Difference between revisions of "Btrfs"

From ArchWiki
Jump to navigation Jump to search
(→‎Compression: add link to package)
m (→‎Scrub: Fix typo)
Line 281: Line 281:
  
 
=== Scrub ===
 
=== Scrub ===
The [https://btrfs.wiki.kernel.org/index.php/Glossary Btrfs Wiki Glossary] says that Btrfs scrub is ''[a]n online filesystem checking tool. Reads all the data and metadata on the filesystem, and uses checksums and the duplicate copies from RAID storage to identify and repair any corrupt data.''
+
The [https://btrfs.wiki.kernel.org/index.php/Glossary Btrfs Wiki Glossary] says that Btrfs scrub is ''an online filesystem checking tool. Reads all the data and metadata on the filesystem, and uses checksums and the duplicate copies from RAID storage to identify and repair any corrupt data.''
  
 
  # btrfs scrub start /
 
  # btrfs scrub start /

Revision as of 10:41, 18 March 2016

zh-CN:Btrfs

From Wikipedia:Btrfs

Btrfs (B-tree file system, pronounced as "butter F S", "better F S", "b-tree F S", or simply by spelling it out) is a file system based on the copy-on-write (COW) principle, initially designed at Oracle Corporation for use in Linux. The development of Btrfs began in 2007, and by August 2014, the file system's on-disk format has been marked as stable.

From Btrfs Wiki

Btrfs is a new 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.
Warning:

Preparation

The official kernels linux and linux-lts include support for Btrfs. Additionally GRUB, mkinitcpio, and Syslinux have support for Btrfs and require no additional configuration.

User space utilities are available by installing the btrfs-progs package.

Partitioning

Tango-inaccurate.pngThe factual accuracy of this article or section is disputed.Tango-inaccurate.png

Reason: Grub#Install_to_partition_or_partitionless_disk warns against using it on a partitionless disk. (Discuss in Talk:Btrfs#)

Btrfs can occupy an entire data storage device, replacing the MBR or GPT partitioning schemes; however, this is not required to simply create a Btrfs filesystem on an existing partition that was created using another method. If using Btrfs for partitioning, one can use subvolumes to simulate partitions. There are some limitations to this approach in single disk setups:

To overwrite the existing partition table with Btrfs, run the following command:

# mkfs.btrfs /dev/sdX

For example, use /dev/sda rather than /dev/sda1. The latter would format an existing partition instead of replacing the entire partitioning scheme.

Install the boot loader like you would for a data storage device with a Master Boot Record. For example, in the case of GRUB:

# grub-install --recheck /dev/sdX
Warning: Using the btrfs subvolume set-default command to change the default sub-volume from anything other than the top level (ID 0) may break Grub. See #Setting a default sub-volume to reset.

File system creation

A Btrfs file system can either be newly created or have one converted.

Creating a new file system

File system on a single device

To format a partition do:

# mkfs.btrfs -L mylabel /dev/partition

The Btrfs default blocksize is 16KB. To use a larger blocksize for data/metadata, specify a value for the nodesize via the -n switch as shown in this example using 16KB blocks:

# mkfs.btrfs -L mylabel -n 16k /dev/partition

Multi-device file system

Multiple devices can be entered to create a RAID. Supported RAID levels include RAID 0, RAID 1, RAID 10, RAID 5 and RAID 6. The RAID levels can be configured separately for data and metadata using the -d and -m options respectively. By default the metadata is mirrored and data is striped. See Using Btrfs with Multiple Devices for more information about how to create a Btrfs RAID volume.

# mkfs.btrfs [options] /dev/part1 /dev/part2 ...
Note: Mounting such a filesystem may result in all but one of the according .device-jobs getting stuck and systemd never finishing startup due to a bug in handling this type of filesystem.

See RAID for advice on maintenance specific to multi-device Btrfs file systems.

Converting a file system from Ext3/4

Warning: As of mid-to-late 2015, there are many reports on the btrfs mailing list about incomplete/corrupt/broken conversions. The situation is improving as patches are being submitted, but proceed very carefully. Make sure you have working backups of any data you cannot afford to lose. See Conversion from Ext3 on the btrfs wiki.

