/ Check-in [e2f08426]
Login
SQLite training in Houston TX on 2019-11-05 (details)
Part of the 2019 Tcl Conference

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Use mkdir() and rmdir() rather than open() and unlink() to create the lock files with the unix-dotlock VFS. The change is backwards compatible and, we are told, works better on some network filesystems.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1: e2f08426d7a84a2ac6148f485b7af377201a175b
User & Date: drh 2011-11-04 02:24:02
Context
2011-11-04
10:37
Add a missing va_end() macro to fts3.c. check-in: e6f82574 user: dan tags: trunk
02:24
Use mkdir() and rmdir() rather than open() and unlink() to create the lock files with the unix-dotlock VFS. The change is backwards compatible and, we are told, works better on some network filesystems. check-in: e2f08426 user: drh tags: trunk
00:35
Enhance the shell so that the ".schema" command works with case insensitive LIKE patterns even with PRAGMA cache_sensitive_like=ON. check-in: b06bf3b3 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/os_unix.c.

403
404
405
406
407
408
409






410
411
412
413
414
415
416
....
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
....
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
....
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
....
1981
1982
1983
1984
1985
1986
1987

1988
1989
1990
1991
1992
1993
1994
....
2002
2003
2004
2005
2006
2007
2008

2009

2010
2011
2012
2013
2014
2015
2016
....
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
....
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761

  { "unlink",       (sqlite3_syscall_ptr)unlink,           0 },
#define osUnlink    ((int(*)(const char*))aSyscall[16].pCurrent)

  { "openDirectory",    (sqlite3_syscall_ptr)openDirectory,      0 },
#define osOpenDirectory ((int(*)(const char*,int*))aSyscall[17].pCurrent)







}; /* End of the overrideable system calls */

/*
** This is the xSetSystemCall() method of sqlite3_vfs for all of the
** "unix" VFSes.  Return SQLITE_OK opon successfully updating the
** system call pointer, or SQLITE_NOTFOUND if there is no configurable
** system call named zName.
................................................................................
/******************* End of the no-op lock implementation *********************
******************************************************************************/

/******************************************************************************
************************* Begin dot-file Locking ******************************
**
** The dotfile locking implementation uses the existance of separate lock
** files in order to control access to the database.  This works on just
** about every filesystem imaginable.  But there are serious downsides:
**
**    (1)  There is zero concurrency.  A single reader blocks all other
**         connections from reading or writing the database.
**
**    (2)  An application crash or power loss can leave stale lock files
**         sitting around that need to be cleared manually.
**
** Nevertheless, a dotlock is an appropriate locking mode for use if no
** other locking strategy is available.
**
** Dotfile locking works by creating a file in the same directory as the
** database and with the same name but with a ".lock" extension added.
** The existance of a lock file implies an EXCLUSIVE lock.  All other lock
** types (SHARED, RESERVED, PENDING) are mapped into EXCLUSIVE.
*/

/*
** The file suffix added to the data base filename in order to create the
** lock file.
*/
#define DOTLOCK_SUFFIX ".lock"

/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, set *pResOut
** to a non-zero value otherwise *pResOut is set to zero.  The return value
................................................................................
** routine to lower a locking level.
**
** With dotfile locking, we really only support state (4): EXCLUSIVE.
** But we track the other locking levels internally.
*/
static int dotlockLock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  int fd;
  char *zLockFile = (char *)pFile->lockingContext;
  int rc = SQLITE_OK;


  /* If we have any lock, then the lock file already exists.  All we have
  ** to do is adjust our internal record of the lock level.
  */
................................................................................
#else
    utimes(zLockFile, NULL);
#endif
    return SQLITE_OK;
  }
  
  /* grab an exclusive lock */
  fd = robust_open(zLockFile,O_RDONLY|O_CREAT|O_EXCL,0600);
  if( fd<0 ){
    /* failed to open/create the file, someone else may have stolen the lock */
    int tErrno = errno;
    if( EEXIST == tErrno ){
      rc = SQLITE_BUSY;
    } else {
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
    }
    return rc;
  } 
  robust_close(pFile, fd, __LINE__);
  
  /* got it, set the type and return ok */
  pFile->eFileLock = eFileLock;
  return rc;
}

/*
................................................................................
** the requested locking level, this routine is a no-op.
**
** When the locking level reaches NO_LOCK, delete the lock file.
*/
static int dotlockUnlock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  char *zLockFile = (char *)pFile->lockingContext;


  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d pid=%d (dotlock)\n", pFile->h, eFileLock,
	   pFile->eFileLock, getpid()));
  assert( eFileLock<=SHARED_LOCK );
  
