PostgreSQL Source Code  git master
xlogutils.h File Reference
#include "access/xlogreader.h"
#include "storage/bufmgr.h"
Include dependency graph for xlogutils.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros

#define InHotStandby   (standbyState >= STANDBY_SNAPSHOT_PENDING)
 

Enumerations

enum  HotStandbyState { STANDBY_DISABLED, STANDBY_INITIALIZED, STANDBY_SNAPSHOT_PENDING, STANDBY_SNAPSHOT_READY }
 
enum  XLogRedoAction { BLK_NEEDS_REDO, BLK_DONE, BLK_RESTORED, BLK_NOTFOUND }
 

Functions

bool XLogHaveInvalidPages (void)
 
void XLogCheckInvalidPages (void)
 
void XLogDropRelation (RelFileNode rnode, ForkNumber forknum)
 
void XLogDropDatabase (Oid dbid)
 
void XLogTruncateRelation (RelFileNode rnode, ForkNumber forkNum, BlockNumber nblocks)
 
XLogRedoAction XLogReadBufferForRedo (XLogReaderState *record, uint8 buffer_id, Buffer *buf)
 
Buffer XLogInitBufferForRedo (XLogReaderState *record, uint8 block_id)
 
XLogRedoAction XLogReadBufferForRedoExtended (XLogReaderState *record, uint8 buffer_id, ReadBufferMode mode, bool get_cleanup_lock, Buffer *buf)
 
Buffer XLogReadBufferExtended (RelFileNode rnode, ForkNumber forknum, BlockNumber blkno, ReadBufferMode mode)
 
Relation CreateFakeRelcacheEntry (RelFileNode rnode)
 
void FreeFakeRelcacheEntry (Relation fakerel)
 
int read_local_xlog_page (XLogReaderState *state, XLogRecPtr targetPagePtr, int reqLen, XLogRecPtr targetRecPtr, char *cur_page)
 
void wal_segment_open (XLogReaderState *state, XLogSegNo nextSegNo, TimeLineID *tli_p)
 
void wal_segment_close (XLogReaderState *state)
 
void XLogReadDetermineTimeline (XLogReaderState *state, XLogRecPtr wantPage, uint32 wantLength)
 
void WALReadRaiseError (WALReadError *errinfo)
 

Variables

bool InRecovery
 
HotStandbyState standbyState
 

Macro Definition Documentation

◆ InHotStandby

Enumeration Type Documentation

◆ HotStandbyState

Enumerator
STANDBY_DISABLED 
STANDBY_INITIALIZED 
STANDBY_SNAPSHOT_PENDING 
STANDBY_SNAPSHOT_READY 

Definition at line 47 of file xlogutils.h.

◆ XLogRedoAction

Enumerator
BLK_NEEDS_REDO 
BLK_DONE 
BLK_RESTORED 
BLK_NOTFOUND 

Definition at line 69 of file xlogutils.h.

70 {
71  BLK_NEEDS_REDO, /* changes from WAL record need to be applied */
72  BLK_DONE, /* block is already up-to-date */
73  BLK_RESTORED, /* block was restored from a full-page image */
74  BLK_NOTFOUND /* block was not found (and hence does not
75  * need to be replayed) */
XLogRedoAction
Definition: xlogutils.h:69

Function Documentation

◆ CreateFakeRelcacheEntry()

Relation CreateFakeRelcacheEntry ( RelFileNode  rnode)

Definition at line 574 of file xlogutils.c.

References LockRelId::dbId, RelFileNode::dbNode, InvalidBackendId, LockInfoData::lockRelId, palloc0(), FakeRelCacheEntryData::pgc, RelationData::rd_backend, RelationData::rd_lockInfo, RelationData::rd_node, RelationData::rd_rel, RelationData::rd_smgr, RelationGetRelationName, LockRelId::relId, RelFileNode::relNode, and sprintf.

Referenced by heap_xlog_delete(), heap_xlog_insert(), heap_xlog_lock(), heap_xlog_lock_updated(), heap_xlog_multi_insert(), heap_xlog_update(), heap_xlog_visible(), smgr_redo(), and smgrDoPendingSyncs().

