File::NFSLock - perl module to do NFS (or not) locking

  use File::NFSLock qw(uncache);
  use Fcntl qw(LOCK_EX LOCK_NB);

  my $file = "somefile";

  ### set up a lock - lasts until object looses scope
  if (my $lock = new File::NFSLock {
    file      => $file,
    lock_type => LOCK_EX|LOCK_NB,
    blocking_timeout   => 10,      # 10 sec
    stale_lock_timeout => 30 * 60, # 30 min
  }) {

    ### OR
    ### my $lock = File::NFSLock->new($file,LOCK_EX|LOCK_NB,10,30*60);

    ### do write protected stuff on $file
    ### at this point $file is uncached from NFS (most recent)
    open(FILE, "+<$file") || die $!;

    ### or open it any way you like
    ### my $fh = IO::File->open( $file, 'w' ) || die $!

    ### update (uncache across NFS) other files
    uncache("someotherfile1");
    uncache("someotherfile2");
    # open(FILE2,"someotherfile1");

    ### unlock it
    $lock->unlock();
    ### OR
    ### undef $lock;
    ### OR let $lock go out of scope
  }else{
    die "I couldn't lock the file [$File::NFSLock::errstr]";
  }

Program based of concept of hard linking of files being atomic across NFS. This concept was mentioned in Mail::Box::Locker (which was originally presented in Mail::Folder::Maildir). Some routine flow is taken from there particularly the idea of creating a random local file, hard linking a common file to the local file, and then checking the nlink status. Some ideologies were not complete (uncache mechanism, shared locking) and some coding was even incorrect (wrong stat index). File::NFSLock was written to be light, generic, and fast.

Locking occurs by creating a File::NFSLock object. If the object is created successfully, a lock is currently in place and remains in place until the lock object goes out of scope (or calls the unlock method).

A lock object is created by calling the new method and passing two to four parameters in the following manner:

  my $lock = File::NFSLock->new($file,
                                $lock_type,
                                $blocking_timeout,
                                $stale_lock_timeout,
                                );

Additionally, parameters may be passed as a hashref:

  my $lock = File::NFSLock->new({
    file               => $file,
    lock_type          => $lock_type,
    blocking_timeout   => $blocking_timeout,
    stale_lock_timeout => $stale_lock_timeout,
  });

Parameter 1: file

Filename of the file upon which it is anticipated that a write will happen to. Locking will provide the most recent version (uncached) of this file upon a successful file lock. It is not necessary for this file to exist.

Parameter 2: lock_type

Lock type must be one of the following:

  BLOCKING
  BL
  EXCLUSIVE (BLOCKING)
  EX
  NONBLOCKING
  NB
  SHARED
  SH

Or else one or more of the following joined with '|':

  Fcntl::LOCK_EX() (BLOCKING)
  Fcntl::LOCK_NB() (NONBLOCKING)
  Fcntl::LOCK_SH() (SHARED)

Lock type determines whether the lock will be blocking, non blocking, or shared. Blocking locks will wait until other locks are removed before the process continues. Non blocking locks will return undef if another process currently has the lock. Shared will allow other process to do a shared lock at the same time as long as there is not already an exclusive lock obtained.

Parameter 3: blocking_timeout (optional)

Timeout is used in conjunction with a blocking timeout. If specified, File::NFSLock will block up to the number of seconds specified in timeout before returning undef (could not get a lock).

Parameter 4: stale_lock_timeout (optional)

Timeout is used to see if an existing lock file is older than the stale lock timeout. If do_lock fails to get a lock, the modified time is checked and do_lock is attempted again. If the stale_lock_timeout is set to low, a recursion load could exist so do_lock will only recurse 10 times (this is only a problem if the stale_lock_timeout is set too low on the order of one or two seconds).

After the CW$lock object is instantiated with new, as outlined above, some methods may be used for additional functionality.

  $lock->unlock;

This method may be used to explicitly release a lock that is aquired. In most cases, it is not necessary to call unlock directly since it will implicitly be called when the object leaves whatever scope it is in.

  $lock->uncache;
  $lock->uncache("otherfile1");
  uncache("otherfile2");

This method is used to freshen up the contents of a file across NFS, ignoring what is contained in the NFS client cache. It is always called from within the new constructor on the file that the lock is being attempted. uncache may be used as either an object method or as a stand alone subroutine.

  my $pid = fork;
  if (defined $pid) {
    # Fork Failed
  } elsif ($pid) {
    $lock->newpid; # Parent
  } else {
    $lock->newpid; # Child
  }

If fork() is called after a lock has been aquired, then when the lock object leaves scope in either the parent or child, it will be released. This behavior may be inappropriate for your application. To delegate ownership of the lock from the parent to the child, both the parent and child process must call the newpid() method after a successful fork() call. This will prevent the parent from releasing the lock when unlock is called or when the lock object leaves scope. This is also useful to allow the parent to fail on subsequent lock attempts if the child lock is still aquired.

On failure, a global variable, CW$File::NFSLock::errstr, should be set and should contain the cause for the failure to get a lock. Useful primarily for debugging.

By default File::NFSLock will use a lock file extenstion of .NFSLock. This is in a global variable CW$File::NFSLock::LOCK_EXTENSION that may be changed to suit other purposes (such as compatibility in mail systems).

Notify paul@seamons.com or bbb@cpan.org if you spot anything.

Locks are not necessarily obtained on a first come first serve basis. Not only does this not seem fair to new processes trying to obtain a lock, but it may cause a process starvation condition on heavily locked files.

Locks cannot be obtained on directory nodes, nor can a directory node be uncached with the uncache routine because hard links do not work with directory nodes. Some other algorithm might be used to uncache a directory, but I am unaware of the best way to do it. The biggest use I can see would be to avoid NFS cache of directory modified and last accessed timestamps.

Download and extract tarball before running these commands in its base directory:

  perl Makefile.PL
  make
  make test
  make install

  rpm -ta SOURCES/File-NFSLock-*.tar.gz
  rpm -ih RPMS/noarch/perl-File-NFSLock-*.rpm

Paul T Seamons (paul@seamons.com) - Performed majority of the programming with copious amounts of input from Rob Brown. Rob B Brown (bbb@cpan.org) - In addition to helping in the programming, Rob Brown provided most of the core testing to make sure implementation worked properly. He is now the current maintainer. Also Mark Overmeer (mark@overmeer.net) - Author of Mail::Box::Locker, from which some key concepts for File::NFSLock were taken. Also Kevin Johnson (kjj@pobox.com) - Author of Mail::Folder::Maildir, from which Mark Overmeer based Mail::Box::Locker.

  Copyright (C) 2001
  Paul T Seamons
  paul@seamons.com
  http://seamons.com/

  Copyright (C) 2002-2003,
  Rob B Brown
  bbb@cpan.org

  This package may be distributed under the terms of either the
  GNU General Public License
    or the
  Perl Artistic License

  All rights reserved.