  /* no-op if possible */
................................................................................
  if( eFileLock==SHARED_LOCK ){
    pFile->eFileLock = SHARED_LOCK;
    return SQLITE_OK;
  }
  
  /* To fully unlock the database, delete the lock file */
  assert( eFileLock==NO_LOCK );

  if( osUnlink(zLockFile) ){

    int rc = 0;
    int tErrno = errno;
    if( ENOENT != tErrno ){
      rc = SQLITE_IOERR_UNLOCK;
    }
    if( IS_LOCK_ERROR(rc) ){
      pFile->lastErrno = tErrno;
................................................................................
  buf[0] = lockPath[0];
  for( i=1; i<len; i++ ){
    if( lockPath[i] == '/' && (i - start > 0) ){
      /* only mkdir if leaf dir != "." or "/" or ".." */
      if( i-start>2 || (i-start==1 && buf[start] != '.' && buf[start] != '/') 
         || (i-start==2 && buf[start] != '.' && buf[start+1] != '.') ){
        buf[i]='\0';
        if( mkdir(buf, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){
          int err=errno;
          if( err!=EEXIST ) {
            OSTRACE(("CREATELOCKPATH  FAILED creating %s, "
                     "'%s' proxy lock path=%s pid=%d\n",
                     buf, strerror(err), lockPath, getpid()));
            return err;
          }
................................................................................
    UNIXVFS("unix-proxy",    proxyIoFinder ),
#endif
  };
  unsigned int i;          /* Loop counter */

  /* Double-check that the aSyscall[] array has been constructed
  ** correctly.  See ticket [bb3a86e890c8e96ab] */
  assert( ArraySize(aSyscall)==18 );

  /* Register all VFSes defined in the aVfs[] array */
  for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){
    sqlite3_vfs_register(&aVfs[i], i==0);
  }
  return SQLITE_OK; 
}







>
>
>
>
>
>







 







|
|










|
|
|
|




|







 







<







 







|
|
|











<







 







>







 







>
|
>







 







|







 







|







403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
....
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
....
1934
1935
1936
1937
1938
1939
1940

1941
1942
1943
1944
1945
1946
1947
....
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973

1974
1975
1976
1977
1978
1979
1980
....
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
....
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
....
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
....
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768

  { "unlink",       (sqlite3_syscall_ptr)unlink,           0 },
#define osUnlink    ((int(*)(const char*))aSyscall[16].pCurrent)

  { "openDirectory",    (sqlite3_syscall_ptr)openDirectory,      0 },
#define osOpenDirectory ((int(*)(const char*,int*))aSyscall[17].pCurrent)

  { "mkdir",        (sqlite3_syscall_ptr)mkdir,           0 },
#define osMkdir     ((int(*)(const char*,mode_t))aSyscall[18].pCurrent)

  { "rmdir",        (sqlite3_syscall_ptr)rmdir,           0 },
#define osRmdir     ((int(*)(const char*))aSyscall[19].pCurrent)

}; /* End of the overrideable system calls */

/*
** This is the xSetSystemCall() method of sqlite3_vfs for all of the
** "unix" VFSes.  Return SQLITE_OK opon successfully updating the
** system call pointer, or SQLITE_NOTFOUND if there is no configurable
** system call named zName.
................................................................................
/******************* End of the no-op lock implementation *********************
******************************************************************************/

/******************************************************************************
************************* Begin dot-file Locking ******************************
**
** The dotfile locking implementation uses the existance of separate lock
** files (really a directory) to control access to the database.  This works
** on just about every filesystem imaginable.  But there are serious downsides:
**
**    (1)  There is zero concurrency.  A single reader blocks all other
**         connections from reading or writing the database.
**
**    (2)  An application crash or power loss can leave stale lock files
**         sitting around that need to be cleared manually.
**
** Nevertheless, a dotlock is an appropriate locking mode for use if no
** other locking strategy is available.
**
** Dotfile locking works by creating a subdirectory in the same directory as
** the database and with the same name but with a ".lock" extension added.
** The existance of a lock directory implies an EXCLUSIVE lock.  All other
** lock types (SHARED, RESERVED, PENDING) are mapped into EXCLUSIVE.
*/

/*
** The file suffix added to the data base filename in order to create the
** lock directory.
*/
#define DOTLOCK_SUFFIX ".lock"