575 {
576  FakeRelCacheEntry fakeentry;
577  Relation rel;
578 
579  /* Allocate the Relation struct and all related space in one block. */
580  fakeentry = palloc0(sizeof(FakeRelCacheEntryData));
581  rel = (Relation) fakeentry;
582 
583  rel->rd_rel = &fakeentry->pgc;
584  rel->rd_node = rnode;
585 
586  /*
587  * We will never be working with temp rels during recovery or while
588  * syncing WAL-skipped files.
589  */
591 
592  /* It must be a permanent table here */
593  rel->rd_rel->relpersistence = RELPERSISTENCE_PERMANENT;
594 
595  /* We don't know the name of the relation; use relfilenode instead */
596  sprintf(RelationGetRelationName(rel), "%u", rnode.relNode);
597 
598  /*
599  * We set up the lockRelId in case anything tries to lock the dummy
600  * relation. Note that this is fairly bogus since relNode may be
601  * different from the relation's OID. It shouldn't really matter though.
602  * In recovery, we are running by ourselves and can't have any lock
603  * conflicts. While syncing, we already hold AccessExclusiveLock.
604  */
605  rel->rd_lockInfo.lockRelId.dbId = rnode.dbNode;
606  rel->rd_lockInfo.lockRelId.relId = rnode.relNode;
607 
608  rel->rd_smgr = NULL;
609 
610  return rel;
611 }
LockRelId lockRelId
Definition: rel.h:45
Oid dbId
Definition: rel.h:40
Form_pg_class rd_rel
Definition: rel.h:109
#define sprintf
Definition: port.h:219
struct RelationData * Relation
Definition: relcache.h:26
LockInfoData rd_lockInfo
Definition: rel.h:112
#define RelationGetRelationName(relation)
Definition: rel.h:511
#define InvalidBackendId
Definition: backendid.h:23
void * palloc0(Size size)
Definition: mcxt.c:1093
RelFileNode rd_node
Definition: rel.h:56
BackendId rd_backend
Definition: rel.h:59
SMgrRelation rd_smgr
Definition: rel.h:57
FormData_pg_class pgc
Definition: xlogutils.c:554
Oid relId
Definition: rel.h:39

◆ FreeFakeRelcacheEntry()

void FreeFakeRelcacheEntry ( Relation  fakerel)

Definition at line 617 of file xlogutils.c.

References pfree(), RelationData::rd_smgr, and smgrclearowner().

Referenced by heap_xlog_delete(), heap_xlog_insert(), heap_xlog_lock(), heap_xlog_lock_updated(), heap_xlog_multi_insert(), heap_xlog_update(), heap_xlog_visible(), smgr_redo(), and smgrDoPendingSyncs().

618 {
619  /* make sure the fakerel is not referenced by the SmgrRelation anymore */
620  if (fakerel->rd_smgr != NULL)
621  smgrclearowner(&fakerel->rd_smgr, fakerel->rd_smgr);
622  pfree(fakerel);
623 }
void smgrclearowner(SMgrRelation *owner, SMgrRelation reln)
Definition: smgr.c:227
void pfree(void *pointer)
Definition: mcxt.c:1169
SMgrRelation rd_smgr
Definition: rel.h:57

◆ read_local_xlog_page()

int read_local_xlog_page ( XLogReaderState state,
XLogRecPtr  targetPagePtr,
int  reqLen,
XLogRecPtr  targetRecPtr,
char *  cur_page 
)

Definition at line 848 of file xlogutils.c.

References CHECK_FOR_INTERRUPTS, XLogReaderState::currTLI, XLogReaderState::currTLIValidUntil, GetFlushRecPtr(), GetXLogReplayRecPtr(), pg_usleep(), RecoveryInProgress(), ThisTimeLineID, WALRead(), WALReadRaiseError(), and XLogReadDetermineTimeline().

Referenced by create_logical_replication_slot(), pg_logical_replication_slot_advance(), pg_logical_slot_get_changes_guts(), and XlogReadTwoPhaseData().

