer to page offset %u with length %u ends beyond maximum page offset %d",
605 0 : ctx.lp_off,
606 0 : ctx.lp_len,
607 : BLCKSZ));
608 0 : continue;
609 : }
610 :
611 : /* It should be safe to examine the tuple's header, at least */
612 0 : lp_valid[ctx.offnum] = true;
613 0 : ctx.tuphdr = (HeapTupleHeader) PageGetItem(ctx.page, ctx.itemid);
614 0 : ctx.natts = HeapTupleHeaderGetNatts(ctx.tuphdr);
615 :
616 : /* Ok, ready to check this next tuple */
617 0 : check_tuple(&ctx,
618 0 : &xmin_commit_status_ok[ctx.offnum],
619 0 : &xmin_commit_status[ctx.offnum]);
620 :
621 : /*
622 : * If the CTID field of this tuple seems to point to another tuple
623 : * on the same page, record that tuple as the successor of this
624 : * one.
625 : */
626 0 : nextblkno = ItemPointerGetBlockNumber(&(ctx.tuphdr)->t_ctid);
627 0 : nextoffnum = ItemPointerGetOffsetNumber(&(ctx.tuphdr)->t_ctid);
628 0 : if (nextblkno == ctx.blkno && nextoffnum != ctx.offnum &&
629 0 : nextoffnum >= FirstOffsetNumber && nextoffnum <= maxoff)
630 0 : successor[ctx.offnum] = nextoffnum;
631 0 : }
632 :
633 : /*
634 : * Update chain validation. Check each line pointer that's got a valid
635 : * successor against that successor.
636 : */
637 0 : ctx.attnum = -1;
638 0 : for (ctx.offnum = FirstOffsetNumber; ctx.offnum <= maxoff;
639 0 : ctx.offnum = OffsetNumberNext(ctx.offnum))
640 : {
641 0 : ItemId curr_lp;
642 0 : ItemId next_lp;
643 0 : HeapTupleHeader curr_htup;
644 0 : HeapTupleHeader next_htup;
645 0 : TransactionId curr_xmin;
646 0 : TransactionId curr_xmax;
647 0 : TransactionId next_xmin;
648 0 : OffsetNumber nextoffnum = successor[ctx.offnum];
649 :
650 : /*
651 : * The current line pointer may not have a successor, either
652 : * because it's not valid or because it didn't point to anything.
653 : * In either case, we have to give up.
654 : *
655 : * If the current line pointer does point to something, it's
656 : * possible that the target line pointer isn't valid. We have to
657 : * give up in that case, too.
658 : */
659 0 : if (nextoffnum == InvalidOffsetNumber || !lp_valid[nextoffnum])
660 0 : continue;
661 :
662 : /* We have two valid line pointers that we can examine. */
663 0 : curr_lp = PageGetItemId(ctx.page, ctx.offnum);
664 0 : next_lp = PageGetItemId(ctx.page, nextoffnum);
665 :
666 : /* Handle the cases where the current line pointer is a redirect. */
667 0 : if (ItemIdIsRedirected(curr_lp))
668 : {
669 : /*
670 : * We should not have set successor[ctx.offnum] to a value
671 : * other than InvalidOffsetNumber unless that line pointer is
672 : * LP_NORMAL.
673 : */
674 0 : Assert(ItemIdIsNormal(next_lp));
675 :
676 : /* Can only redirect to a HOT tuple. */
677 0 : next_htup = (HeapTupleHeader) PageGetItem(ctx.page, next_lp);
678 0 : if (!HeapTupleHeaderIsHeapOnly(next_htup))
679 : {
680 0 : report_corruption(&ctx,
681 0 : psprintf("redirected line pointer points to a non-heap-only tuple at offset %d",
682 0 : nextoffnum));
683 0 : }
684 :
685 : /* HOT chains should not intersect. */
686 0 : if (predecessor[nextoffnum] != InvalidOffsetNumber)
687 : {
688 0 : report_corruption(&ctx,
689 0 : psprintf("redirect line pointer points to offset %d, but offset %d also points there",
690 0 : nextoffnum, predecessor[nextoffnum]));
691 0 : continue;
692 : }
693 :
694 : /*
695 : * This redirect and the tuple to which it points seem to be
696 : * part of an update chain.
697 : */
698 0 : predecessor[nextoffnum] = ctx.offnum;
699 0 : continue;
700 : }
701 :
702 : /*
703 : * If the next line pointer is a redirect, or if it's a tuple but
704 : * the XMAX of this tuple doesn't match the XMIN of the next
705 : * tuple, then the two aren't part of the same update chain and
706 : * there is nothing more to do.
707 : */
708 0 : if (ItemIdIsRedirected(next_lp))
709 0 : continue;
710 0 : curr_htup = (HeapTupleHeader) PageGetItem(ctx.page, curr_lp);
711 0 : curr_xmax = HeapTupleHeaderGetUpdateXid(curr_htup);
712 0 : next_htup = (HeapTupleHeader) PageGetItem(ctx.page, next_lp);
713 0 : next_xmin = HeapTupleHeaderGetXmin(next_htup);
714 0 : if (!TransactionIdIsValid(curr_xmax) ||
715 0 : !TransactionIdEquals(curr_xmax, next_xmin))
716 0 : continue;
717 :
718 : /* HOT chains should not intersect. */
719 0 : if (predecessor[nextoffnum] != InvalidOffsetNumber)
720 : {
721 0 : report_corruption(&ctx,
722 0 : psprintf("tuple points to new version at offset %d, but offset %d also points there",
723 0 : nextoffnum, predecessor[nextoffnum]));
724 0 : continue;
725 : }
726 :
727 : /*
728 : * This tuple and the tuple to which it points seem to be part of
729 : * an update chain.
730 : */
731 0 : predecessor[nextoffnum] = ctx.offnum;
732 :
733 : /*
734 : * If the current tuple is marked as HOT-updated, then the next
735 : * tuple should be marked as a heap-only tuple. Conversely, if the
736 : * current tuple isn't marked as HOT-updated, then the next tuple
737 : * shouldn't be marked as a heap-only tuple.
738 : *
739 : * NB: Can't use HeapTupleHeaderIsHotUpdated() as it checks if
740 : * hint bits indicate xmin/xmax aborted.
741 : */
742 0 : if (!(curr_htup->t_infomask2 & HEAP_HOT_UPDATED) &&
743 0 : HeapTupleHeaderIsHeapOnly(next_htup))
744 : {
745 0 : report_corruption(&ctx,
746 0 : psprintf("non-heap-only update produced a heap-only tuple at offset %d",
747 0 : nextoffnum));
748 0 : }
749 0 : if ((curr_htup->t_infomask2 & HEAP_HOT_UPDATED) &&
750 0 : !HeapTupleHeaderIsHeapOnly(next_htup))
751 : {
752 0 : report_corruption(&ctx,
753 0 : psprintf("heap-only update produced a non-heap only tuple at offset %d",
754 0 : nextoffnum));
755 0 : }
756 :
757 : /*
758 : * If the current tuple's xmin is still in progress but the
759 : * successor tuple's xmin is committed, that's corruption.
760 : *
761 : * NB: We recheck the commit status of the current tuple's xmin
762 : * here, because it might have committed after we checked it and
763 : * before we checked the commit status of the successor tuple's
764 : * xmin. This should be safe because the xmin itself can't have
765 : * changed, only its commit status.
766 : */
767 0 : curr_xmin = HeapTupleHeaderGetXmin(curr_htup);
768 0 : if (xmin_commit_status_ok[ctx.offnum] &&
769 0 : xmin_commit_status[ctx.offnum] == XID_IN_PROGRESS &&
770 0 : xmin_commit_status_ok[nextoffnum] &&
771 0 : xmin_commit_status[nextoffnum] == XID_COMMITTED &&
772 0 : TransactionIdIsInProgress(curr_xmin))
773 : {
774 0 : report_corruption(&ctx,
775 0 : psprintf("tuple with in-progress xmin %u was updated to produce a tuple at offset %d with committed xmin %u",
776 0 : curr_xmin,
777 0 : ctx.offnum,
778 0 : next_xmin));
779 0 : }
780 :
781 : /*
782 : * If the current tuple's xmin is aborted but the successor
783 : * tuple's xmin is in-progress or committed, that's corruption.