/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, set *pResOut
** to a non-zero value otherwise *pResOut is set to zero.  The return value
................................................................................
** routine to lower a locking level.
**
** With dotfile locking, we really only support state (4): EXCLUSIVE.
** But we track the other locking levels internally.
*/
static int dotlockLock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;

  char *zLockFile = (char *)pFile->lockingContext;
  int rc = SQLITE_OK;


  /* If we have any lock, then the lock file already exists.  All we have
  ** to do is adjust our internal record of the lock level.
  */
................................................................................
#else
    utimes(zLockFile, NULL);
#endif
    return SQLITE_OK;
  }
  
  /* grab an exclusive lock */
  rc = osMkdir(zLockFile, 0777);
  if( rc<0 ){
    /* failed to open/create the lock directory */
    int tErrno = errno;
    if( EEXIST == tErrno ){
      rc = SQLITE_BUSY;
    } else {
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
    }
    return rc;
  } 

  
  /* got it, set the type and return ok */
  pFile->eFileLock = eFileLock;
  return rc;
}

/*
................................................................................
** the requested locking level, this routine is a no-op.
**
** When the locking level reaches NO_LOCK, delete the lock file.
*/
static int dotlockUnlock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  char *zLockFile = (char *)pFile->lockingContext;
  int rc;

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d pid=%d (dotlock)\n", pFile->h, eFileLock,
	   pFile->eFileLock, getpid()));
  assert( eFileLock<=SHARED_LOCK );
  
  /* no-op if possible */
................................................................................
  if( eFileLock==SHARED_LOCK ){
    pFile->eFileLock = SHARED_LOCK;
    return SQLITE_OK;
  }
  
  /* To fully unlock the database, delete the lock file */
  assert( eFileLock==NO_LOCK );
  rc = osRmdir(zLockFile);
  if( rc<0 && errno==ENOTDIR ) rc = osUnlink(zLockFile);
  if( rc<0 ){
    int rc = 0;
    int tErrno = errno;
    if( ENOENT != tErrno ){
      rc = SQLITE_IOERR_UNLOCK;
    }
    if( IS_LOCK_ERROR(rc) ){
      pFile->lastErrno = tErrno;
................................................................................
  buf[0] = lockPath[0];
  for( i=1; i<len; i++ ){
    if( lockPath[i] == '/' && (i - start > 0) ){
      /* only mkdir if leaf dir != "." or "/" or ".." */
      if( i-start>2 || (i-start==1 && buf[start] != '.' && buf[start] != '/') 
         || (i-start==2 && buf[start] != '.' && buf[start+1] != '.') ){
        buf[i]='\0';
        if( osMkdir(buf, SQLITE_DEFAULT_PROXYDIR_PERMISSIONS) ){
          int err=errno;
          if( err!=EEXIST ) {
            OSTRACE(("CREATELOCKPATH  FAILED creating %s, "
                     "'%s' proxy lock path=%s pid=%d\n",
                     buf, strerror(err), lockPath, getpid()));
            return err;
          }
................................................................................
    UNIXVFS("unix-proxy",    proxyIoFinder ),
#endif
  };
  unsigned int i;          /* Loop counter */

  /* Double-check that the aSyscall[] array has been constructed
  ** correctly.  See ticket [bb3a86e890c8e96ab] */
  assert( ArraySize(aSyscall)==20 );

  /* Register all VFSes defined in the aVfs[] array */
  for(i=0; i<(sizeof(aVfs)/sizeof(sqlite3_vfs)); i++){
    sqlite3_vfs_register(&aVfs[i], i==0);
  }
  return SQLITE_OK; 
}

Changes to test/syscall.test.

55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

#-------------------------------------------------------------------------
# Tests for the xNextSystemCall method.
#
foreach s {
    open close access getcwd stat fstat ftruncate
    fcntl read pread write pwrite fchmod fallocate
    pread64 pwrite64 unlink openDirectory
} {
  if {[test_syscall exists $s]} {lappend syscall_list $s}
}
do_test 3.1 { lsort [test_syscall list] } [lsort $syscall_list]

#-------------------------------------------------------------------------
# This test verifies that if a call to open() fails and errno is set to







|







55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

#-------------------------------------------------------------------------
# Tests for the xNextSystemCall method.
#
foreach s {
    open close access getcwd stat fstat ftruncate
    fcntl read pread write pwrite fchmod fallocate
    pread64 pwrite64 unlink openDirectory mkdir rmdir
} {
  if {[test_syscall exists $s]} {lappend syscall_list $s}
}
do_test 3.1 { lsort [test_syscall list] } [lsort $syscall_list]

#-------------------------------------------------------------------------
# This test verifies that if a call to open() fails and errno is set to