PostgreSQL Source Code  git master
md.h File Reference
#include "storage/block.h"
#include "storage/relfilenode.h"
#include "storage/smgr.h"
#include "storage/sync.h"
Include dependency graph for md.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Functions

void mdinit (void)
 
void mdopen (SMgrRelation reln)
 
void mdclose (SMgrRelation reln, ForkNumber forknum)
 
void mdcreate (SMgrRelation reln, ForkNumber forknum, bool isRedo)
 
bool mdexists (SMgrRelation reln, ForkNumber forknum)
 
void mdunlink (RelFileNodeBackend rnode, ForkNumber forknum, bool isRedo)
 
void mdextend (SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
 
bool mdprefetch (SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
 
void mdread (SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer)
 
void mdwrite (SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
 
void mdwriteback (SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, BlockNumber nblocks)
 
BlockNumber mdnblocks (SMgrRelation reln, ForkNumber forknum)
 
void mdtruncate (SMgrRelation reln, ForkNumber forknum, BlockNumber nblocks)
 
void mdimmedsync (SMgrRelation reln, ForkNumber forknum)
 
void ForgetDatabaseSyncRequests (Oid dbid)
 
void DropRelationFiles (RelFileNode *delrels, int ndelrels, bool isRedo)
 
int mdsyncfiletag (const FileTag *ftag, char *path)
 
int mdunlinkfiletag (const FileTag *ftag, char *path)
 
bool mdfiletagmatches (const FileTag *ftag, const FileTag *candidate)
 

Function Documentation

◆ DropRelationFiles()

void DropRelationFiles ( RelFileNode delrels,
int  ndelrels,
bool  isRedo 
)

Definition at line 1049 of file md.c.

References i, InvalidBackendId, MAX_FORKNUM, palloc(), pfree(), smgrclose(), smgrdounlinkall(), smgropen(), and XLogDropRelation().

Referenced by FinishPreparedTransaction(), xact_redo_abort(), and xact_redo_commit().

1050 {
1051  SMgrRelation *srels;
1052  int i;
1053 
1054  srels = palloc(sizeof(SMgrRelation) * ndelrels);
1055  for (i = 0; i < ndelrels; i++)
1056  {
1057  SMgrRelation srel = smgropen(delrels[i], InvalidBackendId);
1058 
1059  if (isRedo)
1060  {
1061  ForkNumber fork;
1062 
1063  for (fork = 0; fork <= MAX_FORKNUM; fork++)
1064  XLogDropRelation(delrels[i], fork);
1065  }
1066  srels[i] = srel;
1067  }
1068 
1069  smgrdounlinkall(srels, ndelrels, isRedo);
1070 
1071  for (i = 0; i < ndelrels; i++)
1072  smgrclose(srels[i]);
1073  pfree(srels);
1074 }
void smgrclose(SMgrRelation reln)
Definition: smgr.c:256
void smgrdounlinkall(SMgrRelation *rels, int nrels, bool isRedo)
Definition: smgr.c:384
void pfree(void *pointer)
Definition: mcxt.c:1169
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:146
ForkNumber
Definition: relpath.h:40
#define InvalidBackendId
Definition: backendid.h:23
#define MAX_FORKNUM
Definition: relpath.h:55
void XLogDropRelation(RelFileNode rnode, ForkNumber forknum)
Definition: xlogutils.c:612
void * palloc(Size size)
Definition: mcxt.c:1062
int i

◆ ForgetDatabaseSyncRequests()

void ForgetDatabaseSyncRequests ( Oid  dbid)

Definition at line 1031 of file md.c.

References RelFileNode::dbNode, INIT_MD_FILETAG, InvalidBlockNumber, InvalidForkNumber, RegisterSyncRequest(), RelFileNode::relNode, RelFileNode::spcNode, and SYNC_FILTER_REQUEST.

Referenced by dbase_redo(), and dropdb().

1032 {
1033  FileTag tag;
1034  RelFileNode rnode;
1035 
1036  rnode.dbNode = dbid;
1037  rnode.spcNode = 0;
1038  rnode.relNode = 0;
1039 
1041 
1042  RegisterSyncRequest(&tag, SYNC_FILTER_REQUEST, true /* retryOnError */ );
1043 }
#define INIT_MD_FILETAG(a, xx_rnode, xx_forknum, xx_segno)
Definition: md.c:92
bool RegisterSyncRequest(const FileTag *ftag, SyncRequestType type, bool retryOnError)
Definition: sync.c:553
#define InvalidBlockNumber
Definition: block.h:33
Definition: sync.h:50

◆ mdclose()

void mdclose ( SMgrRelation  reln,
ForkNumber  forknum 
)

Definition at line 532 of file md.c.

References _fdvec_resize(), FileClose(), SMgrRelationData::md_num_open_segs, SMgrRelationData::md_seg_fds, and _MdfdVec::mdfd_vfd.

Referenced by mdexists().

