scripts/qcow2-to-stdout.py: Add script to write qcow2 images to stdout

Commit Message

Alberto Garcia June 10, 2024, 2:47 p.m. UTC

This tool converts a disk image to qcow2, writing the result directly
to stdout. This can be used for example to send the generated file
over the network.

This is equivalent to using qemu-img to convert a file to qcow2 and
then writing the result to stdout, with the difference that this tool
does not need to create this temporary qcow2 file and therefore does
not need any additional disk space.

The input file is read twice. The first pass is used to determine
which clusters contain non-zero data and that information is used to
create the qcow2 header, refcount table and blocks, and L1 and L2
tables. After all that metadata is created then the second pass is
used to write the guest data.

By default qcow2-to-stdout.py expects the input to be a raw file, but
if qemu-storage-daemon is available then it can also be used to read
images in other formats. Alternatively the user can also run qemu-ndb
or qemu-storage-daemon manually instead.

Signed-off-by: Alberto Garcia <[e-mail]>
Signed-off-by: Madeeha Javed <[e-mail]>
---
 scripts/qcow2-to-stdout.py | 330 +++++++++++++++++++++++++++++++++++++
 1 file changed, 330 insertions(+)
 create mode 100755 scripts/qcow2-to-stdout.py

Comments

Manos Pitsidianakis June 12, 2024, 6:01 a.m. UTC | #1

Hello Alberto,

On Mon, 10 Jun 2024 17:47, Alberto Garcia <[e-mail]> wrote:
>This tool converts a disk image to qcow2, writing the result directly
>to stdout. This can be used for example to send the generated file
>over the network.
>
>This is equivalent to using qemu-img to convert a file to qcow2 and
>then writing the result to stdout, with the difference that this tool
>does not need to create this temporary qcow2 file and therefore does
>not need any additional disk space.


Can you expand on this a little bit? Would modifying qemu-img to write 
to stdout if given, say, - instead of a file output path be enough to 
make this script unnecessary?


>The input file is read twice. The first pass is used to determine
>which clusters contain non-zero data and that information is used to
>create the qcow2 header, refcount table and blocks, and L1 and L2
>tables. After all that metadata is created then the second pass is
>used to write the guest data.
>
>By default qcow2-to-stdout.py expects the input to be a raw file, but
>if qemu-storage-daemon is available then it can also be used to read
>images in other formats. Alternatively the user can also run qemu-ndb
>or qemu-storage-daemon manually instead.
>
>Signed-off-by: Alberto Garcia <[e-mail]>
>Signed-off-by: Madeeha Javed <[e-mail]>
>---

Alberto Garcia June 12, 2024, 9:21 a.m. UTC | #2

On Wed 12 Jun 2024 09:01:01 AM +03, Manos Pitsidianakis wrote:
> Hello Alberto,

Hello Manos!

> > This is equivalent to using qemu-img to convert a file to qcow2 and
> > then writing the result to stdout, with the difference that this
> > tool does not need to create this temporary qcow2 file and therefore
> > does not need any additional disk space.
>
> Can you expand on this a little bit? Would modifying qemu-img to write
> to stdout if given, say, - instead of a file output path be enough to
> make this script unnecessary?

Yes, it would be enough. Allowing qemu-img convert to write to stdout
would indeed be very nice for the end user but it's a bit of a niche use
case and it's also not a trivial task so I don't think that it's worth
the effort. The output files that you pass to qemu-img convert need to
be seekable because the only way to produce a qcow2 file without doing
that is by precalculating all the metadata in advance before starting to
write anything (that's why this script reads the input file twice).

This is fundamentally different to what qemu-img convert does, which is
to read the input file from start to finish and write it to the output
file, relying on the relevant driver's existing write operations. All
those assume random access to the output file.

qemu-img is also much more generic in the sense that it supports many
different output formats and image options.

In contrast, writing the algorithm for a basic subset of qcow2 is quite
simple and that's why I think that it makes sense to do it in a separate
tool.

Berto

Manos Pitsidianakis June 12, 2024, 11 a.m. UTC | #3

On Mon, 10 Jun 2024 17:47, Alberto Garcia <[e-mail]> wrote:
>This tool converts a disk image to qcow2, writing the result directly
>to stdout. This can be used for example to send the generated file
>over the network.
>
>This is equivalent to using qemu-img to convert a file to qcow2 and
>then writing the result to stdout, with the difference that this tool
>does not need to create this temporary qcow2 file and therefore does
>not need any additional disk space.
>
>The input file is read twice. The first pass is used to determine
>which clusters contain non-zero data and that information is used to
>create the qcow2 header, refcount table and blocks, and L1 and L2
>tables. After all that metadata is created then the second pass is
>used to write the guest data.
>
>By default qcow2-to-stdout.py expects the input to be a raw file, but
>if qemu-storage-daemon is available then it can also be used to read
>images in other formats. Alternatively the user can also run qemu-ndb
>or qemu-storage-daemon manually instead.
>
>Signed-off-by: Alberto Garcia <[e-mail]>
>Signed-off-by: Madeeha Javed <[e-mail]>
>---
> scripts/qcow2-to-stdout.py | 330 +++++++++++++++++++++++++++++++++++++
> 1 file changed, 330 insertions(+)
> create mode 100755 scripts/qcow2-to-stdout.py
>

I recommend running the `black` formatter on this script, it makes the 
code more diff-friendly and uniform. Also it has become the de-facto 
python style.

Also, it's more pythonic to name constants in uppercase, like 
allocated_l2_tables. You can check such lints with pylint 
scripts/qcow2-to-stdout.py