784 : */
785 0 : if (xmin_commit_status_ok[ctx.offnum] &&
786 0 : xmin_commit_status[ctx.offnum] == XID_ABORTED &&
787 0 : xmin_commit_status_ok[nextoffnum])
788 : {
789 0 : if (xmin_commit_status[nextoffnum] == XID_IN_PROGRESS)
790 0 : report_corruption(&ctx,
791 0 : psprintf("tuple with aborted xmin %u was updated to produce a tuple at offset %d with in-progress xmin %u",
792 0 : curr_xmin,
793 0 : ctx.offnum,
794 0 : next_xmin));
795 0 : else if (xmin_commit_status[nextoffnum] == XID_COMMITTED)
796 0 : report_corruption(&ctx,
797 0 : psprintf("tuple with aborted xmin %u was updated to produce a tuple at offset %d with committed xmin %u",
798 0 : curr_xmin,
799 0 : ctx.offnum,
800 0 : next_xmin));
801 0 : }
802 0 : }
803 :
804 : /*
805 : * An update chain can start either with a non-heap-only tuple or with
806 : * a redirect line pointer, but not with a heap-only tuple.
807 : *
808 : * (This check is in a separate loop because we need the predecessor
809 : * array to be fully populated before we can perform it.)
810 : */
811 0 : for (ctx.offnum = FirstOffsetNumber;
812 0 : ctx.offnum <= maxoff;
813 0 : ctx.offnum = OffsetNumberNext(ctx.offnum))
814 : {
815 0 : if (xmin_commit_status_ok[ctx.offnum] &&
816 0 : (xmin_commit_status[ctx.offnum] == XID_COMMITTED ||
817 0 : xmin_commit_status[ctx.offnum] == XID_IN_PROGRESS) &&
818 0 : predecessor[ctx.offnum] == InvalidOffsetNumber)
819 : {
820 0 : ItemId curr_lp;
821 :
822 0 : curr_lp = PageGetItemId(ctx.page, ctx.offnum);
823 0 : if (!ItemIdIsRedirected(curr_lp))
824 : {
825 0 : HeapTupleHeader curr_htup;
826 :
827 0 : curr_htup = (HeapTupleHeader)
828 0 : PageGetItem(ctx.page, curr_lp);
829 0 : if (HeapTupleHeaderIsHeapOnly(curr_htup))
830 0 : report_corruption(&ctx,
831 0 : psprintf("tuple is root of chain but is marked as heap-only tuple"));
832 0 : }
833 0 : }
834 0 : }
835 :
836 : /* clean up */
837 0 : UnlockReleaseBuffer(ctx.buffer);
838 :
839 : /*
840 : * Check any toast pointers from the page whose lock we just released
841 : */
842 0 : if (ctx.toasted_attributes != NIL)
843 : {
844 0 : ListCell *cell;
845 :
846 0 : foreach(cell, ctx.toasted_attributes)
847 0 : check_toasted_attribute(&ctx, lfirst(cell));
848 0 : list_free_deep(ctx.toasted_attributes);
849 0 : ctx.toasted_attributes = NIL;
850 0 : }
851 :
852 0 : if (on_error_stop && ctx.is_corrupt)
853 0 : break;
854 0 : }
855 :
856 0 : read_stream_end(stream);
857 :
858 0 : if (vmbuffer != InvalidBuffer)
859 0 : ReleaseBuffer(vmbuffer);
860 :
861 : /* Close the associated toast table and indexes, if any. */
862 0 : if (ctx.toast_indexes)
863 0 : toast_close_indexes(ctx.toast_indexes, ctx.num_toast_indexes,
864 : AccessShareLock);
865 0 : if (ctx.toast_rel)
866 0 : table_close(ctx.toast_rel, AccessShareLock);
867 :
868 : /* Close the main relation */
869 0 : relation_close(ctx.rel, AccessShareLock);
870 :
871 0 : PG_RETURN_NULL();
872 0 : }
873 :
874 : /*
875 : * Heap amcheck's read stream callback for getting the next unskippable block.
876 : * This callback is only used when 'all-visible' or 'all-frozen' is provided
877 : * as the skip option to verify_heapam(). With the default 'none',
878 : * block_range_read_stream_cb() is used instead.
879 : */
880 : static BlockNumber
881 0 : heapcheck_read_stream_next_unskippable(ReadStream *stream,
882 : void *callback_private_data,
883 : void *per_buffer_data)
884 : {
885 0 : HeapCheckReadStreamData *p = callback_private_data;
886 :
887 : /* Loops over [current_blocknum, last_exclusive) blocks */
888 0 : for (BlockNumber i; (i = p->range.current_blocknum++) < p->range.last_exclusive;)
889 : {
890 0 : uint8 mapbits = visibilitymap_get_status(p->rel, i, p->vmbuffer);
891 :
892 0 : if (p->skip_option == SKIP_PAGES_ALL_FROZEN)
893 : {
894 0 : if ((mapbits & VISIBILITYMAP_ALL_FROZEN) != 0)
895 0 : continue;
896 0 : }
897 :
898 0 : if (p->skip_option == SKIP_PAGES_ALL_VISIBLE)
899 : {
900 0 : if ((mapbits & VISIBILITYMAP_ALL_VISIBLE) != 0)
901 0 : continue;
902 0 : }
903 :
904 0 : return i;
905 0 : }
906 :
907 0 : return InvalidBlockNumber;
908 0 : }
909 :
910 : /*
911 : * Shared internal implementation for report_corruption and
912 : * report_toast_corruption.
913 : */
914 : static void
915 0 : report_corruption_internal(Tuplestorestate *tupstore, TupleDesc tupdesc,
916 : BlockNumber blkno, OffsetNumber offnum,
917 : AttrNumber attnum, char *msg)
918 : {
919 0 : Datum values[HEAPCHECK_RELATION_COLS] = {0};
920 0 : bool nulls[HEAPCHECK_RELATION_COLS] = {0};
921 0 : HeapTuple tuple;
922 :
923 0 : values[0] = Int64GetDatum(blkno);
924 0 : values[1] = Int32GetDatum(offnum);
925 0 : values[2] = Int32GetDatum(attnum);
926 0 : nulls[2] = (attnum < 0);
927 0 : values[3] = CStringGetTextDatum(msg);
928 :
929 : /*
930 : * In principle, there is nothing to prevent a scan over a large, highly
931 : * corrupted table from using work_mem worth of memory building up the
932 : * tuplestore. That's ok, but if we also leak the msg argument memory
933 : * until the end of the query, we could exceed work_mem by more than a
934 : * trivial amount. Therefore, free the msg argument each time we are
935 : * called rather than waiting for our current memory context to be freed.
936 : */
937 0 : pfree(msg);
938 :
939 0 : tuple = heap_form_tuple(tupdesc, values, nulls);
940 0 : tuplestore_puttuple(tupstore, tuple);
941 0 : }
942 :
943 : /*
944 : * Record a single corruption found in the main table. The values in ctx should
945 : * indicate the location of the corruption, and the msg argument should contain
946 : * a human-readable description of the corruption.
947 : *
948 : * The msg argument is pfree'd by this function.
949 : */
950 : static void
951 0 : report_corruption(HeapCheckContext *ctx, char *msg)
952 : {
953 0 : report_corruption_internal(ctx->tupstore, ctx->tupdesc, ctx->blkno,
954 0 : ctx->offnum, ctx->attnum, msg);
955 0 : ctx->is_corrupt = true;
956 0 : }
957 :
958 : /*
959 : * Record corruption found in the toast table. The values in ta should
960 : * indicate the location in the main table where the toast pointer was
961 : * encountered, and the msg argument should contain a human-readable
962 : * description of the toast table corruption.
963 : *
964 : * As above, the msg argument is pfree'd by this function.
965 : */
966 : static void
967 0 : report_toast_corruption(HeapCheckContext *ctx, ToastedAttribute *ta,
968 : char *msg)
969 : {
970 0 : report_corruption_internal(ctx->tupstore, ctx->tupdesc, ta->blkno,
971 0 : ta->offnum, ta->attnum, msg);
972 0 : ctx->is_corrupt = true;
973 0 : }
974 :
975 : /*
976 : * Check for tuple header corruption.
977 : *
978 : * Some kinds of corruption make it unsafe to check the tuple attributes, for
979 : * example when the line pointer refers to a range of bytes outside the page.
980 : * In such cases, we return false (not checkable) after recording appropriate
981 : * corruption messages.
982 : *
983 : * Some other kinds of tuple header corruption confuse the question of where
984 : * the tuple attributes begin, or how long the nulls bitmap is, etc., making it
985 : * unreasonable to attempt to check attributes, even if all candidate answers
986 : * to those questions would not result in reading past the end of the line
987 : * pointer or page. In such cases, like above, we record corruption messages
988 : * about the header and then return false.
989 : *
990 : * Other kinds of tuple header corruption do not bear on the question of
991 : * whether the tuple attributes can be checked, so we record corruption
992 : * messages for them but we do not return false merely because we detected
993 : * them.
994 : *
995 : * Returns whether the tuple is sufficiently sensible to undergo visibility and
996 : * attribute checks.