533 {
534  int nopensegs = reln->md_num_open_segs[forknum];
535 
536  /* No work if already closed */
537  if (nopensegs == 0)
538  return;
539 
540  /* close segments starting from the end */
541  while (nopensegs > 0)
542  {
543  MdfdVec *v = &reln->md_seg_fds[forknum][nopensegs - 1];
544 
545  FileClose(v->mdfd_vfd);
546  _fdvec_resize(reln, forknum, nopensegs - 1);
547  nopensegs--;
548  }
549 }
static void _fdvec_resize(SMgrRelation reln, ForkNumber forknum, int nseg)
Definition: md.c:1081
Definition: md.c:82
void FileClose(File file)
Definition: fd.c:1873
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69
File mdfd_vfd
Definition: md.c:84

◆ mdcreate()

void mdcreate ( SMgrRelation  reln,
ForkNumber  forknum,
bool  isRedo 
)

Definition at line 178 of file md.c.

References _fdvec_resize(), Assert, RelFileNode::dbNode, ereport, errcode_for_file_access(), errmsg(), ERROR, fd(), SMgrRelationData::md_num_open_segs, SMgrRelationData::md_seg_fds, _MdfdVec::mdfd_segno, _MdfdVec::mdfd_vfd, RelFileNodeBackend::node, PathNameOpenFile(), pfree(), PG_BINARY, relpath, SMgrRelationData::smgr_rnode, RelFileNode::spcNode, and TablespaceCreateDbspace().

179 {
180  MdfdVec *mdfd;
181  char *path;
182  File fd;
183 
184  if (isRedo && reln->md_num_open_segs[forkNum] > 0)
185  return; /* created and opened already... */
186 
187  Assert(reln->md_num_open_segs[forkNum] == 0);
188 
189  /*
190  * We may be using the target table space for the first time in this
191  * database, so create a per-database subdirectory if needed.
192  *
193  * XXX this is a fairly ugly violation of module layering, but this seems
194  * to be the best place to put the check. Maybe TablespaceCreateDbspace
195  * should be here and not in commands/tablespace.c? But that would imply
196  * importing a lot of stuff that smgr.c oughtn't know, either.
197  */
199  reln->smgr_rnode.node.dbNode,
200  isRedo);
201 
202  path = relpath(reln->smgr_rnode, forkNum);
203 
204  fd = PathNameOpenFile(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
205 
206  if (fd < 0)
207  {
208  int save_errno = errno;
209 
210  if (isRedo)
211  fd = PathNameOpenFile(path, O_RDWR | PG_BINARY);
212  if (fd < 0)
213  {
214  /* be sure to report the error reported by create, not open */
215  errno = save_errno;
216  ereport(ERROR,
218  errmsg("could not create file \"%s\": %m", path)));
219  }
220  }
221 
222  pfree(path);
223 
224  _fdvec_resize(reln, forkNum, 1);
225  mdfd = &reln->md_seg_fds[forkNum][0];
226  mdfd->mdfd_vfd = fd;
227  mdfd->mdfd_segno = 0;
228 }
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1484
BlockNumber mdfd_segno
Definition: md.c:85
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define PG_BINARY
Definition: c.h:1271
void pfree(void *pointer)
Definition: mcxt.c:1169
#define ERROR
Definition: elog.h:46
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
int errcode_for_file_access(void)
Definition: elog.c:721
static void _fdvec_resize(SMgrRelation reln, ForkNumber forknum, int nseg)
Definition: md.c:1081
Definition: md.c:82
#define ereport(elevel,...)
Definition: elog.h:157
RelFileNode node
Definition: relfilenode.h:74
#define Assert(condition)
Definition: c.h:804
void TablespaceCreateDbspace(Oid spcNode, Oid dbNode, bool isRedo)
Definition: tablespace.c:115
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
int errmsg(const char *fmt,...)
Definition: elog.c:909
#define relpath(rnode, forknum)
Definition: relpath.h:87
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69
File mdfd_vfd
Definition: md.c:84
int File
Definition: fd.h:54

◆ mdexists()

bool mdexists ( SMgrRelation  reln,
ForkNumber  forknum 
)

Definition at line 161 of file md.c.

References EXTENSION_RETURN_NULL, mdclose(), and mdopenfork().

162 {
163  /*
164  * Close it first, to ensure that we notice if the fork has been unlinked
165  * since we opened it.
166  */
167  mdclose(reln, forkNum);
168 
169  return (mdopenfork(reln, forkNum, EXTENSION_RETURN_NULL) != NULL);
170 }
static MdfdVec * mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
Definition: md.c:478
#define EXTENSION_RETURN_NULL
Definition: md.c:106
void mdclose(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:532

◆ mdextend()

void mdextend ( SMgrRelation  reln,
ForkNumber  forknum,
BlockNumber  blocknum,
char *  buffer,
bool  skipFsync 
)

Definition at line 414 of file md.c.

References _mdfd_getseg(), _mdnblocks(), Assert, ereport, errcode(), errcode_for_file_access(), errhint(), errmsg(), ERROR, EXTENSION_CREATE, FilePathName(), FileWrite(), InvalidBlockNumber, _MdfdVec::mdfd_vfd, mdnblocks(), register_dirty_segment(), relpath, SMgrRelationData::smgr_rnode, SmgrIsTemp, and WAIT_EVENT_DATA_FILE_EXTEND.

Referenced by _mdfd_getseg().