>diff --git a/scripts/qcow2-to-stdout.py b/scripts/qcow2-to-stdout.py
>new file mode 100755
>index 0000000000..b9f75de690
>--- /dev/null
>+++ b/scripts/qcow2-to-stdout.py
>@@ -0,0 +1,330 @@
>+#!/usr/bin/env python3
>+
>+# This tool reads a disk image in any format and converts it to qcow2,
>+# writing the result directly to stdout.
>+#
>+# Copyright (C) 2024 Igalia, S.L.
>+#
>+# Authors: Alberto Garcia <[e-mail]>
>+#          Madeeha Javed <[e-mail]>
>+#
>+# SPDX-License-Identifier: GPL-2.0-or-later
>+#
>+# qcow2 files produced by this script are always arranged like this:
>+#
>+# - qcow2 header
>+# - refcount table
>+# - refcount blocks
>+# - L1 table
>+# - L2 tables
>+# - Data clusters
>+#
>+# A note about variable names: in qcow2 there is one refcount table
>+# and one (active) L1 table, although each can occupy several
>+# clusters. For the sake of simplicity the code sometimes talks about
>+# refcount tables and L1 tables when referring to those clusters.
>+
>+import argparse
>+import atexit
>+import math
>+import os
>+import signal
>+import struct
>+import sys
>+import subprocess
>+import tempfile
>+import time
>+
>+QCOW2_DEFAULT_CLUSTER_SIZE = 65536
>+QCOW2_DEFAULT_REFCOUNT_BITS = 16
>+QCOW2_DEFAULT_VERSION = 3
>+QCOW_OFLAG_COPIED = 1 << 63
>+
>+def bitmap_set(bitmap, idx):
>+    bitmap[int(idx / 8)] |= (1 << (idx % 8))
>+
>+def bitmap_test(bitmap, idx):
>+    return (bitmap[int(idx / 8)] & (1 << (idx % 8))) != 0
>+
>+# Kill the storage daemon on exit
>+def kill_storage_daemon(pid_file, raw_file, temp_dir):
>+    if os.path.exists(pid_file):
>+        with open(pid_file, 'r') as f:
>+            pid=int(f.readline())
>+        os.kill(pid, signal.SIGTERM)
>+        while os.path.exists(pid_file):
>+            time.sleep(0.1)
>+    os.unlink(raw_file)
>+    os.rmdir(temp_dir)
>+
>+def write_features(header):
>+    qcow2_features = [
>+        # Incompatible
>+        (0, 0, 'dirty bit'),
>+        (0, 1, 'corrupt bit'),
>+        (0, 2, 'external data file'),
>+        (0, 3, 'compression type'),
>+        (0, 4, 'extended L2 entries'),
>+        # Compatible
>+        (1, 0, 'lazy refcounts'),
>+        # Autoclear
>+        (2, 0, 'bitmaps'),
>+        (2, 1, 'raw external data')
>+    ]
>+    struct.pack_into('>I', header, 0x70, 0x6803f857)
>+    struct.pack_into('>I', header, 0x74, len(qcow2_features) * 48)
>+    cur_offset = 0x78

Minor comment: extract magic values/offsets into constant globals with 
descriptive names, it'd help the code be more readable and easier to 
maintain if ported in the future to other formats.

>+    for (feature_type, feature_bit, feature_name) in qcow2_features:
>+        struct.pack_into('>BB46s', header, cur_offset,
>+                         feature_type, feature_bit, feature_name.encode('ascii'))
>+        cur_offset += 48
>+


From here onwards put everything under a main block like so:

  if __name__ == "__main__":
      parser = argparse.ArgumentParser(
          description="This program converts a QEMU disk image to qcow2 "
          "and writes it to the standard output"
      )
      etc...

>+# Command-line arguments
>+parser = argparse.ArgumentParser(description='This program converts a QEMU disk image to qcow2 '
>+                                 'and writes it to the standard output')
>+parser.add_argument('input_file', help='name of the input file')

Suggestion:

-parser.add_argument('input_file', help='name of the input file')
+parser.add_argument('input_file', help='name of the input file', type=pathlib.Path, required=True)


>+parser.add_argument('-f', dest='input_format', metavar='input_format',
>+                    default='raw',
>+                    help='format of the input file (default: raw)')
>+parser.add_argument('-c', dest='cluster_size', metavar='cluster_size',
>+                    help=f'qcow2 cluster size (default: {QCOW2_DEFAULT_CLUSTER_SIZE})',
>+                    default=QCOW2_DEFAULT_CLUSTER_SIZE, type=int,
>+                    choices=[1 << x for x in range(9,22)])
>+parser.add_argument('-r', dest='refcount_bits', metavar='refcount_bits',
>+                    help=f'width of the reference count entries (default: {QCOW2_DEFAULT_REFCOUNT_BITS})',
>+                    default=QCOW2_DEFAULT_REFCOUNT_BITS, type=int,
>+                    choices=[1 << x for x in range(7)])
>+parser.add_argument('-v', dest='qcow2_version', metavar='qcow2_version',

Maybe -q instead of -v? No strong feelings on this one, it's just that 
-v is usually version. -q is also usually --quiet so not sure...

>+                    help=f'qcow2 version (default: {QCOW2_DEFAULT_VERSION})',
>+                    default=QCOW2_DEFAULT_VERSION, type=int, choices=[2, 3])
>+args = parser.parse_args()
>+
>+qcow2_cluster_size = args.cluster_size
>+qcow2_refcount_bits = args.refcount_bits
>+qcow2_version = args.qcow2_version
>+input_format = args.input_format
>+input_file = args.input_file
>+if not os.path.exists(input_file):
>+    print(f'[Error] Input file {input_file} does not exist', file=sys.stderr)
>+    sys.exit(1)
>+
>+if qcow2_refcount_bits != 16 and qcow2_version != 3:
>+    print(f'[Error] refcount_bits={qcow2_refcount_bits} is only supported with qcow2_version=3',
>+          file=sys.stderr)
>+    sys.exit(1)
>+
>+# Some basic values
>+l1_entries_per_table = int(qcow2_cluster_size / 8)
>+l2_entries_per_table = int(qcow2_cluster_size / 8)
>+refcounts_per_table  = int(qcow2_cluster_size / 8)
>+refcounts_per_block  = int(qcow2_cluster_size * 8 / qcow2_refcount_bits)
>+
>+# If the input file is not in raw format we can use
>+# qemu-storage-daemon to make it appear as such
>+if input_format != 'raw':
>+    temp_dir = tempfile.mkdtemp()

Consider using the tempfile.TemporaryDirectory as with... context 
manager so that the temp dir cleanup is performed automatically


>+    pid_file = temp_dir + "/pid"
>+    raw_file = temp_dir + "/raw"
>+    open(raw_file, 'wb').close()

Consider using a with open(...) open manager for opening the file


>+    atexit.register(kill_storage_daemon, pid_file, raw_file, temp_dir)

Hm, this too could be a context manager. Seems very C-like to use atexit 
here.

