Btrfs

Revision as of 12:47, 16 January 2022 by Andreas85 (talk | contribs) (Upercase title)


Page Under Construction.jpg This article or section is under construction
Reason: Rome wasn't built in a day (Discuss)
Page Under Construction.jpg

Btrfs is a modern CoW file system

A modern Copy on Write file system for Linux aimed at implementing advanced features while also focusing on fault tolerance, repair and easy administration. Btrfs not only is a file system, but also is partly a volume manager, software-raid, backup-tool, and it is flash-friendly.

Because Btrfs is different, some things seem unfamiliar and strange. If you want to learn the details and the newest stuff, then btrfs.wiki.kernel.org is the place to go or btrfs.readthedocs.io. Development of Btrfs started in 2007. Since that time, Btrfs is a part of the Linux kernel and is under active development. The Btrfs code base and filesystem-layout is stable . However, new features are still under development. Its main features and benefits are:

  • Snapshots which do not make the full copy of files
  • Volume Manager join partitions, split into subvolumes
  • RAID - support for software-based RAID 0, RAID 1, RAID 10
  • Auto-repair - checksums for data and metadata, automatic detection of silent data corruption

(see btrfs@kernel.org, btrfs.readthedocs.io, Btrfs@ARC-wiki, Btrfs@wikipedia)

Familiar with btrfs-slang ?

Because Btrfs is different, you will find some words that do have a special meaning when used for btrfs. This may be a source of confusion.

▶ Btrfs volume
A volume is a pool of raw storage and consists of one or more devices. The size of the volume will be the addition of all devices that are part of this volume. In most cases you will only use one volume. You are able to add/remove devices at any time. Usually you do not mount a Btrfs volume.
chunk
A chunk is simply a piece of storage that Btrfs can use to put data on. Think of a chunk (usually 1GiB) as of a page in a book. The book is the volume, and the chunk is one page of it. When you start, all pages are empty. When you write data to the volume, one page (="chunk") after the other is written to.
device
A device is some linux device. It may be a partition like /dev/sdz1 or /dev/sdz2. Or it may be a raw disk device like /dev/sdz without any partitioning. A Btrfs volume consists of at least one device.
subvolume
A Btrfs subvolume is an independently mountable POSIX file-tree and not a block device. It is the part of a volume that will be mounted writeable into your Linux system. By convention the names of subvolumes start with @ (@, @home, @snapshots ...). All subvolumes share the space of the Btrfs volume. You may create subvolumes at will. (You may think of subvolumes as sort of "dynamic partitions" inside a Btrfs volume)
default subvolume
The default subvolume of a Btrfs volume is special. When you mount, you normally have to name a subvolume to mount. When you don't name a subvolume, the default subvolume will be used. The default subvolume can be changed to any subvolume. It is advisable to set that subvolume as default, that is used for mounting linux "/" this is often the subvolume with the name "@"
▶ Btrfs volume-root "/", Btrfs layout
A volume contains one ore more subvolumes. But they are not stored in form of a simple list. These subvolumes are stored in a tree-like structure like in a filesystem. Sometimes called the "top-level subvolume" or "root of the volume". But be careful this is not linux-root "/", but Btrfs volume-root "/". There are several basic schemas to layout subvolumes in a volume
snapshot
A snapshot looks nearly the same as a subvolumes. But don´t get confused. When we talk about snapshots we usually mean a "Read-Only (ro) photograph of a subvolume". While the subvolume changes with time. A snapshot stays in the state of the subvolume at the time we made it. You can mount snapshots into your linux system, but you only can read the content. And the content will never change while this snapshot exists. When creating snapshots you have to watch out for the Btrfs-layout in use.

It is possible to make a writeable(rw) subvolume out of a ro-snapshot. This is the way roll back does work.