416 {
417  off_t seekpos;
418  int nbytes;
419  MdfdVec *v;
420 
421  /* This assert is too expensive to have on normally ... */
422 #ifdef CHECK_WRITE_VS_EXTEND
423  Assert(blocknum >= mdnblocks(reln, forknum));
424 #endif
425 
426  /*
427  * If a relation manages to grow to 2^32-1 blocks, refuse to extend it any
428  * more --- we mustn't create a block whose number actually is
429  * InvalidBlockNumber.
430  */
431  if (blocknum == InvalidBlockNumber)
432  ereport(ERROR,
433  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
434  errmsg("cannot extend file \"%s\" beyond %u blocks",
435  relpath(reln->smgr_rnode, forknum),
437 
438  v = _mdfd_getseg(reln, forknum, blocknum, skipFsync, EXTENSION_CREATE);
439 
440  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
441 
442  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
443 
444  if ((nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_EXTEND)) != BLCKSZ)
445  {
446  if (nbytes < 0)
447  ereport(ERROR,
449  errmsg("could not extend file \"%s\": %m",
450  FilePathName(v->mdfd_vfd)),
451  errhint("Check free disk space.")));
452  /* short write: complain appropriately */
453  ereport(ERROR,
454  (errcode(ERRCODE_DISK_FULL),
455  errmsg("could not extend file \"%s\": wrote only %d of %d bytes at block %u",
457  nbytes, BLCKSZ, blocknum),
458  errhint("Check free disk space.")));
459  }
460 
461  if (!skipFsync && !SmgrIsTemp(reln))
462  register_dirty_segment(reln, forknum, v);
463 
464  Assert(_mdnblocks(reln, forknum, v) <= ((BlockNumber) RELSEG_SIZE));
465 }
static MdfdVec * _mdfd_getseg(SMgrRelation reln, ForkNumber forkno, BlockNumber blkno, bool skipFsync, int behavior)
Definition: md.c:1187
int errhint(const char *fmt,...)
Definition: elog.c:1156
BlockNumber mdnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:762
int errcode(int sqlerrcode)
Definition: elog.c:698
uint32 BlockNumber
Definition: block.h:31
char * FilePathName(File file)
Definition: fd.c:2251
#define SmgrIsTemp(smgr)
Definition: smgr.h:77
#define ERROR
Definition: elog.h:46
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
int errcode_for_file_access(void)
Definition: elog.c:721
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2080
Definition: md.c:82
#define ereport(elevel,...)
Definition: elog.h:157
#define EXTENSION_CREATE
Definition: md.c:108
#define Assert(condition)
Definition: c.h:804
static BlockNumber _mdnblocks(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:1312
#define InvalidBlockNumber
Definition: block.h:33
int errmsg(const char *fmt,...)
Definition: elog.c:909
#define relpath(rnode, forknum)
Definition: relpath.h:87
File mdfd_vfd
Definition: md.c:84
static void register_dirty_segment(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:974

◆ mdfiletagmatches()

bool mdfiletagmatches ( const FileTag ftag,
const FileTag candidate 
)

Definition at line 1399 of file md.c.

References RelFileNode::dbNode, and FileTag::rnode.

1400 {
1401  /*
1402  * For now we only use filter requests as a way to drop all scheduled
1403  * callbacks relating to a given database, when dropping the database.
1404  * We'll return true for all candidates that have the same database OID as
1405  * the ftag from the SYNC_FILTER_REQUEST request, so they're forgotten.
1406  */
1407  return ftag->rnode.dbNode == candidate->rnode.dbNode;
1408 }
RelFileNode rnode
Definition: sync.h:54

◆ mdimmedsync()

void mdimmedsync ( SMgrRelation  reln,
ForkNumber  forknum 
)

Definition at line 921 of file md.c.

References _fdvec_resize(), _mdfd_openseg(), data_sync_elevel(), ereport, errcode_for_file_access(), errmsg(), ERROR, FileClose(), FilePathName(), FileSync(), SMgrRelationData::md_num_open_segs, SMgrRelationData::md_seg_fds, _MdfdVec::mdfd_vfd, mdnblocks(), and WAIT_EVENT_DATA_FILE_IMMEDIATE_SYNC.

922 {
923  int segno;
924  int min_inactive_seg;
925 
926  /*
927  * NOTE: mdnblocks makes sure we have opened all active segments, so that
928  * fsync loop will get them all!
929  */
930  mdnblocks(reln, forknum);
931 
932  min_inactive_seg = segno = reln->md_num_open_segs[forknum];
933 
934  /*
935  * Temporarily open inactive segments, then close them after sync. There
936  * may be some inactive segments left opened after fsync() error, but that
937  * is harmless. We don't bother to clean them up and take a risk of
938  * further trouble. The next mdclose() will soon close them.
939  */
940  while (_mdfd_openseg(reln, forknum, segno, 0) != NULL)
941  segno++;
942 
943  while (segno > 0)
944  {
945  MdfdVec *v = &reln->md_seg_fds[forknum][segno - 1];
946 
950  errmsg("could not fsync file \"%s\": %m",
951  FilePathName(v->mdfd_vfd))));
952 
953  /* Close inactive segments immediately */
954  if (segno > min_inactive_seg)
955  {
956  FileClose(v->mdfd_vfd);
957  _fdvec_resize(reln, forknum, segno - 1);
958  }
959 
960  segno--;
961  }
962 }
BlockNumber mdnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:762
char * FilePathName(File file)
Definition: fd.c:2251
#define ERROR
Definition: elog.h:46
int FileSync(File file, uint32 wait_event_info)
Definition: fd.c:2178
int errcode_for_file_access(void)
Definition: elog.c:721
int data_sync_elevel(int elevel)
Definition: fd.c:3714
static void _fdvec_resize(SMgrRelation reln, ForkNumber forknum, int nseg)
Definition: md.c:1081
Definition: md.c:82
static MdfdVec * _mdfd_openseg(SMgrRelation reln, ForkNumber forkno, BlockNumber segno, int oflags)
Definition: md.c:1142
#define ereport(elevel,...)
Definition: elog.h:157
void FileClose(File file)
Definition: fd.c:1873
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
int errmsg(const char *fmt,...)
Definition: elog.c:909
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69
File mdfd_vfd
Definition: md.c:84