See 
https://docs.python.org/3/library/contextlib.html#contextlib.contextmanager

for an example which I paste here:


  from contextlib import contextmanager

  @contextmanager
  def managed_resource(*args, **kwds):
      # Code to acquire resource, e.g.:
      resource = acquire_resource(*args, **kwds)
      try:
          yield resource
      finally:
          # Code to release resource, e.g.:
          release_resource(resource)

  with managed_resource(timeout=3600) as resource:
      # Resource is released at the end of this block,
      # even if code in the block raises an exception


>+    ret = subprocess.run(["qemu-storage-daemon", "--daemonize", "--pidfile", pid_file,
>+                          "--blockdev", f"driver=file,node-name=file0,driver=file,filename={input_file},read-only=on",
>+                          "--blockdev", f"driver={input_format},node-name=disk0,file=file0,read-only=on",
>+                          "--export", f"type=fuse,id=export0,node-name=disk0,mountpoint={raw_file},writable=off"])

You can add shell=True, check=False arguments to subprocess.run() so 
that it captures the outputs. (check=False is the default behavior, but 
better make it explicit)


>+    if ret.returncode != 0:
>+        print('[Error] Could not start the qemu-storage-daemon', file=sys.stderr)
>+        sys.exit(1)
>+    input_file = raw_file
>+
>+# Virtual disk size, number of data clusters and L1 entries
>+disk_size = math.ceil(os.path.getsize(input_file) / 512) * 512 # Round up to the nearest multiple of 512
>+total_data_clusters = math.ceil(disk_size / qcow2_cluster_size)
>+l1_entries = math.ceil(total_data_clusters / l2_entries_per_table)
>+allocated_l1_tables = math.ceil(l1_entries / l1_entries_per_table)
>+
>+# Max L1 table size is 32 MB (QCOW_MAX_L1_SIZE in block/qcow2.h)
>+if ((l1_entries * 8) > (32 * 1024 *1024)):
>+    print('[Error] The image size is too large. Try using a larger cluster size',
>+          file=sys.stderr)
>+    sys.exit(1)
>+
>+# Two bitmaps indicating which L1 and L2 entries are set
>+l1_bitmap = bytearray(int(allocated_l1_tables * l1_entries_per_table / 8))
>+l2_bitmap = bytearray(int(allocated_l1_tables * l1_entries_per_table * l2_entries_per_table / 8))
>+allocated_l2_tables = 0
>+allocated_data_clusters = 0
>+with open(input_file, 'rb') as reader:
>+    zero_cluster = bytes(qcow2_cluster_size)
>+    # Read all clusters from the input file
>+    for idx in range(total_data_clusters):
>+        cluster = reader.read(qcow2_cluster_size)
>+        # If the last cluster is smaller than qcow2_cluster_size pad it with zeroes
>+        if (len(cluster) < qcow2_cluster_size):
>+            cluster += bytes(qcow2_cluster_size - len(cluster))
>+        # If a cluster has non-zero data then it must be allocated
>+        # in the output file and its L2 entry must be set
>+        if cluster != zero_cluster:
>+            bitmap_set(l2_bitmap, idx)
>+            allocated_data_clusters += 1
>+            # Allocated data clusters also need their corresponding L1 entry and L2 table
>+            l1_idx = math.floor(idx / l2_entries_per_table)
>+            if not bitmap_test(l1_bitmap, l1_idx):
>+                bitmap_set(l1_bitmap, l1_idx)
>+                allocated_l2_tables += 1
>+
>+# Total amount of allocated clusters excluding the refcount blocks and table
>+total_allocated_clusters = 1 + allocated_l1_tables + allocated_l2_tables + allocated_data_clusters
>+
>+# Clusters allocated for the refcount blocks and table
>+allocated_refcount_blocks = math.ceil(total_allocated_clusters  / refcounts_per_block)
>+allocated_refcount_tables = math.ceil(allocated_refcount_blocks / refcounts_per_table)
>+
>+# Now we have a problem because allocated_refcount_blocks and allocated_refcount_tables...
>+# (a) increase total_allocated_clusters, and
>+# (b) need to be recalculated when total_allocated_clusters is increased
>+# So we need to repeat the calculation as long as the numbers change
>+while True:
>+    new_total_allocated_clusters = total_allocated_clusters + allocated_refcount_tables + allocated_refcount_blocks
>+    new_allocated_refcount_blocks = math.ceil(new_total_allocated_clusters / refcounts_per_block)
>+    if new_allocated_refcount_blocks > allocated_refcount_blocks:
>+        allocated_refcount_blocks = new_allocated_refcount_blocks
>+        allocated_refcount_tables = math.ceil(allocated_refcount_blocks / refcounts_per_table)
>+    else:
>+        break
>+
>+# Now that we have the final numbers we can update total_allocated_clusters
>+total_allocated_clusters += allocated_refcount_tables + allocated_refcount_blocks
>+
>+# At this point we have the exact number of clusters that the output
>+# image is going to use so we can calculate all the offsets.
>+current_cluster_idx = 1
>+
>+refcount_table_offset = current_cluster_idx * qcow2_cluster_size
>+current_cluster_idx += allocated_refcount_tables
>+
>+refcount_block_offset = current_cluster_idx * qcow2_cluster_size
>+current_cluster_idx += allocated_refcount_blocks
>+
>+l1_table_offset = current_cluster_idx * qcow2_cluster_size
>+current_cluster_idx += allocated_l1_tables
>+
>+l2_table_offset = current_cluster_idx * qcow2_cluster_size
>+current_cluster_idx += allocated_l2_tables
>+
>+data_clusters_offset = current_cluster_idx * qcow2_cluster_size
>+
>+# Calculate some values used in the qcow2 header
>+if allocated_l1_tables == 0:
>+    l1_table_offset = 0
>+
>+hdr_cluster_bits = int(math.log2(qcow2_cluster_size))
>+hdr_refcount_bits = 0
>+hdr_length = 0
>+if qcow2_version == 3:
>+    hdr_refcount_bits = int(math.log2(qcow2_refcount_bits))
>+    hdr_length = 0x70
>+
>+### Write qcow2 header
>+cluster = bytearray(qcow2_cluster_size)
>+struct.pack_into('>4sIQIIQIIQQIIQQQQII', cluster, 0,
>+     b'QFI\xfb',            # QCOW magic string
>+     qcow2_version,
>+     0,                     # backing file offset
>+     0,                     # backing file sizes
>+     hdr_cluster_bits,
>+     disk_size,
>+     0,                     # encryption method
>+     l1_entries,
>+     l1_table_offset,
>+     refcount_table_offset,
>+     allocated_refcount_tables,
>+     0,                     # number of snapshots
>+     0,                     # snapshot table offset
>+     0,                     # compatible features
>+     0,                     # incompatible features
>+     0,                     # autoclear features
>+     hdr_refcount_bits,
>+     hdr_length)
>+
>+if qcow2_version == 3:
>+    write_features(cluster)
>+
>+sys.stdout.buffer.write(cluster)

Would it be a good idea to check if stdout is a tty and not a 
pipe/redirection? You can check it with isatty() and error out to 
prevent printing binary to the terminal.