997 : */
998 : static bool
999 0 : check_tuple_header(HeapCheckContext *ctx)
1000 : {
1001 0 : HeapTupleHeader tuphdr = ctx->tuphdr;
1002 0 : uint16 infomask = tuphdr->t_infomask;
1003 0 : TransactionId curr_xmax = HeapTupleHeaderGetUpdateXid(tuphdr);
1004 0 : bool result = true;
1005 0 : unsigned expected_hoff;
1006 :
1007 0 : if (ctx->tuphdr->t_hoff > ctx->lp_len)
1008 : {
1009 0 : report_corruption(ctx,
1010 0 : psprintf("data begins at offset %u beyond the tuple length %u",
1011 0 : ctx->tuphdr->t_hoff, ctx->lp_len));
1012 0 : result = false;
1013 0 : }
1014 :
1015 0 : if ((ctx->tuphdr->t_infomask & HEAP_XMAX_COMMITTED) &&
1016 0 : (ctx->tuphdr->t_infomask & HEAP_XMAX_IS_MULTI))
1017 : {
1018 0 : report_corruption(ctx,
1019 0 : pstrdup("multixact should not be marked committed"));
1020 :
1021 : /*
1022 : * This condition is clearly wrong, but it's not enough to justify
1023 : * skipping further checks, because we don't rely on this to determine
1024 : * whether the tuple is visible or to interpret other relevant header
1025 : * fields.
1026 : */
1027 0 : }
1028 :
1029 0 : if (!TransactionIdIsValid(curr_xmax) &&
1030 0 : HeapTupleHeaderIsHotUpdated(tuphdr))
1031 : {
1032 0 : report_corruption(ctx,
1033 0 : psprintf("tuple has been HOT updated, but xmax is 0"));
1034 :
1035 : /*
1036 : * As above, even though this shouldn't happen, it's not sufficient
1037 : * justification for skipping further checks, we should still be able
1038 : * to perform sensibly.
1039 : */
1040 0 : }
1041 :
1042 0 : if (HeapTupleHeaderIsHeapOnly(tuphdr) &&
1043 0 : ((tuphdr->t_infomask & HEAP_UPDATED) == 0))
1044 : {
1045 0 : report_corruption(ctx,
1046 0 : psprintf("tuple is heap only, but not the result of an update"));
1047 :
1048 : /* Here again, we can still perform further checks. */
1049 0 : }
1050 :
1051 0 : if (infomask & HEAP_HASNULL)
1052 0 : expected_hoff = MAXALIGN(SizeofHeapTupleHeader + BITMAPLEN(ctx->natts));
1053 : else
1054 0 : expected_hoff = MAXALIGN(SizeofHeapTupleHeader);
1055 0 : if (ctx->tuphdr->t_hoff != expected_hoff)
1056 : {
1057 0 : if ((infomask & HEAP_HASNULL) && ctx->natts == 1)
1058 0 : report_corruption(ctx,
1059 0 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (1 attribute, has nulls)",
1060 0 : expected_hoff, ctx->tuphdr->t_hoff));
1061 0 : else if ((infomask & HEAP_HASNULL))
1062 0 : report_corruption(ctx,
1063 0 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (%u attributes, has nulls)",
1064 0 : expected_hoff, ctx->tuphdr->t_hoff, ctx->natts));
1065 0 : else if (ctx->natts == 1)
1066 0 : report_corruption(ctx,
1067 0 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (1 attribute, no nulls)",
1068 0 : expected_hoff, ctx->tuphdr->t_hoff));
1069 : else
1070 0 : report_corruption(ctx,
1071 0 : psprintf("tuple data should begin at byte %u, but actually begins at byte %u (%u attributes, no nulls)",
1072 0 : expected_hoff, ctx->tuphdr->t_hoff, ctx->natts));
1073 0 : result = false;
1074 0 : }
1075 :
1076 0 : return result;
1077 0 : }
1078 :
1079 : /*
1080 : * Checks tuple visibility so we know which further checks are safe to
1081 : * perform.
1082 : *
1083 : * If a tuple could have been inserted by a transaction that also added a
1084 : * column to the table, but which ultimately did not commit, or which has not
1085 : * yet committed, then the table's current TupleDesc might differ from the one
1086 : * used to construct this tuple, so we must not check it.
1087 : *
1088 : * As a special case, if our own transaction inserted the tuple, even if we
1089 : * added a column to the table, our TupleDesc should match. We could check the
1090 : * tuple, but choose not to do so.
1091 : *
1092 : * If a tuple has been updated or deleted, we can still read the old tuple for
1093 : * corruption checking purposes, as long as we are careful about concurrent
1094 : * vacuums. The main table tuple itself cannot be vacuumed away because we
1095 : * hold a buffer lock on the page, but if the deleting transaction is older
1096 : * than our transaction snapshot's xmin, then vacuum could remove the toast at
1097 : * any time, so we must not try to follow TOAST pointers.
1098 : *
1099 : * If xmin or xmax values are older than can be checked against clog, or appear
1100 : * to be in the future (possibly due to wrap-around), then we cannot make a
1101 : * determination about the visibility of the tuple, so we skip further checks.
1102 : *
1103 : * Returns true if the tuple itself should be checked, false otherwise. Sets
1104 : * ctx->tuple_could_be_pruned if the tuple -- and thus also any associated
1105 : * TOAST tuples -- are eligible for pruning.
1106 : *
1107 : * Sets *xmin_commit_status_ok to true if the commit status of xmin is known
1108 : * and false otherwise. If it's set to true, then also set *xmin_commit_status
1109 : * to the actual commit status.
1110 : */
1111 : static bool
1112 0 : check_tuple_visibility(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
1113 : XidCommitStatus *xmin_commit_status)
1114 : {
1115 0 : TransactionId xmin;
1116 0 : TransactionId xvac;
1117 0 : TransactionId xmax;
1118 0 : XidCommitStatus xmin_status;
1119 0 : XidCommitStatus xvac_status;
1120 0 : XidCommitStatus xmax_status;
1121 0 : HeapTupleHeader tuphdr = ctx->tuphdr;
1122 :
1123 0 : ctx->tuple_could_be_pruned = true; /* have not yet proven otherwise */
1124 0 : *xmin_commit_status_ok = false; /* have not yet proven otherwise */
1125 :
1126 : /* If xmin is normal, it should be within valid range */
1127 0 : xmin = HeapTupleHeaderGetXmin(tuphdr);
1128 0 : switch (get_xid_status(xmin, ctx, &xmin_status))
1129 : {
1130 : case XID_INVALID:
1131 : /* Could be the result of a speculative insertion that aborted. */
1132 0 : return false;
1133 : case XID_BOUNDS_OK:
1134 0 : *xmin_commit_status_ok = true;
1135 0 : *xmin_commit_status = xmin_status;
1136 0 : break;
1137 : case XID_IN_FUTURE:
1138 0 : report_corruption(ctx,
1139 0 : psprintf("xmin %u equals or exceeds next valid transaction ID %u:%u",
1140 0 : xmin,
1141 0 : EpochFromFullTransactionId(ctx->next_fxid),
1142 0 : XidFromFullTransactionId(ctx->next_fxid)));
1143 0 : return false;
1144 : case XID_PRECEDES_CLUSTERMIN:
1145 0 : report_corruption(ctx,
1146 0 : psprintf("xmin %u precedes oldest valid transaction ID %u:%u",
1147 0 : xmin,
1148 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1149 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1150 0 : return false;
1151 : case XID_PRECEDES_RELMIN:
1152 0 : report_corruption(ctx,
1153 0 : psprintf("xmin %u precedes relation freeze threshold %u:%u",
1154 0 : xmin,
1155 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1156 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1157 0 : return false;
1158 : }
1159 :
1160 : /*
1161 : * Has inserting transaction committed?
