Btrfs

Resources

Home

Userspace

Kernel

Wikipedia

Open Hub

Btrfs is a copy-on-write (CoW) filesystem for Linux aimed at implementing advanced features while focusing on fault tolerance, repair, and easy administration. Jointly developed at Oracle, Red Hat, Fujitsu, Intel, SUSE, STRATO, and many others. Btrfs is licensed under the GPL and open for contribution from anyone.

特徴

Ext4 が安全確実で、広い領域をもつ巨大なファイルシステムも扱えるのにもかかわらず、なぜ Btrfs に切り替えるのでしょうか。たしかに Btrfs は未だ試験的なものと思われており、安定性の向上途上です。しかし、Linux システムのデフォルトのファイルシステムとなる時が近づきつつあります。Linux のディストリビューションのなかには、最新のリリースにおいて既に Btrfs に切り替え始めたものもあります。Btrfs は、ZFS にも備わっている一般的な先進機能に対応しています。こうした機能は、 BSD 系ディストリビューションや NAS 機器において ZFS の人気が高くなっている理由でもあります。

• Copy on Write (CoW) and snapshotting - Make incremental backups painless even from a "hot" filesystem or virtual machine (VM).
• File level checksums - Metadata for each file includes a checksum that is used to detect and repair errors.
• Compression - Files may be compressed and decompressed on the fly, which speeds up read performance.
• Auto defragmentation - The filesystems are tuned by a background thread while they are in use.
• Subvolumes - Filesystems can share a single pool of space instead of being put into their own partitions.
• RAID - Btrfs does its own RAID implementations so LVM or mdadm are not required in to have RAID. Currently RAID 0 and 1 are supported; RAID 5 and 6 are upcoming.
• Partitions are optional - While Btrfs can work with partitions, it has the potential to use raw devices (/dev/<device>) directly.
• Data deduplication - There is limited data deduplication support; however, deduplication will eventually become a standard feature in Btrfs. This enables Btrfs to save space by comparing files via binary diffs.

Tip
For an up-to-date and somewhat exhaustive listing of features see the upstream wiki's status page. Not all features are sufficiently mature for wide use though.

Down the road, new clustered filesystems will readily take advantage of Btrfs with its copy on write and other advanced features for their object stores. Ceph is one example of a clustered filesystem that looks very promising, and can take advantage of Btrfs.

インストール

カーネル

Btrfs に対応するには、つぎのカーネルオプションを有効にしてください:

KERNEL カーネルで Btrfs を有効にする

File systems  --->
    <*> Btrfs filesystem

Emerge

sys-fs/btrfs-progs パッケージには、 Btrfs を扱うために必要なユーティリティが入っています。

root #emerge --ask sys-fs/btrfs-progs

使い方

Typing long Btrfs commands can quickly become a hassle. Each command (besides the initial btrfs command) can be reduced to a very short set of instructions. This method is helpful when working from the command line to reduce the amount of characters typed.

For example, to defragment a filesystem located at /, the following shows the long command:

root #btrfs filesystem defragment -v /

Shorten each of the longer commands after the btrfs command by reducing them to their unique, shortest prefix. In this context, unique means that no other btrfs commands will match the command at the command's shortest length. The shortened version of the above command is:

root #btrfs fi de -v /

No other btrfs commands start with fi; filesystem is the only one. The same goes for the de sub-command under the filesystem command.

作成

Warning
The mkfs.btrfs command irreversibly destroys any content of the partition it is told to format. Please be sure the correct drive and partition have been selected before running any mkfs command!

/dev/sdXN パーティションを Btrfs で作成するには:

root #mkfs.btrfs /dev/sdXN

In the example above, replace N with the partition number and X with the disk letter that is to be formatted. For example, to format the third partition of the first drive in the system with Btrfs, run:

root #mkfs.btrfs /dev/sda3

Important
The last number column in /etc/fstab should be 0 for all Btrfs partitions. fsck.btrfs and btrfsck should not be run during each system boot.

マウント

ファイルシステムを作成したら、さまざまな方法でマウント可能です:

• mount - Manual mount.
• fstab - Defining mount points in /etc/fstab enables automatic mounts on system boot.
• Removable media - Automatic mounts on demand (useful for USB drives).
• AutoFS - Automatic mount on filesystem access.

Converting ext* based file systems

It is possible to convert ext2, ext3, and ext4 filesystems to Btrfs using the btrfs-convert utility.

