=head1 NAME

nbdkit-data-plugin - nbdkit plugin for serving data from the command line

=head1 SYNOPSIS

 nbdkit data [data=]'0 1 2 3 @0x1fe 0x55 0xaa'
             [size=SIZE] [allocator=sparse|malloc|zstd]

=for paragraph

 nbdkit data base64='aGVsbG8gbmJka2l0IHVzZXI='
             [size=SIZE] [allocator=sparse|malloc|zstd]

=for paragraph

 nbdkit data raw='binary_data'
             [size=SIZE] [allocator=sparse|malloc|zstd]

=head1 DESCRIPTION

C<nbdkit-data-plugin> is a plugin for L<nbdkit(1)> which serves a
small amount of data specified directly on the command line.  The
plugin gets its name from the C<data:> URI scheme used by web
browsers.  This is mainly useful for testing NBD clients.

You can serve data read-only using the I<-r> flag, or read-write.  Any
writes are thrown away when nbdkit exits.

Most operating systems have command line size limits which are quite a
lot smaller than any desirable disk image, so specifying a large,
fully populated disk image on the command line would not be possible.
However you can specify a small amount of data at the beginning of the
image, possibly followed by zeroes (using the C<size> parameter to pad
the image to the full size), or use the C<data> parameter creatively
to make mostly sparse disk images.

The C<size> parameter can specify any virtual size up to the maximum
supported by nbdkit (S<2⁶³-1 bytes>).

=head1 EXAMPLES

=head2 Create small disks filled with test patterns

 nbdkit data ' ( 0x55 0xAA )*2048 '
 nbdkit data ' ( "Hello" )*2000 ' size=8192

The first command creates a disk containing 4096 bytes filled with the
repeating bytes 0x55 0xAA.  The second command repeats
C<HelloHelloHello...>, truncating the disk to exactly 8192 bytes.

See also L<nbdkit-pattern-plugin(3)>.

=head2 Create a 1 MB disk with MBR-format partition table

 nbdkit data '
   @0x1be    # MBR first partition entry
     0           # Partition status
     0 2 0       # CHS start
     0x83        # Partition type (Linux)
     0x20 0x20 0 # CHS last sector
     le32:1      # LBA first sector
     le32:0x7ff  # LBA number of sectors
   @0x1fe    # Boot signature
     0x55 0xaa
   ' size=1M

A more flexible way to create partitions is to use
L<nbdkit-partitioning-plugin(1)>.  To create a data string from an
existing disk use the C<disk2data.pl> script provided in the nbdkit
sources
(L<https://gitlab.com/nbdkit/nbdkit/blob/master/plugins/data/disk2data.pl>).

=head2 Create a disk image with sector-aligned data

 nbdkit data ' <file1 @^512 <file2 @^512 <file3 @^512 '

Local binary files F<file1>, F<file2> and F<file3> are copied into the
disk image.  Regardless of the size of these files, they will all be
aligned to 512-byte sector boundaries.  Furthermore because of the
final alignment operation (C<@^512>) the total size of the disk will
also be rounded to a whole number of sectors.

=head2 Create a disk with the same random data in each sector

 nbdkit data ' </dev/urandom[:512]*16 '

The expression C<E<lt>/dev/urandom[:512]> reads 512 bytes (one sector)
of randomness from the system.  The same random data is repeated over
16 sectors.

=head2 Create a 1 MB disk with some nonsense data at the beginning

 nbdkit data base64=MTIz size=1M

The above command serves the bytes C<0x31 0x32 0x33> (which is the
base64 decoding of C<MTIz>), followed by S<1M - 3 bytes> of zeroes.

=head2 "Hello, world" using this plugin

 $ nbdkit data raw='Hello, world!' --run 'nbdcopy "$uri" - | cat'
 Hello, world!

This works by creating a disk containing the string
C<"Hello, world!">.  L<nbdcopy(1)> connects to the server using an NBD
URI (C<"$uri">) and copies the disk to stdout (C<->).  The extra
L<cat(1)> is needed because nbdcopy refuses to write raw disk data to
a terminal.

=head1 PARAMETERS

Exactly one of the C<data>, C<base64> or C<raw> parameters must be
supplied.

=over 4

=item [B<data=>]DATA

Specify the disk data using a simple compact format.  See
L</DATA FORMAT> below.

C<data=> __IS_MAGIC__

=item B<base64=>BASE64

The C<base64> parameter can be used to supply binary data encoded in
base64 on the command line.