1162 : */
1163 0 : if (!HeapTupleHeaderXminCommitted(tuphdr))
1164 : {
1165 0 : if (HeapTupleHeaderXminInvalid(tuphdr))
1166 0 : return false; /* inserter aborted, don't check */
1167 : /* Used by pre-9.0 binary upgrades */
1168 0 : else if (tuphdr->t_infomask & HEAP_MOVED_OFF)
1169 : {
1170 0 : xvac = HeapTupleHeaderGetXvac(tuphdr);
1171 :
1172 0 : switch (get_xid_status(xvac, ctx, &xvac_status))
1173 : {
1174 : case XID_INVALID:
1175 0 : report_corruption(ctx,
1176 0 : pstrdup("old-style VACUUM FULL transaction ID for moved off tuple is invalid"));
1177 0 : return false;
1178 : case XID_IN_FUTURE:
1179 0 : report_corruption(ctx,
1180 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple equals or exceeds next valid transaction ID %u:%u",
1181 0 : xvac,
1182 0 : EpochFromFullTransactionId(ctx->next_fxid),
1183 0 : XidFromFullTransactionId(ctx->next_fxid)));
1184 0 : return false;
1185 : case XID_PRECEDES_RELMIN:
1186 0 : report_corruption(ctx,
1187 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple precedes relation freeze threshold %u:%u",
1188 0 : xvac,
1189 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1190 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1191 0 : return false;
1192 : case XID_PRECEDES_CLUSTERMIN:
1193 0 : report_corruption(ctx,
1194 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple precedes oldest valid transaction ID %u:%u",
1195 0 : xvac,
1196 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1197 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1198 0 : return false;
1199 : case XID_BOUNDS_OK:
1200 : break;
1201 : }
1202 :
1203 0 : switch (xvac_status)
1204 : {
1205 : case XID_IS_CURRENT_XID:
1206 0 : report_corruption(ctx,
1207 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple matches our current transaction ID",
1208 0 : xvac));
1209 0 : return false;
1210 : case XID_IN_PROGRESS:
1211 0 : report_corruption(ctx,
1212 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved off tuple appears to be in progress",
1213 0 : xvac));
1214 0 : return false;
1215 :
1216 : case XID_COMMITTED:
1217 :
1218 : /*
1219 : * The tuple is dead, because the xvac transaction moved
1220 : * it off and committed. It's checkable, but also
1221 : * prunable.
1222 : */
1223 0 : return true;
1224 :
1225 : case XID_ABORTED:
1226 :
1227 : /*
1228 : * The original xmin must have committed, because the xvac
1229 : * transaction tried to move it later. Since xvac is
1230 : * aborted, whether it's still alive now depends on the
1231 : * status of xmax.
1232 : */
1233 : break;
1234 : }
1235 0 : }
1236 : /* Used by pre-9.0 binary upgrades */
1237 0 : else if (tuphdr->t_infomask & HEAP_MOVED_IN)
1238 : {
1239 0 : xvac = HeapTupleHeaderGetXvac(tuphdr);
1240 :
1241 0 : switch (get_xid_status(xvac, ctx, &xvac_status))
1242 : {
1243 : case XID_INVALID:
1244 0 : report_corruption(ctx,
1245 0 : pstrdup("old-style VACUUM FULL transaction ID for moved in tuple is invalid"));
1246 0 : return false;
1247 : case XID_IN_FUTURE:
1248 0 : report_corruption(ctx,
1249 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple equals or exceeds next valid transaction ID %u:%u",
1250 0 : xvac,
1251 0 : EpochFromFullTransactionId(ctx->next_fxid),
1252 0 : XidFromFullTransactionId(ctx->next_fxid)));
1253 0 : return false;
1254 : case XID_PRECEDES_RELMIN:
1255 0 : report_corruption(ctx,
1256 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple precedes relation freeze threshold %u:%u",
1257 0 : xvac,
1258 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1259 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1260 0 : return false;
1261 : case XID_PRECEDES_CLUSTERMIN:
1262 0 : report_corruption(ctx,
1263 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple precedes oldest valid transaction ID %u:%u",
1264 0 : xvac,
1265 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1266 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1267 0 : return false;
1268 : case XID_BOUNDS_OK:
1269 : break;
1270 : }
1271 :
1272 0 : switch (xvac_status)
1273 : {
1274 : case XID_IS_CURRENT_XID:
1275 0 : report_corruption(ctx,
1276 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple matches our current transaction ID",
1277 0 : xvac));
1278 0 : return false;
1279 : case XID_IN_PROGRESS:
1280 0 : report_corruption(ctx,
1281 0 : psprintf("old-style VACUUM FULL transaction ID %u for moved in tuple appears to be in progress",
1282 0 : xvac));
1283 0 : return false;
1284 :
1285 : case XID_COMMITTED:
1286 :
1287 : /*
1288 : * The original xmin must have committed, because the xvac
1289 : * transaction moved it later. Whether it's still alive
1290 : * now depends on the status of xmax.
1291 : */
1292 : break;
1293 :
1294 : case XID_ABORTED:
1295 :
1296 : /*
1297 : * The tuple is dead, because the xvac transaction moved
1298 : * it off and committed. It's checkable, but also
1299 : * prunable.
1300 : */
1301 0 : return true;
1302 : }
1303 0 : }
1304 0 : else if (xmin_status != XID_COMMITTED)
1305 : {
1306 : /*
1307 : * Inserting transaction is not in progress, and not committed, so
1308 : * it might have changed the TupleDesc in ways we don't know
1309 : * about. Thus, don't try to check the tuple structure.
1310 : *
1311 : * If xmin_status happens to be XID_IS_CURRENT_XID, then in theory
1312 : * any such DDL changes ought to be visible to us, so perhaps we
1313 : * could check anyway in that case. But, for now, let's be
1314 : * conservative and treat this like any other uncommitted insert.
1315 : */
1316 0 : return false;
1317 : }
1318 0 : }
1319 :
1320 : /*
1321 : * Okay, the inserter committed, so it was good at some point. Now what
1322 : * about the deleting transaction?
1323 : */
1324 :
1325 0 : if (tuphdr->t_infomask & HEAP_XMAX_IS_MULTI)
1326 : {
1327 : /*
1328 : * xmax is a multixact, so sanity-check the MXID. Note that we do this
1329 : * prior to checking for HEAP_XMAX_INVALID or
1330 : * HEAP_XMAX_IS_LOCKED_ONLY. This might therefore complain about
1331 : * things that wouldn't actually be a problem during a normal scan,
1332 : * but eventually we're going to have to freeze, and that process will
1333 : * ignore hint bits.
1334 : *
1335 : * Even if the MXID is out of range, we still know that the original
1336 : * insert committed, so we can check the tuple itself. However, we
1337 : * can't rule out the possibility that this tuple is dead, so don't
1338 : * clear ctx->tuple_could_be_pruned. Possibly we should go ahead and
1339 : * clear that flag anyway if HEAP_XMAX_INVALID is set or if
1340 : * HEAP_XMAX_IS_LOCKED_ONLY is true, but for now we err on the side of
1341 : * avoiding possibly-bogus complaints about missing TOAST entries.
1342 : */
1343 0 : xmax = HeapTupleHeaderGetRawXmax(tuphdr);
1344 0 : switch (check_mxid_valid_in_rel(xmax, ctx))
1345 : {
1346 : case XID_INVALID:
1347 0 : report_corruption(ctx,
1348 0 : pstrdup("multitransaction ID is invalid"));
1349 0 : return true;
1350 : case XID_PRECEDES_RELMIN:
1351 0 : report_corruption(ctx,
1352 0 : psprintf("multitransaction ID %u precedes relation minimum multitransaction ID threshold %u",
1353 0 : xmax, ctx->relminmxid));
1354 0 : return true;
1355 : case XID_PRECEDES_CLUSTERMIN:
1356 0 : report_corruption(ctx,
1357 0 : psprintf("multitransaction ID %u precedes oldest valid multitransaction ID threshold %u",
1358 0 : xmax, ctx->oldest_mxact));
1359 0 : return true;
1360 : case XID_IN_FUTURE:
1361 0 : report_corruption(ctx,
1362 0 : psprintf("multitransaction ID %u equals or exceeds next valid multitransaction ID %u",
1363 0 : xmax,
1364 0 : ctx->next_mxact));
1365 0 : return true;
1366 : case XID_BOUNDS_OK:
1367 : break;
1368 : }
1369 0 : }
1370 :
1371 0 : if (tuphdr->t_infomask & HEAP_XMAX_INVALID)
1372 : {
1373 : /*
1374 : * This tuple is live. A concurrently running transaction could
1375 : * delete it before we get around to checking the toast, but any such
1376 : * running transaction is surely not less than our safe_xmin, so the
1377 : * toast cannot be vacuumed out from under us.
1378 : */
1379 0 : ctx->tuple_could_be_pruned = false;
1380 0 : return true;
1381 : }
1382 :
1383 0 : if (HEAP_XMAX_IS_LOCKED_ONLY(tuphdr->t_infomask))
1384 : {
1385 : /*
1386 : * "Deleting" xact really only locked it, so the tuple is live in any
1387 : * case. As above, a concurrently running transaction could delete
1388 : * it, but it cannot be vacuumed out from under us.
1389 : */
1390 0 : ctx->tuple_could_be_pruned = false;
1391 0 : return true;
1392 : }
1393 :
1394 0 : if (tuphdr->t_infomask & HEAP_XMAX_IS_MULTI)
1395 : {
1396 : /*
1397 : * We already checked above that this multixact is within limits for
1398 : * this table. Now check the update xid from this multixact.
