=head1 NAME

nbdkit-file-plugin - nbdkit file plugin

=head1 SYNOPSIS

 nbdkit file [file=]FILENAME
             [cache=default|none] [fadvise=normal|random|sequential]

=for paragraph

 nbdkit file dir=DIRECTORY

=for paragraph

 nbdkit file fd=FILE_DESCRIPTOR

=for paragraph

 nbdkit file dirfd=FILE_DESCRIPTOR

=head1 DESCRIPTION

C<nbdkit-file-plugin> is a file serving plugin for L<nbdkit(1)>.

It serves the named C<FILENAME> over NBD.  Local block devices
(eg. F</dev/sda>) may also be served.

If you use the C<dir> parameter the plugin works in a different mode
where it serves files from the given C<DIRECTORY>, chosen by the
client using the NBD export name.

If you use the C<fd> or C<dirfd> parameter then you can pass the file
descriptor of a single disk or a directory to the plugin, inherited
from the parent process.  This can be useful where special permissions
or capabilities are needed to open the file descriptor, or you want to
run nbdkit in a sandboxed environment.

The file is writable unless either the I<-r> command line option, or
L<nbdkit-readonly-filter(1)> is used.

=head1 PARAMETERS

Exactly one of B<file>, B<dir>, B<fd> or B<dirfd> must be given.  This
controls the mode of the plugin, either serving a single file, the
files in a directory, a single file descriptor, or the files in the
directory of the file descriptor.

=over 4

=item B<cache=default>

=item B<cache=none>

(nbdkit E<ge> 1.22, not Windows)

If the file is very large and the plugin will read or write the whole
file, this can cause the file to be cached in memory (in the page
cache on Linux), causing other files and processes to be evicted,
leading to overall slow system performance.  If you know that the file
will only be read/written once and will not be needed afterwards so
caching it is pointless, use C<cache=none>.  This causes reads and
writes to minimize use of the page cache.  The default is to use the
kernel's default page cache strategy.

See also C<fadvise> below which can be used to hint the access
pattern.

=item B<dir=>DIRECTORY

(nbdkit E<ge> 1.22, not Windows)

Serve all regular files and block devices located directly inside the
directory named C<DIRECTORY>, including those found by following
symbolic links.  Other special files in the directory (such as
subdirectories, pipes, or Unix sockets) are ignored.

See L</Serving multiple files and block devices> below.

=item B<dirfd=>FILE_DESCRIPTOR

(nbdkit E<ge> 1.34, not Windows)

This is like the C<dir=> option, but instead of specifying the
directory by name, the parent process should open the directory and
pass this file descriptor by inheritance to nbdkit.

See L</Serving multiple files and block devices> below.

=item B<fadvise=normal>

=item B<fadvise=random>

=item B<fadvise=sequential>

(nbdkit E<ge> 1.22, not Windows)

This optional flag hints to the kernel that you will access the file
normally, or in a random order, or sequentially.  The exact behaviour
depends on your operating system, but for Linux using C<normal> causes
the kernel to read-ahead, C<sequential> causes the kernel to
read-ahead twice as much as C<normal>, and C<random> turns off
read-ahead.  See also L<posix_fadvise(2)>.

The default is C<normal>.

=item B<fd=>FILE_DESCRIPTOR

(nbdkit E<ge> 1.34, not Windows)

The parameter is the number of a file descriptor.  Serve the file or
device already open on this file descriptor.  The file descriptor is
usually inherited from the parent process.

=item [B<file=>]FILENAME

Serve the file named C<FILENAME>.  A local block device name can also
be used here.  When this mode is used, the export name requested by
the client is ignored.

C<file=> __IS_MAGIC__

=item [B<file=>]B<\\.\>CB<:>

=item [B<file=>]B<\\.\>Volume

=item [B<file=>]B<\\.\PhysicalDisk>I<N>

=item [B<file=>]B<\\.\CdRom>I<N>

(Windows only)

Serve the Windows volume specified by the device name.  See:
L<https://docs.microsoft.com/en-us/windows/win32/fileio/naming-a-file#win32-device-namespaces>.

=back

=head1 NOTES

=head2 Serving multiple files and block devices

Using C<dir=DIRECTORY> (or C<dirfd=DIRFD>) you can serve all regular
files and block devices located directly inside the directory named
C<DIRECTORY>, including those found by following symbolic links.
Other special files in the directory (such as subdirectories, pipes,
or Unix sockets) are ignored.

When this mode is used, the file to be served is chosen by the export
name passed by the client.  For security, when using directory mode,
this plugin will not accept export names containing slash (C</>).

