PostgreSQL Source Code  git master
fd.h File Reference
#include "port/pg_iovec.h"
#include <dirent.h>
#include <fcntl.h>
Include dependency graph for fd.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros

#define IO_DIRECT_DATA   0x01
 
#define IO_DIRECT_WAL   0x02
 
#define IO_DIRECT_WAL_INIT   0x04
 
#define FILE_POSSIBLY_DELETED(err)   ((err) == ENOENT)
 
#define PG_O_DIRECT   0
 

Typedefs

typedef int File
 

Functions

File PathNameOpenFile (const char *fileName, int fileFlags)
 
File PathNameOpenFilePerm (const char *fileName, int fileFlags, mode_t fileMode)
 
File OpenTemporaryFile (bool interXact)
 
void FileClose (File file)
 
int FilePrefetch (File file, off_t offset, off_t amount, uint32 wait_event_info)
 
ssize_t FileReadV (File file, const struct iovec *ioc, int iovcnt, off_t offset, uint32 wait_event_info)
 
ssize_t FileWriteV (File file, const struct iovec *ioc, int iovcnt, off_t offset, uint32 wait_event_info)
 
int FileSync (File file, uint32 wait_event_info)
 
int FileZero (File file, off_t offset, off_t amount, uint32 wait_event_info)
 
int FileFallocate (File file, off_t offset, off_t amount, uint32 wait_event_info)
 
off_t FileSize (File file)
 
int FileTruncate (File file, off_t offset, uint32 wait_event_info)
 
void FileWriteback (File file, off_t offset, off_t nbytes, uint32 wait_event_info)
 
char * FilePathName (File file)
 
int FileGetRawDesc (File file)
 
int FileGetRawFlags (File file)
 
mode_t FileGetRawMode (File file)
 
File PathNameCreateTemporaryFile (const char *path, bool error_on_failure)
 
File PathNameOpenTemporaryFile (const char *path, int mode)
 
bool PathNameDeleteTemporaryFile (const char *path, bool error_on_failure)
 
void PathNameCreateTemporaryDir (const char *basedir, const char *directory)
 
void PathNameDeleteTemporaryDir (const char *dirname)
 
void TempTablespacePath (char *path, Oid tablespace)
 
FILE * AllocateFile (const char *name, const char *mode)
 
int FreeFile (FILE *file)
 
FILE * OpenPipeStream (const char *command, const char *mode)
 
int ClosePipeStream (FILE *file)
 
DIRAllocateDir (const char *dirname)
 
struct direntReadDir (DIR *dir, const char *dirname)
 
struct direntReadDirExtended (DIR *dir, const char *dirname, int elevel)
 
int FreeDir (DIR *dir)
 
int OpenTransientFile (const char *fileName, int fileFlags)
 
int OpenTransientFilePerm (const char *fileName, int fileFlags, mode_t fileMode)
 
int CloseTransientFile (int fd)
 
int BasicOpenFile (const char *fileName, int fileFlags)
 
int BasicOpenFilePerm (const char *fileName, int fileFlags, mode_t fileMode)
 
bool AcquireExternalFD (void)
 
void ReserveExternalFD (void)
 
void ReleaseExternalFD (void)
 
int MakePGDirectory (const char *directoryName)
 
void InitFileAccess (void)
 
void InitTemporaryFileAccess (void)
 
void set_max_safe_fds (void)
 
void closeAllVfds (void)
 
void SetTempTablespaces (Oid *tableSpaces, int numSpaces)
 
bool TempTablespacesAreSet (void)
 
int GetTempTablespaces (Oid *tableSpaces, int numSpaces)
 
Oid GetNextTempTableSpace (void)
 
void AtEOXact_Files (bool isCommit)
 
void AtEOSubXact_Files (bool isCommit, SubTransactionId mySubid, SubTransactionId parentSubid)
 
void RemovePgTempFiles (void)
 
void RemovePgTempFilesInDir (const char *tmpdirname, bool missing_ok, bool unlink_all)
 
bool looks_like_temp_rel_name (const char *name)
 
int pg_fsync (int fd)
 
int pg_fsync_no_writethrough (int fd)
 
int pg_fsync_writethrough (int fd)
 
int pg_fdatasync (int fd)
 
bool pg_file_exists (const char *fname)
 
void pg_flush_data (int fd, off_t offset, off_t nbytes)
 
int pg_truncate (const char *path, off_t length)
 
void fsync_fname (const char *fname, bool isdir)
 
int fsync_fname_ext (const char *fname, bool isdir, bool ignore_perm, int elevel)
 
int durable_rename (const char *oldfile, const char *newfile, int elevel)
 
int durable_unlink (const char *fname, int elevel)
 
void SyncDataDirectory (void)
 
int data_sync_elevel (int elevel)
 
static ssize_t FileRead (File file, void *buffer, size_t amount, off_t offset, uint32 wait_event_info)
 
static ssize_t FileWrite (File file, const void *buffer, size_t amount, off_t offset, uint32 wait_event_info)
 

Variables

PGDLLIMPORT int max_files_per_process
 
PGDLLIMPORT bool data_sync_retry
 
PGDLLIMPORT int recovery_init_sync_method
 
PGDLLIMPORT int io_direct_flags
 
PGDLLIMPORT int max_safe_fds
 

Macro Definition Documentation

◆ FILE_POSSIBLY_DELETED

#define FILE_POSSIBLY_DELETED (   err)    ((err) == ENOENT)

Definition at line 78 of file fd.h.

◆ IO_DIRECT_DATA

#define IO_DIRECT_DATA   0x01

Definition at line 54 of file fd.h.

◆ IO_DIRECT_WAL

#define IO_DIRECT_WAL   0x02

Definition at line 55 of file fd.h.

◆ IO_DIRECT_WAL_INIT

#define IO_DIRECT_WAL_INIT   0x04

Definition at line 56 of file fd.h.

◆ PG_O_DIRECT

#define PG_O_DIRECT   0

Definition at line 97 of file fd.h.

Typedef Documentation

◆ File

typedef int File

Definition at line 51 of file fd.h.

Function Documentation

◆ AcquireExternalFD()

bool AcquireExternalFD ( void  )

Definition at line 1186 of file fd.c.

1187 {
1188  /*
1189  * We don't want more than max_safe_fds / 3 FDs to be consumed for
1190  * "external" FDs.
1191  */
1192  if (numExternalFDs < max_safe_fds / 3)
1193  {
1195  return true;
1196  }
1197  errno = EMFILE;
1198  return false;
1199 }
int max_safe_fds
Definition: fd.c:159
void ReserveExternalFD(void)
Definition: fd.c:1221
static int numExternalFDs
Definition: fd.c:274

References max_safe_fds, numExternalFDs, and ReserveExternalFD().

Referenced by CreateWaitEventSet(), and libpqsrv_connect_prepare().

◆ AllocateDir()

DIR* AllocateDir ( const char *  dirname)

Definition at line 2843 of file fd.c.

2844 {
2845  DIR *dir;
2846 
2847  DO_DB(elog(LOG, "AllocateDir: Allocated %d (%s)",
2848  numAllocatedDescs, dirname));
2849 
2850  /* Can we allocate another non-virtual FD? */
2851  if (!reserveAllocatedDesc())
2852  ereport(ERROR,
2853  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2854  errmsg("exceeded maxAllocatedDescs (%d) while trying to open directory \"%s\"",
2855  maxAllocatedDescs, dirname)));
2856 
2857  /* Close excess kernel FDs. */
2858  ReleaseLruFiles();
2859 
2860 TryAgain:
2861  if ((dir = opendir(dirname)) != NULL)
2862  {
2864 
2865  desc->kind = AllocateDescDir;
2866  desc->desc.dir = dir;
2869  return desc->desc.dir;
2870  }
2871 
2872  if (errno == EMFILE || errno == ENFILE)
2873  {
2874  int save_errno = errno;
2875 
2876  ereport(LOG,
2877  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2878  errmsg("out of file descriptors: %m; release and retry")));
2879  errno = 0;
2880  if (ReleaseLruFile())
2881  goto TryAgain;
2882  errno = save_errno;
2883  }
2884 
2885  return NULL;
2886 }
DIR * opendir(const char *)
Definition: dirent.c:33
int errcode(int sqlerrcode)
Definition: elog.c:859
int errmsg(const char *fmt,...)
Definition: elog.c:1072
#define LOG
Definition: elog.h:31
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:224
#define ereport(elevel,...)
Definition: elog.h:149
static bool ReleaseLruFile(void)
Definition: fd.c:1382
static int maxAllocatedDescs
Definition: fd.c:268
static int numAllocatedDescs
Definition: fd.c:267
#define DO_DB(A)
Definition: fd.c:180
static AllocateDesc * allocatedDescs
Definition: fd.c:269
@ AllocateDescDir
Definition: fd.c:251
static bool reserveAllocatedDesc(void)
Definition: fd.c:2508
static void ReleaseLruFiles(void)
Definition: fd.c:1404
SubTransactionId create_subid
Definition: fd.c:258
DIR * dir
Definition: fd.c:262
union AllocateDesc::@22 desc
AllocateDescKind kind
Definition: fd.c:257
Definition: dirent.c:26
SubTransactionId GetCurrentSubTransactionId(void)
Definition: xact.c:788

References allocatedDescs, AllocateDescDir, AllocateDesc::create_subid, AllocateDesc::desc, AllocateDesc::dir, DO_DB, elog, ereport, errcode(), errmsg(), ERROR, GetCurrentSubTransactionId(), AllocateDesc::kind, LOG, maxAllocatedDescs, numAllocatedDescs, opendir(), ReleaseLruFile(), ReleaseLruFiles(), and reserveAllocatedDesc().

Referenced by calculate_database_size(), calculate_tablespace_size(), CheckPointLogicalRewriteHeap(), CheckPointSnapBuild(), CheckTablespaceDirectory(), CleanupBackupHistory(), copydir(), db_dir_size(), DeleteAllExportedSnapshotFiles(), destroy_tablespace_directories(), directory_is_empty(), do_pg_backup_start(), dsm_cleanup_for_mmap(), extension_file_exists(), get_ext_ver_list(), GetConfFilesInDir(), getInstallationPaths(), GetWalSummaries(), movedb(), ParseTzFile(), perform_base_backup(), pg_available_extension_versions(), pg_available_extensions(), pg_ls_dir(), pg_ls_dir_files(), pg_tablespace_databases(), pg_tzenumerate_next(), pg_tzenumerate_start(), pgarch_readyXlog(), RelationCacheInitFileRemove(), RelationCacheInitFileRemoveInDir(), RemoveNonParentXlogFiles(), RemoveOldXlogFiles(), RemovePgTempFiles(), RemovePgTempFilesInDir(), RemovePgTempRelationFiles(), RemovePgTempRelationFilesInDbspace(), RemoveTempXlogFiles(), ReorderBufferCleanupSerializedTXNs(), ResetUnloggedRelations(), ResetUnloggedRelationsInDbspaceDir(), ResetUnloggedRelationsInTablespaceDir(), restoreTwoPhaseData(), scan_directory_ci(), sendDir(), SlruScanDirectory(), StartupReorderBuffer(), StartupReplicationSlots(), SyncDataDirectory(), UpdateLogicalMappings(), walkdir(), and XLogGetOldestSegno().

◆ AllocateFile()

FILE* AllocateFile ( const char *  name,
const char *  mode 
)

Definition at line 2583 of file fd.c.

2584 {
2585  FILE *file;
2586 
2587  DO_DB(elog(LOG, "AllocateFile: Allocated %d (%s)",
2589 
2590  /* Can we allocate another non-virtual FD? */
2591  if (!reserveAllocatedDesc())
2592  ereport(ERROR,
2593  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2594  errmsg("exceeded maxAllocatedDescs (%d) while trying to open file \"%s\"",
2595  maxAllocatedDescs, name)));
2596 
2597  /* Close excess kernel FDs. */
2598  ReleaseLruFiles();
2599 
2600 TryAgain:
2601  if ((file = fopen(name, mode)) != NULL)
2602  {
2604 
2605  desc->kind = AllocateDescFile;
2606  desc->desc.file = file;
2609  return desc->desc.file;
2610  }
2611 
2612  if (errno == EMFILE || errno == ENFILE)
2613  {
2614  int save_errno = errno;
2615 
2616  ereport(LOG,
2617  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2618  errmsg("out of file descriptors: %m; release and retry")));
2619  errno = 0;
2620  if (ReleaseLruFile())
2621  goto TryAgain;
2622  errno = save_errno;
2623  }
2624 
2625  return NULL;
2626 }
@ AllocateDescFile
Definition: fd.c:249
static PgChecksumMode mode
Definition: pg_checksums.c:56
FILE * file
Definition: fd.c:261
const char * name

References allocatedDescs, AllocateDescFile, AllocateDesc::create_subid, AllocateDesc::desc, DO_DB, elog, ereport, errcode(), errmsg(), ERROR, AllocateDesc::file, GetCurrentSubTransactionId(), AllocateDesc::kind, LOG, maxAllocatedDescs, mode, name, numAllocatedDescs, ReleaseLruFile(), ReleaseLruFiles(), and reserveAllocatedDesc().

Referenced by AlterSystemSetConfigFile(), apw_dump_now(), apw_load_buffers(), BeginCopyFrom(), BeginCopyTo(), checkControlFile(), do_pg_backup_stop(), entry_reset(), existsTimeLineHistory(), ExportSnapshot(), gc_qtexts(), GetHugePageSize(), ImportSnapshot(), load_dh_file(), load_relcache_init_file(), open_auth_file(), parse_extension_control_file(), ParseTzFile(), pg_current_logfile(), pg_promote(), pgss_shmem_shutdown(), pgss_shmem_startup(), pgstat_read_statsfile(), pgstat_write_statsfile(), read_backup_label(), read_binary_file(), read_tablespace_map(), read_whole_file(), readTimeLineHistory(), tsearch_readline_begin(), ValidatePgVersion(), write_relcache_init_file(), XLogArchiveForceDone(), and XLogArchiveNotify().