◆ mdinit()

void mdinit ( void  )

Definition at line 148 of file md.c.

References ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate, and TopMemoryContext.

149 {
151  "MdSmgr",
153 }
#define AllocSetContextCreate
Definition: memutils.h:173
static MemoryContext MdCxt
Definition: md.c:88
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:195
MemoryContext TopMemoryContext
Definition: mcxt.c:48

◆ mdnblocks()

BlockNumber mdnblocks ( SMgrRelation  reln,
ForkNumber  forknum 
)

Definition at line 762 of file md.c.

References _mdfd_openseg(), _mdnblocks(), Assert, elog, EXTENSION_FAIL, FATAL, SMgrRelationData::md_num_open_segs, SMgrRelationData::md_seg_fds, and mdopenfork().

Referenced by mdextend(), mdimmedsync(), mdtruncate(), and mdwrite().

763 {
764  MdfdVec *v;
765  BlockNumber nblocks;
766  BlockNumber segno;
767 
768  mdopenfork(reln, forknum, EXTENSION_FAIL);
769 
770  /* mdopen has opened the first segment */
771  Assert(reln->md_num_open_segs[forknum] > 0);
772 
773  /*
774  * Start from the last open segments, to avoid redundant seeks. We have
775  * previously verified that these segments are exactly RELSEG_SIZE long,
776  * and it's useless to recheck that each time.
777  *
778  * NOTE: this assumption could only be wrong if another backend has
779  * truncated the relation. We rely on higher code levels to handle that
780  * scenario by closing and re-opening the md fd, which is handled via
781  * relcache flush. (Since the checkpointer doesn't participate in
782  * relcache flush, it could have segment entries for inactive segments;
783  * that's OK because the checkpointer never needs to compute relation
784  * size.)
785  */
786  segno = reln->md_num_open_segs[forknum] - 1;
787  v = &reln->md_seg_fds[forknum][segno];
788 
789  for (;;)
790  {
791  nblocks = _mdnblocks(reln, forknum, v);
792  if (nblocks > ((BlockNumber) RELSEG_SIZE))
793  elog(FATAL, "segment too big");
794  if (nblocks < ((BlockNumber) RELSEG_SIZE))
795  return (segno * ((BlockNumber) RELSEG_SIZE)) + nblocks;
796 
797  /*
798  * If segment is exactly RELSEG_SIZE, advance to next one.
799  */
800  segno++;
801 
802  /*
803  * We used to pass O_CREAT here, but that has the disadvantage that it
804  * might create a segment which has vanished through some operating
805  * system misadventure. In such a case, creating the segment here
806  * undermines _mdfd_getseg's attempts to notice and report an error
807  * upon access to a missing segment.
808  */
809  v = _mdfd_openseg(reln, forknum, segno, 0);
810  if (v == NULL)
811  return segno * ((BlockNumber) RELSEG_SIZE);
812  }
813 }
uint32 BlockNumber
Definition: block.h:31
#define EXTENSION_FAIL
Definition: md.c:104
static MdfdVec * mdopenfork(SMgrRelation reln, ForkNumber forknum, int behavior)
Definition: md.c:478
#define FATAL
Definition: elog.h:49
Definition: md.c:82
static MdfdVec * _mdfd_openseg(SMgrRelation reln, ForkNumber forkno, BlockNumber segno, int oflags)
Definition: md.c:1142
#define Assert(condition)
Definition: c.h:804
static BlockNumber _mdnblocks(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:1312
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
#define elog(elevel,...)
Definition: elog.h:232
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69

◆ mdopen()

void mdopen ( SMgrRelation  reln)

Definition at line 521 of file md.c.

References MAX_FORKNUM, and SMgrRelationData::md_num_open_segs.

522 {
523  /* mark it not open */
524  for (int forknum = 0; forknum <= MAX_FORKNUM; forknum++)
525  reln->md_num_open_segs[forknum] = 0;
526 }
#define MAX_FORKNUM
Definition: relpath.h:55
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68

◆ mdprefetch()

bool mdprefetch ( SMgrRelation  reln,
ForkNumber  forknum,
BlockNumber  blocknum 
)

Definition at line 555 of file md.c.

References _mdfd_getseg(), Assert, EXTENSION_FAIL, EXTENSION_RETURN_NULL, FilePrefetch(), InRecovery, _MdfdVec::mdfd_vfd, and WAIT_EVENT_DATA_FILE_PREFETCH.