1399 : */
1400 0 : xmax = HeapTupleGetUpdateXid(tuphdr);
1401 0 : switch (get_xid_status(xmax, ctx, &xmax_status))
1402 : {
1403 : case XID_INVALID:
1404 : /* not LOCKED_ONLY, so it has to have an xmax */
1405 0 : report_corruption(ctx,
1406 0 : pstrdup("update xid is invalid"));
1407 0 : return true;
1408 : case XID_IN_FUTURE:
1409 0 : report_corruption(ctx,
1410 0 : psprintf("update xid %u equals or exceeds next valid transaction ID %u:%u",
1411 0 : xmax,
1412 0 : EpochFromFullTransactionId(ctx->next_fxid),
1413 0 : XidFromFullTransactionId(ctx->next_fxid)));
1414 0 : return true;
1415 : case XID_PRECEDES_RELMIN:
1416 0 : report_corruption(ctx,
1417 0 : psprintf("update xid %u precedes relation freeze threshold %u:%u",
1418 0 : xmax,
1419 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1420 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1421 0 : return true;
1422 : case XID_PRECEDES_CLUSTERMIN:
1423 0 : report_corruption(ctx,
1424 0 : psprintf("update xid %u precedes oldest valid transaction ID %u:%u",
1425 0 : xmax,
1426 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1427 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1428 0 : return true;
1429 : case XID_BOUNDS_OK:
1430 : break;
1431 : }
1432 :
1433 0 : switch (xmax_status)
1434 : {
1435 : case XID_IS_CURRENT_XID:
1436 : case XID_IN_PROGRESS:
1437 :
1438 : /*
1439 : * The delete is in progress, so it cannot be visible to our
1440 : * snapshot.
1441 : */
1442 0 : ctx->tuple_could_be_pruned = false;
1443 0 : break;
1444 : case XID_COMMITTED:
1445 :
1446 : /*
1447 : * The delete committed. Whether the toast can be vacuumed
1448 : * away depends on how old the deleting transaction is.
1449 : */
1450 0 : ctx->tuple_could_be_pruned = TransactionIdPrecedes(xmax,
1451 0 : ctx->safe_xmin);
1452 0 : break;
1453 : case XID_ABORTED:
1454 :
1455 : /*
1456 : * The delete aborted or crashed. The tuple is still live.
1457 : */
1458 0 : ctx->tuple_could_be_pruned = false;
1459 0 : break;
1460 : }
1461 :
1462 : /* Tuple itself is checkable even if it's dead. */
1463 0 : return true;
1464 : }
1465 :
1466 : /* xmax is an XID, not a MXID. Sanity check it. */
1467 0 : xmax = HeapTupleHeaderGetRawXmax(tuphdr);
1468 0 : switch (get_xid_status(xmax, ctx, &xmax_status))
1469 : {
1470 : case XID_INVALID:
1471 0 : ctx->tuple_could_be_pruned = false;
1472 0 : return true;
1473 : case XID_IN_FUTURE:
1474 0 : report_corruption(ctx,
1475 0 : psprintf("xmax %u equals or exceeds next valid transaction ID %u:%u",
1476 0 : xmax,
1477 0 : EpochFromFullTransactionId(ctx->next_fxid),
1478 0 : XidFromFullTransactionId(ctx->next_fxid)));
1479 0 : return false; /* corrupt */
1480 : case XID_PRECEDES_RELMIN:
1481 0 : report_corruption(ctx,
1482 0 : psprintf("xmax %u precedes relation freeze threshold %u:%u",
1483 0 : xmax,
1484 0 : EpochFromFullTransactionId(ctx->relfrozenfxid),
1485 0 : XidFromFullTransactionId(ctx->relfrozenfxid)));
1486 0 : return false; /* corrupt */
1487 : case XID_PRECEDES_CLUSTERMIN:
1488 0 : report_corruption(ctx,
1489 0 : psprintf("xmax %u precedes oldest valid transaction ID %u:%u",
1490 0 : xmax,
1491 0 : EpochFromFullTransactionId(ctx->oldest_fxid),
1492 0 : XidFromFullTransactionId(ctx->oldest_fxid)));
1493 0 : return false; /* corrupt */
1494 : case XID_BOUNDS_OK:
1495 : break;
1496 : }
1497 :
1498 : /*
1499 : * Whether the toast can be vacuumed away depends on how old the deleting
1500 : * transaction is.
1501 : */
1502 0 : switch (xmax_status)
1503 : {
1504 : case XID_IS_CURRENT_XID:
1505 : case XID_IN_PROGRESS:
1506 :
1507 : /*
1508 : * The delete is in progress, so it cannot be visible to our
1509 : * snapshot.
1510 : */
1511 0 : ctx->tuple_could_be_pruned = false;
1512 0 : break;
1513 :
1514 : case XID_COMMITTED:
1515 :
1516 : /*
1517 : * The delete committed. Whether the toast can be vacuumed away
1518 : * depends on how old the deleting transaction is.
1519 : */
1520 0 : ctx->tuple_could_be_pruned = TransactionIdPrecedes(xmax,
1521 0 : ctx->safe_xmin);
1522 0 : break;
1523 :
1524 : case XID_ABORTED:
1525 :
1526 : /*
1527 : * The delete aborted or crashed. The tuple is still live.
1528 : */
1529 0 : ctx->tuple_could_be_pruned = false;
1530 0 : break;
1531 : }
1532 :
1533 : /* Tuple itself is checkable even if it's dead. */
1534 0 : return true;
1535 0 : }
1536 :
1537 :
1538 : /*
1539 : * Check the current toast tuple against the state tracked in ctx, recording
1540 : * any corruption found in ctx->tupstore.
1541 : *
1542 : * This is not equivalent to running verify_heapam on the toast table itself,
1543 : * and is not hardened against corruption of the toast table. Rather, when
1544 : * validating a toasted attribute in the main table, the sequence of toast
1545 : * tuples that store the toasted value are retrieved and checked in order, with
1546 : * each toast tuple being checked against where we are in the sequence, as well
1547 : * as each toast tuple having its varlena structure sanity checked.
1548 : *
1549 : * On entry, *expected_chunk_seq should be the chunk_seq value that we expect
1550 : * to find in toasttup. On exit, it will be updated to the value the next call
1551 : * to this function should expect to see.
1552 : */
1553 : static void
1554 0 : check_toast_tuple(HeapTuple toasttup, HeapCheckContext *ctx,
1555 : ToastedAttribute *ta, int32 *expected_chunk_seq,
1556 : uint32 extsize)
1557 : {
1558 0 : int32 chunk_seq;
1559 0 : int32 last_chunk_seq = (extsize - 1) / TOAST_MAX_CHUNK_SIZE;
1560 0 : Pointer chunk;
1561 0 : bool isnull;
1562 0 : int32 chunksize;
1563 0 : int32 expected_size;
1564 :
1565 : /* Sanity-check the sequence number. */
1566 0 : chunk_seq = DatumGetInt32(fastgetattr(toasttup, 2,
1567 0 : ctx->toast_rel->rd_att, &isnull));
1568 0 : if (isnull)
1569 : {
1570 0 : report_toast_corruption(ctx, ta,
1571 0 : psprintf("toast value %u has toast chunk with null sequence number",
1572 0 : ta->toast_pointer.va_valueid));
1573 0 : return;
1574 : }
1575 0 : if (chunk_seq != *expected_chunk_seq)
1576 : {
1577 : /* Either the TOAST index is corrupt, or we don't have all chunks. */
1578 0 : report_toast_corruption(ctx, ta,
1579 0 : psprintf("toast value %u index scan returned chunk %d when expecting chunk %d",
1580 0 : ta->toast_pointer.va_valueid,
1581 0 : chunk_seq, *expected_chunk_seq));
1582 0 : }
1583 0 : *expected_chunk_seq = chunk_seq + 1;
1584 :
1585 : /* Sanity-check the chunk data. */
1586 0 : chunk = DatumGetPointer(fastgetattr(toasttup, 3,
1587 0 : ctx->toast_rel->rd_att, &isnull));
1588 0 : if (isnull)
1589 : {
1590 0 : report_toast_corruption(ctx, ta,
1591 0 : psprintf("toast value %u chunk %d has null data",
1592 0 : ta->toast_pointer.va_valueid,
1593 0 : chunk_seq));
1594 0 : return;
1595 : }
1596 0 : if (!VARATT_IS_EXTENDED(chunk))
1597 0 : chunksize = VARSIZE(chunk) - VARHDRSZ;
1598 0 : else if (VARATT_IS_SHORT(chunk))
1599 : {
1600 : /*
1601 : * could happen due to heap_form_tuple doing its thing
1602 : */
1603 0 : chunksize = VARSIZE_SHORT(chunk) - VARHDRSZ_SHORT;
1604 0 : }
1605 : else
1606 : {
1607 : /* should never happen */
1608 0 : uint32 header = ((varattrib_4b *) chunk)->va_4byte.va_header;
1609 :
1610 0 : report_toast_corruption(ctx, ta,
1611 0 : psprintf("toast value %u chunk %d has invalid varlena header %0x",
1612 0 : ta->toast_pointer.va_valueid,
1613 0 : chunk_seq, header));
1614 : return;
1615 0 : }
1616 :
1617 : /*
1618 : * Some checks on the data we've found
1619 : */
1620 0 : if (chunk_seq > last_chunk_seq)
1621 : {
1622 0 : report_toast_corruption(ctx, ta,
1623 0 : psprintf("toast value %u chunk %d follows last expected chunk %d",
1624 0 : ta->toast_pointer.va_valueid,
1625 0 : chunk_seq, last_chunk_seq));
1626 0 : return;
1627 : }
1628 :
1629 0 : expected_size = chunk_seq < last_chunk_seq ? TOAST_MAX_CHUNK_SIZE
1630 0 : : extsize - (last_chunk_seq * TOAST_MAX_CHUNK_SIZE);
1631 :
1632 0 : if (chunksize != expected_size)
1633 0 : report_toast_corruption(ctx, ta,
1634 0 : psprintf("toast value %u chunk %d has size %u, but expected size %u",
1635 0 : ta->toast_pointer.va_valueid,
1636 0 : chunk_seq, chunksize, expected_size));
1637 0 : }
1638 :
1639 : /*
1640 : * Check the current attribute as tracked in ctx, recording any corruption
1641 : * found in ctx->tupstore.