◆ AtEOSubXact_Files()

void AtEOSubXact_Files ( bool  isCommit,
SubTransactionId  mySubid,
SubTransactionId  parentSubid 
)

Definition at line 3132 of file fd.c.

3134 {
3135  Index i;
3136 
3137  for (i = 0; i < numAllocatedDescs; i++)
3138  {
3139  if (allocatedDescs[i].create_subid == mySubid)
3140  {
3141  if (isCommit)
3142  allocatedDescs[i].create_subid = parentSubid;
3143  else
3144  {
3145  /* have to recheck the item after FreeDesc (ugly) */
3146  FreeDesc(&allocatedDescs[i--]);
3147  }
3148  }
3149  }
3150 }
unsigned int Index
Definition: c.h:614
static int FreeDesc(AllocateDesc *desc)
Definition: fd.c:2742
int i
Definition: isn.c:73

References allocatedDescs, AllocateDesc::create_subid, FreeDesc(), i, and numAllocatedDescs.

Referenced by AbortSubTransaction(), and CommitSubTransaction().

◆ AtEOXact_Files()

void AtEOXact_Files ( bool  isCommit)

Definition at line 3165 of file fd.c.

3166 {
3167  CleanupTempFiles(isCommit, false);
3168  tempTableSpaces = NULL;
3169  numTempTableSpaces = -1;
3170 }
static int numTempTableSpaces
Definition: fd.c:289
static Oid * tempTableSpaces
Definition: fd.c:288
static void CleanupTempFiles(bool isCommit, bool isProcExit)
Definition: fd.c:3202

References CleanupTempFiles(), numTempTableSpaces, and tempTableSpaces.

Referenced by AbortTransaction(), BackgroundWriterMain(), CheckpointerMain(), CommitTransaction(), pgarch_archiveXlog(), PrepareTransaction(), WalSummarizerMain(), and WalWriterMain().

◆ BasicOpenFile()

int BasicOpenFile ( const char *  fileName,
int  fileFlags 
)

Definition at line 1087 of file fd.c.

1088 {
1089  return BasicOpenFilePerm(fileName, fileFlags, pg_file_create_mode);
1090 }
int BasicOpenFilePerm(const char *fileName, int fileFlags, mode_t fileMode)
Definition: fd.c:1109
int pg_file_create_mode
Definition: file_perm.c:19

References BasicOpenFilePerm(), and pg_file_create_mode.

Referenced by AlterSystemSetConfigFile(), ReadControlFile(), update_controlfile(), wal_segment_open(), WalSndSegmentOpen(), WriteControlFile(), XLogFileInit(), XLogFileInitInternal(), XLogFileOpen(), and XLogFileRead().

◆ BasicOpenFilePerm()

int BasicOpenFilePerm ( const char *  fileName,
int  fileFlags,
mode_t  fileMode 
)

Definition at line 1109 of file fd.c.

1110 {
1111  int fd;
1112 
1113 tryAgain:
1114 #ifdef PG_O_DIRECT_USE_F_NOCACHE
1115 
1116  /*
1117  * The value we defined to stand in for O_DIRECT when simulating it with
1118  * F_NOCACHE had better not collide with any of the standard flags.
1119  */
1121  (O_APPEND |
1122  O_CLOEXEC |
1123  O_CREAT |
1124  O_DSYNC |
1125  O_EXCL |
1126  O_RDWR |
1127  O_RDONLY |
1128  O_SYNC |
1129  O_TRUNC |
1130  O_WRONLY)) == 0,
1131  "PG_O_DIRECT value collides with standard flag");
1132  fd = open(fileName, fileFlags & ~PG_O_DIRECT, fileMode);
1133 #else
1134  fd = open(fileName, fileFlags, fileMode);
1135 #endif
1136 
1137  if (fd >= 0)
1138  {
1139 #ifdef PG_O_DIRECT_USE_F_NOCACHE
1140  if (fileFlags & PG_O_DIRECT)
1141  {
1142  if (fcntl(fd, F_NOCACHE, 1) < 0)
1143  {
1144  int save_errno = errno;
1145 
1146  close(fd);
1147  errno = save_errno;
1148  return -1;
1149  }
1150  }
1151 #endif
1152 
1153  return fd; /* success! */
1154  }
1155 
1156  if (errno == EMFILE || errno == ENFILE)
1157  {
1158  int save_errno = errno;
1159 
1160  ereport(LOG,
1161  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
1162  errmsg("out of file descriptors: %m; release and retry")));
1163  errno = 0;
1164  if (ReleaseLruFile())
1165  goto tryAgain;
1166  errno = save_errno;
1167  }
1168 
1169  return -1; /* failure */
1170 }
#define StaticAssertStmt(condition, errmessage)
Definition: c.h:938
#define PG_O_DIRECT
Definition: fd.h:97
#define close(a)
Definition: win32.h:12
static int fd(const char *x, int i)
Definition: preproc-init.c:105
#define O_CLOEXEC
Definition: win32_port.h:359
#define O_DSYNC
Definition: win32_port.h:352

References close, ereport, errcode(), errmsg(), fd(), LOG, O_CLOEXEC, O_DSYNC, PG_O_DIRECT, ReleaseLruFile(), and StaticAssertStmt.

Referenced by BasicOpenFile(), LruInsert(), OpenTransientFilePerm(), PathNameOpenFilePerm(), and readRecoverySignalFile().

◆ closeAllVfds()

void closeAllVfds ( void  )

Definition at line 3020 of file fd.c.

3021 {
3022  Index i;
3023 
3024  if (SizeVfdCache > 0)
3025  {
3026  Assert(FileIsNotOpen(0)); /* Make sure ring not corrupted */
3027  for (i = 1; i < SizeVfdCache; i++)
3028  {
3029  if (!FileIsNotOpen(i))
3030  LruDelete(i);
3031  }
3032  }
3033 }
#define Assert(condition)
Definition: c.h:858
static void LruDelete(File file)
Definition: fd.c:1287
static Size SizeVfdCache
Definition: fd.c:217
#define FileIsNotOpen(file)
Definition: fd.c:189

References Assert, FileIsNotOpen, i, LruDelete(), and SizeVfdCache.

Referenced by standard_ProcessUtility().

◆ ClosePipeStream()

int ClosePipeStream ( FILE *  file)

Definition at line 2991 of file fd.c.

2992 {
2993  int i;
2994 
2995  DO_DB(elog(LOG, "ClosePipeStream: Allocated %d", numAllocatedDescs));
2996 
2997  /* Remove file from list of allocated files, if it's present */
2998  for (i = numAllocatedDescs; --i >= 0;)
2999  {
3000  AllocateDesc *desc = &allocatedDescs[i];
3001 
3002  if (desc->kind == AllocateDescPipe && desc->desc.file == file)
3003  return FreeDesc(desc);
3004  }
3005 
3006  /* Only get here if someone passes us a file not in allocatedDescs */
3007  elog(WARNING, "file passed to ClosePipeStream was not obtained from OpenPipeStream");
3008 
3009  return pclose(file);
3010 }
#define WARNING
Definition: elog.h:36
@ AllocateDescPipe
Definition: fd.c:250

References allocatedDescs, AllocateDescPipe, AllocateDesc::desc, DO_DB, elog, AllocateDesc::file, FreeDesc(), i, AllocateDesc::kind, LOG, numAllocatedDescs, and WARNING.

Referenced by ClosePipeFromProgram(), ClosePipeToProgram(), pg_import_system_collations(), run_ssl_passphrase_command(), and shell_finish_command().

◆ CloseTransientFile()

int CloseTransientFile ( int  fd)

Definition at line 2809 of file fd.c.

2810 {
2811  int i;
2812 
2813  DO_DB(elog(LOG, "CloseTransientFile: Allocated %d", numAllocatedDescs));
2814 
2815  /* Remove fd from list of allocated files, if it's present */
2816  for (i = numAllocatedDescs; --i >= 0;)
2817  {
2818  AllocateDesc *desc = &allocatedDescs[i];
2819 
2820  if (desc->kind == AllocateDescRawFD && desc->desc.fd == fd)
2821  return FreeDesc(desc);
2822  }
2823 
2824  /* Only get here if someone passes us a file not in allocatedDescs */
2825  elog(WARNING, "fd passed to CloseTransientFile was not obtained from OpenTransientFile");
2826 
2827  return close(fd);
2828 }
@ AllocateDescRawFD
Definition: fd.c:252
int fd
Definition: fd.c:263

References allocatedDescs, AllocateDescRawFD, close, AllocateDesc::desc, DO_DB, elog, AllocateDesc::fd, fd(), FreeDesc(), i, AllocateDesc::kind, LOG, numAllocatedDescs, and WARNING.

Referenced by ApplyLogicalMappingFile(), be_lo_export(), CheckPointLogicalRewriteHeap(), CheckPointReplicationOrigin(), compare_files(), copy_file(), CreateDirAndVersionFile(), dsm_impl_mmap(), durable_rename(), fsync_fname_ext(), get_controlfile_by_exact_path(), heap_xlog_logical_rewrite(), lo_import_internal(), perform_base_backup(), pg_truncate(), qtext_load_file(), qtext_store(), read_relmap_file(), ReadTwoPhaseFile(), RecreateTwoPhaseFile(), ReorderBufferSerializeChange(), ReorderBufferSerializeTXN(), RestoreSlotFromDisk(), SaveSlotToPath(), sendFile(), SendTimeLineHistory(), SimpleLruDoesPhysicalPageExist(), SimpleLruWriteAll(), SlruInternalWritePage(), SlruPhysicalReadPage(), SlruPhysicalWritePage(), SlruSyncFileTag(), SnapBuildRestore(), SnapBuildRestoreContents(), SnapBuildSerialize(), StartupReplicationOrigin(), write_relmap_file(), writeTimeLineHistory(), writeTimeLineHistoryFile(), and XLogFileCopy().

◆ data_sync_elevel()

◆ durable_rename()

int durable_rename ( const char *  oldfile,
const char *  newfile,
int  elevel 
)

Definition at line 782 of file fd.c.

783 {
784  int fd;
785 
786  /*
787  * First fsync the old and target path (if it exists), to ensure that they
788  * are properly persistent on disk. Syncing the target file is not
789  * strictly necessary, but it makes it easier to reason about crashes;
790  * because it's then guaranteed that either source or target file exists
791  * after a crash.
792  */
793  if (fsync_fname_ext(oldfile, false, false, elevel) != 0)
794  return -1;
795 
796  fd = OpenTransientFile(newfile, PG_BINARY | O_RDWR);
797  if (fd < 0)
798  {
799  if (errno != ENOENT)
800  {
801  ereport(elevel,
803  errmsg("could not open file \"%s\": %m", newfile)));
804  return -1;
805  }
806  }
807  else
808  {
809  if (pg_fsync(fd) != 0)
810  {
811  int save_errno;
812 
813  /* close file upon error, might not be in transaction context */
814  save_errno = errno;
816  errno = save_errno;
817 
818  ereport(elevel,
820  errmsg("could not fsync file \"%s\": %m", newfile)));
821  return -1;
822  }
823 
824  if (CloseTransientFile(fd) != 0)
825  {
826  ereport(elevel,
828  errmsg("could not close file \"%s\": %m", newfile)));
829  return -1;
830  }
831  }
832 
833  /* Time to do the real deal... */
834  if (rename(oldfile, newfile) < 0)
835  {
836  ereport(elevel,
838  errmsg("could not rename file \"%s\" to \"%s\": %m",
839  oldfile, newfile)));
840  return -1;
841  }
842 
843  /*
844  * To guarantee renaming the file is persistent, fsync the file with its
845  * new name, and its containing directory.
846  */
847  if (fsync_fname_ext(newfile, false, false, elevel) != 0)
848  return -1;
849 
850  if (fsync_parent_path(newfile, elevel) != 0)
851  return -1;
852 
853  return 0;
854 }
#define PG_BINARY
Definition: c.h:1273
int errcode_for_file_access(void)
Definition: elog.c:882
int CloseTransientFile(int fd)
Definition: fd.c:2809
int fsync_fname_ext(const char *fname, bool isdir, bool ignore_perm, int elevel)
Definition: fd.c:3797
int pg_fsync(int fd)
Definition: fd.c:386
static int fsync_parent_path(const char *fname, int elevel)
Definition: fd.c:3873
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2633

References CloseTransientFile(), ereport, errcode_for_file_access(), errmsg(), fd(), fsync_fname_ext(), fsync_parent_path(), OpenTransientFile(), PG_BINARY, and pg_fsync().

Referenced by AlterSystemSetConfigFile(), apw_dump_now(), BaseBackup(), basic_archive_file(), bbsink_server_end_manifest(), CheckPointReplicationOrigin(), CleanupAfterArchiveRecovery(), dir_close(), InitWalRecovery(), InstallXLogFileSegment(), KeepFileRestoredFromArchive(), pgss_shmem_shutdown(), StartupXLOG(), SummarizeWAL(), write_relmap_file(), writeTimeLineHistory(), writeTimeLineHistoryFile(), and XLogArchiveForceDone().

◆ durable_unlink()

int durable_unlink ( const char *  fname,
int  elevel 
)

Definition at line 872 of file fd.c.