The following instructions only supports the conversion of filesystems that are unmounted. To convert the root partition, boot to a system rescue disk (SystemRescueCD works nicely) and run on the conversion commands on the root partition.

First be sure the the mount point is unmounted:

root #umount <mounted_device>

Check the integrity of the filesystem using the appropriate fsck tool. In the next example, the filesystem is ext4:

root #fsck.ext4 -f <unmounted_device>

Use btrfs-convert to convert the ext* formatted device into a Btrfs-formatted device:

root #btrfs-convert <unmounted_device>

Be sure to edit /etc/fstab after the device has been formatted to change the filesystem column from ext4 to Btrfs:

FILE /etc/fstabChanging ext4 to btrfs

<device>   <mountpoint>  btrfs  defaults  0 0

デフラグ

Another feature of Btrfs is online defragmentation. To defragment a root Btrfs filesystem run:

root #btrfs filesystem defragment -r -v /

Warning
Defragmenting with kernel versions < 3.9 or ≥ 3.14-rc2 as well as with Linux stable kernel versions ≥ 3.10.31, ≥ 3.12.12 or ≥ 3.13.4 breaks up ref-links between files and their COW copies and thus may increase space usage considerably. Make sure to have enough free space available and not too many snapshots on the drive as full btrfs partitions can get really slow.

圧縮

Btrfs supports transparent compression using the zlib, lzo, and zstd compression algorithms.

ファイルの属性を設定することで特定のファイルのみ圧縮することも可能です:

user $chattr +c

The compress mount option sets the default behavior to compress all the newly created files. To re-compress the whole filesystem, run the following command:

root #btrfs filesystem defragment -r -v -clzo /

Depending on the CPU and disk performance, using lzo compression could improve the overall throughput.

It is possible to use the zlib compression algorithm instead of lzo. Zlib is slower but has a higher compression ratio:

root #btrfs filesystem defragment -r -v -czlib /

Compression level

Since kernel version 4.15.0, zlib compression can now be set by levels 1-9. For example, to set zlib to maximum compression at mount time:

root #mount -o compress=zlib:9 /dev/sdXY /path/to/btrfs/mountpoint

Or to set minimal compression:

root #mount -o compress=zlib:1 /dev/sdXY /path/to/btrfs/mountpoint

Or adjust compression by remounting:

root #mount -o remount,compress=zlib:3 /path/to/btrfs/mountpoint

The compression level should be visible in /proc/mounts or by checking the most recent output of dmesg |grep -i btrfs.

root #dmesg |grep -i btrfs

[    0.495284] Btrfs loaded, crc32c=crc32c-intel
[ 3010.727383] BTRFS: device label My Passport devid 1 transid 31 /dev/sdd1
[ 3111.930960] BTRFS info (device sdd1): disk space caching is enabled
[ 3111.930973] BTRFS info (device sdd1): has skinny extents
[ 9428.918325] BTRFS info (device sdd1): use zlib compression, level 3

Compression ratio and disk usage

The usual userspace tools for determining used and free space like du and df may provide inaccurate results on a Btrfs partition due to inherent design differences in the way files are written compared to ,for example, ext2/3/4.

It is therefore advised to use the du/df alternatives provided by the btrfs userspace tool btrfs filesystem. In Addition to that, The compsize tool found from the sys-fs/compsize package can be helpful in providing additional information regarding compression ratios and the disk usage of compressed files. The following are example uses of these tools for a btrfs partition mounted under /media/drive.

user $btrfs filesystem du -s /media/drive

     Total   Exclusive  Set shared  Filename
 848.12GiB   848.12GiB       0.00B  /media/drive/

user $btrfs filesystem df /media/drive

Data, single: total=846.00GiB, used=845.61GiB
System, DUP: total=8.00MiB, used=112.00KiB
Metadata, DUP: total=2.00GiB, used=904.30MiB
GlobalReserve, single: total=512.00MiB, used=0.00B

user $compsize /media/drive

Processed 2262 files, 112115 regular extents (112115 refs), 174 inline.
Type       Perc     Disk Usage   Uncompressed Referenced  
TOTAL       99%      845G         848G         848G       
none       100%      844G         844G         844G       
zlib        16%      532M         3.2G         3.2G 

RAID

Btrfs can be used with multiple devices in order to create RAIDs. Using Btrfs to create filesystems that span multiple devices is much easier than creating using mdadm; there is no initialization time needed for creation.

