=head1 NAME
File::FcntlLock - File locking with L<fcntl(2)>
This text also documents the following sub-packages:
=over 2
=item File::FcntlLock::XS
=item File::FcntlLock::Pure
=item File::FcntlLock::Inline
=back
=head1 SYNOPSIS
my $fs = new File::FcntlLock;
$fs->l_type( F_RDLCK );
$fs->l_whence( SEEK_CUR );
$fs->l_start( 100 );
$fs->l_len( 123 );
open my $fh, '<', 'file_name' or die "Can't open file: $!\n";
$fs->lock( $fh, F_SETLK )
or print "Locking failed: " . $fs->error . "\n";
$fs->l_type( F_UNLCK );
$fs->lock( $fh, F_SETLK )
or print "Unlocking failed: " . $fs->error . "\n";
=head1 DESCRIPTION
File locking in Perl is usually done using the L<flock()> function.
Unfortunately, this only allows locks on whole files and is often
implemented in terms of the L<flock(2)> system function which has
some shortcomings (especially concerning locks on remotely mounted
file systems) and slightly different behaviour than L<fcntl(2)>.
Using this module file locking via L<fcntl(2)> can be done (obviously,
this restricts the use of the module to systems that have a L<fcntl(2)>
system call). Before a file (or parts of a file) can be locked, an
object simulating a flock structure, containing information in a
binary format to be passed to L<fcntl(2)> for locking requests, must
be created and its properties set. Afterwards, by calling the C<lock()>
method a lock can be set and removed or it can be determined if and which
process currently holds the lock.
File::FcntlLock (or its alias File::FcntlLock::XS) uses a shared library,
build during installation, to call the L<fcntl(2)> system function directly.
If this is unsuitable there are two alternatives, File::FcntlLock::Pure and
File::FcntlLock::Inline. Both call the Perl L<fcntl()> function instead and
use Perl code to assemble and disassemble the structure. For this at some
time the (system-dependent) binary layout of the flock structure must
have been determined via a program written in C. The difference between
File::FcntlLock::Pure and File::FcntlLock::Inline is that for the former
this happened when the package is installed while for the latter it is
done each time the package is loaded (e.g., with the L<use> function).
Thus, for File::FcntlLock::Inline to work a C compiler must be available.
There are some minor differences in the functionality and the behaviour
on passing the method for locking invalid arguments to be described
below.
=head2 Creating objects
To create a new object, representing a flock structure, call C<new()>:
$fs = new File::FcntlLock;
The object has a number of properties, reflecting the members of the
flock structure to be passed to L<fcntl(2)> (more below). Per default
on object creation the C<l_type> property is set to C<F_RDLCK>,
C<l_whence> to C<SEEK_SET>, and both C<l_start> and C<l_len> to 0,
i.e., the settings for a read lock on the whole file.
These defaults can be overruled by passing the C<new()> method a set
of key-value pairs to initialize the objects properties, e.g. use
$fs = new File::FcntlLock( l_type => F_WRLCK,
l_whence => SEEK_SET,
l_start => 0,
l_len => 100 );
if you intend to obtain a write lock for the first 100 bytes of a file.
=head2 Object properties
Once the object simulating the flock structure has been created
the following methods allow to query and, in most cases, to also
modify its properties.
=over 4
=item C<l_type()>
If called without an argument the method returns the current setting
of the lock type, otherwise the lock type is set to the argument's value
which must be either C<F_RDLCK>, C<F_WRLCK> or C<F_UNLCK> (for read lock,
write lock or unlock).
=item C<l_whence()>
This method sets, when called with an argument, the C<l_whence>
property of the flock object, determining if the C<l_start> value
is relative to the start of the file, to the current position in
the file or to the end of the file. These values are C<SEEK_SET>,
C<SEEK_CUR> and C<SEEK_END> (also see the man page for L<lseek(2)>).
If called with no argument the current value of the property is
returned.
=item C<l_start()>
Queries or sets the start position (offset) of the lock in the
file according to the mode selected by the C<l_whence> member.
See also the man page for L<lseek(2)>.
=item C<l_len()>
Queries or sets the length of the region (in bytes) in the file
to be locked. A value of 0 is interpreted to mean a lock (starting
at C<l_start>) to the end of the file. E.g., a lock obtained
with C<l_whence> set to C<SEEK_SET> and both C<l_start> and
L<l_len> set to 0 locks the complete file.
According to SUSv3 negative values for C<l_start> are allowed
(resulting in a lock ranging from C<l_start+l_len> to
C<l_start-1>). Unfortunately, not all systems allow negative
arguments and will return an error when you try to obtain the
lock, so please read the L<fcntl(2)> man page of the system
carefully for details.