850 {
851  XLogRecPtr read_upto,
852  loc;
853  TimeLineID tli;
854  int count;
855  WALReadError errinfo;
856 
857  loc = targetPagePtr + reqLen;
858 
859  /* Loop waiting for xlog to be available if necessary */
860  while (1)
861  {
862  /*
863  * Determine the limit of xlog we can currently read to, and what the
864  * most recent timeline is.
865  *
866  * RecoveryInProgress() will update ThisTimeLineID when it first
867  * notices recovery finishes, so we only have to maintain it for the
868  * local process until recovery ends.
869  */
870  if (!RecoveryInProgress())
871  read_upto = GetFlushRecPtr();
872  else
873  read_upto = GetXLogReplayRecPtr(&ThisTimeLineID);
874  tli = ThisTimeLineID;
875 
876  /*
877  * Check which timeline to get the record from.
878  *
879  * We have to do it each time through the loop because if we're in
880  * recovery as a cascading standby, the current timeline might've
881  * become historical. We can't rely on RecoveryInProgress() because in
882  * a standby configuration like
883  *
884  * A => B => C
885  *
886  * if we're a logical decoding session on C, and B gets promoted, our
887  * timeline will change while we remain in recovery.
888  *
889  * We can't just keep reading from the old timeline as the last WAL
890  * archive in the timeline will get renamed to .partial by
891  * StartupXLOG().
892  *
893  * If that happens after our caller updated ThisTimeLineID but before
894  * we actually read the xlog page, we might still try to read from the
895  * old (now renamed) segment and fail. There's not much we can do
896  * about this, but it can only happen when we're a leaf of a cascading
897  * standby whose primary gets promoted while we're decoding, so a
898  * one-off ERROR isn't too bad.
899  */
900  XLogReadDetermineTimeline(state, targetPagePtr, reqLen);
901 
902  if (state->currTLI == ThisTimeLineID)
903  {
904 
905  if (loc <= read_upto)
906  break;
907 
909  pg_usleep(1000L);
910  }
911  else
912  {
913  /*
914  * We're on a historical timeline, so limit reading to the switch
915  * point where we moved to the next timeline.
916  *
917  * We don't need to GetFlushRecPtr or GetXLogReplayRecPtr. We know
918  * about the new timeline, so we must've received past the end of
919  * it.
920  */
921  read_upto = state->currTLIValidUntil;
922 
923  /*
924  * Setting tli to our wanted record's TLI is slightly wrong; the
925  * page might begin on an older timeline if it contains a timeline
926  * switch, since its xlog segment will have been copied from the
927  * prior timeline. This is pretty harmless though, as nothing
928  * cares so long as the timeline doesn't go backwards. We should
929  * read the page header instead; FIXME someday.
930  */
931  tli = state->currTLI;
932 
933  /* No need to wait on a historical timeline */
934  break;
935  }
936  }
937 
938  if (targetPagePtr + XLOG_BLCKSZ <= read_upto)
939  {
940  /*
941  * more than one block available; read only that block, have caller
942  * come back if they need more.
943  */
944  count = XLOG_BLCKSZ;
945  }
946  else if (targetPagePtr + reqLen > read_upto)
947  {
948  /* not enough data there */
949  return -1;
950  }
951  else
952  {
953  /* enough bytes available to satisfy the request */
954  count = read_upto - targetPagePtr;
955  }
956 
957  /*
958  * Even though we just determined how much of the page can be validly read
959  * as 'count', read the whole page anyway. It's guaranteed to be
960  * zero-padded up to the page boundary if it's incomplete.
961  */
962  if (!WALRead(state, cur_page, targetPagePtr, XLOG_BLCKSZ, tli,
963  &errinfo))
964  WALReadRaiseError(&errinfo);
965 
966  /* number of valid bytes in the buffer */
967  return count;
968 }
uint32 TimeLineID
Definition: xlogdefs.h:59
void WALReadRaiseError(WALReadError *errinfo)
Definition: xlogutils.c:975
XLogRecPtr GetFlushRecPtr(void)
Definition: xlog.c:8680
bool RecoveryInProgress(void)
Definition: xlog.c:8328
void XLogReadDetermineTimeline(XLogReaderState *state, XLogRecPtr wantPage, uint32 wantLength)
Definition: xlogutils.c:707
void pg_usleep(long microsec)
Definition: signal.c:53
XLogRecPtr GetXLogReplayRecPtr(TimeLineID *replayTLI)
Definition: xlog.c:11927
XLogRecPtr currTLIValidUntil
Definition: xlogreader.h:248
TimeLineID ThisTimeLineID
Definition: xlog.c:194
TimeLineID currTLI
Definition: xlogreader.h:238
uint64 XLogRecPtr
Definition: xlogdefs.h:21
bool WALRead(XLogReaderState *state, char *buf, XLogRecPtr startptr, Size count, TimeLineID tli, WALReadError *errinfo)
Definition: xlogreader.c:1100
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:120

◆ wal_segment_close()

void wal_segment_close ( XLogReaderState state)

◆ wal_segment_open()

void wal_segment_open ( XLogReaderState state,
XLogSegNo  nextSegNo,
TimeLineID tli_p 
)

Definition at line 804 of file xlogutils.c.

References BasicOpenFile(), ereport, errcode_for_file_access(), errmsg(), ERROR, MAXPGPATH, PG_BINARY, XLogReaderState::seg, XLogReaderState::segcxt, WALOpenSegment::ws_file, WALSegmentContext::ws_segsize, and XLogFilePath.

Referenced by create_logical_replication_slot(), pg_logical_replication_slot_advance(), pg_logical_slot_get_changes_guts(), and XlogReadTwoPhaseData().

806 {
807  TimeLineID tli = *tli_p;
808  char path[MAXPGPATH];
809 
810  XLogFilePath(path, tli, nextSegNo, state->segcxt.ws_segsize);
811  state->seg.ws_file = BasicOpenFile(path, O_RDONLY | PG_BINARY);
812  if (state->seg.ws_file >= 0)
813  return;
814 
815  if (errno == ENOENT)
816  ereport(ERROR,
818  errmsg("requested WAL segment %s has already been removed",
819  path)));
820  else
821  ereport(ERROR,
823  errmsg("could not open file \"%s\": %m",
824  path)));
825 }
uint32 TimeLineID
Definition: xlogdefs.h:59
#define PG_BINARY
Definition: c.h:1271
WALOpenSegment seg
Definition: xlogreader.h:225
#define ERROR
Definition: elog.h:46
#define MAXPGPATH
int errcode_for_file_access(void)
Definition: elog.c:721
#define ereport(elevel,...)
Definition: elog.h:157
int BasicOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1069
#define XLogFilePath(path, tli, logSegNo, wal_segsz_bytes)
int errmsg(const char *fmt,...)
Definition: elog.c:909
WALSegmentContext segcxt
Definition: xlogreader.h:224

◆ WALReadRaiseError()

void WALReadRaiseError ( WALReadError errinfo)

Definition at line 975 of file xlogutils.c.