デバイス全体に対して RAID を形成するのに最も簡単な方法は:

root #mkfs.btrfs -m raid1 <device1> <device2> -d raid1 <device1> <device2>

Converting between RAID modes is possible with the balance sub-command. For example, say a multiple device RAID 1 is mounted at /srv. It is possible to convert this RAID1 to RAID0 with using the following command:

root #btrfs balance start -dconvert=raid0 -mconvert=raid0 --force /srv

RAID mode conversion can be performed while the filesystem is online and in use. Possible RAID modes in btrfs include RAID0, RAID1, RAID5, RAID6, and RAID10. See the upstream BTRFS wiki for more information.

Warning
It is currently not safe to use the RAID 5 or 6 modes. RAID 5 and 6 modes have seen some fixes in Linux 4.12, but overall status is still marked as unstable.. Users who want to use RAID5 or RAID6 functionality of btrfs are encouraged to check the btrfs status page for stability status of said modes before utilizing the modes.

サブボリューム

As mentioned above in the features list, Btrfs can create subvolumes. Subvolumes can be used to better organize and manage data. They become especially powerful when combined with snapshots. Important distinctions must be made between Btrfs subvolumes and subvolumes created by Logical Volume Management (LVM). Btrfs subvolumes are not block level devices, they are POSIX file namespaces. They can be created at any location in the filesystem and will act like any other directory on the system with one caveat: subvolumes can be mounted and unmounted. Subvolumes are nestable (subvolumes can be created inside other subvolumes), and easily created or removed.

Note
A subvolume cannot be created across different Btrfs filesystems. If /dev/sda and /dev/sdb both contain separate (non-RAID) Btrfs filesystems, there is no way a subvolume can expand across the two filesystems. The snapshot can be moved from one filesystem to another, but it cannot span across the two. It must be on /dev/sda or /dev/sdb.

作成

To create a subvolume, issue the following command inside a Btrfs filesystem's name space:

root #btrfs subvolume create <dest-name>

Replace <dest-name> with the desired destination and subvolume name. For example, if a Btrfs filesystem exists at /mnt/btrfs, a subvolume could be created inside it using the following command:

root #btrfs subvolume create /mnt/btrfs/subvolume1

一覧

To see the subvolume(s) that have been created, use the subvolume list command followed by a Btrfs filesystem location. If the current directory is somewhere inside a Btrfs filesystem, the following command will display the subvolume(s) that exist on the filesystem:

root #btrfs subvolume list .

If a Btrfs filesystem with subvolumes exists at the mount point created in the example command above, the output from the list command will look similar to the following:

root #btrfs subvolume list /mnt/btrfs

ID 309 gen 102913 top level 5 path mnt/btrfs/subvolume1

削除

Subvolumes can be properly removed by using the subvolume delete command followed by the path to the subvolume. All available subvolume paths in a Btrfs filesystem can be seen using the list command above.

root #btrfs subvolume delete <subvolume-path>

As above, replace <subvolume-path> with the actual path to the subvolume to be removed. To delete the subvolume used in the examples above, the following command would be issued:

root #btrfs subvolume delete /mnt/btrfs/subvolume1

Delete subvolume (no-commit): '/mnt/btrfs/subvolume1'

スナップショット

Snapshots are subvolumes that share data and metadata with other subvolumes. This is made possible by Btrfs' Copy on Write (CoW) ability. Snapshots can be used for several purposes, one of which is to create backups of file system structures at specific points in time.

ルートのファイルシステムが Btrfs であるときには、 subvolume snapshot コマンドを用いてスナップショットの作成が可能です:

root #mkdir -p /mnt/backup/rootfs

root #btrfs subvolume snapshot / /mnt/backup/rootfs/

The following small shell script can be added to a timed cron job to create a timestamped snapshot backup of a Btrfs formatted root filesystem. The timestamps can be adjusted to whatever is preferred by the user.

FILE btrfs_snapshot.shBtrfs rootfs snapshot cron job example

#!/bin/bash
NOW=$(date +"%Y-%m-%d_%H:%M:%S")
 
if [ ! -e /mnt/backup ]; then
mkdir -p /mnt/backup
fi
 
cd /
/sbin/btrfs subvolume snapshot / "/mnt/backup/backup_${NOW}"

Mounting

A subvolume can be mounted in a location different from where it was created, or users can choose to not mount them at all. For example, a user could create a Btrfs filesystem in /mnt/btrfs and create /mnt/btrfs/home and /mnt/btrfs/portage subvolumes. The subvolumes could then be mounted at /home and /usr/portage, with the original top level subvolume left unmounted. This results in a configuration where the subvolumes' relative path from the top level subvolume is different from their actual path.