556 {
557 #ifdef USE_PREFETCH
558  off_t seekpos;
559  MdfdVec *v;
560 
561  v = _mdfd_getseg(reln, forknum, blocknum, false,
563  if (v == NULL)
564  return false;
565 
566  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
567 
568  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
569 
570  (void) FilePrefetch(v->mdfd_vfd, seekpos, BLCKSZ, WAIT_EVENT_DATA_FILE_PREFETCH);
571 #endif /* USE_PREFETCH */
572 
573  return true;
574 }
static MdfdVec * _mdfd_getseg(SMgrRelation reln, ForkNumber forkno, BlockNumber blkno, bool skipFsync, int behavior)
Definition: md.c:1187
bool InRecovery
Definition: xlog.c:207
uint32 BlockNumber
Definition: block.h:31
#define EXTENSION_FAIL
Definition: md.c:104
#define EXTENSION_RETURN_NULL
Definition: md.c:106
Definition: md.c:82
int FilePrefetch(File file, off_t offset, int amount, uint32 wait_event_info)
Definition: fd.c:1973
#define Assert(condition)
Definition: c.h:804
File mdfd_vfd
Definition: md.c:84

◆ mdread()

void mdread ( SMgrRelation  reln,
ForkNumber  forknum,
BlockNumber  blocknum,
char *  buffer 
)

Definition at line 632 of file md.c.

References _mdfd_getseg(), Assert, RelFileNodeBackend::backend, RelFileNode::dbNode, ereport, errcode(), ERRCODE_DATA_CORRUPTED, errcode_for_file_access(), errmsg(), ERROR, EXTENSION_CREATE_RECOVERY, EXTENSION_FAIL, FilePathName(), FileRead(), InRecovery, _MdfdVec::mdfd_vfd, MemSet, RelFileNodeBackend::node, RelFileNode::relNode, SMgrRelationData::smgr_rnode, RelFileNode::spcNode, WAIT_EVENT_DATA_FILE_READ, and zero_damaged_pages.

634 {
635  off_t seekpos;
636  int nbytes;
637  MdfdVec *v;
638 
639  TRACE_POSTGRESQL_SMGR_MD_READ_START(forknum, blocknum,
640  reln->smgr_rnode.node.spcNode,
641  reln->smgr_rnode.node.dbNode,
642  reln->smgr_rnode.node.relNode,
643  reln->smgr_rnode.backend);
644 
645  v = _mdfd_getseg(reln, forknum, blocknum, false,
647 
648  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
649 
650  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
651 
652  nbytes = FileRead(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_READ);
653 
654  TRACE_POSTGRESQL_SMGR_MD_READ_DONE(forknum, blocknum,
655  reln->smgr_rnode.node.spcNode,
656  reln->smgr_rnode.node.dbNode,
657  reln->smgr_rnode.node.relNode,
658  reln->smgr_rnode.backend,
659  nbytes,
660  BLCKSZ);
661 
662  if (nbytes != BLCKSZ)
663  {
664  if (nbytes < 0)
665  ereport(ERROR,
667  errmsg("could not read block %u in file \"%s\": %m",
668  blocknum, FilePathName(v->mdfd_vfd))));
669 
670  /*
671  * Short read: we are at or past EOF, or we read a partial block at
672  * EOF. Normally this is an error; upper levels should never try to
673  * read a nonexistent block. However, if zero_damaged_pages is ON or
674  * we are InRecovery, we should instead return zeroes without
675  * complaining. This allows, for example, the case of trying to
676  * update a block that was later truncated away.
677  */
679  MemSet(buffer, 0, BLCKSZ);
680  else
681  ereport(ERROR,
683  errmsg("could not read block %u in file \"%s\": read only %d of %d bytes",
684  blocknum, FilePathName(v->mdfd_vfd),
685  nbytes, BLCKSZ)));
686  }
687 }
static MdfdVec * _mdfd_getseg(SMgrRelation reln, ForkNumber forkno, BlockNumber blkno, bool skipFsync, int behavior)
Definition: md.c:1187
bool InRecovery
Definition: xlog.c:207
int errcode(int sqlerrcode)
Definition: elog.c:698
#define MemSet(start, val, len)
Definition: c.h:1008
uint32 BlockNumber
Definition: block.h:31
char * FilePathName(File file)
Definition: fd.c:2251
#define EXTENSION_FAIL
Definition: md.c:104
#define ERROR
Definition: elog.h:46
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
int errcode_for_file_access(void)
Definition: elog.c:721
#define EXTENSION_CREATE_RECOVERY
Definition: md.c:110
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:45
Definition: md.c:82
#define ereport(elevel,...)
Definition: elog.h:157
RelFileNode node
Definition: relfilenode.h:74
#define Assert(condition)
Definition: c.h:804
BackendId backend
Definition: relfilenode.h:75
int errmsg(const char *fmt,...)
Definition: elog.c:909
int FileRead(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2024
File mdfd_vfd
Definition: md.c:84
bool zero_damaged_pages
Definition: bufmgr.c:132

◆ mdsyncfiletag()

int mdsyncfiletag ( const FileTag ftag,
char *  path 
)

Definition at line 1333 of file md.c.

