Dump a Safe btrfs-image for Debugging

When a Btrfs filesystem is too fragile to hand over directly, btrfs-image captures its metadata into a file a developer can safely inspect. This guide covers dumping metadata to an image file, checking the result without treating it as a backup, and restoring to a disposable file when a damaged filesystem needs investigation. It uses btrfs-image from btrfs-progs 6.6.3, installed on Ubuntu as package version 6.6.3-1.1build2.

• Allow about fifteen minutes for the commands, plus however long the source filesystem takes to read.
• You need btrfs-progs, enough free space for the image, and a source device or filesystem image you can access.
• Privilege split. Dumping normally needs elevated privileges because the source is commonly a block device. Restoring always changes the target, so use a disposable target unless you are following a documented recovery plan.

1. Check the installed command

Start with the read-only help output. This also avoids assuming a newer online manual exactly matches the binary on your host.

$ command -v btrfs-image
/usr/bin/btrfs-image
$ dpkg-query -W -f='${Package} ${Version}\n' btrfs-progs
btrfs-progs 6.6.3-1.1build2
$ btrfs-image --help
usage: btrfs-image [options] source target

The installed help lists the short options -r, -c, -t, -o, -s, -w, -m and -d. The local btrfs-image(8) manpage documents the first seven, but not -d: that option is present in this installed 6.6.3 executable and means the dump also includes data. It conflicts with -w. Prefer the documented metadata-only mode unless a debugger explicitly needs file data.

2. Choose a source and destination

In dump mode, the first argument is the btrfs device or file and the second is the output image. Use a clear, absolute destination outside the filesystem being read.

SOURCE_DEVICE='/dev/REPLACE_WITH_BTRFS_DEVICE'
IMAGE_FILE='/var/tmp/REPLACE_WITH_CASE_NAME.btrfs-image'
sudo btrfs-image "$SOURCE_DEVICE" "$IMAGE_FILE"

Replace both placeholders before running the command. A path such as /dev/nvme0n1p2 must be the btrfs filesystem you intend to capture, not merely a disk that happens to contain one. Check the identity first.

$ lsblk -f
$ findmnt -t btrfs

Do not write the image into the source filesystem if that filesystem is short of space or unstable. If the source is mounted and actively changing, the captured metadata may describe different points in time. For a reproducible debugging capture, arrange a maintenance window or use an appropriate read-only source copy.

3. Create the metadata image

Run the plain two-argument form for the normal case.

$ sudo btrfs-image "$SOURCE_DEVICE" "$IMAGE_FILE"
$ printf 'image status: %s\n' "$?"
image status: 0
$ stat -c 'image: %n, bytes: %s' "$IMAGE_FILE"
image: /var/tmp/REPLACE_WITH_CASE_NAME.btrfs-image, bytes: 123456

The byte count is host-specific. A zero exit status means the command reported no error. This is a metadata image for debugging, not a restorable backup of user files: normal output has file data zeroed while metadata and related structures are retained. Protect it as case material anyway, because names and metadata can still disclose information.

4. Tune a dump only when the case needs it

Use -c to select a compression level from 0 to 9 and -t to select 1 to 32 processing threads.

$ sudo btrfs-image -c 6 -t 4 "$SOURCE_DEVICE" "$IMAGE_FILE"
$ printf 'image status: %s\n' "$?"
image status: 0

• These are trade-offs, not repair settings. A higher compression level can consume more CPU, while more threads can increase CPU and I/O pressure. Start with the defaults unless you have measured a reason to change them.
• Use -w only when the extent tree is corrupted and you need btrfs-image to walk all trees manually for referenced blocks. It is a recovery diagnostic, not a general quality improvement.
• Use -d sparingly: it also dumps data, conflicts with -w, and makes the result larger and less suitable for sharing.
• Do not use -s casually. It sanitises file names, changes directory-hash relationships, and can mask problems. A single -s uses garbage names; two attempt hash collisions and can be very CPU-intensive.

Keep the command line in the case notes so another operator can reproduce the capture.

5. Restore to a disposable file

Restore mode reverses the argument roles: the image is the source and the btrfs device or file is the target. Create a separate target file and never substitute a live disk path while experimenting.

RESTORE_TARGET='/var/tmp/REPLACE_WITH_DISPOSABLE_TARGET.img'
truncate -s 4G "$RESTORE_TARGET"
sudo btrfs-image -r "$IMAGE_FILE" "$RESTORE_TARGET"
printf 'restore status: %s\n' "$?"
ls -lh "$RESTORE_TARGET"

The target size must suit the image and its filesystem layout. The command may modify it even when a later mount or check fails, so treat the target as disposable until you have verified the result. If you need to abandon it, unmount it first if it was mounted, then remove that explicitly named test file.

$ sudo umount /mnt/REPLACE_WITH_TEST_MOUNT 2>/dev/null || true
$ rm -f -- "$RESTORE_TARGET"

6. Understand the restore default

With -r, the current restore method fixes the superblock chunk tree by using one stripe pointing at the primary device. This lets a restored filesystem be mounted, or its tree log replayed, when the metadata permits it. A successful restore still does not prove the filesystem is healthy.

-o selects the old restore method and does not fix up the chunk tree. The local manual warns that the resulting filesystem cannot be mounted; use it only when a specific investigation requires the old layout behaviour.

$ sudo btrfs-image -r -o "$IMAGE_FILE" "$RESTORE_TARGET"
$ printf 'old-method restore status: %s\n' "$?"
old-method restore status: 0

Do not combine this example with a target you expect to mount. If a restore status is non-zero, retain the diagnostic output and work from a new target rather than repeatedly overwriting the same test file.

7. Handle multi-device images deliberately

For a filesystem that uses more than one device, -m enables multi-device restore and more than one target device must be provided. Do not infer the device list from a single mount path: record the original device mapping and prepare replacement targets of the correct size before attempting recovery. A multi-device restore is a destructive storage operation and belongs in a maintenance window with a tested rollback plan.

For an ordinary single-device debugging image, leave -m out. If you are unsure whether the source is multi-device, inspect it before writing anything.

$ sudo btrfs filesystem show "$SOURCE_DEVICE"
$ sudo btrfs filesystem usage "$SOURCE_DEVICE"