Boot from an install CD, then convert by doing:

# btrfs-convert /dev/partition

Mount the partion and test the conversion by checking the files. Be sure to change the /etc/fstab to reflect the change (type to btrfs and fs_passno [the last field] to 0 as Btrfs does not do a file system check on boot). Also note that the UUID of the partition will have changed, so update fstab accordingly when using UUIDs. chroot into the system and rebuild the GRUB menu list (see Install from existing Linux and GRUB articles). If converting a root filesystem, while still chrooted run mkinitcpio -p linux to regenerate the initramfs or the system will not successfully boot. If you get stuck in grub with 'unknown filesystem' try reinstalling grub with grub-install /dev/partition and regenerate the config as well grub-mkconfig -o /boot/grub/grub.cfg.

After confirming that there are no problems, complete the conversion by deleting the backup ext2_saved sub-volume. Note that you cannot revert back to ext3/4 without it.

# btrfs subvolume delete /ext2_saved

Finally balance the file system to reclaim the space.

Configuring the file system

Mount options

Warning: Specific mount options can disable safety features and increase the risk of complete file system corruption.

See Btrfs Wiki Mount options and Btrfs Wiki Gotchas for more information.

In addition to configurations that can be made during or after file system creation, the various mount options for Btrfs can drastically change its performance characteristics.

As this is a file system that is still in active development, changes and regressions should be expected. See links in the #See also section for some benchmarks.

Examples

  • Linux 3.15
    • Btrfs on a SSD for system installation and an emphasis on maximizing performance (also read #SSD TRIM)
    noatime,discard,ssd,compress=lzo,space_cache
    • Btrfs on a HDD for archival purposes with an emphasis on maximizing space.
    noatime,autodefrag,compress-force=lzo,space_cache

Copy-On-Write (CoW)

By default, Btrfs performs copy-on-write for all files, at all times: If you write a file that did not exist before, then the data is written to empty space, and some of the metadata blocks that make up the filesystem are copied-on-write. In a traditional filesystem, if you then go back and overwrite a piece of that file, then the piece you are writing is put directly over the data it is replacing. In a CoW filesystem, the new data is written to a piece of free space on the disk, and only then is the file's metadata changed to refer to the new data. The old data that was replaced can then be freed up if nothing points to it any more.

Copy-on-write comes with some advantages, but can negatively affect performance with large files that have small random writes because it will fragment them (even if no "copy" is ever performed!). It is recommended to disable copy-on-write for database files and virtual machine images.

One can disable copy-on-write for the entire block device by mounting it with nodatacow option. However, this will disable copy-on-write for the entire file system.

Note: nodatacow will only affect newly created files. Copy-on-write may still happen for existing files.

To disable copy-on-write for single files/directories do:

$ chattr +C /dir/file

This will disable copy-on-write for those operation in which there is only one reference to the file. If there is more than one reference (e.g. through cp --reflink=always or because of a filesystem snapshot), copy-on-write still occurs.

Note: From chattr man page: For Btrfs, the 'C' flag should be set on new or empty files. If it is set on a file which already has data blocks, it is undefined when the blocks assigned to the file will be fully stable. If the 'C' flag is set on a directory, it will have no effect on the directory, but new files created in that directory will have the No_COW attribute.
Tip: In accordance with the note above, you can use the following trick to disable copy-on-write on existing files in a directory:
$ mv /path/to/dir /path/to/dir_old
$ mkdir /path/to/dir
$ chattr +C /path/to/dir
$ cp -a /path/to/dir_old/* /path/to/dir
$ rm -rf /path/to/dir_old

Make sure that the data are not used during this process. Also note that mv or cp --reflink as described below will not work.

Likewise, to save space by forcing copy-on-write when copying files use:

$ cp --reflink source dest 

As dest file is changed, only those blocks that are changed from source will be written to the disk. One might consider aliasing cp to cp --reflink=auto.

Compression

Btrfs supports transparent compression, meaning every file on the partition is automatically compressed. This not only reduces the size of files, but also improves performance, in particular if using the lzo algorithm, in some specific use cases (e.g. single thread with heavy file IO), while obviously harming performance on other cases (e.g. multithreaded and/or cpu intensive tasks with large file IO).