873 {
874  if (unlink(fname) < 0)
875  {
876  ereport(elevel,
878  errmsg("could not remove file \"%s\": %m",
879  fname)));
880  return -1;
881  }
882 
883  /*
884  * To guarantee that the removal of the file is persistent, fsync its
885  * parent directory.
886  */
887  if (fsync_parent_path(fname, elevel) != 0)
888  return -1;
889 
890  return 0;
891 }

References ereport, errcode_for_file_access(), errmsg(), and fsync_parent_path().

Referenced by InstallXLogFileSegment(), RemoveXlogFile(), and StartupXLOG().

◆ FileClose()

void FileClose ( File  file)

Definition at line 1978 of file fd.c.

1979 {
1980  Vfd *vfdP;
1981 
1982  Assert(FileIsValid(file));
1983 
1984  DO_DB(elog(LOG, "FileClose: %d (%s)",
1985  file, VfdCache[file].fileName));
1986 
1987  vfdP = &VfdCache[file];
1988 
1989  if (!FileIsNotOpen(file))
1990  {
1991  /* close the file */
1992  if (close(vfdP->fd) != 0)
1993  {
1994  /*
1995  * We may need to panic on failure to close non-temporary files;
1996  * see LruDelete.
1997  */
1999  "could not close file \"%s\": %m", vfdP->fileName);
2000  }
2001 
2002  --nfile;
2003  vfdP->fd = VFD_CLOSED;
2004 
2005  /* remove the file from the lru ring */
2006  Delete(file);
2007  }
2008 
2009  if (vfdP->fdstate & FD_TEMP_FILE_LIMIT)
2010  {
2011  /* Subtract its size from current usage (do first in case of error) */
2012  temporary_files_size -= vfdP->fileSize;
2013  vfdP->fileSize = 0;
2014  }
2015 
2016  /*
2017  * Delete the file if it was temporary, and make a log entry if wanted
2018  */
2019  if (vfdP->fdstate & FD_DELETE_AT_CLOSE)
2020  {
2021  struct stat filestats;
2022  int stat_errno;
2023 
2024  /*
2025  * If we get an error, as could happen within the ereport/elog calls,
2026  * we'll come right back here during transaction abort. Reset the
2027  * flag to ensure that we can't get into an infinite loop. This code
2028  * is arranged to ensure that the worst-case consequence is failing to
2029  * emit log message(s), not failing to attempt the unlink.
2030  */
2031  vfdP->fdstate &= ~FD_DELETE_AT_CLOSE;
2032 
2033 
2034  /* first try the stat() */
2035  if (stat(vfdP->fileName, &filestats))
2036  stat_errno = errno;
2037  else
2038  stat_errno = 0;
2039 
2040  /* in any case do the unlink */
2041  if (unlink(vfdP->fileName))
2042  ereport(LOG,
2044  errmsg("could not delete file \"%s\": %m", vfdP->fileName)));
2045 
2046  /* and last report the stat results */
2047  if (stat_errno == 0)
2048  ReportTemporaryFileUsage(vfdP->fileName, filestats.st_size);
2049  else
2050  {
2051  errno = stat_errno;
2052  ereport(LOG,
2054  errmsg("could not stat file \"%s\": %m", vfdP->fileName)));
2055  }
2056  }
2057 
2058  /* Unregister it from the resource owner */
2059  if (vfdP->resowner)
2060  ResourceOwnerForgetFile(vfdP->resowner, file);
2061 
2062  /*
2063  * Return the Vfd slot to the free list
2064  */
2065  FreeVfd(file);
2066 }
#define FD_DELETE_AT_CLOSE
Definition: fd.c:192
static void Delete(File file)
Definition: fd.c:1268
static void ResourceOwnerForgetFile(ResourceOwner owner, File file)
Definition: fd.c:377
#define FileIsValid(file)
Definition: fd.c:186
static int nfile
Definition: fd.c:222
static void FreeVfd(File file)
Definition: fd.c:1472
#define FD_TEMP_FILE_LIMIT
Definition: fd.c:194
int data_sync_elevel(int elevel)
Definition: fd.c:3936
static void ReportTemporaryFileUsage(const char *path, off_t size)
Definition: fd.c:1528
static uint64 temporary_files_size
Definition: fd.c:236
#define VFD_CLOSED
Definition: fd.c:184
static Vfd * VfdCache
Definition: fd.c:216
Definition: fd.c:197
int fd
Definition: fd.c:198
char * fileName
Definition: fd.c:205
ResourceOwner resowner
Definition: fd.c:200
unsigned short fdstate
Definition: fd.c:199
off_t fileSize
Definition: fd.c:204
#define stat
Definition: win32_port.h:284

References Assert, close, data_sync_elevel(), Delete(), DO_DB, elog, ereport, errcode_for_file_access(), errmsg(), vfd::fd, FD_DELETE_AT_CLOSE, FD_TEMP_FILE_LIMIT, vfd::fdstate, FileIsNotOpen, FileIsValid, vfd::fileName, vfd::fileSize, FreeVfd(), LOG, nfile, ReportTemporaryFileUsage(), ResourceOwnerForgetFile(), vfd::resowner, stat::st_size, stat, temporary_files_size, VFD_CLOSED, and VfdCache.

Referenced by bbsink_server_end_archive(), bbsink_server_end_manifest(), BufFileClose(), BufFileTruncateFileSet(), CleanupTempFiles(), logical_end_heap_rewrite(), mdclose(), mdimmedsync(), mdregistersync(), mdsyncfiletag(), mdtruncate(), pg_wal_summary_contents(), PrepareForIncrementalBackup(), ReorderBufferIterTXNFinish(), ReorderBufferRestoreChanges(), ResOwnerReleaseFile(), and SummarizeWAL().

◆ FileFallocate()

int FileFallocate ( File  file,
off_t  offset,
off_t  amount,
uint32  wait_event_info 
)

Definition at line 2369 of file fd.c.

2370 {
2371 #ifdef HAVE_POSIX_FALLOCATE
2372  int returnCode;
2373 
2374  Assert(FileIsValid(file));
2375 
2376  DO_DB(elog(LOG, "FileFallocate: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2377  file, VfdCache[file].fileName,
2378  (int64) offset, (int64) amount));
2379 
2380  returnCode = FileAccess(file);
2381  if (returnCode < 0)
2382  return -1;
2383 
2384 retry:
2385  pgstat_report_wait_start(wait_event_info);
2386  returnCode = posix_fallocate(VfdCache[file].fd, offset, amount);
2388 
2389  if (returnCode == 0)
2390  return 0;
2391  else if (returnCode == EINTR)
2392  goto retry;
2393 
2394  /* for compatibility with %m printing etc */
2395  errno = returnCode;
2396 
2397  /*
2398  * Return in cases of a "real" failure, if fallocate is not supported,
2399  * fall through to the FileZero() backed implementation.
2400  */
2401  if (returnCode != EINVAL && returnCode != EOPNOTSUPP)
2402  return -1;
2403 #endif
2404 
2405  return FileZero(file, offset, amount, wait_event_info);
2406 }
#define INT64_FORMAT
Definition: c.h:548
static int FileAccess(File file)
Definition: fd.c:1492
int FileZero(File file, off_t offset, off_t amount, uint32 wait_event_info)
Definition: fd.c:2324
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: wait_event.h:88
static void pgstat_report_wait_end(void)
Definition: wait_event.h:104
#define EINTR
Definition: win32_port.h:374
#define EOPNOTSUPP
Definition: win32_port.h:398

References Assert, DO_DB, EINTR, elog, EOPNOTSUPP, fd(), FileAccess(), FileIsValid, FileZero(), INT64_FORMAT, LOG, pgstat_report_wait_end(), pgstat_report_wait_start(), and VfdCache.

Referenced by mdzeroextend().

◆ FileGetRawDesc()

int FileGetRawDesc ( File  file)

Definition at line 2477 of file fd.c.

2478 {
2479  Assert(FileIsValid(file));
2480  return VfdCache[file].fd;
2481 }

References Assert, vfd::fd, FileIsValid, and VfdCache.

◆ FileGetRawFlags()

int FileGetRawFlags ( File  file)

Definition at line 2487 of file fd.c.

2488 {
2489  Assert(FileIsValid(file));
2490  return VfdCache[file].fileFlags;
2491 }
int fileFlags
Definition: fd.c:207

References Assert, vfd::fileFlags, FileIsValid, and VfdCache.

◆ FileGetRawMode()

mode_t FileGetRawMode ( File  file)

Definition at line 2497 of file fd.c.

2498 {
2499  Assert(FileIsValid(file));
2500  return VfdCache[file].fileMode;
2501 }
mode_t fileMode
Definition: fd.c:208

References Assert, FileIsValid, vfd::fileMode, and VfdCache.

◆ FilePathName()

◆ FilePrefetch()

int FilePrefetch ( File  file,
off_t  offset,
off_t  amount,
uint32  wait_event_info 
)

Definition at line 2078 of file fd.c.

2079 {
2080 #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_WILLNEED)
2081  int returnCode;
2082 
2083  Assert(FileIsValid(file));
2084 
2085  DO_DB(elog(LOG, "FilePrefetch: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2086  file, VfdCache[file].fileName,
2087  (int64) offset, (int64) amount));
2088 
2089  returnCode = FileAccess(file);
2090  if (returnCode < 0)
2091  return returnCode;
2092 
2093 retry:
2094  pgstat_report_wait_start(wait_event_info);
2095  returnCode = posix_fadvise(VfdCache[file].fd, offset, amount,
2096  POSIX_FADV_WILLNEED);
2098 
2099  if (returnCode == EINTR)
2100  goto retry;
2101 
2102  return returnCode;
2103 #else
2104  Assert(FileIsValid(file));
2105  return 0;
2106 #endif
2107 }

References Assert, DO_DB, EINTR, elog, fd(), FileAccess(), FileIsValid, INT64_FORMAT, LOG, pgstat_report_wait_end(), pgstat_report_wait_start(), and VfdCache.

Referenced by mdprefetch().

◆ FileRead()

static ssize_t FileRead ( File  file,
void *  buffer,
size_t  amount,
off_t  offset,
uint32  wait_event_info 
)
inlinestatic

Definition at line 196 of file fd.h.

198 {
199  struct iovec iov = {
200  .iov_base = buffer,
201  .iov_len = amount
202  };
203 
204  return FileReadV(file, &iov, 1, offset, wait_event_info);
205 }
ssize_t FileReadV(File file, const struct iovec *ioc, int iovcnt, off_t offset, uint32 wait_event_info)
Definition: fd.c:2136

References FileReadV().

Referenced by BufFileLoadBuffer(), ReadWalSummary(), and ReorderBufferRestoreChanges().

◆ FileReadV()

ssize_t FileReadV ( File  file,
const struct iovec *  ioc,
int  iovcnt,
off_t  offset,
uint32  wait_event_info 
)

Definition at line 2136 of file fd.c.

2138 {
2139  ssize_t returnCode;
2140  Vfd *vfdP;
2141 
2142  Assert(FileIsValid(file));
2143 
2144  DO_DB(elog(LOG, "FileReadV: %d (%s) " INT64_FORMAT " %d",
2145  file, VfdCache[file].fileName,
2146  (int64) offset,
2147  iovcnt));
2148 
2149  returnCode = FileAccess(file);
2150  if (returnCode < 0)
2151  return returnCode;
2152 
2153  vfdP = &VfdCache[file];
2154 
2155 retry:
2156  pgstat_report_wait_start(wait_event_info);
2157  returnCode = pg_preadv(vfdP->fd, iov, iovcnt, offset);
2159 
2160  if (returnCode < 0)
2161  {
2162  /*
2163  * Windows may run out of kernel buffers and return "Insufficient
2164  * system resources" error. Wait a bit and retry to solve it.
2165  *
2166  * It is rumored that EINTR is also possible on some Unix filesystems,
2167  * in which case immediate retry is indicated.
2168  */
2169 #ifdef WIN32
2170  DWORD error = GetLastError();
2171 
2172  switch (error)
2173  {
2174  case ERROR_NO_SYSTEM_RESOURCES:
2175  pg_usleep(1000L);
2176  errno = EINTR;
2177  break;
2178  default:
2179  _dosmaperr(error);
2180  break;
2181  }
2182 #endif
2183  /* OK to retry if interrupted */
2184  if (errno == EINTR)
2185  goto retry;
2186  }
2187 
2188  return returnCode;
2189 }
static ssize_t pg_preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset)
Definition: pg_iovec.h:44
void pg_usleep(long microsec)
Definition: signal.c:53
static void error(void)
Definition: sql-dyntest.c:147
void _dosmaperr(unsigned long)
Definition: win32error.c:177

References _dosmaperr(), Assert, DO_DB, EINTR, elog, error(), vfd::fd, FileAccess(), FileIsValid, INT64_FORMAT, LOG, pg_preadv(), pg_usleep(), pgstat_report_wait_end(), pgstat_report_wait_start(), and VfdCache.

Referenced by FileRead(), and mdreadv().

◆ FileSize()

off_t FileSize ( File  file)

Definition at line 2409 of file fd.c.

2410 {
2411  Assert(FileIsValid(file));
2412 
2413  DO_DB(elog(LOG, "FileSize %d (%s)",
2414  file, VfdCache[file].fileName));
2415 
2416  if (FileIsNotOpen(file))
2417  {
2418  if (FileAccess(file) < 0)
2419  return (off_t) -1;
2420  }
2421 
2422  return lseek(VfdCache[file].fd, 0, SEEK_END);
2423 }

References Assert, DO_DB, elog, fd(), FileAccess(), FileIsNotOpen, FileIsValid, LOG, and VfdCache.