For example:

 $ ls -l /var/tmp/exports
 total 0
 -rw-r--r--. 1 rjones rjones 1048576 Dec 14 15:34 disk1
 -rw-r--r--. 1 rjones rjones 2097152 Dec 14 15:34 disk2
 lrwxrwxrwx. 1 rjones rjones       9 Dec 14 15:35 sda1 -> /dev/sda1
 $ nbdkit file dir=/var/tmp/exports

will serve three exports called C<"disk1">, C<"disk2"> and C<"sda1">.
The first two are regular files and the last is a block device.  You
can add or remove files or symbolic links from the directory while
nbdkit is running.

To list exports, use L<nbdinfo(1)> I<--list> option, for example:

 $ nbdinfo --list nbd://localhost
 protocol: newstyle-fixed without TLS, using structured packets
 export="disk1":
     export-size: 1048576 (1M)
     uri: nbd://localhost:10809/disk1
 [etc]

An NBD client can request a list of available exports using
C<NBD_OPT_LIST>.  For libnbd clients see nbd_opt_list(3).

A client that requests the default export (C<"">) will be rejected.
However, you can use L<nbdkit-exportname-filter(1)> to adjust the
default export as well as other transformations of export names.  For
example to make F</var/tmp/exports/disk1> be the default export:

 nbdkit file dir=/var/tmp/exports \
        --filter=exportname default-export=disk1

=head2 Optimizing for random or sequential access

If you know in advance that the NBD client will access the file
randomly or only sequentially then you can hint that to the kernel
using:

 nbdkit file disk.img fadvise=random
 nbdkit file disk.img fadvise=sequential

As described in the L</PARAMETERS> section above, on Linux this
disables or increases the amount of read-ahead that the kernel does.

=head2 Files on tmpfs

If you want to expose a file that resides on a file system known to
have poor C<lseek(2)> performance when searching for holes (C<tmpfs>
is known to be one such file system), you can use
L<nbdkit-noextents-filter(1)> to avoid the penalty of probing for
holes.

=head2 Plugin I<--dump-plugin> output

You can obtain extra information about how the file plugin was
compiled by doing:

 nbdkit file --dump-plugin

Some of the fields which may appear are listed below.  Note these are
for information only and may be changed or removed at any time in the
future.

=over 4

=item C<file_block_size=yes>

If set, the plugin has support for getting the minimum and preferred
I/O size of block devices.

=item C<file_blkrotational=yes>

If set, the plugin has support for getting the rotational property of
block devices.

=item C<file_blksszget=yes>

=item C<file_blkzeroout=yes>

If both set, the plugin may be able to efficiently zero ranges of
block devices, where the driver and block device itself supports this.

=item C<file_blkdiscard=yes>

If set, the plugin may be able to efficiently trim ranges of block
devices, where the driver and block device itself supports this.

=item C<file_extents=yes>

If set, the plugin can read file extents.

=item C<file_falloc_fl_punch_hole=yes>

If set, the plugin may be able to punch holes (make sparse) files and
block devices.

=item C<file_falloc_fl_zero_range=yes>

If set, the plugin may be able to efficiently zero ranges of files and
block devices.

=item C<winfile=yes>

If present, this is the Windows version of the file plugin with
reduced functionality and some special Windows-only features, as noted
in this manual.

=back

=head2 Windows sparse files

This plugin supports sparse files on Windows (with hole punching).
However for this to work the files must already have the sparse
property, the plugin will not make existing files sparse.  Use the
S<C<fsutil sparse>> command to control the sparseness property of
files.

=head2 Old C<rdelay> and C<wdelay> parameters.

Before nbdkit supported filters (E<lt> 1.2) this plugin had extra
parameters C<rdelay> and C<wdelay> to insert delays.  These parameters
have been moved to L<nbdkit-delay-filter(1)>.  Modify the command line
to add I<--filter=delay> in order to use these parameters.

=head2 Concatenating files

To concatenate and export multiple files, use
L<nbdkit-split-plugin(1)>.

=head1 DEBUG FLAG

=over 4

=item B<-D file.zero=1>

This enables very verbose debugging of the NBD zero request.  This can
be used to tell if the file plugin is able to zero ranges in the file
or block device efficiently or not.

=back

=head1 FILES

=over 4

=item F<$plugindir/nbdkit-file-plugin.so>

The plugin.

Use C<nbdkit --dump-config> to find the location of C<$plugindir>.

=back

=head1 VERSION

C<nbdkit-file-plugin> first appeared in nbdkit 1.0.

=head1 SEE ALSO

L<nbdkit(1)>,
L<nbdkit-plugin(3)>,
L<nbdkit-split-plugin(1)>,
L<nbdkit-partitioning-plugin(1)>,
L<nbdkit-tmpdisk-plugin(1)>,
L<nbdkit-exportname-filter(1)>,
L<nbdkit-fua-filter(1)>,
L<nbdkit-luks-filter(1)>,
L<nbdkit-noextents-filter(1)>,
L<nbdkit-readonly-filter(1)>,
L<nbdinfo(1)>.

=head1 AUTHORS

Eric Blake

Nir Soffer

Richard W.M. Jones

=head1 COPYRIGHT

Copyright Red Hat