>+
>+### Write refcount table
>+cur_offset = refcount_block_offset
>+remaining_refcount_table_entries = allocated_refcount_blocks # Each entry is a pointer to a refcount block
>+while remaining_refcount_table_entries > 0:
>+    cluster = bytearray(qcow2_cluster_size)
>+    to_write = min(remaining_refcount_table_entries, refcounts_per_table)
>+    remaining_refcount_table_entries -= to_write
>+    for idx in range(to_write):
>+        struct.pack_into('>Q', cluster, idx * 8, cur_offset)
>+        cur_offset += qcow2_cluster_size
>+    sys.stdout.buffer.write(cluster)
>+
>+### Write refcount blocks
>+remaining_refcount_block_entries = total_allocated_clusters # One entry for each allocated cluster
>+for tbl in range(allocated_refcount_blocks):
>+    cluster = bytearray(qcow2_cluster_size)
>+    to_write = min(remaining_refcount_block_entries, refcounts_per_block)
>+    remaining_refcount_block_entries -= to_write
>+    # All refcount entries contain the number 1. The only difference
>+    # is their bit width, defined when the image is created.
>+    for idx in range(to_write):
>+        if qcow2_refcount_bits == 64:
>+            struct.pack_into('>Q', cluster, idx * 8, 1)
>+        elif qcow2_refcount_bits == 32:
>+            struct.pack_into('>L', cluster, idx * 4, 1)
>+        elif qcow2_refcount_bits == 16:
>+            struct.pack_into('>H', cluster, idx * 2, 1)
>+        elif qcow2_refcount_bits == 8:
>+            cluster[idx] = 1
>+        elif qcow2_refcount_bits == 4:
>+            cluster[int(idx / 2)] |= 1 << ((idx % 2) * 4)
>+        elif qcow2_refcount_bits == 2:
>+            cluster[int(idx / 4)] |= 1 << ((idx % 4) * 2)
>+        elif qcow2_refcount_bits == 1:
>+            cluster[int(idx / 8)] |= 1 << (idx % 8)
>+    sys.stdout.buffer.write(cluster)
>+
>+### Write L1 table
>+cur_offset = l2_table_offset
>+for tbl in range(allocated_l1_tables):
>+    cluster = bytearray(qcow2_cluster_size)
>+    for idx in range(l1_entries_per_table):
>+        l1_idx = tbl * l1_entries_per_table + idx
>+        if bitmap_test(l1_bitmap, l1_idx):
>+            struct.pack_into('>Q', cluster, idx * 8, cur_offset | QCOW_OFLAG_COPIED)
>+            cur_offset += qcow2_cluster_size
>+    sys.stdout.buffer.write(cluster)
>+
>+### Write L2 tables
>+cur_offset = data_clusters_offset
>+for tbl in range(l1_entries):
>+    # Skip the empty L2 tables. We can identify them because
>+    # there is no L1 entry pointing at them.
>+    if bitmap_test(l1_bitmap, tbl):
>+        cluster = bytearray(qcow2_cluster_size)
>+        for idx in range(l2_entries_per_table):
>+            l2_idx = tbl * l2_entries_per_table + idx
>+            if bitmap_test(l2_bitmap, l2_idx):
>+                struct.pack_into('>Q', cluster, idx * 8, cur_offset | QCOW_OFLAG_COPIED)
>+                cur_offset += qcow2_cluster_size
>+        sys.stdout.buffer.write(cluster)
>+
>+### Write data clusters
>+with open(input_file, 'rb') as reader:
>+    skip = 0
>+    for idx in range(total_data_clusters):
>+        if bitmap_test(l2_bitmap, idx):
>+            if skip > 0:
>+                reader.seek(qcow2_cluster_size * skip, 1)
>+                skip = 0
>+            cluster = reader.read(qcow2_cluster_size)
>+            # If the last cluster is smaller than qcow2_cluster_size pad it with zeroes
>+            if (len(cluster) < qcow2_cluster_size):
>+                cluster += bytes(qcow2_cluster_size - len(cluster))
>+            sys.stdout.buffer.write(cluster)
>+        else:
>+            skip += 1
>-- 
>2.39.2
>
>

Manos Pitsidianakis June 12, 2024, 11:25 a.m. UTC | #4

On Wed, 12 Jun 2024 12:21, Alberto Garcia <[e-mail]> wrote:
>On Wed 12 Jun 2024 09:01:01 AM +03, Manos Pitsidianakis wrote:
>> Hello Alberto,
>
>Hello Manos!
>
>> > This is equivalent to using qemu-img to convert a file to qcow2 and
>> > then writing the result to stdout, with the difference that this
>> > tool does not need to create this temporary qcow2 file and therefore
>> > does not need any additional disk space.
>>
>> Can you expand on this a little bit? Would modifying qemu-img to write
>> to stdout if given, say, - instead of a file output path be enough to
>> make this script unnecessary?
>
>Yes, it would be enough. Allowing qemu-img convert to write to stdout
>would indeed be very nice for the end user but it's a bit of a niche use
>case and it's also not a trivial task so I don't think that it's worth
>the effort. The output files that you pass to qemu-img convert need to
>be seekable because the only way to produce a qcow2 file without doing
>that is by precalculating all the metadata in advance before starting to
>write anything (that's why this script reads the input file twice).
>
>This is fundamentally different to what qemu-img convert does, which is
>to read the input file from start to finish and write it to the output
>file, relying on the relevant driver's existing write operations. All
>those assume random access to the output file.
>
>qemu-img is also much more generic in the sense that it supports many
>different output formats and image options.
>
>In contrast, writing the algorithm for a basic subset of qcow2 is quite
>simple and that's why I think that it makes sense to do it in a separate
>tool.
>
>Berto

Thanks for the complete explanation! It makes sense. Maybe add it to the 
commit message as well, it's informative.

Manos

Alberto Garcia July 1, 2024, 8:56 a.m. UTC | #5

On Wed 12 Jun 2024 02:00:19 PM +03, Manos Pitsidianakis wrote:

Hi, thanks for the review and sorry for taking so long to reply, I was
on vacation.

>> scripts/qcow2-to-stdout.py | 330 +++++++++++++++++++++++++++++++++++++
>> 1 file changed, 330 insertions(+)
>> create mode 100755 scripts/qcow2-to-stdout.py
>
> I recommend running the `black` formatter on this script, it makes the
> code more diff-friendly and uniform. Also it has become the de-facto
> python style.