▶ Self-healing
This is no magic. Because of the nature of Btrfs as CopyOnWrite filesystem and because of the checksums, it is possible to check the filesystem and repair some errors. This does happen silently.
  • Without RAID it is possible to correct some small faults that happen because of power outage. (This is done when the filesystem is mounted again)
  • With RAID1 it is possible to repair some parts of files that where damaged by faults on one device. (This is done when the file is read the next time)
scrub
A scrub is like an inspection of your car. The mechanic will look at all parts, and will tell you if something is amiss. If he finds very small problems, he will repair this automatically without asking for permission, and with minimal extra cost.
balance
A balance is like renovating your home. Sometimes it is necessary to renovate a room, sometimes you will renovate your home completely. But you don't do it once in every week ;-) With balance, the furniture will be transported around and rearranged. You need to do this when changing RAID-levels.

Volume management

A volume is a pool of raw storage. Consists of one or more devices. The size of the volume will be the addition of all included devices, unless you use RAID.

If you do use more then one device, please also read the section about RAID. You are able to add/remove devices at any time to increase/decrease the size of the volume. With adding/removing devices it is also possible to move a volume from one device to another (without changing the UUID).

Usually you do not mount the Btrfs volume itself, but you mount subvolumes. There may be times when it is practical to mount the Btrfs volume-root itself. Then you are able to change the volume layout. All (writeable) subvolumes inside a volume are movable inside the volume with mv. Moving subvolumes will not touch the data, but change the volume layout in an instant.

When not otherwise specified, additional devices are handled as Just a Bunch of Disks (JBOD)

Tip
In most cases it is advisable to use only one volume

Extend a volume

This is very easy because btrfs incorporates a volume manager. You only have to look for a free device. You may use a partition like /dev/sdz4 or you may use a raw device like /dev/sdz. Then add this device to your existing volume with btrfs device add. You may need to do a balance afterwards to redistribute some chunks.

Dont´t format the partition!
To add a device it should have NO filesystem on it. The btrfs volume manager will include the free storage of the device into the existing filesystem.

Move a volume (to another disk)

There are a lot of ways you can move a "normal" filesystem from one disk to another. But there are dangers with moving btrfs volumes that do not exist with other filesystems! Don´t ever move a btrfs volume with a tool that does not say it is 100% btrfs-proof. When at any time there are 2 partitions in one computer that have the same filesystem UUID, one ore both filesystems may be destroyed. Under the topic tips you will find an easy way to do move a volume without any danger

Do NOT
  • make a block-level copy of a Btrfs filesystem to another block device
  • use LVM snapshots, or any other kind of block level snapshots
  • turn a copy of a filesystem that is stored in a file into a block device with the loopback driver
  • try to mount either the original or the copy while both are visible to the same kernel
See why at Block-level copies of devices@btrfs.wiki.kernel.org
Tip: move a volume
There is an easy and secure way to move a volume to another disk/device. If you use Btrfs itself to move the volume, there will be no danger. You even can do this while the volume is in use. See how to in "tips"

Subvolume

A subvolume is an independent mountable POSIX file-tree and not a block device. It is the part of a volume that will be mounted writeable into your Linux system. If you dont´t care about snapshots, and you don´t care about backups, it would be possible to use only one subvolume for everything. But then you would not be able to use the powers of Btrfs. Lets assume you do care.

All subvolumes share the space of the Btrfs volume. You may create subvolumes at will. (You may think of subvolumes as sort of "dynamic partitions" inside a Btrfs volume)

When making snapshots (or send/receive) every subvolume will be handled separately. For example when you have 2 subvolumes(@, @home), and make a snapshot of one of them(@), this snapshot will contain every bit of data of all files in this subvolume(@), but none of the data from the other subvolume(@home). So if you make a few subvolumes, you are able to follow different strategies for snapshots of them. And you can restore each of them separately.

By convention the names of subvolumes start with @ (@home, @snapshots ...).


Subvolume @

This is the subvolume where your complete manjaro system will reside. It is mounted at "/" in your filesystem. You may take snapshots of this subvolume (or backups with send/receive) to secure a running manjaro system. When something bad happens, you are able to rollback to one of the snapshots, or to restore one of the backups of this subvolume without loosing your data at /home.