References _mdfd_segpath(), FileClose(), FilePathName(), FileSync(), FileTag::forknum, InvalidBackendId, MAXPGPATH, SMgrRelationData::md_num_open_segs, SMgrRelationData::md_seg_fds, _MdfdVec::mdfd_vfd, PathNameOpenFile(), pfree(), PG_BINARY, FileTag::rnode, FileTag::segno, smgropen(), strlcpy(), and WAIT_EVENT_DATA_FILE_SYNC.

1334 {
1336  File file;
1337  bool need_to_close;
1338  int result,
1339  save_errno;
1340 
1341  /* See if we already have the file open, or need to open it. */
1342  if (ftag->segno < reln->md_num_open_segs[ftag->forknum])
1343  {
1344  file = reln->md_seg_fds[ftag->forknum][ftag->segno].mdfd_vfd;
1345  strlcpy(path, FilePathName(file), MAXPGPATH);
1346  need_to_close = false;
1347  }
1348  else
1349  {
1350  char *p;
1351 
1352  p = _mdfd_segpath(reln, ftag->forknum, ftag->segno);
1353  strlcpy(path, p, MAXPGPATH);
1354  pfree(p);
1355 
1356  file = PathNameOpenFile(path, O_RDWR | PG_BINARY);
1357  if (file < 0)
1358  return -1;
1359  need_to_close = true;
1360  }
1361 
1362  /* Sync the file. */
1363  result = FileSync(file, WAIT_EVENT_DATA_FILE_SYNC);
1364  save_errno = errno;
1365 
1366  if (need_to_close)
1367  FileClose(file);
1368 
1369  errno = save_errno;
1370  return result;
1371 }
uint32 segno
Definition: sync.h:55
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1484
int16 forknum
Definition: sync.h:53
RelFileNode rnode
Definition: sync.h:54
char * FilePathName(File file)
Definition: fd.c:2251
#define PG_BINARY
Definition: c.h:1271
void pfree(void *pointer)
Definition: mcxt.c:1169
#define MAXPGPATH
int FileSync(File file, uint32 wait_event_info)
Definition: fd.c:2178
SMgrRelation smgropen(RelFileNode rnode, BackendId backend)
Definition: smgr.c:146
#define InvalidBackendId
Definition: backendid.h:23
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
void FileClose(File file)
Definition: fd.c:1873
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69
File mdfd_vfd
Definition: md.c:84
int File
Definition: fd.h:54
static char * _mdfd_segpath(SMgrRelation reln, ForkNumber forknum, BlockNumber segno)
Definition: md.c:1119

◆ mdtruncate()

void mdtruncate ( SMgrRelation  reln,
ForkNumber  forknum,
BlockNumber  nblocks 
)

Definition at line 819 of file md.c.

References _fdvec_resize(), Assert, ereport, errcode_for_file_access(), errmsg(), ERROR, FileClose(), FilePathName(), FileTruncate(), InRecovery, SMgrRelationData::md_num_open_segs, SMgrRelationData::md_seg_fds, _MdfdVec::mdfd_vfd, mdnblocks(), register_dirty_segment(), relpath, SMgrRelationData::smgr_rnode, SmgrIsTemp, and WAIT_EVENT_DATA_FILE_TRUNCATE.

820 {
821  BlockNumber curnblk;
822  BlockNumber priorblocks;
823  int curopensegs;
824 
825  /*
826  * NOTE: mdnblocks makes sure we have opened all active segments, so that
827  * truncation loop will get them all!
828  */
829  curnblk = mdnblocks(reln, forknum);
830  if (nblocks > curnblk)
831  {
832  /* Bogus request ... but no complaint if InRecovery */
833  if (InRecovery)
834  return;
835  ereport(ERROR,
836  (errmsg("could not truncate file \"%s\" to %u blocks: it's only %u blocks now",
837  relpath(reln->smgr_rnode, forknum),
838  nblocks, curnblk)));
839  }
840  if (nblocks == curnblk)
841  return; /* no work */
842 
843  /*
844  * Truncate segments, starting at the last one. Starting at the end makes
845  * managing the memory for the fd array easier, should there be errors.
846  */
847  curopensegs = reln->md_num_open_segs[forknum];
848  while (curopensegs > 0)
849  {
850  MdfdVec *v;
851 
852  priorblocks = (curopensegs - 1) * RELSEG_SIZE;
853 
854  v = &reln->md_seg_fds[forknum][curopensegs - 1];
855 
856  if (priorblocks > nblocks)
857  {
858  /*
859  * This segment is no longer active. We truncate the file, but do
860  * not delete it, for reasons explained in the header comments.
861  */
863  ereport(ERROR,
865  errmsg("could not truncate file \"%s\": %m",
866  FilePathName(v->mdfd_vfd))));
867 
868  if (!SmgrIsTemp(reln))
869  register_dirty_segment(reln, forknum, v);
870 
871  /* we never drop the 1st segment */
872  Assert(v != &reln->md_seg_fds[forknum][0]);
873 
874  FileClose(v->mdfd_vfd);
875  _fdvec_resize(reln, forknum, curopensegs - 1);
876  }
877  else if (priorblocks + ((BlockNumber) RELSEG_SIZE) > nblocks)
878  {
879  /*
880  * This is the last segment we want to keep. Truncate the file to
881  * the right length. NOTE: if nblocks is exactly a multiple K of
882  * RELSEG_SIZE, we will truncate the K+1st segment to 0 length but
883  * keep it. This adheres to the invariant given in the header
884  * comments.
885  */
886  BlockNumber lastsegblocks = nblocks - priorblocks;
887 
888  if (FileTruncate(v->mdfd_vfd, (off_t) lastsegblocks * BLCKSZ, WAIT_EVENT_DATA_FILE_TRUNCATE) < 0)
889  ereport(ERROR,
891  errmsg("could not truncate file \"%s\" to %u blocks: %m",
893  nblocks)));
894  if (!SmgrIsTemp(reln))
895  register_dirty_segment(reln, forknum, v);
896  }
897  else
898  {
899  /*
900  * We still need this segment, so nothing to do for this and any
901  * earlier segment.
902  */
903  break;
904  }
905  curopensegs--;
906  }
907 }
BlockNumber mdnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:762
bool InRecovery
Definition: xlog.c:207
uint32 BlockNumber
Definition: block.h:31
char * FilePathName(File file)
Definition: fd.c:2251
#define SmgrIsTemp(smgr)
Definition: smgr.h:77
#define ERROR
Definition: elog.h:46
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
int errcode_for_file_access(void)
Definition: elog.c:721
static void _fdvec_resize(SMgrRelation reln, ForkNumber forknum, int nseg)
Definition: md.c:1081
Definition: md.c:82
#define ereport(elevel,...)
Definition: elog.h:157
void FileClose(File file)
Definition: fd.c:1873
#define Assert(condition)
Definition: c.h:804
int md_num_open_segs[MAX_FORKNUM+1]
Definition: smgr.h:68
int errmsg(const char *fmt,...)
Definition: elog.c:909
#define relpath(rnode, forknum)
Definition: relpath.h:87
struct _MdfdVec * md_seg_fds[MAX_FORKNUM+1]
Definition: smgr.h:69
int FileTruncate(File file, off_t offset, uint32 wait_event_info)
Definition: fd.c:2216
File mdfd_vfd
Definition: md.c:84
static void register_dirty_segment(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:974