1642 : *
1643 : * This function follows the logic performed by heap_deform_tuple(), and in the
1644 : * case of a toasted value, optionally stores the toast pointer so later it can
1645 : * be checked following the logic of detoast_external_attr(), checking for any
1646 : * conditions that would result in either of those functions Asserting or
1647 : * crashing the backend. The checks performed by Asserts present in those two
1648 : * functions are also performed here and in check_toasted_attribute. In cases
1649 : * where those two functions are a bit cavalier in their assumptions about data
1650 : * being correct, we perform additional checks not present in either of those
1651 : * two functions. Where some condition is checked in both of those functions,
1652 : * we perform it here twice, as we parallel the logical flow of those two
1653 : * functions. The presence of duplicate checks seems a reasonable price to pay
1654 : * for keeping this code tightly coupled with the code it protects.
1655 : *
1656 : * Returns true if the tuple attribute is sane enough for processing to
1657 : * continue on to the next attribute, false otherwise.
1658 : */
1659 : static bool
1660 0 : check_tuple_attribute(HeapCheckContext *ctx)
1661 : {
1662 0 : Datum attdatum;
1663 0 : struct varlena *attr;
1664 0 : char *tp; /* pointer to the tuple data */
1665 0 : uint16 infomask;
1666 0 : CompactAttribute *thisatt;
1667 0 : struct varatt_external toast_pointer;
1668 :
1669 0 : infomask = ctx->tuphdr->t_infomask;
1670 0 : thisatt = TupleDescCompactAttr(RelationGetDescr(ctx->rel), ctx->attnum);
1671 :
1672 0 : tp = (char *) ctx->tuphdr + ctx->tuphdr->t_hoff;
1673 :
1674 0 : if (ctx->tuphdr->t_hoff + ctx->offset > ctx->lp_len)
1675 : {
1676 0 : report_corruption(ctx,
1677 0 : psprintf("attribute with length %u starts at offset %u beyond total tuple length %u",
1678 0 : thisatt->attlen,
1679 0 : ctx->tuphdr->t_hoff + ctx->offset,
1680 0 : ctx->lp_len));
1681 0 : return false;
1682 : }
1683 :
1684 : /* Skip null values */
1685 0 : if (infomask & HEAP_HASNULL && att_isnull(ctx->attnum, ctx->tuphdr->t_bits))
1686 0 : return true;
1687 :
1688 : /* Skip non-varlena values, but update offset first */
1689 0 : if (thisatt->attlen != -1)
1690 : {
1691 0 : ctx->offset = att_nominal_alignby(ctx->offset, thisatt->attalignby);
1692 0 : ctx->offset = att_addlength_pointer(ctx->offset, thisatt->attlen,
1693 : tp + ctx->offset);
1694 0 : if (ctx->tuphdr->t_hoff + ctx->offset > ctx->lp_len)
1695 : {
1696 0 : report_corruption(ctx,
1697 0 : psprintf("attribute with length %u ends at offset %u beyond total tuple length %u",
1698 0 : thisatt->attlen,
1699 0 : ctx->tuphdr->t_hoff + ctx->offset,
1700 0 : ctx->lp_len));
1701 0 : return false;
1702 : }
1703 0 : return true;
1704 : }
1705 :
1706 : /* Ok, we're looking at a varlena attribute. */
1707 0 : ctx->offset = att_pointer_alignby(ctx->offset, thisatt->attalignby, -1,
1708 : tp + ctx->offset);
1709 :
1710 : /* Get the (possibly corrupt) varlena datum */
1711 0 : attdatum = fetchatt(thisatt, tp + ctx->offset);
1712 :
1713 : /*
1714 : * We have the datum, but we cannot decode it carelessly, as it may still
1715 : * be corrupt.
1716 : */
1717 :
1718 : /*
1719 : * Check that VARTAG_SIZE won't hit an Assert on a corrupt va_tag before
1720 : * risking a call into att_addlength_pointer
1721 : */
1722 0 : if (VARATT_IS_EXTERNAL(tp + ctx->offset))
1723 : {
1724 0 : uint8 va_tag = VARTAG_EXTERNAL(tp + ctx->offset);
1725 :
1726 0 : if (va_tag != VARTAG_ONDISK)
1727 : {
1728 0 : report_corruption(ctx,
1729 0 : psprintf("toasted attribute has unexpected TOAST tag %u",
1730 0 : va_tag));
1731 : /* We can't know where the next attribute begins */
1732 0 : return false;
1733 : }
1734 0 : }
1735 :
1736 : /* Ok, should be safe now */
1737 0 : ctx->offset = att_addlength_pointer(ctx->offset, thisatt->attlen,
1738 : tp + ctx->offset);
1739 :
1740 0 : if (ctx->tuphdr->t_hoff + ctx->offset > ctx->lp_len)
1741 : {
1742 0 : report_corruption(ctx,
1743 0 : psprintf("attribute with length %u ends at offset %u beyond total tuple length %u",
1744 0 : thisatt->attlen,
1745 0 : ctx->tuphdr->t_hoff + ctx->offset,
1746 0 : ctx->lp_len));
1747 :
1748 0 : return false;
1749 : }
1750 :
1751 : /*
1752 : * heap_deform_tuple would be done with this attribute at this point,
1753 : * having stored it in values[], and would continue to the next attribute.
1754 : * We go further, because we need to check if the toast datum is corrupt.
1755 : */
1756 :
1757 0 : attr = (struct varlena *) DatumGetPointer(attdatum);
1758 :
1759 : /*
1760 : * Now we follow the logic of detoast_external_attr(), with the same
1761 : * caveats about being paranoid about corruption.
1762 : */
1763 :
1764 : /* Skip values that are not external */
1765 0 : if (!VARATT_IS_EXTERNAL(attr))
1766 0 : return true;
1767 :
1768 : /* It is external, and we're looking at a page on disk */
1769 :
1770 : /*
1771 : * Must copy attr into toast_pointer for alignment considerations
1772 : */
1773 0 : VARATT_EXTERNAL_GET_POINTER(toast_pointer, attr);
1774 :
1775 : /* Toasted attributes too large to be untoasted should never be stored */
1776 0 : if (toast_pointer.va_rawsize > VARLENA_SIZE_LIMIT)
1777 0 : report_corruption(ctx,
1778 0 : psprintf("toast value %u rawsize %d exceeds limit %d",
1779 0 : toast_pointer.va_valueid,
1780 0 : toast_pointer.va_rawsize,
1781 : VARLENA_SIZE_LIMIT));
1782 :
1783 0 : if (VARATT_EXTERNAL_IS_COMPRESSED(toast_pointer))
1784 : {
1785 0 : ToastCompressionId cmid;
1786 0 : bool valid = false;
1787 :
1788 : /* Compressed attributes should have a valid compression method */
1789 0 : cmid = TOAST_COMPRESS_METHOD(&toast_pointer);
1790 0 : switch (cmid)
1791 : {
1792 : /* List of all valid compression method IDs */
1793 : case TOAST_PGLZ_COMPRESSION_ID:
1794 : case TOAST_LZ4_COMPRESSION_ID:
1795 0 : valid = true;
1796 0 : break;
1797 :
1798 : /* Recognized but invalid compression method ID */
1799 : case TOAST_INVALID_COMPRESSION_ID:
1800 : break;
1801 :
1802 : /* Intentionally no default here */
1803 : }
1804 0 : if (!valid)
1805 0 : report_corruption(ctx,
1806 0 : psprintf("toast value %u has invalid compression method id %d",
1807 0 : toast_pointer.va_valueid, cmid));
1808 0 : }
1809 :
1810 : /* The tuple header better claim to contain toasted values */
1811 0 : if (!(infomask & HEAP_HASEXTERNAL))
1812 : {
1813 0 : report_corruption(ctx,
1814 0 : psprintf("toast value %u is external but tuple header flag HEAP_HASEXTERNAL not set",
1815 0 : toast_pointer.va_valueid));
1816 0 : return true;
1817 : }
1818 :
1819 : /* The relation better have a toast table */
1820 0 : if (!ctx->rel->rd_rel->reltoastrelid)
1821 : {
1822 0 : report_corruption(ctx,
1823 0 : psprintf("toast value %u is external but relation has no toast relation",
1824 0 : toast_pointer.va_valueid));
1825 0 : return true;
1826 : }
1827 :
1828 : /* If we were told to skip toast checking, then we're done. */
1829 0 : if (ctx->toast_rel == NULL)
1830 0 : return true;
1831 :
1832 : /*
1833 : * If this tuple is eligible to be pruned, we cannot check the toast.