In order to make a rollback possible, this has to contain all and every data that is needed for your manjaro to work properly! This includes:

  • config of your bootloader (/boot/brub/grub.cfg)
  • initramdisk (/boot/initramfs-5.10-x86_64.img)
  • kernel (/boot/vmlinuz-5.10-x86_64)
  • kernel-modules (/usr/lib/modules/5.10.59-1-MANJARO/*)
  • programs (/usr/bin/*)
  • configs (/etc/*)
  • libraries (/usr/lib/*)
  • your root account (/root/*)
  • rest of system files (/usr/*)

Subvolume @home

This is the subvolume where all user data ist stored. When you rollback your "@", this will not change at all. You may take snapshots of /home at a different rate and for different reasons. While snapshots of "@" are good for rollback, snapshots of @home are good for undeleting accidentally by users deleted (or overwritten) files.

Subvolume @snapshots, @home.snapshots

It is wise to "store" snapshots NOT inside the subvolume they where taken from. So this may be the right place to store your snapshots of @ or @home.

Subvolume @...

Sometimes it is desired to have other special snapshot strategies (or no snapshots at all) for some parts of the filesystem. If you need this, make another subvolume.

snapshot

A snapshot looks nearly the same as a subvolume. But snapshots really are "read-only photographs of a subvolume". While the subvolume changes with time. The snapshot is frozen in the state of the subvolume at the time you made it. A snapshot is read-only. Therefore it is guaranteed not to change. In a snapshot you will find all files of the subvolume frozen in time. A snapshot is not a substitute for a backup!

Where to place snapshots
When creating snapshots you have to watch out for the volume layout in use

Snapshots (if regularly made) may be used for:

  • Comparing config files from different "times"
  • Merging config files
  • Recovering accidentally deleted/overwritten files
  • Rollback your system
  • Anchor for a backup with send/receive
  • Basis for a seed
  • What do you use snapshots for ?

Making and deleting snapshots is best done automatically:

  • snapper
  • timeshift

Taking snapshots

Taking a snapshot is very fast, and nearly priceless. After the snapshot is taken, all future writes will go as in CoW usual. But none of the space occupied by files in the snapshot will be reusable. As you write more and more new files the filesystem will grow because it can not reuse the files in the snapshot. A new snapshot will freeze additionaly all created or modified files since the last snapshot and so on. If you don´t release(delete) any snapshot you will eventually run out of space soon(disk full)

Releasing snapshots

Deleting a snapshot does not delete any files that are actually in use by other snapshots or the subvolume they where taken from. To free some space, Btrfs has to test for every file in the snapshot, whether it is in use, or it is not. If it is not, the space of this file/version will be freed.(This is greatly simplified) Therefore it is costly to remove snapshots. And Btrfs will do this work in the background. You may notice this, because when you delete a snapshot there will be no immediate gain in freed space. After a while you will notice that some space was freed.

Don´t forget to remove snapshots, or you will get in "out of space" -trouble soon. Btrfs needs some free space to manage its work. If your volume is more then 80% full you have to think(fast) what to do.

  • ++++ Add some partition/device to the volume
  • +++ Remove some snapshots you don't need
  • + Delete some files (this does only help if they are not part of any snapshot)

Rollback to a snapshot

If you need to roll back into a snapshot you have to replace the actual subvolume by the chosen snapshot.

  1. Make a snapshot of the actual subvolume (for later reference)
  2. Delete the subvolume out of its actual place
  3. Create a new subvolume out of the snapshot chosen for rollback
  4. Make this new subvolume the default (optional)
  5. Don't forget to remove the snapshot you made in a few days


Don't forget to remove snapshots before you run out of space
Every time you take a snapshot you only use the "rest of the volume" for storing everything you change from this time on. At some point in time you have to release the snapshot. Don´t use your space up. Btrfs needs space to breath. Don't use more then 80% or you will get into trouble."