◆ mdunlink()

void mdunlink ( RelFileNodeBackend  rnode,
ForkNumber  forknum,
bool  isRedo 
)

Definition at line 277 of file md.c.

References InvalidForkNumber, MAX_FORKNUM, and mdunlinkfork().

278 {
279  /* Now do the per-fork work */
280  if (forkNum == InvalidForkNumber)
281  {
282  for (forkNum = 0; forkNum <= MAX_FORKNUM; forkNum++)
283  mdunlinkfork(rnode, forkNum, isRedo);
284  }
285  else
286  mdunlinkfork(rnode, forkNum, isRedo);
287 }
static void mdunlinkfork(RelFileNodeBackend rnode, ForkNumber forkNum, bool isRedo)
Definition: md.c:314
#define MAX_FORKNUM
Definition: relpath.h:55

◆ mdunlinkfiletag()

int mdunlinkfiletag ( const FileTag ftag,
char *  path 
)

Definition at line 1380 of file md.c.

References MAIN_FORKNUM, MAXPGPATH, pfree(), relpathperm, FileTag::rnode, and strlcpy().

1381 {
1382  char *p;
1383 
1384  /* Compute the path. */
1385  p = relpathperm(ftag->rnode, MAIN_FORKNUM);
1386  strlcpy(path, p, MAXPGPATH);
1387  pfree(p);
1388 
1389  /* Try to unlink the file. */
1390  return unlink(path);
1391 }
#define relpathperm(rnode, forknum)
Definition: relpath.h:83
RelFileNode rnode
Definition: sync.h:54
void pfree(void *pointer)
Definition: mcxt.c:1169
#define MAXPGPATH
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45

◆ mdwrite()

void mdwrite ( SMgrRelation  reln,
ForkNumber  forknum,
BlockNumber  blocknum,
char *  buffer,
bool  skipFsync 
)

Definition at line 697 of file md.c.

References _mdfd_getseg(), Assert, RelFileNodeBackend::backend, RelFileNode::dbNode, ereport, errcode(), errcode_for_file_access(), errhint(), errmsg(), ERROR, EXTENSION_CREATE_RECOVERY, EXTENSION_FAIL, FilePathName(), FileWrite(), _MdfdVec::mdfd_vfd, mdnblocks(), RelFileNodeBackend::node, register_dirty_segment(), RelFileNode::relNode, SMgrRelationData::smgr_rnode, SmgrIsTemp, RelFileNode::spcNode, and WAIT_EVENT_DATA_FILE_WRITE.