References ereport, errcode(), ERRCODE_DATA_CORRUPTED, errcode_for_file_access(), errmsg(), ERROR, MAXFNAMELEN, wal_segment_size, WALReadError::wre_errno, WALReadError::wre_off, WALReadError::wre_read, WALReadError::wre_req, WALReadError::wre_seg, WALOpenSegment::ws_segno, WALOpenSegment::ws_tli, and XLogFileName.

Referenced by logical_read_xlog_page(), read_local_xlog_page(), and XLogSendPhysical().

976 {
977  WALOpenSegment *seg = &errinfo->wre_seg;
978  char fname[MAXFNAMELEN];
979 
980  XLogFileName(fname, seg->ws_tli, seg->ws_segno, wal_segment_size);
981 
982  if (errinfo->wre_read < 0)
983  {
984  errno = errinfo->wre_errno;
985  ereport(ERROR,
987  errmsg("could not read from log segment %s, offset %u: %m",
988  fname, errinfo->wre_off)));
989  }
990  else if (errinfo->wre_read == 0)
991  {
992  ereport(ERROR,
994  errmsg("could not read from log segment %s, offset %u: read %d of %d",
995  fname, errinfo->wre_off, errinfo->wre_read,
996  errinfo->wre_req)));
997  }
998 }
WALOpenSegment wre_seg
Definition: xlogreader.h:301
int wal_segment_size
Definition: xlog.c:119
int errcode(int sqlerrcode)
Definition: elog.c:698
#define ERROR
Definition: elog.h:46
XLogSegNo ws_segno
Definition: xlogreader.h:47
int errcode_for_file_access(void)
Definition: elog.c:721
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:47
#define MAXFNAMELEN
#define ereport(elevel,...)
Definition: elog.h:157
#define XLogFileName(fname, tli, logSegNo, wal_segsz_bytes)
TimeLineID ws_tli
Definition: xlogreader.h:48
int errmsg(const char *fmt,...)
Definition: elog.c:909

◆ XLogCheckInvalidPages()

void XLogCheckInvalidPages ( void  )

Definition at line 243 of file xlogutils.c.

References xl_invalid_page_key::blkno, elog, xl_invalid_page_key::forkno, hash_destroy(), hash_seq_init(), hash_seq_search(), ignore_invalid_pages, xl_invalid_page::key, xl_invalid_page_key::node, PANIC, xl_invalid_page::present, report_invalid_page(), status(), and WARNING.

Referenced by CheckRecoveryConsistency().

244 {
246  xl_invalid_page *hentry;
247  bool foundone = false;
248 
249  if (invalid_page_tab == NULL)
250  return; /* nothing to do */
251 
253 
254  /*
255  * Our strategy is to emit WARNING messages for all remaining entries and
256  * only PANIC after we've dumped all the available info.
257  */
258  while ((hentry = (xl_invalid_page *) hash_seq_search(&status)) != NULL)
259  {
260  report_invalid_page(WARNING, hentry->key.node, hentry->key.forkno,
261  hentry->key.blkno, hentry->present);
262  foundone = true;
263  }
264 
265  if (foundone)
267  "WAL contains references to invalid pages");
268 
270  invalid_page_tab = NULL;
271 }
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:862
ForkNumber forkno
Definition: xlogutils.c:70
#define PANIC
Definition: elog.h:50
RelFileNode node
Definition: xlogutils.c:69
static void report_invalid_page(int elevel, RelFileNode node, ForkNumber forkno, BlockNumber blkno, bool present)
Definition: xlogutils.c:85
BlockNumber blkno
Definition: xlogutils.c:71
xl_invalid_page_key key
Definition: xlogutils.c:76
bool ignore_invalid_pages
Definition: xlogutils.c:36
#define WARNING
Definition: elog.h:40
void * hash_seq_search(HASH_SEQ_STATUS *status)
Definition: dynahash.c:1436
void hash_seq_init(HASH_SEQ_STATUS *status, HTAB *hashp)
Definition: dynahash.c:1426
#define elog(elevel,...)
Definition: elog.h:232
static HTAB * invalid_page_tab
Definition: xlogutils.c:80
static void static void status(const char *fmt,...) pg_attribute_printf(1
Definition: pg_regress.c:227

◆ XLogDropDatabase()

void XLogDropDatabase ( Oid  dbid)

Definition at line 643 of file xlogutils.c.

References forget_invalid_pages_db(), and smgrcloseall().

Referenced by dbase_redo().

644 {
645  /*
646  * This is unnecessarily heavy-handed, as it will close SMgrRelation
647  * objects for other databases as well. DROP DATABASE occurs seldom enough
648  * that it's not worth introducing a variant of smgrclose for just this
649  * purpose. XXX: Or should we rather leave the smgr entries dangling?
650  */
651  smgrcloseall();
652 
654 }
static void forget_invalid_pages_db(Oid dbid)
Definition: xlogutils.c:200
void smgrcloseall(void)
Definition: smgr.c:286

◆ XLogDropRelation()

void XLogDropRelation ( RelFileNode  rnode,
ForkNumber  forknum 
)

Definition at line 632 of file xlogutils.c.