To mount a subvolume, perform the following command, where <rel-path> is the relative path of the subvolume from the top level subvolume, obtainable through the subvolume list command:

root #mount -o subvol=<rel-path> <device> <mountpoint>

Similarly, one can update the filesystem tab to mount their Btrfs subvolumes like so:

FILE /etc/fstabMounting Subvolumes

<device>  <mountpoint>  btrfs  subvol=<rel-path>  0 2

トラブルシューティング

Using with VM disk images

When using Btrfs with virtual machine disk images, it is best to disable copy-on-write on the disk images in order to speed up IO performance. This can only be performed on files that are newly created. It also possible to disable CoW on all files created within a certain directory. For example, using the chattr command:

root #chattr +C /var/lib/libvirt/images

Clear the free space cache

clear_cache マウントオプションを指定してファイルシステムをマウントすることで、Btrfs 内の空きのキャッシュ領域を開放することが可能です。例えば:

root #mount -o clear_cache /path/to/device /path/to/mountpoint

Btrfs hogging memory (disk cache)

When utilizing some of Btrfs' special abilities (like making many --reflink copies or creating a crazy amount of snapshots), lot of memory can be eaten and not freed fast enough by the kernel's inode cache. This issue can go undiscovered since memory dedicated to the disk cache might not be clearly visible in traditional system monitoring utilities. The slabtop utility (available as part of the sys-process/procps package) was specifically created to determine how much memory kernel objects are consuming:

root #slabtop

If the inode cache is consuming too much memory, the kernel can be manually instructed to drop the cache by echoing an integer value to the /proc/sys/vm/drop_caches file.

To be safe, and to help the kernel determine the maximum amount of freeable memory, be sure to run a sync before running the echo commands below:

user $sync

Most of the time Btrfs users will probably want to echo 2 to reclaim just the slab objects (dentries and btrfs_inodes):

root #echo 2 > /proc/sys/vm/drop_caches

To clear the entire disk cache (slab objects and the page cache) use echo 3 instead:

root #echo 3 > /proc/sys/vm/drop_caches

Warning
While the above commands are non-destructive (as long as a sync was completed before running them), they could seriously but temporarily slow down the system while the kernel loads only the necessary items back into memory. Think twice before running the above commands for systems under heavy load!

More information on kernel slabs can be found in this dedoimedo blog entry.

Mounting Btrfs fails, returning mount: unknown filesystem type 'btrfs'

The original solution by Tim on Stack Exchange inspired the following solution: build the kernel manually instead of using genkernel:

#cd /usr/src/linux

#make menuconfig

#make && make modules_install

#cp arch/x86_64/boot/bzImage /boot

#mv /boot/bzImage /boot/whatever_kernel_filename

#genkernel --install initramfs

Btrfs root doesn't boot

Genkernel's initramfs as created with the command below doesn't load btrfs:

root #genkernel --btrfs initramfs

Compile support for btrfs in the kernel rather than as a module, or use genkernel-next or Dracut to generate the initramfs.

参考

• Btrfs snapshots - Script that creates snapshots when files have changed
• Btrfs/System Root Guide - Use the Btrfs filesystem as a collection of subvolumes including one as a system root.
• Btrfs native system root guide - An alternative guide on using a subvolume in a Btrfs filesystem as the system's root.
• ext4 - The default filesystem for most Linux distributions.
• Samba shadow copies - Using Samba to expose Shadow Copies as 'Previous Versions' to Windows clients.
• Snapper - A command-line program capable of managing Btrfs filesystem snapshots.
• ZFS - A filesystem that shares much in common with Btrfs, but has licensing issues.

外部の情報

• https://wiki.debian.org/Btrfs - As described by the Debian wiki.
• https://wiki.archlinux.org/index.php/Btrfs Btrfs article - As described by the Arch Linux wiki.
• http://www.funtoo.org/BTRFS_Fun - BTRFS Fun on the Funtoo wiki.
• http://marc.merlins.org/perso/btrfs/post_2014-05-04_Fixing-Btrfs-Filesystem-Full-Problems.html - Tips and tricks on fixing niche Btrfs filesystem problems in some situations.