This is only supported if nbdkit was compiled with GnuTLS E<ge> 3.6.0.
You can find out by checking if:

 $ nbdkit data --dump-plugin

contains:

 data_base64=yes

=item B<raw=>BINARY

The C<raw> parameter can be used to supply raw binary data directly on
the command line.

It is usually quite difficult to do this unless you are running nbdkit
from another program (see L<nbdkit-captive(1)>).  One particular
problem is that the data must not contain zero bytes (ie. C<\0>) since
those will be processed in C to mean the end of the string.  In almost
all cases it is better to use base64 encoding or the custom C<data>
format.

=item B<size=>SIZE

The data is truncated or extended to the size specified.

This parameter is optional: If omitted the size is defined by the size
of the C<data>, C<raw> or C<base64> parameter.

=item B<allocator=sparse>

=item B<allocator=malloc>[,B<mlock=true>]

=item B<allocator=zstd>

(nbdkit E<ge> 1.22)

Select the backend allocation strategy.  See
L<nbdkit-memory-plugin(1)/ALLOCATORS>.  The default is sparse.

=back

=head1 DATA FORMAT

The C<data> parameter lets you specify small disk images in a simple,
compact format.  It is a string containing a list of bytes which are
written into the disk image sequentially.  You can move the virtual
offset where bytes are written using C<@offset>.

 nbdkit data '0 1 2 3 @0x1fe 0x55 0xaa'

creates:

             total size 0x200 = 512 bytes (1 sector)
┌──────┬──────┬──────┬──────┬───────── ── ── ───┬──────┬──────┐
│   0  │   1  │   2  │   3  │  0     0  ...  0  │ 0x55 │ 0xaa │
└──────┴──────┴──────┴──────┴───────── ── ── ───┴──────┴──────┘
                                                  ↑
                                            offset 0x1fe

In this example the size is implied by the data.  But you could also
use the C<size> parameter to either truncate or extend (with zeroes)
the disk image.  Another way to write the same disk would be this,
where we align the offset to the end of the sector and move back 2
bytes to write the signature:

 nbdkit data '0 1 2 3 @^0x200 @-2 le16:0xaa55'

Fields in the string can be:

=over 4

=item BYTE

Write a byte at the current offset and advance the offset by 1.  The
byte may be specified as either decimal, octal (prefixed by C<0>) or
hexadecimal (prefixed by C<0x>).  To add repeated bytes use the C<*>
operator (eg. C<0xFF*16>).

=item B<le16:>WORD

=item B<be16:>WORD

=item B<le32:>WORD

=item B<be32:>WORD

=item B<le64:>WORD

=item B<be64:>WORD

(nbdkit E<ge> 1.28)

Write a word expressed in little endian (le) or big endian (be) of
length 16/32/64 bits and advance the offset by 2/4/8.  The word may be
specified in decimal, octal or hexadecimal.  For example:

 nbdkit data ' be32:0x1 '

generates the 4 byte sequence C<0 0 0 1>.

=item B<@>OFFSET

Moves the current offset to C<OFFSET>.  The offset may be specified as
either decimal, octal (prefixed by C<0>) or hexadecimal (prefixed by
C<0x>).  Offset C<@0> is the first byte of the disk.

=item B<@+>N

=item B<@->N

(nbdkit E<ge> 1.22)

Add or subtract C<N> from the current offset.

=item B<@^>ALIGNMENT

(nbdkit E<ge> 1.22)

If the current offset is not a multiple of C<ALIGNMENT> then the
offset is moved forward to the next multiple.  The next byte written
will be aligned to C<ALIGNMENT>.

=item B<E<lt>>FILE

(nbdkit E<ge> 1.8)

Read the contents of binary F<FILE> into the disk image at the current
offset.  The offset is incremented by the size of the file.  The
filename can be a relative or absolute path, but cannot contain
whitespace in the name.

=item B<E<lt>(>SCRIPTB<)>

(nbdkit E<ge> 1.24, not Windows)

Substitute the output of the shell script or external program as a
binary blob and advance the offset by the length in bytes of the
output.  You can use this to create more complex test patterns.  For
example this produces a 32K disk image with an incrementing test
pattern in groups of 4 bytes:

 nbdkit data ' <( i=0
                  while :; do
                      printf "%04d" $i; i=$((i+1))
                  done )[:32768] '

The script may contain C<(> and C<)> characters, but they must be in
matching pairs.  A script can produce a finite amount of output; or
(as in the example) an infinite amount which must be truncated using
the C<[:len]> slice operator.