References forget_invalid_pages().

Referenced by DropRelationFiles().

633 {
634  forget_invalid_pages(rnode, forknum, 0);
635 }
static void forget_invalid_pages(RelFileNode node, ForkNumber forkno, BlockNumber minblkno)
Definition: xlogutils.c:165

◆ XLogHaveInvalidPages()

bool XLogHaveInvalidPages ( void  )

Definition at line 233 of file xlogutils.c.

References hash_get_num_entries().

Referenced by RecoveryRestartPoint().

234 {
235  if (invalid_page_tab != NULL &&
237  return true;
238  return false;
239 }
long hash_get_num_entries(HTAB *hashp)
Definition: dynahash.c:1382
static HTAB * invalid_page_tab
Definition: xlogutils.c:80

◆ XLogInitBufferForRedo()

◆ XLogReadBufferExtended()

Buffer XLogReadBufferExtended ( RelFileNode  rnode,
ForkNumber  forknum,
BlockNumber  blkno,
ReadBufferMode  mode 
)

Definition at line 459 of file xlogutils.c.

References Assert, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage, InRecovery, InvalidBackendId, InvalidBuffer, LockBuffer(), log_invalid_page(), P_NEW, PageIsNew, RBM_NORMAL, RBM_NORMAL_NO_LOG, RBM_ZERO_AND_CLEANUP_LOCK, RBM_ZERO_AND_LOCK, ReadBufferWithoutRelcache(), ReleaseBuffer(), smgrcreate(), smgrnblocks(), and smgropen().

Referenced by checkXLogConsistency(), XLogReadBufferForRedoExtended(), and XLogRecordPageWithFreeSpace().

461 {
462  BlockNumber lastblock;
463  Buffer buffer;
464  SMgrRelation smgr;
465 
466  Assert(blkno != P_NEW);
467 
468  /* Open the relation at smgr level */
469  smgr = smgropen(rnode, InvalidBackendId);
470 
471  /*
472  * Create the target file if it doesn't already exist. This lets us cope
473  * if the replay sequence contains writes to a relation that is later
474  * deleted. (The original coding of this routine would instead suppress
475  * the writes, but that seems like it risks losing valuable data if the
476  * filesystem loses an inode during a crash. Better to write the data
477  * until we are actually told to delete the file.)
478  */
479  smgrcreate(smgr, forknum, true);
480 
481  lastblock = smgrnblocks(smgr, forknum);
482 
483  if (blkno < lastblock)
484  {
485  /* page exists in file */
486  buffer = ReadBufferWithoutRelcache(rnode, forknum, blkno,
487  mode, NULL);
488  }
489  else
490  {
491  /* hm, page doesn't exist in file */
492  if (mode == RBM_NORMAL)
493  {
494  log_invalid_page(rnode, forknum, blkno, false);
495  return InvalidBuffer;
496  }
497  if (mode == RBM_NORMAL_NO_LOG)
498  return InvalidBuffer;
499  /* OK to extend the file */
500  /* we do this in recovery only - no rel-extension lock needed */
502  buffer = InvalidBuffer;
503  do
504  {
505  if (buffer != InvalidBuffer)
506  {
509  ReleaseBuffer(buffer);
510  }
511  buffer = ReadBufferWithoutRelcache(rnode, forknum,
512  P_NEW, mode, NULL);
513  }
514  while (BufferGetBlockNumber(buffer) < blkno);
515  /* Handle the corner case that P_NEW returns non-consecutive pages */
516  if (BufferGetBlockNumber(buffer) != blkno)
517  {
520  ReleaseBuffer(buffer);
521  buffer = ReadBufferWithoutRelcache(rnode, forknum, blkno,
522  mode, NULL);
523  }
524  }
525 
526  if (mode == RBM_NORMAL)
527  {
528  /* check that page has been initialized */
529  Page page = (Page) BufferGetPage(buffer);
530 
531  /*
532  * We assume that PageIsNew is safe without a lock. During recovery,
533  * there should be no other backends that could modify the buffer at
534  * the same time.
535  */
536  if (PageIsNew(page))
537  {
538  ReleaseBuffer(buffer);
539  log_invalid_page(rnode, forknum, blkno, true);
540  return InvalidBuffer;
541  }
542  }
543 
544  return buffer;
545 }
static PgChecksumMode mode
Definition: pg_checksums.c:65
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:96
void smgrcreate(SMgrRelation reln, ForkNumber forknum, bool isRedo)
Definition: smgr.c:333
#define InvalidBuffer
Definition: buf.h:25
Buffer ReadBufferWithoutRelcache(RelFileNode rnode, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:780
uint32 BlockNumber
Definition: block.h:31
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3768
#define P_NEW
Definition: bufmgr.h:91
static void log_invalid_page(RelFileNode node, ForkNumber forkno, BlockNumber blkno, bool present)
Definition: xlogutils.c:101
#define BufferGetPage(buffer)
Definition: bufmgr.h:169
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:146
#define InvalidBackendId
Definition: backendid.h:23
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:4007
bool InRecovery
Definition: xlogutils.c:52
BlockNumber smgrnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:548
#define Assert(condition)
Definition: c.h:804
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2748
#define PageIsNew(page)
Definition: bufpage.h:229
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:78

◆ XLogReadBufferForRedo()

XLogRedoAction XLogReadBufferForRedo ( XLogReaderState record,
uint8  buffer_id,
Buffer buf 
)

Definition at line 312 of file xlogutils.c.