Referenced by _mdnblocks(), BufFileSeek(), and BufFileSize().

◆ FileSync()

int FileSync ( File  file,
uint32  wait_event_info 
)

Definition at line 2297 of file fd.c.

2298 {
2299  int returnCode;
2300 
2301  Assert(FileIsValid(file));
2302 
2303  DO_DB(elog(LOG, "FileSync: %d (%s)",
2304  file, VfdCache[file].fileName));
2305 
2306  returnCode = FileAccess(file);
2307  if (returnCode < 0)
2308  return returnCode;
2309 
2310  pgstat_report_wait_start(wait_event_info);
2311  returnCode = pg_fsync(VfdCache[file].fd);
2313 
2314  return returnCode;
2315 }

References Assert, DO_DB, elog, fd(), FileAccess(), FileIsValid, LOG, pg_fsync(), pgstat_report_wait_end(), pgstat_report_wait_start(), and VfdCache.

Referenced by bbsink_server_end_archive(), logical_end_heap_rewrite(), mdimmedsync(), mdsyncfiletag(), and register_dirty_segment().

◆ FileTruncate()

int FileTruncate ( File  file,
off_t  offset,
uint32  wait_event_info 
)

Definition at line 2426 of file fd.c.

2427 {
2428  int returnCode;
2429 
2430  Assert(FileIsValid(file));
2431 
2432  DO_DB(elog(LOG, "FileTruncate %d (%s)",
2433  file, VfdCache[file].fileName));
2434 
2435  returnCode = FileAccess(file);
2436  if (returnCode < 0)
2437  return returnCode;
2438 
2439  pgstat_report_wait_start(wait_event_info);
2440  returnCode = pg_ftruncate(VfdCache[file].fd, offset);
2442 
2443  if (returnCode == 0 && VfdCache[file].fileSize > offset)
2444  {
2445  /* adjust our state for truncation of a temp file */
2446  Assert(VfdCache[file].fdstate & FD_TEMP_FILE_LIMIT);
2447  temporary_files_size -= VfdCache[file].fileSize - offset;
2448  VfdCache[file].fileSize = offset;
2449  }
2450 
2451  return returnCode;
2452 }
static int pg_ftruncate(int fd, off_t length)
Definition: fd.c:703

References Assert, DO_DB, elog, fd(), FD_TEMP_FILE_LIMIT, FileAccess(), FileIsValid, vfd::fileSize, LOG, pg_ftruncate(), pgstat_report_wait_end(), pgstat_report_wait_start(), temporary_files_size, and VfdCache.

Referenced by BufFileTruncateFileSet(), and mdtruncate().

◆ FileWrite()

static ssize_t FileWrite ( File  file,
const void *  buffer,
size_t  amount,
off_t  offset,
uint32  wait_event_info 
)
inlinestatic

Definition at line 208 of file fd.h.

210 {
211  struct iovec iov = {
212  .iov_base = unconstify(void *, buffer),
213  .iov_len = amount
214  };
215 
216  return FileWriteV(file, &iov, 1, offset, wait_event_info);
217 }
#define unconstify(underlying_type, expr)
Definition: c.h:1245
ssize_t FileWriteV(File file, const struct iovec *ioc, int iovcnt, off_t offset, uint32 wait_event_info)
Definition: fd.c:2192

References FileWriteV(), and unconstify.

Referenced by bbsink_server_archive_contents(), bbsink_server_manifest_contents(), BufFileDumpBuffer(), logical_heap_rewrite_flush_mappings(), mdextend(), and WriteWalSummary().

◆ FileWriteback()

void FileWriteback ( File  file,
off_t  offset,
off_t  nbytes,
uint32  wait_event_info 
)

Definition at line 2110 of file fd.c.

2111 {
2112  int returnCode;
2113 
2114  Assert(FileIsValid(file));
2115 
2116  DO_DB(elog(LOG, "FileWriteback: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2117  file, VfdCache[file].fileName,
2118  (int64) offset, (int64) nbytes));
2119 
2120  if (nbytes <= 0)
2121  return;
2122 
2123  if (VfdCache[file].fileFlags & PG_O_DIRECT)
2124  return;
2125 
2126  returnCode = FileAccess(file);
2127  if (returnCode < 0)
2128  return;
2129 
2130  pgstat_report_wait_start(wait_event_info);
2131  pg_flush_data(VfdCache[file].fd, offset, nbytes);
2133 }
void pg_flush_data(int fd, off_t offset, off_t nbytes)
Definition: fd.c:525

References Assert, DO_DB, elog, fd(), FileAccess(), FileIsValid, INT64_FORMAT, LOG, pg_flush_data(), PG_O_DIRECT, pgstat_report_wait_end(), pgstat_report_wait_start(), and VfdCache.

Referenced by mdwriteback().

◆ FileWriteV()

ssize_t FileWriteV ( File  file,
const struct iovec *  ioc,
int  iovcnt,
off_t  offset,
uint32  wait_event_info 
)

Definition at line 2192 of file fd.c.

2194 {
2195  ssize_t returnCode;
2196  Vfd *vfdP;
2197 
2198  Assert(FileIsValid(file));
2199 
2200  DO_DB(elog(LOG, "FileWriteV: %d (%s) " INT64_FORMAT " %d",
2201  file, VfdCache[file].fileName,
2202  (int64) offset,
2203  iovcnt));
2204 
2205  returnCode = FileAccess(file);
2206  if (returnCode < 0)
2207  return returnCode;
2208 
2209  vfdP = &VfdCache[file];
2210 
2211  /*
2212  * If enforcing temp_file_limit and it's a temp file, check to see if the
2213  * write would overrun temp_file_limit, and throw error if so. Note: it's
2214  * really a modularity violation to throw error here; we should set errno
2215  * and return -1. However, there's no way to report a suitable error
2216  * message if we do that. All current callers would just throw error
2217  * immediately anyway, so this is safe at present.
2218  */
2219  if (temp_file_limit >= 0 && (vfdP->fdstate & FD_TEMP_FILE_LIMIT))
2220  {
2221  off_t past_write = offset;
2222 
2223  for (int i = 0; i < iovcnt; ++i)
2224  past_write += iov[i].iov_len;
2225 
2226  if (past_write > vfdP->fileSize)
2227  {
2228  uint64 newTotal = temporary_files_size;
2229 
2230  newTotal += past_write - vfdP->fileSize;
2231  if (newTotal > (uint64) temp_file_limit * (uint64) 1024)
2232  ereport(ERROR,
2233  (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
2234  errmsg("temporary file size exceeds temp_file_limit (%dkB)",
2235  temp_file_limit)));
2236  }
2237  }
2238 
2239 retry:
2240  pgstat_report_wait_start(wait_event_info);
2241  returnCode = pg_pwritev(vfdP->fd, iov, iovcnt, offset);
2243 
2244  if (returnCode >= 0)
2245  {
2246  /*
2247  * Some callers expect short writes to set errno, and traditionally we
2248  * have assumed that they imply disk space shortage. We don't want to
2249  * waste CPU cycles adding up the total size here, so we'll just set
2250  * it for all successful writes in case such a caller determines that
2251  * the write was short and ereports "%m".
2252  */
2253  errno = ENOSPC;
2254 
2255  /*
2256  * Maintain fileSize and temporary_files_size if it's a temp file.
2257  */
2258  if (vfdP->fdstate & FD_TEMP_FILE_LIMIT)
2259  {
2260  off_t past_write = offset + returnCode;
2261 
2262  if (past_write > vfdP->fileSize)
2263  {
2264  temporary_files_size += past_write - vfdP->fileSize;
2265  vfdP->fileSize = past_write;
2266  }
2267  }
2268  }
2269  else
2270  {
2271  /*
2272  * See comments in FileReadV()
2273  */
2274 #ifdef WIN32
2275  DWORD error = GetLastError();
2276 
2277  switch (error)
2278  {
2279  case ERROR_NO_SYSTEM_RESOURCES:
2280  pg_usleep(1000L);
2281  errno = EINTR;
2282  break;
2283  default:
2284  _dosmaperr(error);
2285  break;
2286  }
2287 #endif
2288  /* OK to retry if interrupted */
2289  if (errno == EINTR)
2290  goto retry;
2291  }
2292 
2293  return returnCode;
2294 }
int temp_file_limit
Definition: guc_tables.c:536
static ssize_t pg_pwritev(int fd, const struct iovec *iov, int iovcnt, off_t offset)
Definition: pg_iovec.h:83

References _dosmaperr(), Assert, DO_DB, EINTR, elog, ereport, errcode(), errmsg(), ERROR, error(), vfd::fd, FD_TEMP_FILE_LIMIT, vfd::fdstate, FileAccess(), FileIsValid, vfd::fileSize, i, INT64_FORMAT, LOG, pg_pwritev(), pg_usleep(), pgstat_report_wait_end(), pgstat_report_wait_start(), temp_file_limit, temporary_files_size, and VfdCache.

Referenced by FileWrite(), and mdwritev().

◆ FileZero()

int FileZero ( File  file,
off_t  offset,
off_t  amount,
uint32  wait_event_info 
)

Definition at line 2324 of file fd.c.

2325 {
2326  int returnCode;
2327  ssize_t written;
2328 
2329  Assert(FileIsValid(file));
2330 
2331  DO_DB(elog(LOG, "FileZero: %d (%s) " INT64_FORMAT " " INT64_FORMAT,
2332  file, VfdCache[file].fileName,
2333  (int64) offset, (int64) amount));
2334 
2335  returnCode = FileAccess(file);
2336  if (returnCode < 0)
2337  return returnCode;
2338 
2339  pgstat_report_wait_start(wait_event_info);
2340  written = pg_pwrite_zeros(VfdCache[file].fd, amount, offset);
2342 
2343  if (written < 0)
2344  return -1;
2345  else if (written != amount)
2346  {
2347  /* if errno is unset, assume problem is no disk space */
2348  if (errno == 0)
2349  errno = ENOSPC;
2350  return -1;
2351  }
2352 
2353  return 0;
2354 }
ssize_t pg_pwrite_zeros(int fd, size_t size, off_t offset)
Definition: file_utils.c:687

References Assert, DO_DB, elog, fd(), FileAccess(), FileIsValid, INT64_FORMAT, LOG, pg_pwrite_zeros(), pgstat_report_wait_end(), pgstat_report_wait_start(), and VfdCache.

Referenced by FileFallocate(), and mdzeroextend().

◆ FreeDir()

int FreeDir ( DIR dir)

Definition at line 2961 of file fd.c.

2962 {
2963  int i;
2964 
2965  /* Nothing to do if AllocateDir failed */
2966  if (dir == NULL)
2967  return 0;
2968 
2969  DO_DB(elog(LOG, "FreeDir: Allocated %d", numAllocatedDescs));
2970 
2971  /* Remove dir from list of allocated dirs, if it's present */
2972  for (i = numAllocatedDescs; --i >= 0;)
2973  {
2974  AllocateDesc *desc = &allocatedDescs[i];
2975 
2976  if (desc->kind == AllocateDescDir && desc->desc.dir == dir)
2977  return FreeDesc(desc);
2978  }
2979 
2980  /* Only get here if someone passes us a dir not in allocatedDescs */
2981  elog(WARNING, "dir passed to FreeDir was not obtained from AllocateDir");
2982 
2983  return closedir(dir);
2984 }
int closedir(DIR *)
Definition: dirent.c:127

References allocatedDescs, AllocateDescDir, closedir(), AllocateDesc::desc, AllocateDesc::dir, DO_DB, elog, FreeDesc(), i, AllocateDesc::kind, LOG, numAllocatedDescs, and WARNING.

Referenced by calculate_database_size(), calculate_tablespace_size(), CheckPointLogicalRewriteHeap(), CheckPointSnapBuild(), CleanupBackupHistory(), copydir(), db_dir_size(), DeleteAllExportedSnapshotFiles(), destroy_tablespace_directories(), directory_is_empty(), do_pg_backup_start(), dsm_cleanup_for_mmap(), extension_file_exists(), get_ext_ver_list(), GetConfFilesInDir(), getInstallationPaths(), GetWalSummaries(), movedb(), ParseTzFile(), perform_base_backup(), pg_available_extension_versions(), pg_available_extensions(), pg_ls_dir(), pg_ls_dir_files(), pg_tablespace_databases(), pg_tzenumerate_end(), pg_tzenumerate_next(), pgarch_readyXlog(), RelationCacheInitFileRemove(), RelationCacheInitFileRemoveInDir(), RemoveNonParentXlogFiles(), RemoveOldXlogFiles(), RemovePgTempFiles(), RemovePgTempFilesInDir(), RemovePgTempRelationFiles(), RemovePgTempRelationFilesInDbspace(), RemoveTempXlogFiles(), ReorderBufferCleanupSerializedTXNs(), ResetUnloggedRelations(), ResetUnloggedRelationsInDbspaceDir(), ResetUnloggedRelationsInTablespaceDir(), restoreTwoPhaseData(), scan_directory_ci(), sendDir(), SlruScanDirectory(), StartupReorderBuffer(), StartupReplicationSlots(), SyncDataDirectory(), UpdateLogicalMappings(), walkdir(), and XLogGetOldestSegno().

◆ FreeFile()

int FreeFile ( FILE *  file)

Definition at line 2781 of file fd.c.

2782 {
2783  int i;
2784 
2785  DO_DB(elog(LOG, "FreeFile: Allocated %d", numAllocatedDescs));
2786 
2787  /* Remove file from list of allocated files, if it's present */
2788  for (i = numAllocatedDescs; --i >= 0;)
2789  {
2790  AllocateDesc *desc = &allocatedDescs[i];
2791 
2792  if (desc->kind == AllocateDescFile && desc->desc.file == file)
2793  return FreeDesc(desc);
2794  }
2795 
2796  /* Only get here if someone passes us a file not in allocatedDescs */
2797  elog(WARNING, "file passed to FreeFile was not obtained from AllocateFile");
2798 
2799  return fclose(file);
2800 }

References allocatedDescs, AllocateDescFile, AllocateDesc::desc, DO_DB, elog, AllocateDesc::file, FreeDesc(), i, AllocateDesc::kind, LOG, numAllocatedDescs, and WARNING.

Referenced by AlterSystemSetConfigFile(), apw_dump_now(), apw_load_buffers(), checkControlFile(), do_pg_backup_stop(), EndCopy(), EndCopyFrom(), entry_reset(), existsTimeLineHistory(), ExportSnapshot(), free_auth_file(), gc_qtexts(), GetHugePageSize(), ImportSnapshot(), load_dh_file(), load_relcache_init_file(), parse_extension_control_file(), ParseTzFile(), pg_current_logfile(), pg_promote(), pgss_shmem_shutdown(), pgss_shmem_startup(), pgstat_read_statsfile(), pgstat_write_statsfile(), read_backup_label(), read_binary_file(), read_tablespace_map(), read_whole_file(), readTimeLineHistory(), tsearch_readline_end(), ValidatePgVersion(), write_relcache_init_file(), XLogArchiveForceDone(), and XLogArchiveNotify().

◆ fsync_fname()

◆ fsync_fname_ext()

int fsync_fname_ext ( const char *  fname,
bool  isdir,
bool  ignore_perm,
int  elevel 
)

Definition at line 3797 of file fd.c.

3798 {
3799  int fd;
3800  int flags;
3801  int returncode;
3802 
3803  /*
3804  * Some OSs require directories to be opened read-only whereas other
3805  * systems don't allow us to fsync files opened read-only; so we need both
3806  * cases here. Using O_RDWR will cause us to fail to fsync files that are
3807  * not writable by our userid, but we assume that's OK.
3808  */
3809  flags = PG_BINARY;
3810  if (!isdir)
3811  flags |= O_RDWR;
3812  else
3813  flags |= O_RDONLY;
3814 
3815  fd = OpenTransientFile(fname, flags);
3816 
3817  /*
3818  * Some OSs don't allow us to open directories at all (Windows returns
3819  * EACCES), just ignore the error in that case. If desired also silently
3820  * ignoring errors about unreadable files. Log others.
3821  */
3822  if (fd < 0 && isdir && (errno == EISDIR || errno == EACCES))
3823  return 0;
3824  else if (fd < 0 && ignore_perm && errno == EACCES)
3825  return 0;
3826  else if (fd < 0)
3827  {
3828  ereport(elevel,
3830  errmsg("could not open file \"%s\": %m", fname)));
3831  return -1;
3832  }
3833 
3834  returncode = pg_fsync(fd);
3835 
3836  /*
3837  * Some OSes don't allow us to fsync directories at all, so we can ignore
3838  * those errors. Anything else needs to be logged.
3839  */
3840  if (returncode != 0 && !(isdir && (errno == EBADF || errno == EINVAL)))
3841  {
3842  int save_errno;
3843 
3844  /* close file upon error, might not be in transaction context */
3845  save_errno = errno;
3846  (void) CloseTransientFile(fd);
3847  errno = save_errno;
3848 
3849  ereport(elevel,
3851  errmsg("could not fsync file \"%s\": %m", fname)));
3852  return -1;
3853  }
3854 
3855  if (CloseTransientFile(fd) != 0)
3856  {
3857  ereport(elevel,
3859  errmsg("could not close file \"%s\": %m", fname)));
3860  return -1;
3861  }
3862 
3863  return 0;
3864 }

References CloseTransientFile(), ereport, errcode_for_file_access(), errmsg(), fd(), OpenTransientFile(), PG_BINARY, and pg_fsync().

Referenced by datadir_fsync_fname(), durable_rename(), fsync_fname(), and fsync_parent_path().

◆ GetNextTempTableSpace()

Oid GetNextTempTableSpace ( void  )

Definition at line 3111 of file fd.c.

3112 {
3113  if (numTempTableSpaces > 0)
3114  {
3115  /* Advance nextTempTableSpace counter with wraparound */
3117  nextTempTableSpace = 0;
3119  }
3120  return InvalidOid;
3121 }
static int nextTempTableSpace
Definition: fd.c:290
#define InvalidOid
Definition: postgres_ext.h:36

References InvalidOid, nextTempTableSpace, numTempTableSpaces, and tempTableSpaces.

Referenced by GetDefaultTablespace(), and OpenTemporaryFile().

◆ GetTempTablespaces()

int GetTempTablespaces ( Oid tableSpaces,
int  numSpaces 
)

Definition at line 3093 of file fd.c.

3094 {
3095  int i;
3096 
3098  for (i = 0; i < numTempTableSpaces && i < numSpaces; ++i)
3099  tableSpaces[i] = tempTableSpaces[i];
3100 
3101  return i;
3102 }
bool TempTablespacesAreSet(void)
Definition: fd.c:3078

References Assert, i, numTempTableSpaces, tempTableSpaces, and TempTablespacesAreSet().

Referenced by FileSetInit().

◆ InitFileAccess()

void InitFileAccess ( void  )

Definition at line 903 of file fd.c.

904 {
905  Assert(SizeVfdCache == 0); /* call me only once */
906 
907  /* initialize cache header entry */
908  VfdCache = (Vfd *) malloc(sizeof(Vfd));
909  if (VfdCache == NULL)
910  ereport(FATAL,
911  (errcode(ERRCODE_OUT_OF_MEMORY),
912  errmsg("out of memory")));
913 
914  MemSet((char *) &(VfdCache[0]), 0, sizeof(Vfd));
916 
917  SizeVfdCache = 1;
918 }
#define MemSet(start, val, len)
Definition: c.h:1020
#define FATAL
Definition: elog.h:41
#define malloc(a)
Definition: header.h:50

References Assert, ereport, errcode(), errmsg(), FATAL, vfd::fd, malloc, MemSet, SizeVfdCache, VFD_CLOSED, and VfdCache.

Referenced by BaseInit().

◆ InitTemporaryFileAccess()

void InitTemporaryFileAccess ( void  )

Definition at line 933 of file fd.c.

934 {
935  Assert(SizeVfdCache != 0); /* InitFileAccess() needs to have run */
936  Assert(!temporary_files_allowed); /* call me only once */
937 
938  /*
939  * Register before-shmem-exit hook to ensure temp files are dropped while
940  * we can still report stats.
941  */
943 
944 #ifdef USE_ASSERT_CHECKING
945  temporary_files_allowed = true;
946 #endif
947 }
static void BeforeShmemExit_Files(int code, Datum arg)
Definition: fd.c:3179
void before_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:337

References Assert, before_shmem_exit(), BeforeShmemExit_Files(), and SizeVfdCache.

Referenced by BaseInit().

◆ looks_like_temp_rel_name()

bool looks_like_temp_rel_name ( const char *  name)

Definition at line 3449 of file fd.c.

3450 {
3451  int pos;
3452  int savepos;
3453 
3454  /* Must start with "t". */
3455  if (name[0] != 't')
3456  return false;
3457 
3458  /* Followed by a non-empty string of digits and then an underscore. */
3459  for (pos = 1; isdigit((unsigned char) name[pos]); ++pos)
3460  ;
3461  if (pos == 1 || name[pos] != '_')
3462  return false;
3463 
3464  /* Followed by another nonempty string of digits. */
3465  for (savepos = ++pos; isdigit((unsigned char) name[pos]); ++pos)
3466  ;
3467  if (savepos == pos)
3468  return false;
3469 
3470  /* We might have _forkname or .segment or both. */
3471  if (name[pos] == '_')
3472  {
3473  int forkchar = forkname_chars(&name[pos + 1], NULL);
3474 
3475  if (forkchar <= 0)
3476  return false;
3477  pos += forkchar + 1;
3478  }
3479  if (name[pos] == '.')
3480  {
3481  int segchar;
3482 
3483  for (segchar = 1; isdigit((unsigned char) name[pos + segchar]); ++segchar)
3484  ;
3485  if (segchar <= 1)
3486  return false;
3487  pos += segchar;
3488  }
3489 
3490  /* Now we should be at the end. */
3491  if (name[pos] != '\0')
3492  return false;
3493  return true;
3494 }
int forkname_chars(const char *str, ForkNumber *fork)
Definition: relpath.c:81

References forkname_chars(), and name.

Referenced by RemovePgTempRelationFilesInDbspace(), and sendDir().

◆ MakePGDirectory()

int MakePGDirectory ( const char *  directoryName)

◆ OpenPipeStream()

FILE* OpenPipeStream ( const char *  command,
const char *  mode 
)

Definition at line 2686 of file fd.c.

2687 {
2688  FILE *file;
2689  int save_errno;
2690 
2691  DO_DB(elog(LOG, "OpenPipeStream: Allocated %d (%s)",
2692  numAllocatedDescs, command));
2693 
2694  /* Can we allocate another non-virtual FD? */
2695  if (!reserveAllocatedDesc())
2696  ereport(ERROR,
2697  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2698  errmsg("exceeded maxAllocatedDescs (%d) while trying to execute command \"%s\"",
2699  maxAllocatedDescs, command)));
2700 
2701  /* Close excess kernel FDs. */
2702  ReleaseLruFiles();
2703 
2704 TryAgain:
2705  fflush(NULL);
2707  errno = 0;
2708  file = popen(command, mode);
2709  save_errno = errno;
2711  errno = save_errno;
2712  if (file != NULL)
2713  {
2715 
2716  desc->kind = AllocateDescPipe;
2717  desc->desc.file = file;
2720  return desc->desc.file;
2721  }
2722 
2723  if (errno == EMFILE || errno == ENFILE)
2724  {
2725  ereport(LOG,
2726  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2727  errmsg("out of file descriptors: %m; release and retry")));
2728  if (ReleaseLruFile())
2729  goto TryAgain;
2730  errno = save_errno;
2731  }
2732 
2733  return NULL;
2734 }
static void const char fflush(stdout)
pqsigfunc pqsignal(int signo, pqsigfunc func)
#define SIG_DFL
Definition: win32_port.h:163
#define SIGPIPE
Definition: win32_port.h:173
#define SIG_IGN
Definition: win32_port.h:165

References allocatedDescs, AllocateDescPipe, AllocateDesc::create_subid, AllocateDesc::desc, DO_DB, elog, ereport, errcode(), errmsg(), ERROR, fflush(), AllocateDesc::file, GetCurrentSubTransactionId(), AllocateDesc::kind, LOG, maxAllocatedDescs, mode, numAllocatedDescs, pqsignal(), ReleaseLruFile(), ReleaseLruFiles(), reserveAllocatedDesc(), SIG_DFL, SIG_IGN, and SIGPIPE.

Referenced by BeginCopyFrom(), BeginCopyTo(), pg_import_system_collations(), run_ssl_passphrase_command(), and shell_run_command().

◆ OpenTemporaryFile()

File OpenTemporaryFile ( bool  interXact)

Definition at line 1724 of file fd.c.

1725 {
1726  File file = 0;
1727 
1728  Assert(temporary_files_allowed); /* check temp file access is up */
1729 
1730  /*
1731  * Make sure the current resource owner has space for this File before we
1732  * open it, if we'll be registering it below.
1733  */
1734  if (!interXact)
1736 
1737  /*
1738  * If some temp tablespace(s) have been given to us, try to use the next
1739  * one. If a given tablespace can't be found, we silently fall back to
1740  * the database's default tablespace.
1741  *
1742  * BUT: if the temp file is slated to outlive the current transaction,
1743  * force it into the database's default tablespace, so that it will not
1744  * pose a threat to possible tablespace drop attempts.
1745  */
1746  if (numTempTableSpaces > 0 && !interXact)
1747  {
1748  Oid tblspcOid = GetNextTempTableSpace();
1749 
1750  if (OidIsValid(tblspcOid))
1751  file = OpenTemporaryFileInTablespace(tblspcOid, false);
1752  }
1753 
1754  /*
1755  * If not, or if tablespace is bad, create in database's default
1756  * tablespace. MyDatabaseTableSpace should normally be set before we get
1757  * here, but just in case it isn't, fall back to pg_default tablespace.
1758  */
1759  if (file <= 0)
1762  DEFAULTTABLESPACE_OID,
1763  true);
1764 
1765  /* Mark it for deletion at close and temporary file size limit */
1767 
1768  /* Register it with the current resource owner */
1769  if (!interXact)
1770  RegisterTemporaryFile(file);
1771 
1772  return file;
1773 }
#define OidIsValid(objectId)
Definition: c.h:775
Oid GetNextTempTableSpace(void)
Definition: fd.c:3111
static File OpenTemporaryFileInTablespace(Oid tblspcOid, bool rejectError)
Definition: fd.c:1804
static void RegisterTemporaryFile(File file)
Definition: fd.c:1547
int File
Definition: fd.h:51
Oid MyDatabaseTableSpace
Definition: globals.c:93
unsigned int Oid
Definition: postgres_ext.h:31
ResourceOwner CurrentResourceOwner
Definition: resowner.c:165
void ResourceOwnerEnlarge(ResourceOwner owner)
Definition: resowner.c:442

References Assert, CurrentResourceOwner, FD_DELETE_AT_CLOSE, FD_TEMP_FILE_LIMIT, vfd::fdstate, GetNextTempTableSpace(), MyDatabaseTableSpace, numTempTableSpaces, OidIsValid, OpenTemporaryFileInTablespace(), RegisterTemporaryFile(), ResourceOwnerEnlarge(), and VfdCache.

Referenced by BufFileCreateTemp(), and extendBufFile().

◆ OpenTransientFile()

◆ OpenTransientFilePerm()

int OpenTransientFilePerm ( const char *  fileName,
int  fileFlags,
mode_t  fileMode 
)

Definition at line 2642 of file fd.c.

2643 {
2644  int fd;
2645 
2646  DO_DB(elog(LOG, "OpenTransientFile: Allocated %d (%s)",
2647  numAllocatedDescs, fileName));
2648 
2649  /* Can we allocate another non-virtual FD? */
2650  if (!reserveAllocatedDesc())
2651  ereport(ERROR,
2652  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
2653  errmsg("exceeded maxAllocatedDescs (%d) while trying to open file \"%s\"",
2654  maxAllocatedDescs, fileName)));
2655 
2656  /* Close excess kernel FDs. */
2657  ReleaseLruFiles();
2658 
2659  fd = BasicOpenFilePerm(fileName, fileFlags, fileMode);
2660 
2661  if (fd >= 0)
2662  {
2664 
2665  desc->kind = AllocateDescRawFD;
2666  desc->desc.fd = fd;
2669 
2670  return fd;
2671  }
2672 
2673  return -1; /* failure */
2674 }

References allocatedDescs, AllocateDescRawFD, BasicOpenFilePerm(), AllocateDesc::create_subid, AllocateDesc::desc, DO_DB, elog, ereport, errcode(), errmsg(), ERROR, AllocateDesc::fd, fd(), GetCurrentSubTransactionId(), AllocateDesc::kind, LOG, maxAllocatedDescs, numAllocatedDescs, ReleaseLruFiles(), and reserveAllocatedDesc().

Referenced by be_lo_export(), and OpenTransientFile().

◆ PathNameCreateTemporaryDir()

void PathNameCreateTemporaryDir ( const char *  basedir,
const char *  directory 
)

Definition at line 1660 of file fd.c.

1661 {
1662  if (MakePGDirectory(directory) < 0)
1663  {
1664  if (errno == EEXIST)
1665  return;
1666 
1667  /*
1668  * Failed. Try to create basedir first in case it's missing. Tolerate
1669  * EEXIST to close a race against another process following the same
1670  * algorithm.
1671  */
1672  if (MakePGDirectory(basedir) < 0 && errno != EEXIST)
1673  ereport(ERROR,
1675  errmsg("cannot create temporary directory \"%s\": %m",
1676  basedir)));
1677 
1678  /* Try again. */
1679  if (MakePGDirectory(directory) < 0 && errno != EEXIST)
1680  ereport(ERROR,
1682  errmsg("cannot create temporary subdirectory \"%s\": %m",
1683  directory)));
1684  }
1685 }
int MakePGDirectory(const char *directoryName)
Definition: fd.c:3913
static char * basedir
static const char * directory
Definition: zic.c:634

References basedir, directory, ereport, errcode_for_file_access(), errmsg(), ERROR, and MakePGDirectory().

Referenced by FileSetCreate().

◆ PathNameCreateTemporaryFile()

File PathNameCreateTemporaryFile ( const char *  path,
bool  error_on_failure 
)

Definition at line 1861 of file fd.c.

1862 {
1863  File file;
1864 
1865  Assert(temporary_files_allowed); /* check temp file access is up */
1866 
1868 
1869  /*
1870  * Open the file. Note: we don't use O_EXCL, in case there is an orphaned
1871  * temp file that can be reused.
1872  */
1873  file = PathNameOpenFile(path, O_RDWR | O_CREAT | O_TRUNC | PG_BINARY);
1874  if (file <= 0)
1875  {
1876  if (error_on_failure)
1877  ereport(ERROR,
1879  errmsg("could not create temporary file \"%s\": %m",
1880  path)));
1881  else
1882  return file;
1883  }
1884 
1885  /* Mark it for temp_file_limit accounting. */
1887 
1888  /* Register it for automatic close. */
1889  RegisterTemporaryFile(file);
1890 
1891  return file;
1892 }
File PathNameOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1575

References Assert, CurrentResourceOwner, ereport, errcode_for_file_access(), errmsg(), ERROR, FD_TEMP_FILE_LIMIT, vfd::fdstate, PathNameOpenFile(), PG_BINARY, RegisterTemporaryFile(), ResourceOwnerEnlarge(), and VfdCache.

Referenced by FileSetCreate().

◆ PathNameDeleteTemporaryDir()

void PathNameDeleteTemporaryDir ( const char *  dirname)

Definition at line 1691 of file fd.c.

1692 {
1693  struct stat statbuf;
1694 
1695  /* Silently ignore missing directory. */
1696  if (stat(dirname, &statbuf) != 0 && errno == ENOENT)
1697  return;
1698 
1699  /*
1700  * Currently, walkdir doesn't offer a way for our passed in function to
1701  * maintain state. Perhaps it should, so that we could tell the caller
1702  * whether this operation succeeded or failed. Since this operation is
1703  * used in a cleanup path, we wouldn't actually behave differently: we'll
1704  * just log failures.
1705  */
1706  walkdir(dirname, unlink_if_exists_fname, false, LOG);
1707 }
static void unlink_if_exists_fname(const char *fname, bool isdir, int elevel)
Definition: fd.c:3772
static void walkdir(const char *path, void(*action)(const char *fname, bool isdir, int elevel), bool process_symlinks, int elevel)
Definition: fd.c:3658

References LOG, stat, unlink_if_exists_fname(), and walkdir().

Referenced by FileSetDeleteAll().

◆ PathNameDeleteTemporaryFile()

bool PathNameDeleteTemporaryFile ( const char *  path,
bool  error_on_failure 
)

Definition at line 1932 of file fd.c.

1933 {
1934  struct stat filestats;
1935  int stat_errno;
1936 
1937  /* Get the final size for pgstat reporting. */
1938  if (stat(path, &filestats) != 0)
1939  stat_errno = errno;
1940  else
1941  stat_errno = 0;
1942 
1943  /*
1944  * Unlike FileClose's automatic file deletion code, we tolerate
1945  * non-existence to support BufFileDeleteFileSet which doesn't know how
1946  * many segments it has to delete until it runs out.
1947  */
1948  if (stat_errno == ENOENT)
1949  return false;
1950 
1951  if (unlink(path) < 0)
1952  {
1953  if (errno != ENOENT)
1954  ereport(error_on_failure ? ERROR : LOG,
1956  errmsg("could not unlink temporary file \"%s\": %m",
1957  path)));
1958  return false;
1959  }
1960 
1961  if (stat_errno == 0)
1962  ReportTemporaryFileUsage(path, filestats.st_size);
1963  else
1964  {
1965  errno = stat_errno;
1966  ereport(LOG,
1968  errmsg("could not stat file \"%s\": %m", path)));
1969  }
1970 
1971  return true;
1972 }

References ereport, errcode_for_file_access(), errmsg(), ERROR, LOG, ReportTemporaryFileUsage(), stat::st_size, and stat.

Referenced by FileSetDelete(), and unlink_if_exists_fname().

◆ PathNameOpenFile()

File PathNameOpenFile ( const char *  fileName,
int  fileFlags 
)

◆ PathNameOpenFilePerm()

File PathNameOpenFilePerm ( const char *  fileName,
int  fileFlags,
mode_t  fileMode 
)

Definition at line 1588 of file fd.c.

1589 {
1590  char *fnamecopy;
1591  File file;
1592  Vfd *vfdP;
1593 
1594  DO_DB(elog(LOG, "PathNameOpenFilePerm: %s %x %o",
1595  fileName, fileFlags, fileMode));
1596 
1597  /*
1598  * We need a malloc'd copy of the file name; fail cleanly if no room.
1599  */
1600  fnamecopy = strdup(fileName);
1601  if (fnamecopy == NULL)
1602  ereport(ERROR,
1603  (errcode(ERRCODE_OUT_OF_MEMORY),
1604  errmsg("out of memory")));
1605 
1606  file = AllocateVfd();
1607  vfdP = &VfdCache[file];
1608 
1609  /* Close excess kernel FDs. */
1610  ReleaseLruFiles();
1611 
1612  /*
1613  * Descriptors managed by VFDs are implicitly marked O_CLOEXEC. The
1614  * client shouldn't be expected to know which kernel descriptors are
1615  * currently open, so it wouldn't make sense for them to be inherited by
1616  * executed subprograms.
1617  */
1618  fileFlags |= O_CLOEXEC;
1619 
1620  vfdP->fd = BasicOpenFilePerm(fileName, fileFlags, fileMode);
1621 
1622  if (vfdP->fd < 0)
1623  {
1624  int save_errno = errno;
1625 
1626  FreeVfd(file);
1627  free(fnamecopy);
1628  errno = save_errno;
1629  return -1;
1630  }
1631  ++nfile;
1632  DO_DB(elog(LOG, "PathNameOpenFile: success %d",
1633  vfdP->fd));
1634 
1635  vfdP->fileName = fnamecopy;
1636  /* Saved flags are adjusted to be OK for re-opening file */
1637  vfdP->fileFlags = fileFlags & ~(O_CREAT | O_TRUNC | O_EXCL);
1638  vfdP->fileMode = fileMode;
1639  vfdP->fileSize = 0;
1640  vfdP->fdstate = 0x0;
1641  vfdP->resowner = NULL;
1642 
1643  Insert(file);
1644 
1645  return file;
1646 }
static File AllocateVfd(void)
Definition: fd.c:1414
static void Insert(File file)
Definition: fd.c:1313
#define free(a)
Definition: header.h:65

References AllocateVfd(), BasicOpenFilePerm(), DO_DB, elog, ereport, errcode(), errmsg(), ERROR, vfd::fd, vfd::fdstate, vfd::fileFlags, vfd::fileMode, vfd::fileName, vfd::fileSize, free, FreeVfd(), Insert(), LOG, nfile, O_CLOEXEC, ReleaseLruFiles(), vfd::resowner, and VfdCache.

Referenced by PathNameOpenFile().

◆ PathNameOpenTemporaryFile()

File PathNameOpenTemporaryFile ( const char *  path,
int  mode 
)

Definition at line 1901 of file fd.c.

1902 {
1903  File file;
1904 
1905  Assert(temporary_files_allowed); /* check temp file access is up */
1906 
1908 
1909  file = PathNameOpenFile(path, mode | PG_BINARY);
1910 
1911  /* If no such file, then we don't raise an error. */
1912  if (file <= 0 && errno != ENOENT)
1913  ereport(ERROR,
1915  errmsg("could not open temporary file \"%s\": %m",
1916  path)));
1917 
1918  if (file > 0)
1919  {
1920  /* Register it for automatic close. */
1921  RegisterTemporaryFile(file);
1922  }
1923 
1924  return file;
1925 }

References Assert, CurrentResourceOwner, ereport, errcode_for_file_access(), errmsg(), ERROR, mode, PathNameOpenFile(), PG_BINARY, RegisterTemporaryFile(), and ResourceOwnerEnlarge().

Referenced by FileSetOpen().

◆ pg_fdatasync()

int pg_fdatasync ( int  fd)

Definition at line 480 of file fd.c.

481 {
482  int rc;
483 
484  if (!enableFsync)
485  return 0;
486 
487 retry:
488  rc = fdatasync(fd);
489 
490  if (rc == -1 && errno == EINTR)
491  goto retry;
492 
493  return rc;
494 }
int fdatasync(int fildes)
bool enableFsync
Definition: globals.c:126

References EINTR, enableFsync, fd(), and fdatasync().

Referenced by issue_xlog_fsync().

◆ pg_file_exists()

bool pg_file_exists ( const char *  fname)

Definition at line 503 of file fd.c.

504 {
505  struct stat st;
506 
507  Assert(name != NULL);
508 
509  if (stat(name, &st) == 0)
510  return !S_ISDIR(st.st_mode);
511  else if (!(errno == ENOENT || errno == ENOTDIR || errno == EACCES))
512  ereport(ERROR,
514  errmsg("could not access file \"%s\": %m", name)));
515 
516  return false;
517 }
#define S_ISDIR(m)
Definition: win32_port.h:325

References Assert, ereport, errcode_for_file_access(), errmsg(), ERROR, name, S_ISDIR, stat::st_mode, and stat.

Referenced by expand_dynamic_library_name(), find_in_dynamic_libpath(), InjectionPointRun(), and provider_init().

◆ pg_flush_data()

void pg_flush_data ( int  fd,
off_t  offset,
off_t  nbytes 
)

Definition at line 525 of file fd.c.

526 {
527  /*
528  * Right now file flushing is primarily used to avoid making later
529  * fsync()/fdatasync() calls have less impact. Thus don't trigger flushes
530  * if fsyncs are disabled - that's a decision we might want to make
531  * configurable at some point.
532  */
533  if (!enableFsync)
534  return;
535 
536  /*
537  * We compile all alternatives that are supported on the current platform,
538  * to find portability problems more easily.
539  */
540 #if defined(HAVE_SYNC_FILE_RANGE)
541  {
542  int rc;
543  static bool not_implemented_by_kernel = false;
544 
545  if (not_implemented_by_kernel)
546  return;
547 
548 retry:
549 
550  /*
551  * sync_file_range(SYNC_FILE_RANGE_WRITE), currently linux specific,
552  * tells the OS that writeback for the specified blocks should be
553  * started, but that we don't want to wait for completion. Note that
554  * this call might block if too much dirty data exists in the range.
555  * This is the preferable method on OSs supporting it, as it works
556  * reliably when available (contrast to msync()) and doesn't flush out
557  * clean data (like FADV_DONTNEED).
558  */
559  rc = sync_file_range(fd, offset, nbytes,
560  SYNC_FILE_RANGE_WRITE);
561  if (rc != 0)
562  {
563  int elevel;
564 
565  if (rc == EINTR)
566  goto retry;
567 
568  /*
569  * For systems that don't have an implementation of
570  * sync_file_range() such as Windows WSL, generate only one
571  * warning and then suppress all further attempts by this process.
572  */
573  if (errno == ENOSYS)
574  {
575  elevel = WARNING;
576  not_implemented_by_kernel = true;
577  }
578  else
579  elevel = data_sync_elevel(WARNING);
580 
581  ereport(elevel,
583  errmsg("could not flush dirty data: %m")));
584  }
585 
586  return;
587  }
588 #endif
589 #if !defined(WIN32) && defined(MS_ASYNC)
590  {
591  void *p;
592  static int pagesize = 0;
593 
594  /*
595  * On several OSs msync(MS_ASYNC) on a mmap'ed file triggers
596  * writeback. On linux it only does so if MS_SYNC is specified, but
597  * then it does the writeback synchronously. Luckily all common linux
598  * systems have sync_file_range(). This is preferable over
599  * FADV_DONTNEED because it doesn't flush out clean data.
600  *
601  * We map the file (mmap()), tell the kernel to sync back the contents
602  * (msync()), and then remove the mapping again (munmap()).
603  */
604 
605  /* mmap() needs actual length if we want to map whole file */
606  if (offset == 0 && nbytes == 0)
607  {
608  nbytes = lseek(fd, 0, SEEK_END);
609  if (nbytes < 0)
610  {
613  errmsg("could not determine dirty data size: %m")));
614  return;
615  }
616  }
617 
618  /*
619  * Some platforms reject partial-page mmap() attempts. To deal with
620  * that, just truncate the request to a page boundary. If any extra
621  * bytes don't get flushed, well, it's only a hint anyway.
622  */
623 
624  /* fetch pagesize only once */
625  if (pagesize == 0)
626  pagesize = sysconf(_SC_PAGESIZE);
627 
628  /* align length to pagesize, dropping any fractional page */
629  if (pagesize > 0)
630  nbytes = (nbytes / pagesize) * pagesize;
631 
632  /* fractional-page request is a no-op */
633  if (nbytes <= 0)
634  return;
635 
636  /*
637  * mmap could well fail, particularly on 32-bit platforms where there
638  * may simply not be enough address space. If so, silently fall
639  * through to the next implementation.
640  */
641  if (nbytes <= (off_t) SSIZE_MAX)
642  p = mmap(NULL, nbytes, PROT_READ, MAP_SHARED, fd, offset);
643  else
644  p = MAP_FAILED;
645 
646  if (p != MAP_FAILED)
647  {
648  int rc;
649 
650  rc = msync(p, (size_t) nbytes, MS_ASYNC);
651  if (rc != 0)
652  {
655  errmsg("could not flush dirty data: %m")));
656  /* NB: need to fall through to munmap()! */
657  }
658 
659  rc = munmap(p, (size_t) nbytes);
660  if (rc != 0)
661  {
662  /* FATAL error because mapping would remain */
663  ereport(FATAL,
665  errmsg("could not munmap() while flushing data: %m")));
666  }
667 
668  return;
669  }
670  }
671 #endif
672 #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
673  {
674  int rc;
675 
676  /*
677  * Signal the kernel that the passed in range should not be cached
678  * anymore. This has the, desired, side effect of writing out dirty
679  * data, and the, undesired, side effect of likely discarding useful
680  * clean cached blocks. For the latter reason this is the least
681  * preferable method.
682  */
683 
684  rc = posix_fadvise(fd, offset, nbytes, POSIX_FADV_DONTNEED);
685 
686  if (rc != 0)
687  {
688  /* don't error out, this is just a performance optimization */
691  errmsg("could not flush dirty data: %m")));
692  }
693 
694  return;
695  }
696 #endif
697 }
#define MAP_FAILED
Definition: mem.h:45

References data_sync_elevel(), EINTR, enableFsync, ereport, errcode_for_file_access(), errmsg(), FATAL, fd(), MAP_FAILED, and WARNING.

Referenced by copy_file(), and FileWriteback().

◆ pg_fsync()

int pg_fsync ( int  fd)

Definition at line 386 of file fd.c.

387 {
388 #if !defined(WIN32) && defined(USE_ASSERT_CHECKING)
389  struct stat st;
390 
391  /*
392  * Some operating system implementations of fsync() have requirements
393  * about the file access modes that were used when their file descriptor
394  * argument was opened, and these requirements differ depending on whether
395  * the file descriptor is for a directory.
396  *
397  * For any file descriptor that may eventually be handed to fsync(), we
398  * should have opened it with access modes that are compatible with
399  * fsync() on all supported systems, otherwise the code may not be
400  * portable, even if it runs ok on the current system.
401  *
402  * We assert here that a descriptor for a file was opened with write
403  * permissions (either O_RDWR or O_WRONLY) and for a directory without
404  * write permissions (O_RDONLY).
405  *
406  * Ignore any fstat errors and let the follow-up fsync() do its work.
407  * Doing this sanity check here counts for the case where fsync() is
408  * disabled.
409  */
410  if (fstat(fd, &st) == 0)
411  {
412  int desc_flags = fcntl(fd, F_GETFL);
413 
414  /*
415  * O_RDONLY is historically 0, so just make sure that for directories
416  * no write flags are used.
417  */
418  if (S_ISDIR(st.st_mode))
419  Assert((desc_flags & (O_RDWR | O_WRONLY)) == 0);
420  else
421  Assert((desc_flags & (O_RDWR | O_WRONLY)) != 0);
422  }
423  errno = 0;
424 #endif
425 
426  /* #if is to skip the wal_sync_method test if there's no need for it */
427 #if defined(HAVE_FSYNC_WRITETHROUGH)
429  return pg_fsync_writethrough(fd);
430  else
431 #endif
433 }
int pg_fsync_no_writethrough(int fd)
Definition: fd.c:441
int pg_fsync_writethrough(int fd)
Definition: fd.c:461
#define fstat
Definition: win32_port.h:283
int wal_sync_method
Definition: xlog.c:130
@ WAL_SYNC_METHOD_FSYNC_WRITETHROUGH
Definition: xlog.h:27

References Assert, fd(), fstat, pg_fsync_no_writethrough(), pg_fsync_writethrough(), S_ISDIR, stat::st_mode, wal_sync_method, and WAL_SYNC_METHOD_FSYNC_WRITETHROUGH.

Referenced by AddToDataDirLockFile(), assign_wal_sync_method(), BootStrapXLOG(), CheckPointLogicalRewriteHeap(), CreateDirAndVersionFile(), CreateLockFile(), durable_rename(), FileSync(), fsync_fname_ext(), heap_xlog_logical_rewrite(), readRecoverySignalFile(), RecreateTwoPhaseFile(), RestoreSlotFromDisk(), SaveSlotToPath(), SlruPhysicalWritePage(), SlruSyncFileTag(), SnapBuildSerialize(), update_controlfile(), write_auto_conf_file(), WriteControlFile(), writeTimeLineHistory(), writeTimeLineHistoryFile(), XLogFileCopy(), and XLogFileInitInternal().

◆ pg_fsync_no_writethrough()

int pg_fsync_no_writethrough ( int  fd)

Definition at line 441 of file fd.c.

442 {
443  int rc;
444 
445  if (!enableFsync)
446  return 0;
447 
448 retry:
449  rc = fsync(fd);
450 
451  if (rc == -1 && errno == EINTR)
452  goto retry;
453 
454  return rc;
455 }
#define fsync(fd)
Definition: win32_port.h:85

References EINTR, enableFsync, fd(), and fsync.

Referenced by issue_xlog_fsync(), and pg_fsync().

◆ pg_fsync_writethrough()

int pg_fsync_writethrough ( int  fd)

Definition at line 461 of file fd.c.

462 {
463  if (enableFsync)
464  {
465 #if defined(F_FULLFSYNC)
466  return (fcntl(fd, F_FULLFSYNC, 0) == -1) ? -1 : 0;
467 #else
468  errno = ENOSYS;
469  return -1;
470 #endif
471  }
472  else
473  return 0;
474 }

References enableFsync, and fd().

Referenced by issue_xlog_fsync(), pg_fsync(), and test_sync().

◆ pg_truncate()

int pg_truncate ( const char *  path,
off_t  length 
)

Definition at line 720 of file fd.c.

721 {
722  int ret;
723 #ifdef WIN32
724  int save_errno;
725  int fd;
726 
727  fd = OpenTransientFile(path, O_RDWR | PG_BINARY);
728  if (fd >= 0)
729  {
730  ret = pg_ftruncate(fd, length);
731  save_errno = errno;
733  errno = save_errno;
734  }
735  else
736  ret = -1;
737 #else
738 
739 retry:
740  ret = truncate(path, length);
741 
742  if (ret == -1 && errno == EINTR)
743  goto retry;
744 #endif
745 
746  return ret;
747 }

References CloseTransientFile(), EINTR, fd(), OpenTransientFile(), PG_BINARY, and pg_ftruncate().

Referenced by do_truncate().

◆ ReadDir()

◆ ReadDirExtended()

struct dirent* ReadDirExtended ( DIR dir,
const char *  dirname,
int  elevel 
)

Definition at line 2924 of file fd.c.

2925 {
2926  struct dirent *dent;
2927 
2928  /* Give a generic message for AllocateDir failure, if caller didn't */
2929  if (dir == NULL)
2930  {
2931  ereport(elevel,
2933  errmsg("could not open directory \"%s\": %m",
2934  dirname)));
2935  return NULL;
2936  }
2937 
2938  errno = 0;
2939  if ((dent = readdir(dir)) != NULL)
2940  return dent;
2941 
2942  if (errno)
2943  ereport(elevel,
2945  errmsg("could not read directory \"%s\": %m",
2946  dirname)));
2947  return NULL;
2948 }
struct dirent * readdir(DIR *)
Definition: dirent.c:78
Definition: dirent.h:10

References ereport, errcode_for_file_access(), errmsg(), and readdir().

Referenced by DeleteAllExportedSnapshotFiles(), ReadDir(), RelationCacheInitFileRemove(), RelationCacheInitFileRemoveInDir(), RemovePgTempFiles(), RemovePgTempFilesInDir(), RemovePgTempRelationFiles(), RemovePgTempRelationFilesInDbspace(), ReorderBufferCleanupSerializedTXNs(), scan_directory_ci(), SyncDataDirectory(), and walkdir().

◆ ReleaseExternalFD()

◆ RemovePgTempFiles()

void RemovePgTempFiles ( void  )

Definition at line 3274 of file fd.c.

3275 {
3276  char temp_path[MAXPGPATH + 10 + sizeof(TABLESPACE_VERSION_DIRECTORY) + sizeof(PG_TEMP_FILES_DIR)];
3277  DIR *spc_dir;
3278  struct dirent *spc_de;
3279 
3280  /*
3281  * First process temp files in pg_default ($PGDATA/base)
3282  */
3283  snprintf(temp_path, sizeof(temp_path), "base/%s", PG_TEMP_FILES_DIR);
3284  RemovePgTempFilesInDir(temp_path, true, false);
3285  RemovePgTempRelationFiles("base");
3286 
3287  /*
3288  * Cycle through temp directories for all non-default tablespaces.
3289  */
3290  spc_dir = AllocateDir("pg_tblspc");
3291 
3292  while ((spc_de = ReadDirExtended(spc_dir, "pg_tblspc", LOG)) != NULL)
3293  {
3294  if (strcmp(spc_de->d_name, ".") == 0 ||
3295  strcmp(spc_de->d_name, "..") == 0)
3296  continue;
3297 
3298  snprintf(temp_path, sizeof(temp_path), "pg_tblspc/%s/%s/%s",
3300  RemovePgTempFilesInDir(temp_path, true, false);
3301 
3302  snprintf(temp_path, sizeof(temp_path), "pg_tblspc/%s/%s",
3304  RemovePgTempRelationFiles(temp_path);
3305  }
3306 
3307  FreeDir(spc_dir);
3308 
3309  /*
3310  * In EXEC_BACKEND case there is a pgsql_tmp directory at the top level of
3311  * DataDir as well. However, that is *not* cleaned here because doing so
3312  * would create a race condition. It's done separately, earlier in
3313  * postmaster startup.
3314  */
3315 }
int FreeDir(DIR *dir)
Definition: fd.c:2961
static void RemovePgTempRelationFiles(const char *tsdirname)
Definition: fd.c:3393
void RemovePgTempFilesInDir(const char *tmpdirname, bool missing_ok, bool unlink_all)
Definition: fd.c:3333
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2843
#define PG_TEMP_FILES_DIR
Definition: file_utils.h:62
#define MAXPGPATH
#define snprintf
Definition: port.h:238
#define TABLESPACE_VERSION_DIRECTORY
Definition: relpath.h:33
char d_name[MAX_PATH]
Definition: dirent.h:15

References AllocateDir(), dirent::d_name, FreeDir(), LOG, MAXPGPATH, PG_TEMP_FILES_DIR, ReadDirExtended(), RemovePgTempFilesInDir(), RemovePgTempRelationFiles(), snprintf, and TABLESPACE_VERSION_DIRECTORY.

Referenced by PostmasterMain(), and PostmasterStateMachine().

◆ RemovePgTempFilesInDir()

void RemovePgTempFilesInDir ( const char *  tmpdirname,
bool  missing_ok,
bool  unlink_all 
)

Definition at line 3333 of file fd.c.

3334 {
3335  DIR *temp_dir;
3336  struct dirent *temp_de;
3337  char rm_path[MAXPGPATH * 2];
3338 
3339  temp_dir = AllocateDir(tmpdirname);
3340 
3341  if (temp_dir == NULL && errno == ENOENT && missing_ok)
3342  return;
3343 
3344  while ((temp_de = ReadDirExtended(temp_dir, tmpdirname, LOG)) != NULL)
3345  {
3346  if (strcmp(temp_de->d_name, ".") == 0 ||
3347  strcmp(temp_de->d_name, "..") == 0)
3348  continue;
3349 
3350  snprintf(rm_path, sizeof(rm_path), "%s/%s",
3351  tmpdirname, temp_de->d_name);
3352 
3353  if (unlink_all ||
3354  strncmp(temp_de->d_name,
3356  strlen(PG_TEMP_FILE_PREFIX)) == 0)
3357  {
3358  PGFileType type = get_dirent_type(rm_path, temp_de, false, LOG);
3359 
3360  if (type == PGFILETYPE_ERROR)
3361  continue;
3362  else if (type == PGFILETYPE_DIR)
3363  {
3364  /* recursively remove contents, then directory itself */
3365  RemovePgTempFilesInDir(rm_path, false, true);
3366 
3367  if (rmdir(rm_path) < 0)
3368  ereport(LOG,
3370  errmsg("could not remove directory \"%s\": %m",
3371  rm_path)));
3372  }
3373  else
3374  {
3375  if (unlink(rm_path) < 0)
3376  ereport(LOG,
3378  errmsg("could not remove file \"%s\": %m",
3379  rm_path)));
3380  }
3381  }
3382  else
3383  ereport(LOG,
3384  (errmsg("unexpected file found in temporary-files directory: \"%s\"",
3385  rm_path)));
3386  }
3387 
3388  FreeDir(temp_dir);
3389 }
PGFileType get_dirent_type(const char *path, const struct dirent *de, bool look_through_symlinks, int elevel)
Definition: file_utils.c:525
#define PG_TEMP_FILE_PREFIX
Definition: file_utils.h:63
PGFileType
Definition: file_utils.h:19
@ PGFILETYPE_DIR
Definition: file_utils.h:23
@ PGFILETYPE_ERROR
Definition: file_utils.h:20
const char * type

References AllocateDir(), dirent::d_name, ereport, errcode_for_file_access(), errmsg(), FreeDir(), get_dirent_type(), LOG, MAXPGPATH, PG_TEMP_FILE_PREFIX, PGFILETYPE_DIR, PGFILETYPE_ERROR, ReadDirExtended(), snprintf, and type.

Referenced by PostmasterMain(), and RemovePgTempFiles().

◆ ReserveExternalFD()

void ReserveExternalFD ( void  )

Definition at line 1221 of file fd.c.

1222 {
1223  /*
1224  * Release VFDs if needed to stay safe. Because we do this before
1225  * incrementing numExternalFDs, the final state will be as desired, i.e.,
1226  * nfile + numAllocatedDescs + numExternalFDs <= max_safe_fds.
1227  */
1228  ReleaseLruFiles();
1229 
1230  numExternalFDs++;
1231 }

References numExternalFDs, and ReleaseLruFiles().

Referenced by AcquireExternalFD(), BackendInitialize(), dsm_impl_posix(), InitializeLatchSupport(), InitPostmasterDeathWatchHandle(), and XLogWrite().

◆ set_max_safe_fds()

void set_max_safe_fds ( void  )

Definition at line 1044 of file fd.c.

1045 {
1046  int usable_fds;
1047  int already_open;
1048 
1049  /*----------
1050  * We want to set max_safe_fds to
1051  * MIN(usable_fds, max_files_per_process - already_open)
1052  * less the slop factor for files that are opened without consulting
1053  * fd.c. This ensures that we won't exceed either max_files_per_process
1054  * or the experimentally-determined EMFILE limit.
1055  *----------
1056  */
1058  &usable_fds, &already_open);
1059 
1060  max_safe_fds = Min(usable_fds, max_files_per_process - already_open);
1061 
1062  /*
1063  * Take off the FDs reserved for system() etc.
1064  */
1066 
1067  /*
1068  * Make sure we still have enough to get by.
1069  */
1070  if (max_safe_fds < FD_MINFREE)
1071  ereport(FATAL,
1072  (errcode(ERRCODE_INSUFFICIENT_RESOURCES),
1073  errmsg("insufficient file descriptors available to start server process"),
1074  errdetail("System allows %d, server needs at least %d.",
1077 
1078  elog(DEBUG2, "max_safe_fds = %d, usable_fds = %d, already_open = %d",
1079  max_safe_fds, usable_fds, already_open);
1080 }
#define Min(x, y)
Definition: c.h:1004
int errdetail(const char *fmt,...)
Definition: elog.c:1205
#define DEBUG2
Definition: elog.h:29
int max_files_per_process
Definition: fd.c:146
#define FD_MINFREE
Definition: fd.c:138
static void count_usable_fds(int max_to_probe, int *usable_fds, int *already_open)
Definition: fd.c:964
#define NUM_RESERVED_FDS
Definition: fd.c:129

References count_usable_fds(), DEBUG2, elog, ereport, errcode(), errdetail(), errmsg(), FATAL, FD_MINFREE, max_files_per_process, max_safe_fds, Min, and NUM_RESERVED_FDS.

Referenced by PostmasterMain().

◆ SetTempTablespaces()

void SetTempTablespaces ( Oid tableSpaces,
int  numSpaces 
)

Definition at line 3049 of file fd.c.

3050 {
3051  Assert(numSpaces >= 0);
3052  tempTableSpaces = tableSpaces;
3053  numTempTableSpaces = numSpaces;
3054 
3055  /*
3056  * Select a random starting point in the list. This is to minimize
3057  * conflicts between backends that are most likely sharing the same list
3058  * of temp tablespaces. Note that if we create multiple temp files in the
3059  * same transaction, we'll advance circularly through the list --- this
3060  * ensures that large temporary sort files are nicely spread across all
3061  * available tablespaces.
3062  */
3063  if (numSpaces > 1)
3065  0, numSpaces - 1);
3066  else
3067  nextTempTableSpace = 0;
3068 }
uint64 pg_prng_uint64_range(pg_prng_state *state, uint64 rmin, uint64 rmax)
Definition: pg_prng.c:144
pg_prng_state pg_global_prng_state
Definition: pg_prng.c:34

References Assert, nextTempTableSpace, numTempTableSpaces, pg_global_prng_state, pg_prng_uint64_range(), and tempTableSpaces.

Referenced by assign_temp_tablespaces(), and PrepareTempTablespaces().

◆ SyncDataDirectory()

void SyncDataDirectory ( void  )

Definition at line 3544 of file fd.c.

3545 {
3546  bool xlog_is_symlink;
3547 
3548  /* We can skip this whole thing if fsync is disabled. */
3549  if (!enableFsync)
3550  return;
3551 
3552  /*
3553  * If pg_wal is a symlink, we'll need to recurse into it separately,
3554  * because the first walkdir below will ignore it.
3555  */
3556  xlog_is_symlink = false;
3557 
3558  {
3559  struct stat st;
3560 
3561  if (lstat("pg_wal", &st) < 0)
3562  ereport(LOG,
3564  errmsg("could not stat file \"%s\": %m",
3565  "pg_wal")));
3566  else if (S_ISLNK(st.st_mode))
3567  xlog_is_symlink = true;
3568  }
3569 
3570 #ifdef HAVE_SYNCFS
3572  {
3573  DIR *dir;
3574  struct dirent *de;
3575 
3576  /*
3577  * On Linux, we don't have to open every single file one by one. We
3578  * can use syncfs() to sync whole filesystems. We only expect
3579  * filesystem boundaries to exist where we tolerate symlinks, namely
3580  * pg_wal and the tablespaces, so we call syncfs() for each of those
3581  * directories.
3582  */
3583 
3584  /* Prepare to report progress syncing the data directory via syncfs. */
3586 
3587  /* Sync the top level pgdata directory. */
3588  do_syncfs(".");
3589  /* If any tablespaces are configured, sync each of those. */
3590  dir = AllocateDir("pg_tblspc");
3591  while ((de = ReadDirExtended(dir, "pg_tblspc", LOG)))
3592  {
3593  char path[MAXPGPATH];
3594 
3595  if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
3596  continue;
3597 
3598  snprintf(path, MAXPGPATH, "pg_tblspc/%s", de->d_name);
3599  do_syncfs(path);
3600  }
3601  FreeDir(dir);
3602  /* If pg_wal is a symlink, process that too. */
3603  if (xlog_is_symlink)
3604  do_syncfs("pg_wal");
3605  return;
3606  }
3607 #endif /* !HAVE_SYNCFS */
3608 
3609 #ifdef PG_FLUSH_DATA_WORKS
3610  /* Prepare to report progress of the pre-fsync phase. */
3612 
3613  /*
3614  * If possible, hint to the kernel that we're soon going to fsync the data
3615  * directory and its contents. Errors in this step are even less
3616  * interesting than normal, so log them only at DEBUG1.
3617  */
3618  walkdir(".", pre_sync_fname, false, DEBUG1);
3619  if (xlog_is_symlink)
3620  walkdir("pg_wal", pre_sync_fname, false, DEBUG1);
3621  walkdir("pg_tblspc", pre_sync_fname, true, DEBUG1);
3622 #endif
3623 
3624  /* Prepare to report progress syncing the data directory via fsync. */
3626 
3627  /*
3628  * Now we do the fsync()s in the same order.
3629  *
3630  * The main call ignores symlinks, so in addition to specially processing
3631  * pg_wal if it's a symlink, pg_tblspc has to be visited separately with
3632  * process_symlinks = true. Note that if there are any plain directories
3633  * in pg_tblspc, they'll get fsync'd twice. That's not an expected case
3634  * so we don't worry about optimizing it.
3635  */
3636  walkdir(".", datadir_fsync_fname, false, LOG);
3637  if (xlog_is_symlink)
3638  walkdir("pg_wal", datadir_fsync_fname, false, LOG);
3639  walkdir("pg_tblspc", datadir_fsync_fname, true, LOG);
3640 }
void begin_startup_progress_phase(void)
Definition: startup.c:343
#define DEBUG1
Definition: elog.h:30
int recovery_init_sync_method
Definition: fd.c:165
static void datadir_fsync_fname(const char *fname, bool isdir, int elevel)
Definition: fd.c:3759
@ DATA_DIR_SYNC_METHOD_SYNCFS
Definition: file_utils.h:30
#define lstat(path, sb)
Definition: win32_port.h:285
#define S_ISLNK(m)
Definition: win32_port.h:344

References AllocateDir(), begin_startup_progress_phase(), dirent::d_name, DATA_DIR_SYNC_METHOD_SYNCFS, datadir_fsync_fname(), DEBUG1, enableFsync, ereport, errcode_for_file_access(), errmsg(), FreeDir(), LOG, lstat, MAXPGPATH, ReadDirExtended(), recovery_init_sync_method, S_ISLNK, snprintf, stat::st_mode, and walkdir().

Referenced by StartupXLOG().

◆ TempTablespacePath()

void TempTablespacePath ( char *  path,
Oid  tablespace 
)

Definition at line 1779 of file fd.c.

1780 {
1781  /*
1782  * Identify the tempfile directory for this tablespace.
1783  *
1784  * If someone tries to specify pg_global, use pg_default instead.
1785  */
1786  if (tablespace == InvalidOid ||
1787  tablespace == DEFAULTTABLESPACE_OID ||
1788  tablespace == GLOBALTABLESPACE_OID)
1789  snprintf(path, MAXPGPATH, "base/%s", PG_TEMP_FILES_DIR);
1790  else
1791  {
1792  /* All other tablespaces are accessed via symlinks */
1793  snprintf(path, MAXPGPATH, "pg_tblspc/%u/%s/%s",
1796  }
1797 }
char * tablespace
Definition: pgbench.c:216

References InvalidOid, MAXPGPATH, PG_TEMP_FILES_DIR, snprintf, tablespace, and TABLESPACE_VERSION_DIRECTORY.

Referenced by FileSetCreate(), FileSetPath(), OpenTemporaryFileInTablespace(), and pg_ls_tmpdir().

◆ TempTablespacesAreSet()

bool TempTablespacesAreSet ( void  )

Definition at line 3078 of file fd.c.

3079 {
3080  return (numTempTableSpaces >= 0);
3081 }

References numTempTableSpaces.

Referenced by GetTempTablespaces(), and PrepareTempTablespaces().

Variable Documentation

◆ data_sync_retry

PGDLLIMPORT bool data_sync_retry
extern

Definition at line 162 of file fd.c.

Referenced by data_sync_elevel().

◆ io_direct_flags

◆ max_files_per_process

PGDLLIMPORT int max_files_per_process
extern

Definition at line 146 of file fd.c.

Referenced by set_max_safe_fds().

◆ max_safe_fds

PGDLLIMPORT int max_safe_fds
extern

Definition at line 159 of file fd.c.

Referenced by AcquireExternalFD(), ReleaseLruFiles(), reserveAllocatedDesc(), and set_max_safe_fds().

◆ recovery_init_sync_method

PGDLLIMPORT int recovery_init_sync_method
extern

Definition at line 165 of file fd.c.

Referenced by SyncDataDirectory().