1834 : * Otherwise, we push a copy of the toast tuple so we can check it after
1835 : * releasing the main table buffer lock.
1836 : */
1837 0 : if (!ctx->tuple_could_be_pruned)
1838 : {
1839 0 : ToastedAttribute *ta;
1840 :
1841 0 : ta = palloc0_object(ToastedAttribute);
1842 :
1843 0 : VARATT_EXTERNAL_GET_POINTER(ta->toast_pointer, attr);
1844 0 : ta->blkno = ctx->blkno;
1845 0 : ta->offnum = ctx->offnum;
1846 0 : ta->attnum = ctx->attnum;
1847 0 : ctx->toasted_attributes = lappend(ctx->toasted_attributes, ta);
1848 0 : }
1849 :
1850 0 : return true;
1851 0 : }
1852 :
1853 : /*
1854 : * For each attribute collected in ctx->toasted_attributes, look up the value
1855 : * in the toast table and perform checks on it. This function should only be
1856 : * called on toast pointers which cannot be vacuumed away during our
1857 : * processing.
1858 : */
1859 : static void
1860 0 : check_toasted_attribute(HeapCheckContext *ctx, ToastedAttribute *ta)
1861 : {
1862 0 : ScanKeyData toastkey;
1863 0 : SysScanDesc toastscan;
1864 0 : bool found_toasttup;
1865 0 : HeapTuple toasttup;
1866 0 : uint32 extsize;
1867 0 : int32 expected_chunk_seq = 0;
1868 0 : int32 last_chunk_seq;
1869 :
1870 0 : extsize = VARATT_EXTERNAL_GET_EXTSIZE(ta->toast_pointer);
1871 0 : last_chunk_seq = (extsize - 1) / TOAST_MAX_CHUNK_SIZE;
1872 :
1873 : /*
1874 : * Setup a scan key to find chunks in toast table with matching va_valueid
1875 : */
1876 0 : ScanKeyInit(&toastkey,
1877 : (AttrNumber) 1,
1878 : BTEqualStrategyNumber, F_OIDEQ,
1879 0 : ObjectIdGetDatum(ta->toast_pointer.va_valueid));
1880 :
1881 : /*
1882 : * Check if any chunks for this toasted object exist in the toast table,
1883 : * accessible via the index.
1884 : */
1885 0 : toastscan = systable_beginscan_ordered(ctx->toast_rel,
1886 0 : ctx->valid_toast_index,
1887 0 : get_toast_snapshot(), 1,
1888 : &toastkey);
1889 0 : found_toasttup = false;
1890 0 : while ((toasttup =
1891 0 : systable_getnext_ordered(toastscan,
1892 0 : ForwardScanDirection)) != NULL)
1893 : {
1894 0 : found_toasttup = true;
1895 0 : check_toast_tuple(toasttup, ctx, ta, &expected_chunk_seq, extsize);
1896 : }
1897 0 : systable_endscan_ordered(toastscan);
1898 :
1899 0 : if (!found_toasttup)
1900 0 : report_toast_corruption(ctx, ta,
1901 0 : psprintf("toast value %u not found in toast table",
1902 0 : ta->toast_pointer.va_valueid));
1903 0 : else if (expected_chunk_seq <= last_chunk_seq)
1904 0 : report_toast_corruption(ctx, ta,
1905 0 : psprintf("toast value %u was expected to end at chunk %d, but ended while expecting chunk %d",
1906 0 : ta->toast_pointer.va_valueid,
1907 0 : last_chunk_seq, expected_chunk_seq));
1908 0 : }
1909 :
1910 : /*
1911 : * Check the current tuple as tracked in ctx, recording any corruption found in
1912 : * ctx->tupstore.
1913 : *
1914 : * We return some information about the status of xmin to aid in validating
1915 : * update chains.
1916 : */
1917 : static void
1918 0 : check_tuple(HeapCheckContext *ctx, bool *xmin_commit_status_ok,
1919 : XidCommitStatus *xmin_commit_status)
1920 : {
1921 : /*
1922 : * Check various forms of tuple header corruption, and if the header is
1923 : * too corrupt, do not continue with other checks.
1924 : */
1925 0 : if (!check_tuple_header(ctx))
1926 0 : return;
1927 :
1928 : /*
1929 : * Check tuple visibility. If the inserting transaction aborted, we
1930 : * cannot assume our relation description matches the tuple structure, and
1931 : * therefore cannot check it.
1932 : */
1933 0 : if (!check_tuple_visibility(ctx, xmin_commit_status_ok,
1934 0 : xmin_commit_status))
1935 0 : return;
1936 :
1937 : /*
1938 : * The tuple is visible, so it must be compatible with the current version
1939 : * of the relation descriptor. It might have fewer columns than are
1940 : * present in the relation descriptor, but it cannot have more.
1941 : */
1942 0 : if (RelationGetDescr(ctx->rel)->natts < ctx->natts)
1943 : {
1944 0 : report_corruption(ctx,
1945 0 : psprintf("number of attributes %u exceeds maximum %u expected for table",
1946 0 : ctx->natts,
1947 0 : RelationGetDescr(ctx->rel)->natts));
1948 0 : return;
1949 : }
1950 :
1951 : /*
1952 : * Check each attribute unless we hit corruption that confuses what to do
1953 : * next, at which point we abort further attribute checks for this tuple.
1954 : * Note that we don't abort for all types of corruption, only for those
1955 : * types where we don't know how to continue. We also don't abort the
1956 : * checking of toasted attributes collected from the tuple prior to
1957 : * aborting. Those will still be checked later along with other toasted
1958 : * attributes collected from the page.
1959 : */
1960 0 : ctx->offset = 0;
1961 0 : for (ctx->attnum = 0; ctx->attnum < ctx->natts; ctx->attnum++)
1962 0 : if (!check_tuple_attribute(ctx))
1963 0 : break; /* cannot continue */
1964 :
1965 : /* revert attnum to -1 until we again examine individual attributes */
1966 0 : ctx->attnum = -1;
1967 0 : }
1968 :
1969 : /*
1970 : * Convert a TransactionId into a FullTransactionId using our cached values of
1971 : * the valid transaction ID range. It is the caller's responsibility to have
1972 : * already updated the cached values, if necessary. This is akin to
1973 : * FullTransactionIdFromAllowableAt(), but it tolerates corruption in the form
1974 : * of an xid before epoch 0.
1975 : */
1976 : static FullTransactionId
1977 0 : FullTransactionIdFromXidAndCtx(TransactionId xid, const HeapCheckContext *ctx)
1978 : {
1979 0 : uint64 nextfxid_i;
1980 0 : int32 diff;
1981 0 : FullTransactionId fxid;
1982 :
1983 0 : Assert(TransactionIdIsNormal(ctx->next_xid));
1984 0 : Assert(FullTransactionIdIsNormal(ctx->next_fxid));
1985 0 : Assert(XidFromFullTransactionId(ctx->next_fxid) == ctx->next_xid);
1986 :
1987 0 : if (!TransactionIdIsNormal(xid))
1988 0 : return FullTransactionIdFromEpochAndXid(0, xid);
1989 :
1990 0 : nextfxid_i = U64FromFullTransactionId(ctx->next_fxid);
1991 :
1992 : /* compute the 32bit modulo difference */
1993 0 : diff = (int32) (ctx->next_xid - xid);
1994 :
1995 : /*
1996 : * In cases of corruption we might see a 32bit xid that is before epoch 0.