Compression is enabled using the compress=zlib or compress=lzo mount options. Only files created or modified after the mount option is added will be compressed. However, it can be applied quite easily to existing files (e.g. after a conversion from ext3/4) using the btrfs filesystem defragment -calg command, where alg is either zlib or lzo. In order to re-compress the whole file system with lzo, run the following command:

# btrfs filesystem defragment -r -v -clzo /
Tip: Compression can also be enabled per-file without using the compress mount option; simply apply chattr +c to the file. When applied to directories, it will cause new files to be automatically compressed as they come.

When installing Arch to an empty Btrfs partition, use the compress option when mounting the filesystem: mount -o compress=lzo /dev/sdxY /mnt/. During configuration, add compress=lzo to the mount options of the root file system in fstab.

Subvolumes

A btrfs subvolume is not a block device (and cannot be treated as one) instead, a btrfs subvolume can be thought of as a POSIX file namespace. This namespace can be accessed via the top-level subvolume of the filesystem, or it can be mounted in its own right.[1]

Each Btrfs file system has a top-level subvolume with ID 5. It can be mounted itself as "/" (by default), or another subvolume can be mounted.

See the following links for more details:

Creating a subvolume

To create a subvolume:

# btrfs subvolume create /path/to/subvolume

Listing subvolumes

To see a list of current subvolumes:

# btrfs subvolume list -p .

Deleting a subvolume

To delete a subvolume:

# btrfs subvolume delete /path/to/subvolume

Attempting to remove the directory /path/to/subvolume without using the above command will not delete the subvolume.

Mounting subvolumes

Subvolumes can be mounted like file system partitions using the subvol=/path/to/subvolume or subvolid=objectid mount flags. For example, you could have a subvolume named 'subvol_root' and mount it as '/'. One can mimic traditional file system partitions by creating various subvolumes under the top level of the file system and then mounting them at the appropriate mount points. Thus one can easily restore a file system (or part of it) to a previous state easily using #Snapshots.

Tip: Changing subvolume layouts is made simpler by not using the toplevel subvolume (ID=5) as "/" (which is done by default). Instead, consider creating a subvolume to house your actual data and mounting it as "/".

See Snapper#Suggested_Filesystem_Layout, Btrfs SysadminGuide#Managing Snapshots, and Btrfs SysadminGuide#Layout for example file system layouts using subvolumes.

Changing the default sub-volume

Warning: Changing the default subvolume with btrfs subvolume set-default will make the top level of the filesystem inaccessible, except by use of the subvolid=5 mount option. Reference: Btrfs Wiki Sysadmin Guide.

The default sub-volume is mounted if no subvol= mount option is provided.

# btrfs subvolume set-default subvolume-id /.

Example:

# btrfs subvolume list .
ID 258 gen 9512 top level 5 path root_subvolume
ID 259 gen 9512 top level 258 path home
ID 260 gen 9512 top level 258 path var
ID 261 gen 9512 top level 258 path usr
# btrfs subvolume set-default 258 .

Reset:

# btrfs subvolume set-default 5 .

Commit interval settings

The resolution at which data are written to the filesystem is dictated by Btrfs itself and by system-wide settings. Btrfs defaults to a 30 seconds checkpoint interval in which new data are committed to the filesystem. This is tuneable using mount options (see below)

System-wide settings also affect commit intervals. They include the files under /proc/sys/vm/* and are out-of-scope of this wiki article. The kernel documentation for them resides in Documentation/sysctl/vm.txt.

Checkpoint interval

Starting with Linux 3.12, users are able to change the checkpoint interval from the default 30 s to any value by appending the commit mount option in /etc/fstab for the btrfs partition.

LABEL=arch64 / btrfs defaults,noatime,ssd,compress=lzo,commit=120 0 0

SSD TRIM

A Btrfs filesystem will automatically free unused blocks from an SSD drive supporting the TRIM command.

More information about enabling and using TRIM can be found in Solid State Drives#TRIM.

Usage

Displaying used/free space

General linux userspace tools such as /usr/bin/df will inaccurately report free space on a Btrfs partition since it does not take into account space allocated for and used by the metadata. It is recommended to use /usr/bin/btrfs to query a Btrfs partition. Below is an illustration of this effect, first querying using df -h, and then using btrfs filesystem df:

$ df -h /
Filesystem      Size  Used Avail Use% Mounted on
/dev/sda3       119G  3.0G  116G   3% /
$ btrfs filesystem df /
Data: total=3.01GB, used=2.73GB
System: total=4.00MB, used=16.00KB
Metadata: total=1.01GB, used=181.83MB

Notice that df -h reports 3.0GB used but btrfs filesystem df reports 2.73GB for the data. This is due to the way Btrfs allocates space into the pool. The true disk usage is the sum of all three 'used' values which is inferior to 3.0GB as reported by df -h.