NO Snapshots together with quotas
There are reports about massive problems when using quotas together with snapshots (snapper, timeshift). Please have a look at: Known_issues@btrfs.kernel.org

Btrfs RAID

With Btrfs you no longer need to use mdadm to create mirrored volumes or to create RAIDs. This is already included in btrfs, and very easy to use. There are even advanced features bult in:

  • Add devices to the volume This will integrate a device into the mounted volume
    root # btrfs device add /dev/sdz7 / COPY TO CLIPBOARD


  • Remove devices from the volume. This will not delete any data, but remove the device from the volume. Bevorehand all data will be copied to the remaining devices of the volume.
    root # btrfs device delete /dev/sdz8 / COPY TO CLIPBOARD


  • Use devices with different sizes in one volume
  • Switch the volume between RAID levels
  • Convert data to different RAID levels
  • Do this while the volume is mounted and being used

see RAID@wikipedia

RAID 0 (not Just a Bunch of Disks)

Using one ore more devices to build a volume. This volume has the capacity of all the used devices together(1+2+3+4...). This is an very easy way to expand your volume when you need more space. You even can add 2 or 3 devices at a time. When you want to replace a device, you can add the new device, then remove the old device. Btrfs will move all data as necessary. To distribute all data to all devices you may want to balance the volume. Btrfs will stripe the data to all devices.

If one device fails, everything is lost
Be aware that when one of the devices fails your complete volume may be lost if you use RAID 0

1 device

In most setups you will start a volume with 1 device. If only one device is present, metadata will be duplicated on that device. Even with this simple setup you benefit from most features of Btrfs.

2 or more devices

By default, metadata will be mirrored across two devices and data will be striped across all of the devices present. But if you have 2 or more devices in your volume you should consider using RAID 1.

RAID 1 (mirrored), 1C3, 1C4

Automatic repair

In order to preserve the integrity of the volume, Btrfs does separate CRC-checksums of metadata blocks and of data blocks. Every time a data block is read, the checksum is verified. When the checksum shows that the data is not good, Btrfs tries to get a good copy from the mirrored block. Then the bad block is written again with the good data from the mirrored block. This happens in background. The filesystem has been repaired, and this is logged into syslog. This can be forced by using btrfs scrub.

RAID 10 (automatic)

When using enough devices with RAID 1, Btrfs will distribute all data, so that it not only is mirrored but also striped.

RAID 5

Not save yet
This is not recommended see Parity RAID@btrfs.wiki.kernel.org

RAID 6

Not save yet
This is not recommended see Parity RAID@btrfs.wiki.kernel.org

Btrfs maintenance

Btrfs needs maintenace like every filesystem. The more you use advanced features of btrfs, the more you have to watch for the right maintenance.

Fragmentation

Files on btrfs tend to get somewhat fragmented, when they are appended or changed often. If you use a mechanical disk with slow seeks, it may be advisable to defragment some heavy used files from time to time (like once in a month). When the files are only written to, or are only read seldom, don't worry.

Databases

It may be good to defragment a database if there is a noticeable slowdown when using it.

Logfiles

It may be good to defragment a logfile, if there is a noticeable slowdown when booting your system.

Balance

When adding devices or changing the RAID-level of your volume, it may be necessary to balance your volume. While balancing, btrfs will read ALL (or a subset of all) chunks in, and write them out again using the actual RAID-level. It will stripe these files over all available devices equally. While this happens, the volume will keep being usable, but you may see some heavy load on it. Also this may take a very long time because ALL data must be read AND written again. Don´t worry about shutdown. When you shutdown your computer while the balance is running, the balance will pause. After you restart the computer the balance will restart and continue until it is finished.

Filters

Because a complete balance may take a long time, there is the possibility of filters in balance. When using a filter only those chunks will be balanced, that are named by the filter.

Merging chunks

As you use your volume, you will be creating some files, deleting some, modifying some. Then some parts of the chunks are empty. But this is not a coherent space that can be easily reused. Usually this is not a problem for btrfs and will be cleaned up automatically over time. However, if space is scarce (> 80% full), it is advisable to merge free areas together. This can be done by using balance with a filter.

After removing a device

When you remove a device from a volume, btrfs will automatically balance all "chunks" that where on the removed device. These chunks are placed on another device of the volume. So you don't need to balance by yourself after removing a device.