1997 : * We can't represent that as a 64bit xid, due to 64bit xids being
1998 : * unsigned integers, without the modulo arithmetic of 32bit xid. There's
1999 : * no really nice way to deal with that, but it works ok enough to use
2000 : * FirstNormalFullTransactionId in that case, as a freshly initdb'd
2001 : * cluster already has a newer horizon.
2002 : */
2003 0 : if (diff > 0 && (nextfxid_i - FirstNormalTransactionId) < (int64) diff)
2004 : {
2005 0 : Assert(EpochFromFullTransactionId(ctx->next_fxid) == 0);
2006 0 : fxid = FirstNormalFullTransactionId;
2007 0 : }
2008 : else
2009 0 : fxid = FullTransactionIdFromU64(nextfxid_i - diff);
2010 :
2011 0 : Assert(FullTransactionIdIsNormal(fxid));
2012 0 : return fxid;
2013 0 : }
2014 :
2015 : /*
2016 : * Update our cached range of valid transaction IDs.
2017 : */
2018 : static void
2019 0 : update_cached_xid_range(HeapCheckContext *ctx)
2020 : {
2021 : /* Make cached copies */
2022 0 : LWLockAcquire(XidGenLock, LW_SHARED);
2023 0 : ctx->next_fxid = TransamVariables->nextXid;
2024 0 : ctx->oldest_xid = TransamVariables->oldestXid;
2025 0 : LWLockRelease(XidGenLock);
2026 :
2027 : /* And compute alternate versions of the same */
2028 0 : ctx->next_xid = XidFromFullTransactionId(ctx->next_fxid);
2029 0 : ctx->oldest_fxid = FullTransactionIdFromXidAndCtx(ctx->oldest_xid, ctx);
2030 0 : }
2031 :
2032 : /*
2033 : * Update our cached range of valid multitransaction IDs.
2034 : */
2035 : static void
2036 0 : update_cached_mxid_range(HeapCheckContext *ctx)
2037 : {
2038 0 : ReadMultiXactIdRange(&ctx->oldest_mxact, &ctx->next_mxact);
2039 0 : }
2040 :
2041 : /*
2042 : * Return whether the given FullTransactionId is within our cached valid
2043 : * transaction ID range.
2044 : */
2045 : static inline bool
2046 0 : fxid_in_cached_range(FullTransactionId fxid, const HeapCheckContext *ctx)
2047 : {
2048 0 : return (FullTransactionIdPrecedesOrEquals(ctx->oldest_fxid, fxid) &&
2049 0 : FullTransactionIdPrecedes(fxid, ctx->next_fxid));
2050 : }
2051 :
2052 : /*
2053 : * Checks whether a multitransaction ID is in the cached valid range, returning
2054 : * the nature of the range violation, if any.
2055 : */
2056 : static XidBoundsViolation
2057 0 : check_mxid_in_range(MultiXactId mxid, HeapCheckContext *ctx)
2058 : {
2059 0 : if (!TransactionIdIsValid(mxid))
2060 0 : return XID_INVALID;
2061 0 : if (MultiXactIdPrecedes(mxid, ctx->relminmxid))
2062 0 : return XID_PRECEDES_RELMIN;
2063 0 : if (MultiXactIdPrecedes(mxid, ctx->oldest_mxact))
2064 0 : return XID_PRECEDES_CLUSTERMIN;
2065 0 : if (MultiXactIdPrecedesOrEquals(ctx->next_mxact, mxid))
2066 0 : return XID_IN_FUTURE;
2067 0 : return XID_BOUNDS_OK;
2068 0 : }
2069 :
2070 : /*
2071 : * Checks whether the given mxid is valid to appear in the heap being checked,
2072 : * returning the nature of the range violation, if any.
2073 : *
2074 : * This function attempts to return quickly by caching the known valid mxid
2075 : * range in ctx. Callers should already have performed the initial setup of
2076 : * the cache prior to the first call to this function.
2077 : */
2078 : static XidBoundsViolation
2079 0 : check_mxid_valid_in_rel(MultiXactId mxid, HeapCheckContext *ctx)
2080 : {
2081 0 : XidBoundsViolation result;
2082 :
2083 0 : result = check_mxid_in_range(mxid, ctx);
2084 0 : if (result == XID_BOUNDS_OK)
2085 0 : return XID_BOUNDS_OK;
2086 :
2087 : /* The range may have advanced. Recheck. */
2088 0 : update_cached_mxid_range(ctx);
2089 0 : return check_mxid_in_range(mxid, ctx);
2090 0 : }
2091 :
2092 : /*
2093 : * Checks whether the given transaction ID is (or was recently) valid to appear
2094 : * in the heap being checked, or whether it is too old or too new to appear in
2095 : * the relation, returning information about the nature of the bounds violation.
2096 : *
2097 : * We cache the range of valid transaction IDs. If xid is in that range, we
2098 : * conclude that it is valid, even though concurrent changes to the table might
2099 : * invalidate it under certain corrupt conditions. (For example, if the table
2100 : * contains corrupt all-frozen bits, a concurrent vacuum might skip the page(s)
2101 : * containing the xid and then truncate clog and advance the relfrozenxid
2102 : * beyond xid.) Reporting the xid as valid under such conditions seems
2103 : * acceptable, since if we had checked it earlier in our scan it would have
2104 : * truly been valid at that time.
2105 : *
2106 : * If the status argument is not NULL, and if and only if the transaction ID
2107 : * appears to be valid in this relation, the status argument will be set with
2108 : * the commit status of the transaction ID.
2109 : */
2110 : static XidBoundsViolation
2111 0 : get_xid_status(TransactionId xid, HeapCheckContext *ctx,
2112 : XidCommitStatus *status)
2113 : {
2114 0 : FullTransactionId fxid;
2115 0 : FullTransactionId clog_horizon;
2116 :
2117 : /* Quick check for special xids */
2118 0 : if (!TransactionIdIsValid(xid))
2119 0 : return XID_INVALID;
2120 0 : else if (xid == BootstrapTransactionId || xid == FrozenTransactionId)
2121 : {
2122 0 : if (status != NULL)
2123 0 : *status = XID_COMMITTED;
2124 0 : return XID_BOUNDS_OK;
2125 : }
2126 :
2127 : /* Check if the xid is within bounds */
2128 0 : fxid = FullTransactionIdFromXidAndCtx(xid, ctx);
2129 0 : if (!fxid_in_cached_range(fxid, ctx))
2130 : {
2131 : /*
2132 : * We may have been checking against stale values. Update the cached
2133 : * range to be sure, and since we relied on the cached range when we
2134 : * performed the full xid conversion, reconvert.
2135 : */
2136 0 : update_cached_xid_range(ctx);
2137 0 : fxid = FullTransactionIdFromXidAndCtx(xid, ctx);
2138 0 : }
2139 :
2140 0 : if (FullTransactionIdPrecedesOrEquals(ctx->next_fxid, fxid))
2141 0 : return XID_IN_FUTURE;
2142 0 : if (FullTransactionIdPrecedes(fxid, ctx->oldest_fxid))
2143 0 : return XID_PRECEDES_CLUSTERMIN;
2144 0 : if (FullTransactionIdPrecedes(fxid, ctx->relfrozenfxid))
2145 0 : return XID_PRECEDES_RELMIN;
2146 :
2147 : /* Early return if the caller does not request clog checking */
2148 0 : if (status == NULL)
2149 0 : return XID_BOUNDS_OK;
2150 :
2151 : /* Early return if we just checked this xid in a prior call */
2152 0 : if (xid == ctx->cached_xid)
2153 : {
2154 0 : *status = ctx->cached_status;
2155 0 : return XID_BOUNDS_OK;
2156 : }
2157 :
2158 0 : *status = XID_COMMITTED;
2159 0 : LWLockAcquire(XactTruncationLock, LW_SHARED);
2160 : clog_horizon =
2161 0 : FullTransactionIdFromXidAndCtx(TransamVariables->oldestClogXid,
2162 0 : ctx);
2163 0 : if (FullTransactionIdPrecedesOrEquals(clog_horizon, fxid))
2164 : {
2165 0 : if (TransactionIdIsCurrentTransactionId(xid))
2166 0 : *status = XID_IS_CURRENT_XID;
2167 0 : else if (TransactionIdIsInProgress(xid))
2168 0 : *status = XID_IN_PROGRESS;
2169 0 : else if (TransactionIdDidCommit(xid))
2170 0 : *status = XID_COMMITTED;
2171 : else
2172 0 : *status = XID_ABORTED;
2173 0 : }
2174 0 : LWLockRelease(XactTruncationLock);
2175 0 : ctx->cached_xid = xid;
2176 0 : ctx->cached_status = *status;
2177 0 : return XID_BOUNDS_OK;
2178 0 : }
|