699 {
700  off_t seekpos;
701  int nbytes;
702  MdfdVec *v;
703 
704  /* This assert is too expensive to have on normally ... */
705 #ifdef CHECK_WRITE_VS_EXTEND
706  Assert(blocknum < mdnblocks(reln, forknum));
707 #endif
708 
709  TRACE_POSTGRESQL_SMGR_MD_WRITE_START(forknum, blocknum,
710  reln->smgr_rnode.node.spcNode,
711  reln->smgr_rnode.node.dbNode,
712  reln->smgr_rnode.node.relNode,
713  reln->smgr_rnode.backend);
714 
715  v = _mdfd_getseg(reln, forknum, blocknum, skipFsync,
717 
718  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
719 
720  Assert(seekpos < (off_t) BLCKSZ * RELSEG_SIZE);
721 
722  nbytes = FileWrite(v->mdfd_vfd, buffer, BLCKSZ, seekpos, WAIT_EVENT_DATA_FILE_WRITE);
723 
724  TRACE_POSTGRESQL_SMGR_MD_WRITE_DONE(forknum, blocknum,
725  reln->smgr_rnode.node.spcNode,
726  reln->smgr_rnode.node.dbNode,
727  reln->smgr_rnode.node.relNode,
728  reln->smgr_rnode.backend,
729  nbytes,
730  BLCKSZ);
731 
732  if (nbytes != BLCKSZ)
733  {
734  if (nbytes < 0)
735  ereport(ERROR,
737  errmsg("could not write block %u in file \"%s\": %m",
738  blocknum, FilePathName(v->mdfd_vfd))));
739  /* short write: complain appropriately */
740  ereport(ERROR,
741  (errcode(ERRCODE_DISK_FULL),
742  errmsg("could not write block %u in file \"%s\": wrote only %d of %d bytes",
743  blocknum,
745  nbytes, BLCKSZ),
746  errhint("Check free disk space.")));
747  }
748 
749  if (!skipFsync && !SmgrIsTemp(reln))
750  register_dirty_segment(reln, forknum, v);
751 }
static MdfdVec * _mdfd_getseg(SMgrRelation reln, ForkNumber forkno, BlockNumber blkno, bool skipFsync, int behavior)
Definition: md.c:1187
int errhint(const char *fmt,...)
Definition: elog.c:1156
BlockNumber mdnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: md.c:762
int errcode(int sqlerrcode)
Definition: elog.c:698
uint32 BlockNumber
Definition: block.h:31
char * FilePathName(File file)
Definition: fd.c:2251
#define EXTENSION_FAIL
Definition: md.c:104
#define SmgrIsTemp(smgr)
Definition: smgr.h:77
#define ERROR
Definition: elog.h:46
RelFileNodeBackend smgr_rnode
Definition: smgr.h:42
int errcode_for_file_access(void)
Definition: elog.c:721
#define EXTENSION_CREATE_RECOVERY
Definition: md.c:110
int FileWrite(File file, char *buffer, int amount, off_t offset, uint32 wait_event_info)
Definition: fd.c:2080
Definition: md.c:82
#define ereport(elevel,...)
Definition: elog.h:157
RelFileNode node
Definition: relfilenode.h:74
#define Assert(condition)
Definition: c.h:804
BackendId backend
Definition: relfilenode.h:75
int errmsg(const char *fmt,...)
Definition: elog.c:909
File mdfd_vfd
Definition: md.c:84
static void register_dirty_segment(SMgrRelation reln, ForkNumber forknum, MdfdVec *seg)
Definition: md.c:974

◆ mdwriteback()

void mdwriteback ( SMgrRelation  reln,
ForkNumber  forknum,
BlockNumber  blocknum,
BlockNumber  nblocks 
)

Definition at line 583 of file md.c.

References _mdfd_getseg(), Assert, EXTENSION_RETURN_NULL, FileWriteback(), _MdfdVec::mdfd_vfd, and WAIT_EVENT_DATA_FILE_FLUSH.

585 {
586  /*
587  * Issue flush requests in as few requests as possible; have to split at
588  * segment boundaries though, since those are actually separate files.
589  */
590  while (nblocks > 0)
591  {
592  BlockNumber nflush = nblocks;
593  off_t seekpos;
594  MdfdVec *v;
595  int segnum_start,
596  segnum_end;
597 
598  v = _mdfd_getseg(reln, forknum, blocknum, true /* not used */ ,
600 
601  /*
602  * We might be flushing buffers of already removed relations, that's
603  * ok, just ignore that case.
604  */
605  if (!v)
606  return;
607 
608  /* compute offset inside the current segment */
609  segnum_start = blocknum / RELSEG_SIZE;
610 
611  /* compute number of desired writes within the current segment */
612  segnum_end = (blocknum + nblocks - 1) / RELSEG_SIZE;
613  if (segnum_start != segnum_end)
614  nflush = RELSEG_SIZE - (blocknum % ((BlockNumber) RELSEG_SIZE));
615 
616  Assert(nflush >= 1);
617  Assert(nflush <= nblocks);
618 
619  seekpos = (off_t) BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE));
620 
621  FileWriteback(v->mdfd_vfd, seekpos, (off_t) BLCKSZ * nflush, WAIT_EVENT_DATA_FILE_FLUSH);
622 
623  nblocks -= nflush;
624  blocknum += nflush;
625  }
626 }
static MdfdVec * _mdfd_getseg(SMgrRelation reln, ForkNumber forkno, BlockNumber blkno, bool skipFsync, int behavior)
Definition: md.c:1187
uint32 BlockNumber
Definition: block.h:31
#define EXTENSION_RETURN_NULL
Definition: md.c:106
Definition: md.c:82
#define Assert(condition)
Definition: c.h:804
void FileWriteback(File file, off_t offset, off_t nbytes, uint32 wait_event_info)
Definition: fd.c:2001
File mdfd_vfd
Definition: md.c:84