References RBM_NORMAL, and XLogReadBufferForRedoExtended().

Referenced by _bt_clear_incomplete_split(), brin_xlog_desummarize_page(), brin_xlog_insert_update(), brin_xlog_revmap_extend(), brin_xlog_samepage_update(), brin_xlog_update(), btree_xlog_dedup(), btree_xlog_delete(), btree_xlog_insert(), btree_xlog_mark_page_halfdead(), btree_xlog_split(), btree_xlog_unlink_page(), generic_redo(), ginRedoClearIncompleteSplit(), ginRedoDeletePage(), ginRedoInsert(), ginRedoSplit(), ginRedoUpdateMetapage(), ginRedoVacuumDataLeafPage(), ginRedoVacuumPage(), gistRedoClearFollowRight(), gistRedoDeleteRecord(), gistRedoPageDelete(), gistRedoPageUpdateRecord(), hash_xlog_add_ovfl_page(), hash_xlog_delete(), hash_xlog_init_bitmap_page(), hash_xlog_insert(), hash_xlog_move_page_contents(), hash_xlog_split_allocate_page(), hash_xlog_split_cleanup(), hash_xlog_split_complete(), hash_xlog_split_page(), hash_xlog_squeeze_page(), hash_xlog_update_meta_page(), hash_xlog_vacuum_one_page(), heap_xlog_confirm(), heap_xlog_delete(), heap_xlog_freeze_page(), heap_xlog_inplace(), heap_xlog_insert(), heap_xlog_lock(), heap_xlog_lock_updated(), heap_xlog_multi_insert(), heap_xlog_update(), heap_xlog_visible(), spgRedoAddLeaf(), spgRedoAddNode(), spgRedoMoveLeafs(), spgRedoPickSplit(), spgRedoSplitTuple(), spgRedoVacuumLeaf(), spgRedoVacuumRedirect(), spgRedoVacuumRoot(), and xlog_redo().

314 {
315  return XLogReadBufferForRedoExtended(record, block_id, RBM_NORMAL,
316  false, buf);
317 }
static char * buf
Definition: pg_test_fsync.c:68
XLogRedoAction XLogReadBufferForRedoExtended(XLogReaderState *record, uint8 block_id, ReadBufferMode mode, bool get_cleanup_lock, Buffer *buf)
Definition: xlogutils.c:349

◆ XLogReadBufferForRedoExtended()

XLogRedoAction XLogReadBufferForRedoExtended ( XLogReaderState record,
uint8  buffer_id,
ReadBufferMode  mode,
bool  get_cleanup_lock,
Buffer buf 
)

Definition at line 349 of file xlogutils.c.

References Assert, BKPBLOCK_WILL_INIT, BLK_DONE, BLK_NEEDS_REDO, BLK_NOTFOUND, BLK_RESTORED, xl_invalid_page_key::blkno, XLogReaderState::blocks, BUFFER_LOCK_EXCLUSIVE, BufferGetPage, BufferIsValid, elog, XLogReaderState::EndRecPtr, ERROR, DecodedBkpBlock::flags, FlushOneBuffer(), INIT_FORKNUM, LockBuffer(), LockBufferForCleanup(), MarkBufferDirty(), PageGetLSN, PageIsNew, PageSetLSN, PANIC, RBM_ZERO_AND_CLEANUP_LOCK, RBM_ZERO_AND_LOCK, RestoreBlockImage(), XLogReadBufferExtended(), XLogRecBlockImageApply, XLogRecGetBlockTag(), and XLogRecHasBlockImage.

Referenced by btree_xlog_vacuum(), hash_xlog_delete(), hash_xlog_move_page_contents(), hash_xlog_split_allocate_page(), hash_xlog_squeeze_page(), hash_xlog_vacuum_one_page(), heap_xlog_prune(), heap_xlog_vacuum(), heap_xlog_visible(), XLogInitBufferForRedo(), and XLogReadBufferForRedo().

