Record and File Locking

Chapter Updated 2/1/99


Locking Overview

Xbase DBMS supports multi-user processing through file and record locks. Record locking restricts multiple cooperating programs from simultaneously accessing the same data and corrupting it. Without record and file locking in a multi-user environment, simultaneous access to the data and index files can cause the files to become inaccurate and unusable.

Record locking is on by default in the Xbase DBMS library. To disable it, comment out the XB_LOCKING_ON option in the options.h file in the xbase/src directory.

The current Xbase DBMS record locking does not co-exist with other Xbase products and there is not yet support for locking in a DOS/Windows environment. The locking functions do work correctly for a Xbase DBMS only configuration. Future version of Xbase DBMS will have enhanced locking features for co-existing with other Xbase products and also include DOS/Windows support.

The locking methods return either XB_LOCK_FAILED or XB_NO_ERROR. If they return XB_LOCK_FAILED the actual reason can be found in the global variable errno or function perror() can be executed to view the results.

The errno field may contain one of the following values if the lock was not successful.

Error CodeDescription
EBADFInvalid file descriptor
EINVALInvalid lock information or file does not support locks
EACCESS
EAGAIN
Lock can not be set because it is blocked by an existing lock on the file.
ENOLCKThe system is out of lock resources, too many file locks in place.
EDEADLKDeadlock condition
EINTRProcess was interrupted by a signal while it was waiting


Types of Locks

  • Write or Exclusive Locks provide exclusive access to a particular file location. No other process can lock the same location.

  • Read or Shared Locks prohibit any process from requesting a write lock on a specified part of the file. Other processes can request simultaneous read locks.


    DBF File Locking Techniques

    Xbase DBMS uses the following protocol for DBF file and record locking:

    To lock a record - the first byte of the record is locked.
    To lock the file - the header bytes of the file are locked.

    When a record is being appended to the file, the header bytes are locked.
    When a record is being updated, the header bytes and the specific record are locked.

    This locking protocol is probably not compatable with other Xbase type products. However, Xbase can be safely used for multi-user access when it is not simultaneously updating DBF or NDX files while other products/programs are.


    NDX File Locking Techniques

    Xbase DBMS locks indexes by locking the first 512 bytes of the index file. The entire index is locked because any updates to the index potentially can modify significant portions of the index tree.


    DBT File Locking Techniques

    Xbase DBMS locks memo files by locking the first 4 bytes of the memo file. This effectively locks the entire file. The entire file is locked because any updates to the free block chain can significantly change the structure of the file.


    AutoLocking Features

    If XB_LOCKING_ON is set in the options.h file, the locking methods execute any appropriate locking logic. If XB_LOCKING_ON is not set in the options.h file, all locking methods return XB_NO_ERROR without performing any actual record or file locking. This enables the application program to always call locking routines regardless of the XB_LOCKING_ON switch in the options.h file.

    By leaving the autolocking features enabled, the application program does not need to address record, file or index locking. All locking is handled automatically by the Xbase routines. However, if access to the locking routines is required, they are available to the applciation programmer.

    When the files are automatically locked by the Xbase routines, the database file is locked first, then it locks the indexes in alphabetical order. To avoid deadlock conditions, files and record locks should always be done in the same order. When the files are unlocked, then indexes are unlocked first, then the database is unlocked.

    Auto-locking works well in an on-line transaction based environment. However, it does not function efficiently in batch mode. If you will be writing programs which process files in a batch mode, disabling auto-lock and locking the entire file at the beginning of the process and unlocking the file at the end of the process will significantly reduce process time. On a 586-200 class machine, a file with 45000 records can be read thru in a few seconds with the file locked in batch mode. In record-lock mode it takes about six minutes with the same processor.

    For processing large files, locking the file instead of locking each record is far more efficient. This is how you do it.

    For reading the file in batch mode:
    xbDbf.AutoLockOff();
    xbDbf.LockDatabase( F_SETLKW, F_RDLCK, 0L );

    For updating the file in batch mode:
    xbDbf.AutoLockOff();
    xbDbf.LockDatabase( F_SETLKW, F_WRLCK, 0L );