Scripts must be idempotent, producing the same output each time they
are run.  This is because optimizations might change the order of
evaluation or number of times the script is called and you could get
different output in a future version of nbdkit.

Note that the script is passed to F</bin/sh>.  On some platforms like
Debian this might not be a full-featured shell.

=item B<">STRINGB<">

(nbdkit E<ge> 1.22)

Write a string into the image at the current offset and advance the
offset by the length of the string.  To include special characters in
the string you can escape them in the same way as C strings (eg. a
double quote character within the string should be written C<\">).  Be
careful with shell quoting around the whole data parameter.

=item B<(> ... B<)>

(nbdkit E<ge> 1.24)

Group a set of expressions into a single expression.

S<C<( ... )>> recursively creates a new data parser so any expression
can appear inside, including nested S<C<( ... )>>.  Note that offsets
and alignments within the subpattern are relative to the start of the
subpattern, not relative to the final disk image.

=item expression B<*> N

(nbdkit E<ge> 1.24)

Repeat the expression C<N> times.  The offset is incremented by the
length of the expression × N.  For example to create a repeating
pattern of 0x55, 0xAA for 512 (2×256) bytes do:

 nbdkit data '( 0x55 0xAA ) * 256'

=item expression B<[>NB<:>MB<]>

=item expression B<[>startB<:]>

=item expression B<[:>lenB<]>

(nbdkit E<ge> 1.24)

Take a slice of the expression.  Slices are S<B<[>startB<:>end+1B<]>>
where I<start> and I<end> are the first and last byte offsets of the
expression desired.  Either or both may be omitted.

S<B<[>startB<:]>> means to take bytes from offset I<start> to the end
of the expression.  S<B<[:>lenB<]>> means truncate the expression to
I<len> bytes.

=item expression B<-E<gt> \>NAME

=item B<\>NAME

(nbdkit E<ge> 1.24)

Assign an expression to a name which can be used later.  Names can be
used in the current scope (or any scopes nested within the current
scope), but disappear at the end of the current scope.  Names start
with a backslash character followed by one or more alphanumeric, dash
and underscore.  For example this makes two identical sectors both
containing a boot signature at the end:

 nbdkit data ' ( 0x55 0xAA ) -> \boot-signature
               ( @0x1fe \boot-signature ) -> \sector
               \sector \sector '

=item B<$>VAR

(nbdkit E<ge> 1.24)

Substitute command line parameters or environment variables.  The
variable is written in the same language as the C<data> parameter, and
when substituted it creates a nested scope like S<C<( ... )>>
expressions.  These three commands do the same thing:

 nbdkit data '$pattern*16' pattern='0x55 0xAA'

=for paragraph

 export pattern='0x55 0xAA' ; nbdkit data '$pattern*16'

=for paragraph

 nbdkit data '( 0x55 0xAA )*16'

=item B<#> COMMENT

(nbdkit E<ge> 1.24)

C<#> begins a comment stretching to the end of the current line.

=back

=head2 disk2data.pl script

This script can convert from small disk images into the data format
described above.

It is provided in the nbdkit sources.  See
L<https://gitlab.com/nbdkit/nbdkit/blob/master/plugins/data/disk2data.pl>

=head1 FILES

=over 4

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

The plugin.

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

=back

=head1 VERSION

C<nbdkit-data-plugin> first appeared in nbdkit 1.6.

=head1 SEE ALSO

L<nbdkit(1)>,
L<nbdkit-captive(1)>,
L<nbdkit-plugin(3)>,
L<nbdkit-info-plugin(1)>,
L<nbdkit-memory-plugin(1)>,
L<nbdkit-null-plugin(1)>,
L<nbdkit-ones-plugin(1)>,
L<nbdkit-partitioning-plugin(1)>,
L<nbdkit-pattern-plugin(1)>,
L<nbdkit-random-plugin(1)>,
L<nbdkit-sparse-random-plugin(1)>,
L<nbdkit-tmpdisk-plugin(1)>,
L<nbdkit-zero-plugin(1)>,
L<https://gitlab.com/nbdkit/nbdkit/blob/master/plugins/data/disk2data.pl>,
L<https://en.wikipedia.org/wiki/Base64>.

=head1 AUTHORS

Richard W.M. Jones

=head1 COPYRIGHT

Copyright Red Hat