Hmmm, I don't like how it reformats some of the lines. However other
changes do make sense, so I'll apply those.

> Also, it's more pythonic to name constants in uppercase, like
> allocated_l2_tables. You can check such lints with pylint
> scripts/qcow2-to-stdout.py

allocated_l2_tables is not a constant :-?

>>+    struct.pack_into('>I', header, 0x70, 0x6803f857)
>>+    struct.pack_into('>I', header, 0x74, len(qcow2_features) * 48)
>>+    cur_offset = 0x78
>
> Minor comment: extract magic values/offsets into constant globals with
> descriptive names, it'd help the code be more readable and easier to
> maintain if ported in the future to other formats.

Good idea, will do.

>>+    for (feature_type, feature_bit, feature_name) in qcow2_features:
>>+        struct.pack_into('>BB46s', header, cur_offset,
>>+                         feature_type, feature_bit, feature_name.encode('ascii'))
>>+        cur_offset += 48
>>+
>
>>From here onwards put everything under a main block like so:

Ok.

>>+# Command-line arguments
>>+parser = argparse.ArgumentParser(description='This program converts a QEMU disk image to qcow2 '
>>+                                 'and writes it to the standard output')
>>+parser.add_argument('input_file', help='name of the input file')
>
> Suggestion:
>
> -parser.add_argument('input_file', help='name of the input file')
> +parser.add_argument('input_file', help='name of the input file', type=pathlib.Path, required=True)

'required' is not valid in positional arguments, and I'm not sure what
benefits using pathlib brings in this case.

>>+parser.add_argument('-v', dest='qcow2_version', metavar='qcow2_version',
>
> Maybe -q instead of -v? No strong feelings on this one, it's just that 
> -v is usually version. -q is also usually --quiet so not sure...

Yeah, I thought the same but I didn't want to complicate this too much,
this is just a helper script.

>>+# If the input file is not in raw format we can use
>>+# qemu-storage-daemon to make it appear as such
>>+if input_format != 'raw':
>>+    temp_dir = tempfile.mkdtemp()
>
> Consider using the tempfile.TemporaryDirectory as with... context 
> manager so that the temp dir cleanup is performed automatically

I don't think I can do that directly here because the temp dir has to
live until the very end (qemu-storage-daemon needs it).

>>+    pid_file = temp_dir + "/pid"
>>+    raw_file = temp_dir + "/raw"
>>+    open(raw_file, 'wb').close()
>
> Consider using a with open(...) open manager for opening the file

How would that be? Like this?

    with open(raw_file, 'wb'):
        pass

If so I don't see the benefit, I just need to create an empty file and
close it immediately.

>>+    atexit.register(kill_storage_daemon, pid_file, raw_file, temp_dir)
>
> Hm, this too could be a context manager. Seems very C-like to use
> atexit here.

Yeah it is, but I think that using the context manager would require me
to split the main function in two, and I'm not sure that it's worth it
for this case. Other Python scripts in the QEMU repo use atexit already.

>>+    ret = subprocess.run(["qemu-storage-daemon", "--daemonize", "--pidfile", pid_file,
>>+                          "--blockdev", f"driver=file,node-name=file0,driver=file,filename={input_file},read-only=on",
>>+                          "--blockdev", f"driver={input_format},node-name=disk0,file=file0,read-only=on",
>>+                          "--export", f"type=fuse,id=export0,node-name=disk0,mountpoint={raw_file},writable=off"])
>
> You can add shell=True, check=False arguments to subprocess.run() so 
> that it captures the outputs. (check=False is the default behavior, but 
> better make it explicit)

I'm not sure that I understand, why would I need to use a shell here?

>>+sys.stdout.buffer.write(cluster)
>
> Would it be a good idea to check if stdout is a tty and not a
> pipe/redirection? You can check it with isatty() and error out to
> prevent printing binary to the terminal.

Yeah this is a good idea, thanks.

Berto

Manos Pitsidianakis July 1, 2024, 11:07 a.m. UTC | #6

Hi Berto :)

On Mon, 1 Jul 2024 at 11:56, Alberto Garcia <[e-mail]> wrote:
>
> On Wed 12 Jun 2024 02:00:19 PM +03, Manos Pitsidianakis wrote:
>
> Hi, thanks for the review and sorry for taking so long to reply, I was
> on vacation.
>
> >> scripts/qcow2-to-stdout.py | 330 +++++++++++++++++++++++++++++++++++++
> >> 1 file changed, 330 insertions(+)
> >> create mode 100755 scripts/qcow2-to-stdout.py
> >
> > I recommend running the `black` formatter on this script, it makes the
> > code more diff-friendly and uniform. Also it has become the de-facto
> > python style.
>
> Hmmm, I don't like how it reformats some of the lines. However other
> changes do make sense, so I'll apply those.

It's not a project coding style requirement (for now) so it's fine.

>
> > Also, it's more pythonic to name constants in uppercase, like
> > allocated_l2_tables. You can check such lints with pylint
> > scripts/qcow2-to-stdout.py
>
> allocated_l2_tables is not a constant :-?

Eeeh right, correct. `pylint`'s error message said it was a constant,
my bad. It says it is a constant because it is declared as a global
(module-level), `__all__` is not defined with any globals, and
according to PEP-8 non-public globals start with an underscore in the
name.

>
> >>+    struct.pack_into('>I', header, 0x70, 0x6803f857)
> >>+    struct.pack_into('>I', header, 0x74, len(qcow2_features) * 48)
> >>+    cur_offset = 0x78
> >
> > Minor comment: extract magic values/offsets into constant globals with
> > descriptive names, it'd help the code be more readable and easier to
> > maintain if ported in the future to other formats.
>
> Good idea, will do.
>
> >>+    for (feature_type, feature_bit, feature_name) in qcow2_features:
> >>+        struct.pack_into('>BB46s', header, cur_offset,
> >>+                         feature_type, feature_bit, feature_name.encode('ascii'))
> >>+        cur_offset += 48
> >>+
> >
> >>From here onwards put everything under a main block like so:
>
> Ok.
>
> >>+# Command-line arguments
> >>+parser = argparse.ArgumentParser(description='This program converts a QEMU disk image to qcow2 '
> >>+                                 'and writes it to the standard output')
> >>+parser.add_argument('input_file', help='name of the input file')
> >
> > Suggestion:
> >
> > -parser.add_argument('input_file', help='name of the input file')
> > +parser.add_argument('input_file', help='name of the input file', type=pathlib.Path, required=True)
>
> 'required' is not valid in positional arguments,

Sorry did not notice it's a positional!