353 {
354  XLogRecPtr lsn = record->EndRecPtr;
355  RelFileNode rnode;
356  ForkNumber forknum;
357  BlockNumber blkno;
358  Page page;
359  bool zeromode;
360  bool willinit;
361 
362  if (!XLogRecGetBlockTag(record, block_id, &rnode, &forknum, &blkno))
363  {
364  /* Caller specified a bogus block_id */
365  elog(PANIC, "failed to locate backup block with ID %d", block_id);
366  }
367 
368  /*
369  * Make sure that if the block is marked with WILL_INIT, the caller is
370  * going to initialize it. And vice versa.
371  */
373  willinit = (record->blocks[block_id].flags & BKPBLOCK_WILL_INIT) != 0;
374  if (willinit && !zeromode)
375  elog(PANIC, "block with WILL_INIT flag in WAL record must be zeroed by redo routine");
376  if (!willinit && zeromode)
377  elog(PANIC, "block to be initialized in redo routine must be marked with WILL_INIT flag in the WAL record");
378 
379  /* If it has a full-page image and it should be restored, do it. */
380  if (XLogRecBlockImageApply(record, block_id))
381  {
382  Assert(XLogRecHasBlockImage(record, block_id));
383  *buf = XLogReadBufferExtended(rnode, forknum, blkno,
384  get_cleanup_lock ? RBM_ZERO_AND_CLEANUP_LOCK : RBM_ZERO_AND_LOCK);
385  page = BufferGetPage(*buf);
386  if (!RestoreBlockImage(record, block_id, page))
387  elog(ERROR, "failed to restore block image");
388 
389  /*
390  * The page may be uninitialized. If so, we can't set the LSN because
391  * that would corrupt the page.
392  */
393  if (!PageIsNew(page))
394  {
395  PageSetLSN(page, lsn);
396  }
397 
399 
400  /*
401  * At the end of crash recovery the init forks of unlogged relations
402  * are copied, without going through shared buffers. So we need to
403  * force the on-disk state of init forks to always be in sync with the
404  * state in shared buffers.
405  */
406  if (forknum == INIT_FORKNUM)
408 
409  return BLK_RESTORED;
410  }
411  else
412  {
413  *buf = XLogReadBufferExtended(rnode, forknum, blkno, mode);
414  if (BufferIsValid(*buf))
415  {
417  {
418  if (get_cleanup_lock)
420  else
422  }
423  if (lsn <= PageGetLSN(BufferGetPage(*buf)))
424  return BLK_DONE;
425  else
426  return BLK_NEEDS_REDO;
427  }
428  else
429  return BLK_NOTFOUND;
430  }
431 }
static PgChecksumMode mode
Definition: pg_checksums.c:65
void LockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:4064
#define XLogRecHasBlockImage(decoder, block_id)
Definition: xlogreader.h:325
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1565
Buffer XLogReadBufferExtended(RelFileNode rnode, ForkNumber forknum, BlockNumber blkno, ReadBufferMode mode)
Definition: xlogutils.c:459
uint32 BlockNumber
Definition: block.h:31
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:98
#define PANIC
Definition: elog.h:50
XLogRecPtr EndRecPtr
Definition: xlogreader.h:176
#define ERROR
Definition: elog.h:46
static char * buf
Definition: pg_test_fsync.c:68
#define BufferGetPage(buffer)
Definition: bufmgr.h:169
#define BKPBLOCK_WILL_INIT
Definition: xlogrecord.h:186
ForkNumber
Definition: relpath.h:40
bool XLogRecGetBlockTag(XLogReaderState *record, uint8 block_id, RelFileNode *rnode, ForkNumber *forknum, BlockNumber *blknum)
Definition: xlogreader.c:1531
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:4007
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:804
#define BufferIsValid(bufnum)
Definition: bufmgr.h:123
void FlushOneBuffer(Buffer buffer)
Definition: bufmgr.c:3748
bool RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page)
Definition: xlogreader.c:1584
#define PageGetLSN(page)
Definition: bufpage.h:366
#define PageIsNew(page)
Definition: bufpage.h:229
#define elog(elevel,...)
Definition: elog.h:232
#define XLogRecBlockImageApply(decoder, block_id)
Definition: xlogreader.h:327
#define PageSetLSN(page, lsn)
Definition: bufpage.h:368
Pointer Page
Definition: bufpage.h:78
DecodedBkpBlock blocks[XLR_MAX_BLOCK_ID+1]
Definition: xlogreader.h:207

◆ XLogReadDetermineTimeline()

void XLogReadDetermineTimeline ( XLogReaderState state,
XLogRecPtr  wantPage,
uint32  wantLength 
)

Definition at line 707 of file xlogutils.c.

References Assert, XLogReaderState::currTLI, XLogReaderState::currTLIValidUntil, DEBUG3, elog, InvalidXLogRecPtr, list_free_deep(), LSN_FORMAT_ARGS, Min, XLogReaderState::nextTLI, XLogReaderState::readLen, readTimeLineHistory(), XLogReaderState::seg, XLogReaderState::segcxt, XLogReaderState::segoff, ThisTimeLineID, tliOfPointInHistory(), tliSwitchPoint(), WALOpenSegment::ws_segno, and WALSegmentContext::ws_segsize.

Referenced by logical_read_xlog_page(), and read_local_xlog_page().