Note: If you see an entry of type unknown in the output of btrfs filesystem df at kernel >= 3.15, this is a display bug. As of this patch, the entry means GlobalReserve, which is kind of a buffer for changes not yet flushed. This entry is displayed as unknown, single in RAID setups and is not possible to re-balance.

Another useful command to show a less verbose readout of used space is btrfs filesystem show:

# btrfs filesystem show /dev/sda3

The newest command to get information on free/used space of a is btrfs filesystem usage:

# btrfs filesystem usage
Note: The btrfs filesystem usage command does not currently work correctly with RAID5/RAID6 RAID levels.

Defragmentation

Btrfs supports online defragmentation. To defragment the metadata of the root folder:

# btrfs filesystem defragment /

This will not defragment the entire file system. For more information read this page on the Btrfs wiki.

To defragment the entire file system verbosely:

# btrfs filesystem defragment -r -v /

Scrub

The Btrfs Wiki Glossary says that Btrfs scrub is an online filesystem checking tool. Reads all the data and metadata on the filesystem, and uses checksums and the duplicate copies from RAID storage to identify and repair any corrupt data.

# btrfs scrub start /
# btrfs scrub status /
Warning: The running scrub process will prevent the system from suspending, see this thread for details.

The btrfs-progs package brings the btrfs-scrub@.timer unit for monthly scrubbing the specified mountpoint. Enable the timer with an encoded path, e.g. btrfs-scrub@-.timer for / and btrfs-scrub@home.timer for /home.

You can also run the scrub manually by starting btrfs-scrub@.service (with the same encoded path). The advantage of this over # btrfs scrub is that the results of the scrub will be logged in the systemd journal.

Balance

A balance passes all data in the filesystem through the allocator again. It is primarily intended to rebalance the data in the filesystem across the devices when a device is added or removed. A balance will regenerate missing copies for the redundant RAID levels, if a device has failed. [2] See Upstream FAQ page.

# btrfs balance start /
# btrfs balance status /

RAID

Btrfs works in block-pairs for raid0, raid1, and raid10. This means:

raid0 - block-pair striped across 2 devices

raid1 - block-pair written to 2 devices

The RAID level can be changed while the disks are online using the btrfs balance command:

# btrfs balance start -mconvert=RAIDLEVEL -dconvert=RAIDLEVEL /path/to/mount

For 2 disk sets, this matches raid levels as defined in md-raid (mdadm). For 3+ disk-sets, the result is entirely different than md-raid.

For example:

  • Three 1TB disks in an md based raid1 yields a /dev/md0 with 1TB free space and the ability to safely lose 2 disks without losing data.
  • Three 1TB disks in a Btrfs volume with data=raid1 will allow the storage of approximately 1.5TB of data before reporting full. Only 1 disk can safely be lost without losing data.

Btrfs uses a round-robin scheme to decide how block-pairs are spread among disks. As of Linux 3.0, a quasi-round-robin scheme is used which prefers larger disks when distributing block pairs. This allows raid0 and raid1 to take advantage of most (and sometimes all) space in a disk set made of multiple disks. For example, a set consisting of a 1TB disk and 2 500GB disks with data=raid1 will place a copy of every block on the 1TB disk and alternate (round-robin) placing blocks on each of the 500GB disks. Full space utilization will be made. A set made from a 1TB disk, a 750GB disk, and a 500GB disk will work the same, but the filesystem will report full with 250GB unusable on the 750GB disk. To always take advantage of the full space (even in the last example), use data=single. (data=single is akin to JBOD defined by some raid controllers) See the Btrfs FAQ for more info.

Snapshots

A snapshot is simply a subvolume that shares its data (and metadata) with some other subvolume, using btrfs's COW capabilities. See Btrfs Wiki SysadminGuide#Snapshots for details.