> and I'm not sure what
> benefits using pathlib brings in this case.

implicit type requirement, argument value validations, path normalization etc.

>
> >>+parser.add_argument('-v', dest='qcow2_version', metavar='qcow2_version',
> >
> > Maybe -q instead of -v? No strong feelings on this one, it's just that
> > -v is usually version. -q is also usually --quiet so not sure...
>
> Yeah, I thought the same but I didn't want to complicate this too much,
> this is just a helper script.
>
> >>+# If the input file is not in raw format we can use
> >>+# qemu-storage-daemon to make it appear as such
> >>+if input_format != 'raw':
> >>+    temp_dir = tempfile.mkdtemp()
> >
> > Consider using the tempfile.TemporaryDirectory as with... context
> > manager so that the temp dir cleanup is performed automatically
>
> I don't think I can do that directly here because the temp dir has to
> live until the very end (qemu-storage-daemon needs it).
>
> >>+    pid_file = temp_dir + "/pid"
> >>+    raw_file = temp_dir + "/raw"
> >>+    open(raw_file, 'wb').close()
> >
> > Consider using a with open(...) open manager for opening the file
>
> How would that be? Like this?
>
>     with open(raw_file, 'wb'):
>         pass
>
> If so I don't see the benefit, I just need to create an empty file and
> close it immediately.

My only argument here is that it's "more pythonic" which I know is of
little value and consequence :) Feel free to ignore! They were mere
suggestions.

>
> >>+    atexit.register(kill_storage_daemon, pid_file, raw_file, temp_dir)
> >
> > Hm, this too could be a context manager. Seems very C-like to use
> > atexit here.
>
> Yeah it is, but I think that using the context manager would require me
> to split the main function in two, and I'm not sure that it's worth it
> for this case. Other Python scripts in the QEMU repo use atexit already.
>
> >>+    ret = subprocess.run(["qemu-storage-daemon", "--daemonize", "--pidfile", pid_file,
> >>+                          "--blockdev", f"driver=file,node-name=file0,driver=file,filename={input_file},read-only=on",
> >>+                          "--blockdev", f"driver={input_format},node-name=disk0,file=file0,read-only=on",
> >>+                          "--export", f"type=fuse,id=export0,node-name=disk0,mountpoint={raw_file},writable=off"])
> >
> > You can add shell=True, check=False arguments to subprocess.run() so
> > that it captures the outputs. (check=False is the default behavior, but
> > better make it explicit)
>
> I'm not sure that I understand, why would I need to use a shell here?

I must have meant capture_output=True, not shell=True, sorry for that

Alberto Garcia July 1, 2024, 1:42 p.m. UTC | #7

On Mon 01 Jul 2024 02:07:01 PM +03, Manos Pitsidianakis wrote:
>> and I'm not sure what benefits using pathlib brings in this case.
>
> implicit type requirement, argument value validations, path
> normalization etc.

Do you have a specific example? I don't see any difference in behavior
if I make input_file a pathlib.Path, I still need to check if the file
exists, etc., I don't see that this is validating anything.

>>     with open(raw_file, 'wb'):
>>         pass
>>
>> If so I don't see the benefit, I just need to create an empty file and
>> close it immediately.
>
> My only argument here is that it's "more pythonic" which I know is of
> little value and consequence :) Feel free to ignore! They were mere
> suggestions.

In general I would agree (that's why I'm opening files this way in other
parts of the script) but for this case I don't think it's worth it.

>> I'm not sure that I understand, why would I need to use a shell here?
>
> I must have meant capture_output=True, not shell=True, sorry for that
> 