After adding a device

When you add a device there will be no automatic balance. Only when further using the volume, btrfs will use the additional free space according to the actual RAID-level.

After changing RAID-level

When you changed RAID-levels (for example from RAID 0 to RAID 1) there is no automatic duplication of the chunks. Only when writing further, btrfs will respect the changed RAID-level. This may not be what you intended. To complete the conversion to another RAID-level you need to tell btrfs to rewrite chunks where needed. You do this with a manual balance:

see: btrfs filesystem balance

Scrub

A scrub of a btrfs volume is like inspecting your home. Does the light work in every room? Is the battery of a smoke alarm empty? Does the fire extinguisher need to be replaced? Is any faucet dripping? Every now and then it is advisable to look for such things. And fix it right away!

check & repair on the fly

Every time btrfs reads a file, it checks the corresponding checksums. When btrfs is in RAID 0 and detects a 'damaged' file, it only can tell you that the file is damaged. You may delete it, or replace it from your backup. When running in RAID 1 btrfs has 2 copies of every file. So if one file seems to be damaged, btrfs will read the other copy. If this other copy is ok, btrfs will automatically create another good copy of the file, and then afterwards delete the defect copy. You won't even notice this 'automatic' repair.

full check

Sometimes you may want to check ALL files, and to get a report of defects, because:

  • Some files are read very seldom, and you don't want "bitrott"
  • You want to check your data completely for any faults
  • Once in a while you want to make sure all is well
  • You want to force your disk/device to verify all data
  • ...

This is when you use btrfs scrub. Scrub will read ALL data of your complete volume. While this is done, all files and metadata will be checked by the checksums and all problems will be reportet. Because this will read and verify your complete volume (eventually a few Terra-byte), it may take some time. Scrub will not waste time in checking unused chunks. While checking you can enable automatic repair for RAID 1 or disable it. If you disable automatic repair, scrub will work completely readonly, and will change nothing on your volume.

Scrub can be done automatically (for example every month), or manually.


Notice
Be aware, that Scrub does not structurally check the filesystem, but only checksums data and tree blocks. You may also have a look at tree checker

A manual scrub is done by:

root # btrfs filesystem scrub / COPY TO CLIPBOARD


Solving Problems

Out of space

You should avoid to get out of space with btrfs!

Don't be stingy on storage space when creating a btrfs volume. A btrfs volume should normally only be 80% full. Then it is advisable to adjust the volume. In an emergency, 90% are okay too. But that is neither advantageous for the Btrfs volume nor for an SSD. And don't forget to include the number of snapshots you want in the bill. To watch out how near you are to "out of space", look with btrfs filesystem usage at Device unallocated: (not at Free (estimated):) 1

Manjaro is a rolling release distribution. There will be a lot of changes over time.

Get out of jail

This said, there is an easy way out. Give btrfs more space ;-)

  • add a partition (for example a extern USB-Stick with 8GB) to the btrfs-volume with btrfs devive add
  • From now on this usb-partition belongs to your volume! Do not boot without it. Do not reboot at all in this stage! This stick has to stay until btrfs device remove has completed.
  • look for old snapshots you don´t need, remove them now
  • or delete some files you do not need
  • Do this until your disk will be not more then 95% full

Do not get confused if btrfs does not immediately display the vacant space. After the next step (balance) it will become visible.

  • balance your volume with a filter(!)
  • remove the added partition from the volume with btrfs device remove
  • This will take some time ! Do not reboot until this step is complete.
  • Now you may remove the USB-Stick.

Stay out of jail

  • Think about how to extend the btrfs volume to double size
  • seek if there is a program filling your disk
  • look for snapshots !
  • don’t forget to remove old snapshots (best done automatically)


tips

move a volume

There is an easy and secure way to move a volume to another disk/device. If you use Btrfs itself to move the volume, there will be no danger. You even can do this while the volume is in use.