=item C<l_pid()>
If a call of the C<lock()> method with C<F_GETLK> indicates that
another process is holding the lock (in which case the L<l_type>
property will be either C<F_WRLCK> or C<F_RDLCK>) a call of the
C<l_pid()> method returns the PID of the process holding the lock.
This method does not accept any arguments.
=back
=head2 Locking
After having set up the object representing a flock structure one
can then try to obtain a lock, release it or determine the current
holder of the lock by invoking the C<lock()> method:
=over 2
=item C<lock()>
It expects two arguments. The first one is a file handle (or typeglob).
File::FcntlLock and thus File::FcntlLock::XS (B<but neither>
File::FcntlLock::Pure B<nor> File::FcntlLock::Inline) also accepts an
integer file descriptor. The second is a flag indicating the action to
be taken, e.g.
$fs->lock( $fh, F_SETLK );
There are three values that can be used as the second argument:
=over 4
=item C<F_SETLK>
With C<F_SETLK> the C<lock()> method tries to obtain a lock (when
C<l_type> is set to either C<F_WRLCK> or C<F_RDLCK>) or releases
it (if C<l_type> is set to C<F_UNLCK>). If an attempt is made to
obtain a lock but a lock is already held by some other process the
method call returns C<undef> and C<errno> is set to C<EACCESS> or
C<EAGAIN> (please see the the man page for L<fcntl(2)> for more
details).
=item C<F_SETLKW>
is similar to C<F_SETLK>, but instead of returning an error if the
lock can't be obtained immediately it puts the calling process to
sleep, i.e., it blocks, until the lock is obtained at some later
time. If a signal is received while waiting for the lock the
method returns C<undef> and C<errno> is set to C<EINTR>.
=item C<F_GETLK>
WithC<F_GETLK> the C<lock()> method determines if and which process
currently is holding the lock. If there's no other lock the C<l_type>
property will be set to C<F_UNLCK>. Otherwise the flock structure object
is set to the values that would prevent us from obtaining a lock. There
may be several processes that keep us from getting a lock, including
some that themselves are blocked waiting to obtain a lock. C<F_GETLK>
will only make details of one of these processes visible, and one has
no control over which process this is.
=back
On success the C<lock()> method returns the string "0 but true". If
the method fails (as indicated by an C<undef> return value) you can
either immediately evaluate the error number (using $!, $ERRNO
or $OS_ERROR) or check for it via the methods discussed below.
=back
=head2 Error handling
There are minor differences between File::FcntlLock on the one hand
and File::FcntlLock::Pure and File::FcntlLock::Inline on the other,
due to the first calling the system function L<fcntl(2)> directly
while the latter two invoke the Perl L<fcntl> function. Perl's
L<fcnrtl> function already returns a Perl error on some types of
invalid arguments. In contrast File::FcntlLock passes them on to the
L<fcntl(2)> system call and then returns the systems response to the
caller.
There are three methods for obtaining information about the
reason the a call of the C<lock()> method failed:
=over 4
=item C<lock_errno()>
Returns the C<errno> error number from the latest call of C<lock()>.
If the last call did not result in an error C<undef> is returned.
=item C<error()>
Returns a short description of the error that happened during the
latest call of C<lock()>. Please take the messages with a grain of
salt, they represent what SUSv3 (IEEE 1003.1-2001) and the Linux,
TRUE64, OpenBSD3 and Solaris8 man pages tell what the error numbers
mean. There could be differences (and additional error numbers) on
other systems. If there was no error the method returns C<undef>.
=item C<system_error()>
While the C<error()> method tries to return a string with some direct
relevance to the locking operation (i.e., "File or segment already
locked by other process(es)" instead of "Permission denied") this method
returns the "normal" system error message associated with C<errno>. The
method returns C<undef> if there was no error.
=back
=head2 EXPORT
The package exports the following constants:
=over 2
=item F_GETLK F_SETLK F_SETLKW
=item F_RDLCK F_WRLCK F_UNLCK
=item SEEK_SET SEEK_CUR SEEK_END
=back
=head1 CREDITS
Thanks to Mark Jason Dominus and Benjamin Goldberg for helpful discussions,
code examples and encouragement. Glenn Herteg pointed out several problems
and also helped improve the documentation. Julian Moreno Patino helped
correcting the documentation and pointed out problems arising on GNU Hurd
(which seems to have only very rudimentary support for locking with fcntl(2),
at least at that time). Niko Tyni and Guillem Jover encouraged and helped
with implementing alternatives to a XS-only approach which hopefully will
make the module more useful under certain circumstances.
=head1 AUTHOR
Jens Thoms Toerring <jt@toerring.de>
=head1 SEE ALSO
L<perl(1)>, L<fcntl(2)>, L<lseek(2)>.
=head1 LICENSE
This library is free software. You can redistribute it and/or modify it
under the same terms as Perl itself.