Patch

diff --git a/scripts/qcow2-to-stdout.py b/scripts/qcow2-to-stdout.py
new file mode 100755
index 0000000000..b9f75de690
--- /dev/null
+++ b/scripts/qcow2-to-stdout.py
@@ -0,0 +1,330 @@ 
+#!/usr/bin/env python3
+
+# This tool reads a disk image in any format and converts it to qcow2,
+# writing the result directly to stdout.
+#
+# Copyright (C) 2024 Igalia, S.L.
+#
+# Authors: Alberto Garcia <[e-mail]>
+#          Madeeha Javed <[e-mail]>
+#
+# SPDX-License-Identifier: GPL-2.0-or-later
+#
+# qcow2 files produced by this script are always arranged like this:
+#
+# - qcow2 header
+# - refcount table
+# - refcount blocks
+# - L1 table
+# - L2 tables
+# - Data clusters
+#
+# A note about variable names: in qcow2 there is one refcount table
+# and one (active) L1 table, although each can occupy several
+# clusters. For the sake of simplicity the code sometimes talks about
+# refcount tables and L1 tables when referring to those clusters.
+
+import argparse
+import atexit
+import math
+import os
+import signal
+import struct
+import sys
+import subprocess
+import tempfile
+import time
+
+QCOW2_DEFAULT_CLUSTER_SIZE = 65536
+QCOW2_DEFAULT_REFCOUNT_BITS = 16
+QCOW2_DEFAULT_VERSION = 3
+QCOW_OFLAG_COPIED = 1 << 63
+
+def bitmap_set(bitmap, idx):
+    bitmap[int(idx / 8)] |= (1 << (idx % 8))
+
+def bitmap_test(bitmap, idx):
+    return (bitmap[int(idx / 8)] & (1 << (idx % 8))) != 0
+
+# Kill the storage daemon on exit
+def kill_storage_daemon(pid_file, raw_file, temp_dir):
+    if os.path.exists(pid_file):
+        with open(pid_file, 'r') as f:
+            pid=int(f.readline())
+        os.kill(pid, signal.SIGTERM)
+        while os.path.exists(pid_file):
+            time.sleep(0.1)
+    os.unlink(raw_file)
+    os.rmdir(temp_dir)
+
+def write_features(header):
+    qcow2_features = [
+        # Incompatible
+        (0, 0, 'dirty bit'),
+        (0, 1, 'corrupt bit'),
+        (0, 2, 'external data file'),
+        (0, 3, 'compression type'),
+        (0, 4, 'extended L2 entries'),
+        # Compatible
+        (1, 0, 'lazy refcounts'),
+        # Autoclear
+        (2, 0, 'bitmaps'),
+        (2, 1, 'raw external data')
+    ]
+    struct.pack_into('>I', header, 0x70, 0x6803f857)
+    struct.pack_into('>I', header, 0x74, len(qcow2_features) * 48)
+    cur_offset = 0x78
+    for (feature_type, feature_bit, feature_name) in qcow2_features:
+        struct.pack_into('>BB46s', header, cur_offset,
+                         feature_type, feature_bit, feature_name.encode('ascii'))
+        cur_offset += 48
+
+# Command-line arguments
+parser = argparse.ArgumentParser(description='This program converts a QEMU disk image to qcow2 '
+                                 'and writes it to the standard output')
+parser.add_argument('input_file', help='name of the input file')
+parser.add_argument('-f', dest='input_format', metavar='input_format',
+                    default='raw',
+                    help='format of the input file (default: raw)')
+parser.add_argument('-c', dest='cluster_size', metavar='cluster_size',
+                    help=f'qcow2 cluster size (default: {QCOW2_DEFAULT_CLUSTER_SIZE})',
+                    default=QCOW2_DEFAULT_CLUSTER_SIZE, type=int,
+                    choices=[1 << x for x in range(9,22)])
+parser.add_argument('-r', dest='refcount_bits', metavar='refcount_bits',
+                    help=f'width of the reference count entries (default: {QCOW2_DEFAULT_REFCOUNT_BITS})',
+                    default=QCOW2_DEFAULT_REFCOUNT_BITS, type=int,
+                    choices=[1 << x for x in range(7)])
+parser.add_argument('-v', dest='qcow2_version', metavar='qcow2_version',
+                    help=f'qcow2 version (default: {QCOW2_DEFAULT_VERSION})',
+                    default=QCOW2_DEFAULT_VERSION, type=int, choices=[2, 3])
+args = parser.parse_args()
+
+qcow2_cluster_size = args.cluster_size
+qcow2_refcount_bits = args.refcount_bits
+qcow2_version = args.qcow2_version
+input_format = args.input_format
+input_file = args.input_file
+if not os.path.exists(input_file):
+    print(f'[Error] Input file {input_file} does not exist', file=sys.stderr)
+    sys.exit(1)
+
+if qcow2_refcount_bits != 16 and qcow2_version != 3:
+    print(f'[Error] refcount_bits={qcow2_refcount_bits} is only supported with qcow2_version=3',
+          file=sys.stderr)
+    sys.exit(1)
+
+# Some basic values
+l1_entries_per_table = int(qcow2_cluster_size / 8)
+l2_entries_per_table = int(qcow2_cluster_size / 8)
+refcounts_per_table  = int(qcow2_cluster_size / 8)
+refcounts_per_block  = int(qcow2_cluster_size * 8 / qcow2_refcount_bits)
+
+# If the input file is not in raw format we can use
+# qemu-storage-daemon to make it appear as such
+if input_format != 'raw':
+    temp_dir = tempfile.mkdtemp()
+    pid_file = temp_dir + "/pid"
+    raw_file = temp_dir + "/raw"
+    open(raw_file, 'wb').close()
+    atexit.register(kill_storage_daemon, pid_file, raw_file, temp_dir)
+    ret = subprocess.run(["qemu-storage-daemon", "--daemonize", "--pidfile", pid_file,
+                          "--blockdev", f"driver=file,node-name=file0,driver=file,filename={input_file},read-only=on",
+                          "--blockdev", f"driver={input_format},node-name=disk0,file=file0,read-only=on",
+                          "--export", f"type=fuse,id=export0,node-name=disk0,mountpoint={raw_file},writable=off"])
+    if ret.returncode != 0:
+        print('[Error] Could not start the qemu-storage-daemon', file=sys.stderr)
+        sys.exit(1)
+    input_file = raw_file
+
+# Virtual disk size, number of data clusters and L1 entries
+disk_size = math.ceil(os.path.getsize(input_file) / 512) * 512 # Round up to the nearest multiple of 512
+total_data_clusters = math.ceil(disk_size / qcow2_cluster_size)
+l1_entries = math.ceil(total_data_clusters / l2_entries_per_table)
+allocated_l1_tables = math.ceil(l1_entries / l1_entries_per_table)
+
+# Max L1 table size is 32 MB (QCOW_MAX_L1_SIZE in block/qcow2.h)
+if ((l1_entries * 8) > (32 * 1024 *1024)):
+    print('[Error] The image size is too large. Try using a larger cluster size',
+          file=sys.stderr)
+    sys.exit(1)
+
+# Two bitmaps indicating which L1 and L2 entries are set
+l1_bitmap = bytearray(int(allocated_l1_tables * l1_entries_per_table / 8))
+l2_bitmap = bytearray(int(allocated_l1_tables * l1_entries_per_table * l2_entries_per_table / 8))
+allocated_l2_tables = 0
+allocated_data_clusters = 0
+with open(input_file, 'rb') as reader:
+    zero_cluster = bytes(qcow2_cluster_size)
+    # Read all clusters from the input file
+    for idx in range(total_data_clusters):
+        cluster = reader.read(qcow2_cluster_size)
+        # If the last cluster is smaller than qcow2_cluster_size pad it with zeroes
+        if (len(cluster) < qcow2_cluster_size):
+            cluster += bytes(qcow2_cluster_size - len(cluster))
+        # If a cluster has non-zero data then it must be allocated
+        # in the output file and its L2 entry must be set
+        if cluster != zero_cluster:
+            bitmap_set(l2_bitmap, idx)
+            allocated_data_clusters += 1
+            # Allocated data clusters also need their corresponding L1 entry and L2 table
+            l1_idx = math.floor(idx / l2_entries_per_table)
+            if not bitmap_test(l1_bitmap, l1_idx):
+                bitmap_set(l1_bitmap, l1_idx)
+                allocated_l2_tables += 1
+
+# Total amount of allocated clusters excluding the refcount blocks and table
+total_allocated_clusters = 1 + allocated_l1_tables + allocated_l2_tables + allocated_data_clusters
+
+# Clusters allocated for the refcount blocks and table
+allocated_refcount_blocks = math.ceil(total_allocated_clusters  / refcounts_per_block)
+allocated_refcount_tables = math.ceil(allocated_refcount_blocks / refcounts_per_table)
+
+# Now we have a problem because allocated_refcount_blocks and allocated_refcount_tables...
+# (a) increase total_allocated_clusters, and
+# (b) need to be recalculated when total_allocated_clusters is increased
+# So we need to repeat the calculation as long as the numbers change
+while True:
+    new_total_allocated_clusters = total_allocated_clusters + allocated_refcount_tables + allocated_refcount_blocks
+    new_allocated_refcount_blocks = math.ceil(new_total_allocated_clusters / refcounts_per_block)
+    if new_allocated_refcount_blocks > allocated_refcount_blocks:
+        allocated_refcount_blocks = new_allocated_refcount_blocks
+        allocated_refcount_tables = math.ceil(allocated_refcount_blocks / refcounts_per_table)
+    else:
+        break
+
+# Now that we have the final numbers we can update total_allocated_clusters
+total_allocated_clusters += allocated_refcount_tables + allocated_refcount_blocks
+
+# At this point we have the exact number of clusters that the output
+# image is going to use so we can calculate all the offsets.
+current_cluster_idx = 1
+
+refcount_table_offset = current_cluster_idx * qcow2_cluster_size
+current_cluster_idx += allocated_refcount_tables
+
+refcount_block_offset = current_cluster_idx * qcow2_cluster_size
+current_cluster_idx += allocated_refcount_blocks
+
+l1_table_offset = current_cluster_idx * qcow2_cluster_size
+current_cluster_idx += allocated_l1_tables
+
+l2_table_offset = current_cluster_idx * qcow2_cluster_size
+current_cluster_idx += allocated_l2_tables
+
+data_clusters_offset = current_cluster_idx * qcow2_cluster_size
+
+# Calculate some values used in the qcow2 header
+if allocated_l1_tables == 0:
+    l1_table_offset = 0
+
+hdr_cluster_bits = int(math.log2(qcow2_cluster_size))
+hdr_refcount_bits = 0
+hdr_length = 0
+if qcow2_version == 3:
+    hdr_refcount_bits = int(math.log2(qcow2_refcount_bits))
+    hdr_length = 0x70
+
+### Write qcow2 header
+cluster = bytearray(qcow2_cluster_size)
+struct.pack_into('>4sIQIIQIIQQIIQQQQII', cluster, 0,
+     b'QFI\xfb',            # QCOW magic string
+     qcow2_version,
+     0,                     # backing file offset
+     0,                     # backing file sizes
+     hdr_cluster_bits,
+     disk_size,
+     0,                     # encryption method
+     l1_entries,
+     l1_table_offset,
+     refcount_table_offset,
+     allocated_refcount_tables,
+     0,                     # number of snapshots
+     0,                     # snapshot table offset
+     0,                     # compatible features
+     0,                     # incompatible features
+     0,                     # autoclear features
+     hdr_refcount_bits,
+     hdr_length)
+
+if qcow2_version == 3:
+    write_features(cluster)
+
+sys.stdout.buffer.write(cluster)
+
+### Write refcount table
+cur_offset = refcount_block_offset
+remaining_refcount_table_entries = allocated_refcount_blocks # Each entry is a pointer to a refcount block
+while remaining_refcount_table_entries > 0:
+    cluster = bytearray(qcow2_cluster_size)
+    to_write = min(remaining_refcount_table_entries, refcounts_per_table)
+    remaining_refcount_table_entries -= to_write
+    for idx in range(to_write):
+        struct.pack_into('>Q', cluster, idx * 8, cur_offset)
+        cur_offset += qcow2_cluster_size
+    sys.stdout.buffer.write(cluster)
+
+### Write refcount blocks
+remaining_refcount_block_entries = total_allocated_clusters # One entry for each allocated cluster
+for tbl in range(allocated_refcount_blocks):
+    cluster = bytearray(qcow2_cluster_size)
+    to_write = min(remaining_refcount_block_entries, refcounts_per_block)
+    remaining_refcount_block_entries -= to_write
+    # All refcount entries contain the number 1. The only difference
+    # is their bit width, defined when the image is created.
+    for idx in range(to_write):
+        if qcow2_refcount_bits == 64:
+            struct.pack_into('>Q', cluster, idx * 8, 1)
+        elif qcow2_refcount_bits == 32:
+            struct.pack_into('>L', cluster, idx * 4, 1)
+        elif qcow2_refcount_bits == 16:
+            struct.pack_into('>H', cluster, idx * 2, 1)
+        elif qcow2_refcount_bits == 8:
+            cluster[idx] = 1
+        elif qcow2_refcount_bits == 4:
+            cluster[int(idx / 2)] |= 1 << ((idx % 2) * 4)
+        elif qcow2_refcount_bits == 2:
+            cluster[int(idx / 4)] |= 1 << ((idx % 4) * 2)
+        elif qcow2_refcount_bits == 1:
+            cluster[int(idx / 8)] |= 1 << (idx % 8)
+    sys.stdout.buffer.write(cluster)
+
+### Write L1 table
+cur_offset = l2_table_offset
+for tbl in range(allocated_l1_tables):
+    cluster = bytearray(qcow2_cluster_size)
+    for idx in range(l1_entries_per_table):
+        l1_idx = tbl * l1_entries_per_table + idx
+        if bitmap_test(l1_bitmap, l1_idx):
+            struct.pack_into('>Q', cluster, idx * 8, cur_offset | QCOW_OFLAG_COPIED)
+            cur_offset += qcow2_cluster_size
+    sys.stdout.buffer.write(cluster)
+
+### Write L2 tables
+cur_offset = data_clusters_offset
+for tbl in range(l1_entries):
+    # Skip the empty L2 tables. We can identify them because
+    # there is no L1 entry pointing at them.
+    if bitmap_test(l1_bitmap, tbl):
+        cluster = bytearray(qcow2_cluster_size)
+        for idx in range(l2_entries_per_table):
+            l2_idx = tbl * l2_entries_per_table + idx
+            if bitmap_test(l2_bitmap, l2_idx):
+                struct.pack_into('>Q', cluster, idx * 8, cur_offset | QCOW_OFLAG_COPIED)
+                cur_offset += qcow2_cluster_size
+        sys.stdout.buffer.write(cluster)
+
+### Write data clusters
+with open(input_file, 'rb') as reader:
+    skip = 0
+    for idx in range(total_data_clusters):
+        if bitmap_test(l2_bitmap, idx):
+            if skip > 0:
+                reader.seek(qcow2_cluster_size * skip, 1)
+                skip = 0
+            cluster = reader.read(qcow2_cluster_size)
+            # If the last cluster is smaller than qcow2_cluster_size pad it with zeroes
+            if (len(cluster) < qcow2_cluster_size):
+                cluster += bytes(qcow2_cluster_size - len(cluster))
+            sys.stdout.buffer.write(cluster)
+        else:
+            skip += 1