  • Create the partition you want to use as destination without formatting it. Or remove the filesystem when one is present
  • Add the destination device to your volume by
    root # btrfs device add /dev/[destination] COPY TO CLIPBOARD


  • Remove the source device from your volume by
    root # btrfs device remove /dev/[source] COPY TO CLIPBOARD


Btrfs will notice, that it is necessary for this setup to move all data from the source device to the destination device. And it will start immediately to move data in the background. Meanwhile you can use your PC as you want.

  • Empty Blocks will not be moved
  • Compressed data will remain compressed
  • All Snapshots will remain
  • The UUID of the filesystem will remain the same, but btrfs will be aware of this
  • If you used the UUID to identify your volume, you even wont´t need to edit /boot/grub/grub.cfg and /etc/fstab
  • Only, don't shutdown while the move of the volume is not complete.

If you want to watch the volume move, inside a terminal:

user $ pamac install procps-ng COPY TO CLIPBOARD


root # watch -n 60 btrfs filesystem show / COPY TO CLIPBOARD


Btrfs options

Compression

Grub needs to load the kernel and initrd
When you use compression on kernel, initrd, or grub config files, grub needs to decompress these files. Otherwise you will not be able to boot. GRUB introduced zstd support in 2.04. Maybe you need to update grub and reinstall it

encryption

Send + receive = backup

quotas

Quota support in Btrfs is implemented at the subvolume level.

For more info see Quota_support@btrfs.kernel.org

Reports about problems
There are reports about massive problems when using quotas (especially together with snapshots, snapper, timeshift). Please have a look at: Known_issues@btrfs.kernel.org


Btrfs Tools

For complete info an all tools for btrfs please do an actual search at the arch wiki and at the manjaro forum for "btrfs". In the following section only a few commands are described. Especially those commands that are often misinterpreted. Commands described earlier are omitted.

btrfs (the command)

Be aware that some sub-commands of btrfs will not work as normal user. Other sub-commands do work but will give only partial info. So best use them as root or with sudo.

help, version

help together with man btrfs or info btrfs will get you an overview over the usable options on your install.

user $ btrfs help COPY TO CLIPBOARD


With the version of btrfs given here you can look at changelog.

user $ btrfs version COPY TO CLIPBOARD


device

scan will give no visible results 😜

root # btrfs device scan COPY TO CLIPBOARD


stats will give a list of errors detected in the past. This all should be 0, or you may be in trouble.

root # btrfs device stats / COPY TO CLIPBOARD


What to look for in device usage:

  • RAID-level of Data, Metadata and System
  • Unalocated:

Don't ever let Unalocated: get below 5% of your volume (or double the size you need for your next update)! If this goes too low, you will get into "out of space" trouble

root # btrfs device usage / COPY TO CLIPBOARD


Without sudo this will give very wrong results for everything displayed. Dont´t use this without sudo !


btrfs filesystem df /

btrfs filesystem du

btrfs filesystem show

btrfs filesystem usage

btrfs scrub status

btrfsck

This is not what you may think it is 😜

Recomendations

We recommend using Btrfs with UEFI and GPT
Partition Filesystem Size Partition type
/dev/sda1 Fat32 1GiB EFI system partition
/dev/sda2 Btrfs 1Gib - 8EiB Btrfs Volume
/dev/sda3 swap 4GiB, at least your RAM-size Swap partition (optional)
IF you don't have UEFI, you may use Btrfs with BIOS and GPT
Partition Filesystem Size Partition type
/dev/sda1 (bootloader) 4MiB BIOS boot partition
/dev/sda2 Btrfs 1Gib - 8EiB Btrfs Volume
/dev/sda3 swap 4GiB, at least your RAM-size Swap partition (optional)
user $ example command should be here COPY TO CLIPBOARD



Example codes should be here.


Please be aware that the information on this page is a simplified version of the reality. Is is written to make the reader understand a little of these complex things. To get an in depth understanding it will be neccesary to read further at btrfs.wiki.kernel.org or other places.

Additional Information

Why not btrfs ?

A lot of people say: "I don't use btrfs because it is experimental and is not stable. You can´t use it in production. It is not safe!".

Not stable ?