To create a snapshot:

# btrfs subvolume snapshot source [dest/]name

To create a readonly snapshot add the -r flag. To create writable version of a readonly snapshot, simply create a snapshot of it.

Note: Snapshots are not recursive. Every sub-volume inside sub-volume will be an empty directory inside the snapshot.

Send/receive

A subvolume can be sent to stdout or a file using the send command. This is usually most useful when piped to a Btrfs receive command. For example, to send a snapshot named /root_backup (perhaps of a snapshot you made of / earlier) to /backup you would do the following:

 # btrfs send /root_backup | btrfs receive /backup

The snapshot that is sent must be readonly. The above command is useful for copying a subvolume to an external device (e.g., a USB disk mounted at /backup above).

You can also send only the difference between two snapshots. For example, let's say you've already sent a copy of root_backup above and have made a new readonly snapshot on your system named root_backup_new. To send only the incremental difference to /backup do:

 # btrfs send -p /root_backup /root_backup_new | btrfs send /backup

Now a new subvolume named root_backup_new will be present in /backup.

See Btrfs Wiki's Incremental Backup page on how to use this for an incremental backup and for tools that automate the process.

Limitations

A few limitations should be known before trying.

Encryption

Btrfs has no built-in encryption support, but this may come in future. Users can encrypt the partition before running mkfs.btrfs. See dm-crypt.

Existing Btrfs file systems can use something like EncFS or TrueCrypt, though perhaps without some of Btrfs' features.

Swap file

Btrfs does not yet support swap files. This is due to swap files requiring a function that Btrfs doesn't have for possibility of file system corruption [3]. Patches for swapfile support are already available [4] and may be included in an upcoming kernel release. As an alternative a swap file can be mounted on a loop device with poorer performance but will not be able to hibernate. Install the package systemd-swap from the official repositories to automate this.

Linux-rt kernel

As of version 3.14.12_rt9, the linux-rt kernel does not boot with the Btrfs file system. This is due to the slow development of the rt patchset.

Tips and tricks

See Btrfs - Tips and tricks.

Troubleshooting

See the Btrfs Problem FAQ for general troubleshooting.

GRUB

Partition offset

Note: The offset problem may happen when you try to embed core.img into a partitioned disk. It means that it is OK to embed grub's corg.img into a Btrfs pool on a partitionless disk (e.g. /dev/sdX) directly.

GRUB can boot Btrfs partitions however the module may be larger than other file systems. And the core.img file made by grub-install may not fit in the first 63 sectors (31.5KiB) of the drive between the MBR and the first partition. Up-to-date partitioning tools such as fdisk and gdisk avoid this issue by offsetting the first partition by roughly 1MiB or 2MiB.

Missing root

Users experiencing the following: error no such device: root when booting from a RAID style setup then edit /usr/share/grub/grub-mkconfig_lib and remove both quotes from the line echo " search --no-floppy --fs-uuid --set=root ${hints} ${fs_uuid}". Regenerate the config for grub and the system should boot without an error.

BTRFS: open_ctree failed

As of November 2014 there seems to be a bug in systemd or mkinitcpio causing the following error on systems with multi-device Btrfs filesystem using the btrfs hook in mkinitcpio.conf:

BTRFS: open_ctree failed
mount: wrong fs type, bad option, bad superblock on /dev/sdb2, missing codepage or helper program, or other error

In some cases useful info is found in syslog - try dmesg|tail or so.

You are now being dropped into an emergency shell.

A workaround is to remove btrfs from the HOOKS array in /etc/mkinitcpio.conf and instead add btrfs to the MODULES array. Then regenerate the initramfs with mkinitcpio -p linux (adjust the preset if needed) and reboot.

See the original forums thread and FS#42884 for further information and discussion.

You will get the same error if you try to mount a raid array without one of the devices. In that case you must add the degraded mount option to /etc/fstab. If your root resides on the array, you must also add rootflags=degraded to your kernel parameters.

btrfs check

Warning: Since Btrfs is under heavy development, especially the btrfs check command, it is highly recommended to create a backup and consult the following Btfrs documentation before executing btrfs check with the --repair switch.

The btrfs check command can be used to check or repair an unmounted Btrfs filesystem. However, this repair tool is still immature and not able to repair certain filesystem errors even those that do not render the filesystem unmountable.

See Btrfsck for more information.

See also