708 {
709  const XLogRecPtr lastReadPage = (state->seg.ws_segno *
710  state->segcxt.ws_segsize + state->segoff);
711 
712  Assert(wantPage != InvalidXLogRecPtr && wantPage % XLOG_BLCKSZ == 0);
713  Assert(wantLength <= XLOG_BLCKSZ);
714  Assert(state->readLen == 0 || state->readLen <= XLOG_BLCKSZ);
715 
716  /*
717  * If the desired page is currently read in and valid, we have nothing to
718  * do.
719  *
720  * The caller should've ensured that it didn't previously advance readOff
721  * past the valid limit of this timeline, so it doesn't matter if the
722  * current TLI has since become historical.
723  */
724  if (lastReadPage == wantPage &&
725  state->readLen != 0 &&
726  lastReadPage + state->readLen >= wantPage + Min(wantLength, XLOG_BLCKSZ - 1))
727  return;
728 
729  /*
730  * If we're reading from the current timeline, it hasn't become historical
731  * and the page we're reading is after the last page read, we can again
732  * just carry on. (Seeking backwards requires a check to make sure the
733  * older page isn't on a prior timeline).
734  *
735  * ThisTimeLineID might've become historical since we last looked, but the
736  * caller is required not to read past the flush limit it saw at the time
737  * it looked up the timeline. There's nothing we can do about it if
738  * StartupXLOG() renames it to .partial concurrently.
739  */
740  if (state->currTLI == ThisTimeLineID && wantPage >= lastReadPage)
741  {
743  return;
744  }
745 
746  /*
747  * If we're just reading pages from a previously validated historical
748  * timeline and the timeline we're reading from is valid until the end of
749  * the current segment we can just keep reading.
750  */
751  if (state->currTLIValidUntil != InvalidXLogRecPtr &&
752  state->currTLI != ThisTimeLineID &&
753  state->currTLI != 0 &&
754  ((wantPage + wantLength) / state->segcxt.ws_segsize) <
755  (state->currTLIValidUntil / state->segcxt.ws_segsize))
756  return;
757 
758  /*
759  * If we reach this point we're either looking up a page for random
760  * access, the current timeline just became historical, or we're reading
761  * from a new segment containing a timeline switch. In all cases we need
762  * to determine the newest timeline on the segment.
763  *
764  * If it's the current timeline we can just keep reading from here unless
765  * we detect a timeline switch that makes the current timeline historical.
766  * If it's a historical timeline we can read all the segment on the newest
767  * timeline because it contains all the old timelines' data too. So only
768  * one switch check is required.
769  */
770  {
771  /*
772  * We need to re-read the timeline history in case it's been changed
773  * by a promotion or replay from a cascaded replica.
774  */
775  List *timelineHistory = readTimeLineHistory(ThisTimeLineID);
776  XLogRecPtr endOfSegment;
777 
778  endOfSegment = ((wantPage / state->segcxt.ws_segsize) + 1) *
779  state->segcxt.ws_segsize - 1;
780  Assert(wantPage / state->segcxt.ws_segsize ==
781  endOfSegment / state->segcxt.ws_segsize);
782 
783  /*
784  * Find the timeline of the last LSN on the segment containing
785  * wantPage.
786  */
787  state->currTLI = tliOfPointInHistory(endOfSegment, timelineHistory);
788  state->currTLIValidUntil = tliSwitchPoint(state->currTLI, timelineHistory,
789  &state->nextTLI);
790 
792  wantPage + wantLength < state->currTLIValidUntil);
793 
794  list_free_deep(timelineHistory);
795 
796  elog(DEBUG3, "switched to timeline %u valid until %X/%X",
797  state->currTLI,
799  }
800 }
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
TimeLineID tliOfPointInHistory(XLogRecPtr ptr, List *history)
Definition: timeline.c:552
#define DEBUG3
Definition: elog.h:23
#define Min(x, y)
Definition: c.h:986
List * readTimeLineHistory(TimeLineID targetTLI)
Definition: timeline.c:76
void list_free_deep(List *list)
Definition: list.c:1405
WALOpenSegment seg
Definition: xlogreader.h:225
#define LSN_FORMAT_ARGS(lsn)
Definition: xlogdefs.h:43
XLogSegNo ws_segno
Definition: xlogreader.h:47
XLogRecPtr currTLIValidUntil
Definition: xlogreader.h:248
TimeLineID nextTLI
Definition: xlogreader.h:254
XLogRecPtr tliSwitchPoint(TimeLineID tli, List *history, TimeLineID *nextTLI)
Definition: timeline.c:580
TimeLineID ThisTimeLineID
Definition: xlog.c:194
TimeLineID currTLI
Definition: xlogreader.h:238
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:804
#define elog(elevel,...)
Definition: elog.h:232
WALSegmentContext segcxt
Definition: xlogreader.h:224
Definition: pg_list.h:50

◆ XLogTruncateRelation()

void XLogTruncateRelation ( RelFileNode  rnode,
ForkNumber  forkNum,
BlockNumber  nblocks 
)

Definition at line 662 of file xlogutils.c.

References forget_invalid_pages().

Referenced by smgr_redo().

664 {
665  forget_invalid_pages(rnode, forkNum, nblocks);
666 }
static void forget_invalid_pages(RelFileNode node, ForkNumber forkno, BlockNumber minblkno)
Definition: xlogutils.c:165

Variable Documentation

◆ InRecovery

◆ standbyState