The status of btrfs was experimental for a long time, but the the core functionality is considered good enough for daily use. (from kernel.org)

If you see statements declaring Btrfs as not stable, please look for the date of them. Some seem to date from 10 years ago. So if you want to give Btrfs a chance, you have to look for newer statements. Maybe even look at Btrfs Kernel Wiki as that sure is the best information regarding Btrfs

Experimental ?

Btrfs is feature-rich! There are new features being implemented and these should be considered experimental for a few releases when the bugs get ironed out when number of brave users help stabilizing it.(from kernel.org)

Some features are not implemented yet. Others are only partly implemented. Some are experimental and not suggested for production use. As is always the case in Linux-land you decide what to use, and so you are responsible for your own decisions.

Not usable for production ?

  • Distro support for Btrfs as main filesystem
  • Some companies do use Btrfs in production@wiki.btrfs.kernel.org
  • Some manufacturers do deploy devices where Btrfs is installed by default.

Difficult to repair ?

Indeed, when you search for the usual ways to repair a file system like FAT or Ext4 then you don't find good information. But this is not because it is difficult to repair Btrfs, but because repairing Btrfs does work very differently.

What's this "Copy on Write"

When you want to get the most out of using Btrfs you do need to know some things about this file system. Then you are able to use it properly and to your advantage. Btrfs is not difficult, but different to some extend.

Write in place (FAT32)

Most older file systems do write "in place". This means that some data or metadata will be written "over" the previous data at the same place.

For example this is the case for FAT32 file systems. The File Allocation Table is at a fixed place on this file system. When the "FAT" changes (because a file got bigger and needs more blocks), this new FAT must be written with the new data to the same place as before. When the disk is ejected before (or while) this data is written, the file system will be corrupted. And the FAT does change a lot.

The danger of corruption is especially big while metadata (like filename, permission, usage of disk space ...) is being written.

Write to a metadata-log (Ext4)

There is a solution to this with newer file systems like Ext4. Instead of writing metadata "in place", metadata is written into an "endless" log. Then it is not possible to be corrupted while overwritten. This is possible because metadata is only a very small part of the data in a file system.

There has to be an additional mechanism to make this safe. Sometimes this is called "barriers", and there have to be checksums that tell when a part of the log is corrupted.

This does protect the file system itself, but not the files in it. Because a file may be overwritten in place, and then the old file is lost, and the new one may not have been written completely.

Copy on Write! (Btrfs)

Copy on Write is a "new" concept. It means the file system will try to never write over existing data. How is this even possible?

  • Files are appended at the end of a "data page"
  • Metadata is appended at a "metadata page"
  • Inside a page nothing is ever overwritten
  • When a page is full the file system will use the next free page
  • Deleting a file does not write/clean its data, but writes metadata, that marks this file as deleted
  • Overwriting a file does first append the new file to the "data page", then writes the metadata for this file
  • Changing small parts of a file will write only the new parts, then link the rest to the old file
  • there are checksums for data and metadata

Downsides

  • Management of space is complex
  • There are 2 sorts of pages
  • There has to be a clean-up-process who makes the space of deleted files reusable, so that the disk does not run out of free pages
  • It must be avoided to write data unnecessarily, because then the clean-up would also be very expensive

(Dis)advantages

  • It is possible to detect nearly any corruption because of the checksums
    • When the power is lost, or the disk is disconnected, all old data is save. WHY?
    • Every bit of "old" data from before the power loss or the disconnection is present because it is NOT overwritten
    • Only the newly written data may be partly damaged
    • The metadata may also be partly damaged
    • When mounting the volume it is possible by analysing checksums and metadata to find the point in the file system where all was good
    • Btrfs will automatically roll back to this point, then it can mount the file system writeable
  • CoW is a sound foundation to build upon
    • Snapshots
    • RAID
    • Volume management
    • Compression
    • Encryption (maybe some time in the future)


Don´t disable CoW in Btrfs
It is possible to disable CoW in Btrfs. But then you loose all benefits of Btrfs. It won´t even make checksums. If you don't like CoW, then you better use another filesystem

Use the Forum!

It is a good Idea to search the forum for posts related to btrfs.

Btrfs is fast moving! See Also: