(mention grub/zstd support; 404 fixed)
|Line 286:||Line 286:|
* [https://git.savannah.gnu.org/cgit/grub.git/commit/?id=050e8e908085d69c7ac2b6b99f4361787b046d74 Add support for LZO compression in GRUB] (2011-10-06)
* [https://git.savannah.gnu.org/cgit/grub.git/commit/?id=050e8e908085d69c7ac2b6b99f4361787b046d74 Add support for LZO compression in GRUB] (2011-10-06)
* [https://git.savannah.gnu.org/cgit/grub.git/commit/?id=386128648606a3aa6ae7108d1c9af52258202279 btrfs: Add zstd support to grub btrfs] (2018-11-19)
* [https://git.savannah.gnu.org/cgit/grub.git/commit/?id=386128648606a3aa6ae7108d1c9af52258202279 btrfs: Add zstd support to grub btrfs] (2018-11-19)
Revision as of 06:28, 29 August 2019
I have a problem with my btrfs filesystem!
See the Problem FAQ for commonly-encountered problems and solutions.
Explicitly said: please report bugs and issues to the mailing list (you are not required to subscribe).
Then use Bugzilla which will ensure traceability.
I see a warning in dmesg about barriers being disabled when mounting my filesystem. What does that mean?
Your hard drive has been detected as not supporting barriers. This is a severe condition, which can result in full file-system corruption, not just losing or corrupting data that was being written at the time of the power cut or crash. There is only one certain way to work around this:
Failure to perform this can result in massive and possibly irrecoverable corruption (especially in the case of encrypted filesystems).
Help! I ran out of disk space!
Help! Btrfs claims I'm out of space, but it looks like I should have lots left!
Free space is a tricky concept in Btrfs. This is especially apparent when running low on it. Read "Why is there so many ways to check the amount of free space" below for the blow-by-blow.
You can look at the tips below, and you can also try Marc MERLIN's debugging filesystem full page
if your device is small
The best solution for small devices (under about 16 GB) is to reformat the FS with the
--mixed option to mkfs.btrfs. This needs a kernel 2.6.37 or later, and similarly recent btrfs-progs.
The main issue is that the allocation units (chunk size) are very large compared to the size of the filesystem, and the allocation can very quickly become full. A btrfs fi balance may get you working again, but it's probably only a short term fix, as the metadata to data ratio probably won't match the block allocations.
If you can afford to delete files, you can clobber a file via
echo > /path/to/file
which will recover that space without requiring a new metadata allocation (which would otherwise ENOSPC again).
You might consider remounting with -o compress, and either rewrite particular files in-place, or run a recursive defragmentation which (if an explicit flag is given, or if the filesystem is mounted with compression enabled) will also recompress everything. This may take a while.
if your device is large (>16GiB)
If the filesystem has allocated (but not used) all of the available space, and the metadata is close to full, then df can show lots of free space, but you may still get out of space errors because there isn't enough metadata available.
To see if this is the case, first look for the amount of allocated space with
# sudo btrfs fi show /dev/device
If this shows the "used" value equal to the "total" value on each device, then everything has been allocated, which is the first condition for this problem.
Secondly, look at the amount of space you have in metadata, as reported by
$ btrfs fi df /mountpoint
If the "used" metadata is close to the "total" value, then that's the second condition for this problem, and you should read on. What does "close" mean? If the free space in metadata is less than or equal to the block reserve value (typically 512 MiB, but might be something else on a particularly small or large filesystem), then it's close to full.
If you have full up metadata, and more than 1 GiB of space free in data, as reported by btrfs fi df, then you should be able to free up some of the data allocation with a partial balance:
# btrfs balance start /mountpoint -dlimit=3
We know this isn't ideal, and there are plans to improve the behavior. Running close to empty is rarely the ideal case, but we can get far closer to full than we do.
Free space cache file is invalid. skip it
If you see something like this when mounting:
BTRFS info (device sdc1): The free space cache file (2159324168192) is invalid. skip it
then you should run btrfs check on your fs. I have been seeing this in dmesg and Data would not allocate new blocks at all. Doing btrfs check fixed this at least for me. I still would hit the bug below afterwards (bug #74101) - that is, Data would normally resize, but sometimes it would fail to do that, and the fs crashed and remounted as ro. Before btrfs check I also resized the fs to smaller than it was and then to the max size using btrfs fi resize, which may have helped part way, but until I ran btrfs check, Data would not resize. Try resizing.
When you haven't hit the "usual" problem
If the conditions above aren't true (i.e. there's plenty of unallocated space, or there's lots of unused metadata allocation), then you may have hit a known but unresolved bug. If this is the case, please report it to either the mailing list, or IRC. In some cases, it has been possible to deal with the problem, but the approach is new, and we would like more direct contact with people experiencing this particular bug.
The "Data" field in btrfs fi df fills up and doesn't become larger, and then the bug transpires. This has been reported in the btrfs mailing list (search for the subject 6TB partition, Data only 2TB - aka When you haven't hit the "usual" problem). Here is a likely bug report: https://bugzilla.kernel.org/show_bug.cgi?id=74101
The bug would only happen after some time, so I was able to move a lot of data onto the fs before btrfs crashed, and then it was just a case of unmounting and mounting again.
Are btrfs changes backported to stable kernel releases?
Yes, the obviously critical fixes get to the latest stable kernel and sometimes get also applied to the long-term branches. Please note that there's yet no one appointed and submitting the patches is done on voluntary basis or when the maintainer(s) do that.
Beware that apart from critical fixes, the longterm branches do not receive backports of less important fixes done by the upstream maintainers. You should ask your distro kernel maintainers to do that. A CC of the linux-btrfs mailinglist is a good idea when there are patches selected for a particular longterm kernel and requested for addition to stable trees.
Performance vs Correctness
Does Btrfs have data=ordered mode like Ext3?
In v0.16, Btrfs waits until data extents are on disk before updating metadata. This ensures that stale data isn't exposed after a crash, and that file data is consistent with the checksums stored in the btree after a crash.
Note that you may get zero-length files after a crash, see the next questions for more info.
Btrfs does not force all dirty data to disk on every fsync or O_SYNC operation, fsync is designed to be fast.
What are the crash guarantees of overwrite-by-rename?
Overwriting an existing file using a rename is atomic. That means that either the old content of the file is there or the new content. A sequence like this:
echo "oldcontent" > file # make sure oldcontent is on disk sync echo "newcontent" > file.tmp mv -f file.tmp file # *crash*
Will give either
- file contains "newcontent"; file.tmp does not exist
- file contains "oldcontent"; file.tmp may contain "newcontent", be zero-length or not exists at all.
Why I experience poor performance during file access on filesystem?
By default the file system is mounted with relatime flag, which means it must update files' metadata during first access on each day. Since updates to metadata are done as COW, if one visits a lot o files, it results in massive and scattered write operations on the underlying media.
You need to mount file system with noatime flag to prevent this from happening.
More details are in Mount_options#Performance
What are the crash guarantees of rename?
Renames NOT overwriting existing files do not give additional guarantees. This means, a sequence like
echo "content" > file.tmp mv file.tmp file # *crash*
will most likely give you a zero-length "file". The sequence can give you either
- Neither file nor file.tmp exists
- Either file.tmp or file exists and is 0-size or contains "content"
For more info see this thread: http://thread.gmane.org/gmane.comp.file-systems.btrfs/5599/focus=5623
Can the data=ordered mode be turned off in Btrfs?
No, it is an important part of keeping data and checksums consistent. The Btrfs data=ordered mode is very fast and turning it off is not required for good performance.
What checksum function does Btrfs use?
Currently Btrfs uses crc32c for data and metadata. The disk format has room for 256bits of checksum for metadata and up to a full leaf block (roughly 4k or more) for data blocks. Over time we'll add support for more checksum alternatives.
Can data checksumming be turned off?
Yes, you can disable it by mounting with -o nodatasum. Please note that checksumming is also turned off when the filesystem is mounted with nodatacow.
Can copy-on-write be turned off for data blocks?
Yes, there are several ways how to do that.
Disable it by mounting with nodatacow. This implies nodatasum as well. COW may still happen if a snapshot is taken. However COW will still be maintained for existing files, because the COW status can be modified only for empty or newly created files.
For an empty file, add the NOCOW file attribute (use chattr utility with +C), or you create a new file in a directory with the NOCOW attribute set (then the new file will inherit this attribute).
Now copy the original data into the pre-created file, delete original and rename back.
There is a script you can use to do this .
Shell commands may look like this:
touch vm-image.raw chattr +C vm-image.raw fallocate -l10g vm-image.raw
will produce file suitable for a raw VM image -- the blocks will be updated in-place and are preallocated.
Can I have nodatacow (or chattr +C) but still have checksumming?
With normal CoW operations, the atomicity is achieved by constructing a completely new metadata tree containing both changes (references to the data, and the csum metadata), and then atomically changing the superblock to point to the new tree.
With nodatacow, that approach doesn't work, because the new data replaces the old on the physical medium, so you'd have to make the data write atomic with the superblock write -- which can't be done, because it's (at least) two distinct writes. This means that you have to write the two separately, which then means that there's a period between the two writes where things can go wrong.
When you write data and metadata separately (which you have to do in the nodatacow case), and the system halts between the two writes, then you either have the new data with the old csum, or the old csum with the new data. Both data and csum are "good", but good from different states of the FS. In both orderings (write data first or write metadata first), the csum doesn't match the data, and so you now have an I/O error reported when trying to read that data.
You can't easily fix this, because when the data and csum don't match, you need to know the reason they don't match -- is it because the machine was interrupted during write (in which case you can simply compute a good csum from the data and carry on as normal), or is it because the hard disk has been corrupted in some way, completely independently of the crash, and the data is now toast (in which case you shouldn't fix the I/O error)?
Basically, nodatacow bypasses the very mechanisms that are meant to provide consistency in the filesystem.
(See also the Project ideas page)
When will Btrfs have a fsck like tool?
The first detailed report on what comprises "btrfsck"
Btrfsck has its own page, go check it out.
Note that in many cases, you don't want to run fsck. Btrfs is fairly self healing, but when needed check and recovery can be done several ways. Marc MERLIN has written a page that explains the different ways to check and fix a btrfs filesystem.
The btrfsck tool in the git master branch for btrfs-progs is now capable of repairing some types of filesystem breakage. It is not well-tested in real-life situations yet. If you have a broken filesystem, it is probably better to use btrfsck with advice from one of the btrfs developers, just in case something goes wrong. (But even if it does go badly wrong, you've still got your backups, right?)
Note that there is also a recovery tool in the btrfs-progs git repository which can often be used to copy essential files out of broken filesystems.
What's the difference between btrfsck and fsck.btrfs
- btrfsck is the actual utility that is able to check and repair a filesystem
- fsck.btrfs is a utility that should exist for any filesystem type and is called during system setup when the corresponding /etc/fstab entries contain non-zero value for fs_passno. (See fstab(5) for more.)
Traditional filesystems need to run their respective fsck utility in case the filesystem was not unmounted cleanly and the log needs to be replayed before mount. This is not needed for btrfs. You should set fs_passno to 0.
Note, if the fsck.btrfs utility is in fact btrfsck, then the filesystem is unnecessarily checked upon every boot and slows down the whole operation. It is safe to and recommended to turn fsck.btrfs into a no-op, eg. by cp /bin/true /sbin/fsck.btrfs.
Can I use RAID on my Btrfs filesystem?
The parity RAID feature is mostly implemented, but has some problems in the case of power failure (or other unclean shutdown) which lead to damaged data. It is recommended that parity RAID be used only for testing purposes.
See RAID56 for more details.
Is Btrfs optimized for SSD?
There are some optimizations for SSD drives, and you can enable them by mounting with -o ssd. As of 2.6.31-rc1, this mount option will be enabled if Btrfs is able to detect non-rotating storage.
Note that before 4.14 the ssd mount option has a negative impact on usability and lifetime of modern SSDs which have a FTL (Flash Translation Layer). See the ssd section in Gotchas for more information.
Note that -o ssd will not enable TRIM/discard.
Does Btrfs support TRIM/discard?
There are two ways how to apply the discard:
- during normal operation on any space that's going to be freed, enabled by mount option discard
- on demand via the command fstrim
"-o discard" can have some negative consequences on performance on some SSDs or at least whether it adds worthwhile performance is up for debate depending on who you ask, and makes undeletion/recovery near impossible while being a security problem if you use dm-crypt underneath (see http://asalor.blogspot.com/2011/08/trim-dm-crypt-problems.html ), therefore it is not enabled by default. You are welcome to run your own benchmarks and post them here, with the caveat that they'll be very SSD firmware specific.
The fstrim way is more flexible as it allows to apply trim on a specific block range, or can be scheduled to time when the filesystem perfomace drop is not critical.
Does btrfs support encryption?
There are several different ways in which a filesystem can interoperate with encryption to keep your data secure:
- It can operate on top of an encrypted partition (dm-crypt / LUKS) scheme.
- It can be used as a component of a stacked approach (eg. ecryptfs) where a layer above the filesystem transparently provides the encryption.
- It can natively attempt to encrypt file data and associated information such as the file name.
There are advantages and disadvantages to each method, and care should be taken to make sure that the encryption protects against the right threat. In some situations, more than one approach may be needed.
Typically, partition (or entire disk) encryption is used to protect data in case a computer is stolen, a hard disk has to be returned to the manufacturer if it fails under warranty or due to the difficulty of erasing modern drives. Modern hard disks and SSDs transparently remap logical sectors to physical sectors for various reasons, including bad sectors and wear leveling. This means sensitive data can remain on a sector even after an attempt to erase the whole disk. The only protection against these challenges is full disk encryption from the moment you buy the disk. Full disk encryption requires a password for the computer to boot, but the system operates normally after that. All data (except the boot loader and kernel) is encrypted. Btrfs works safely with partition encryption (luks/dm-crypt) since Linux 3.2. Earlier kernels will start up in this mode, but are known to be unsafe and may corrupt due to problems with dm-crypt write barrier support.
Partition encryption does not protect data accessed by a running system -- after boot, a user sees the computer normally, without having to enter extra passwords. There may also be some performance impact since all IO must be encrypted, not just important files. For this reason, it's often preferable to encrypt individual files or folders, so that important files can't be accessed without the right password while the system is online. If the computer might also be stolen, it may be preferable to use partition encryption as well as file encryption.
Btrfs does not support native file encryption (yet), and there's nobody actively working on it. It could conceivably be added in the future.
As an alternative, it is possible to use a stacked filesystem (eg. ecryptfs) with btrfs. In this mode, the stacked encryption layer is mounted over a portion of a btrfs volume and transparently applies the security before the data is sent to btrfs. Another similar option is to use the fuse-based filesystem encfs as a encrypting layer on top of btrfs.
Note that a stacked encryption layer (especially using fuse) may be slow, and because the encryption happens before btrfs sees the data, btrfs compression won't save space (encrypted data is too scrambled). From the point of view of btrfs, the user is just writing files full of noise.
Also keep in mind that if you use partition level encryption and btrfs RAID on top of multiple encrypted partitions, the partition encryption will have to individually encrypt each copy. This may result in somewhat reduced performance compared to a traditional RAID setup where the encryption might be done on top of RAID. Whether the encryption has a significant impact depends on the workload, and note that many newer CPUs have hardware encryption support.
Does Btrfs work on top of dm-crypt?
This is deemed safe since 3.2 kernels. Corruption has been reported before that, so you want a recent kernel. The reason was improper passing of device barriers that are a requirement of the filesystem to guarantee consistency.
If you are trying to mount a btrfs filesystem based on multiple dm-crypted devices, you can see an example script on Marc's btrfs blog: start-btrfs-dmcrypt.
Does btrfs support deduplication?
Deduplication is supported, with some limitations. See Deduplication.
Does btrfs support swap files?
From kernel 5.0+ btrfs have native swap files support, but with some limitations. Swap file - must be fully allocated as NOCOW with no compression on one device.
For kernels before 5.0, swap files not supported. Just making a file NOCOW does not help, swap file support relies on one function that btrfs intentionally does not implement due to potential corruptions. The swap implementation used to rely on some assumptions which may not hold in btrfs, like block numbers in the swap file while btrfs has a different block number mapping in case of multiple devices. There is a new API that could be used to port swap to btrfs; for more details have a look at project ideas#Swap file support.
A workaround, albeit with poor performance, is to mount a swap file via a loop device.
Does grub support btrfs?
In most cases. Grub supports many btrfs configurations, including zlib and lzo compression (but not zstd, depending on the Grub version) and RAID0/1/10 multi-dev filesystems. If your distribution only provides older versions of Grub, you'll have to build it for yourself.
grubenv write support (used to track failed boot entries) is lacking, Grub needs btrfs to support a reserved area.
- BtrFS zlib compression support (2010-12-03)
- Add support for LZO compression in GRUB (2011-10-06)
- btrfs: Add zstd support to grub btrfs (2018-11-19)
Is it possible to boot to btrfs without an initramfs?
With multiple devices, btrfs normally needs an initramfs to perform a device scan. It may be necessary to modprobe (and then rmmod) scsi-wait-scan to work around a race condition. See using Btrfs with multiple devices.
With grub and a single disk, you might not need an initramfs. Grub generates a root=UUID=… command line that the kernel should handle on its own. Some people have also used GPT and root=PARTUUID= specs instead .
Compression support in btrfs
There's a separate page for compression related questions, Compression.
Will btrfs support LZ4?
Maybe, if there is a clear benefit compared to existing compression support. Technically there are no obstacles, but supporing a new algorithm has impact on the the userpace tools or bootloader, that has to be justified. There's a project idea Compression enhancements that targets more than just adding a new compression algorithm.
How do I do...?
See also the UseCases page.
Is btrfs stable?
Short answer: Maybe.
Long answer: Nobody is going to magically stick a label on the btrfs code and say "yes, this is now stable and bug-free". Different people have different concepts of stability: a home user who wants to keep their ripped CDs on it will have a different requirement for stability than a large financial institution running their trading system on it. If you are concerned about stability in commercial production use, you should test btrfs on a testbed system under production workloads to see if it will do what you want of it. In any case, you should join the mailing list (and hang out in IRC) and read through problem reports and follow them to their conclusion to give yourself a good idea of the types of issues that come up, and the degree to which they can be dealt with. Whatever you do, we recommend keeping good, tested, off-system (and off-site) backups.
Pragmatic answer: Many of the developers and testers run btrfs as their primary filesystem for day-to-day usage, or with various forms of real data. With reliable hardware and up-to-date kernels, we see very few unrecoverable problems showing up. As always, keep backups, test them, and be prepared to use them.
When deciding if Btrfs is the right file system for your use case, don't forget to look at the Status page, which contains an overview of the general status of distinct features of the file system.
What version of btrfs-progs should I use for my kernel?
Simply use the latest version.
The userspace tools versions roughly match the kernel releases and should contain support for features introduced in the respective kernel release. The minor versions are bugfix releases or independent updates (eg. documentation, tests).
Do I have to keep my btrfs-progs at the same version as my kernel?
If your btrfs-progs is newer than your kernel, then you may not be able to use some of the features that the btrfs-progs offers, because the kernel doesn't support them.
If your btrfs-progs is older than your kernel, then you may not be able to use some of the features that the kernel offers, because the btrfs-progs doesn't support them.
Other than that, there should be no restrictions on which versions work together.
I have converted my ext4 partition into Btrfs, how do I delete the ext2_saved folder?
The folder is a normal btrfs subvolume and you can delete it with the command
btrfs subvolume delete /path/to/btrfs/ext2_saved
How much free space do I have?
or My filesystem is full, and I've put almost nothing into it!
Free space in Btrfs is a tricky concept from a traditional viewpoint, owing partly to the features it provides and partly to the difficulty in making too many assumptions about the exact information you need to know at the time.
Understanding free space, using the original tools
Btrfs starts with a pool of raw storage, made up of the space on all the devices in the FS. This is what you see when you run
btrfs fi show:
$ sudo btrfs fi show Label: none uuid: 12345678-1234-5678-1234-1234567890ab Total devices 2 FS bytes used 304.48GB devid 1 size 427.24GB used 197.01GB path /dev/sda1 devid 2 size 465.76GB used 197.01GB path /dev/sdc1
The pool is the total size of /dev/sda1 and /dev/sdc1 in this case. The filesystem automatically allocates space out of the pool as it needs it. In this case, it has allocated 394.02 GiB of space (197.01+197.01), out of a total of 893.00 GiB (427.24+465.76) across two devices. The allocation is simply setting aside a region of a disk for some purpose (e.g. "data will be stored in this region"). The output from
btrfs fi show, however, doesn't tell the whole story. The
btrfs fi show output only shows how much has been allocated out of the total available space. In order to see how that allocation is used, and how much of it contains useful data, you also need
btrfs fi df:
$ btrfs fi df / Metadata, single: total=18.00GB, used=6.10GB Data, single: total=376.00GB, used=298.37GB System, single: total=12.00MB, used=40.00KB
Note that the "total" values here add up to the "used" values in btrfs fi show. The "used" values here tell you how much useful information is stored. The rest is free space available for data (or metadata).
So, in this case, there is 77.63 GiB of space (376-298.37) for data allocated but unused, and a further 498.98 GiB (427.24+465.76-197.01-197.01) unallocated and available for further use. Some of the unallocated space may be used for metadata as the FS grows in size.
The interpretation is slightly more complicated with RAID filesystems. With a RAID filesystem,
btrfs fi show still shows the raw disk space being allocated, but
btrfs fi df shows the space available for files, so it takes account of the extra space needed by RAID. Here's a slightly more complex example:
$ sudo btrfs fi show Label: none uuid: b7c23711-6b9e-46a8-b451-4b3f79c7bc46 Total devices 2 FS bytes used 14.67GiB devid 1 size 40.00GiB used 16.01GiB path /dev/sdc1 devid 2 size 40.00GiB used 16.01GiB path /dev/sdd1
$ sudo btrfs fi df /mnt Data, RAID1: total=15.00GiB, used=14.65GiB System, RAID1: total=8.00MiB, used=16.00KiB Metadata, RAID1: total=1.00GiB, used=15.59MiB GlobalReserve, single: total=16.00MiB, used=0.00
Here, we can see that 32.02 GiB of space has been allocated out of the 80 GiB available. The
btrfs fi df output tells us that each of Data, Metadata and System have RAID-1 allocation. Now, note that the sum of the "total" values here is much smaller than the allocated space reported by
btrfs fi show. This is because RAID-1 stores two copies of every byte written to it, so to store the 14.65 GiB of data in this filesystem, the 15 GiB of data allocation actually takes up 30 GiB of space on the disks. Taking account of this, we find that the total allocation is: (15.00 + 0.008 + 1.00) * 2, which is 32.02 GiB. The GlobalReserve can be ignored in this calculation.
Depending on the RAID level you are using, the "correction" factor will be different:
* single and RAID-0 have no correction * DUP, RAID-1 and RAID-10 store two copies, and need to have the values from
btrfs fi dfdoubled * RAID-5 uses one device for parity, so
btrfs fi dfshould be multiplied by n/(n-1) for an n-device filesystem * RAID-6 uses two devices for parity, so
btrfs fi dfshould be multiplied by n/(n-2) for an n-device filesystem
(Note that the number of devices in use may vary on filesystems with unequal devices, so the values for RAID-5 and RAID-6 may not be correct in that case).
Understanding free space, using the new tools
An all-in-one tool for reporting on free space is available in btrfs-progs from 3.18 onwards. This is
btrfs fi usage:
$ sudo btrfs fi usage /mnt Overall: Device size: 80.00GiB Device allocated: 32.02GiB Device unallocated: 47.98GiB Used: 29.33GiB Free (estimated): 24.34GiB (min: 24.34GiB) Data ratio: 2.00 Metadata ratio: 2.00 Global reserve: 16.00MiB (used: 0.00B) Data,RAID1: Size:15.00GiB, Used:14.65GiB /dev/sdc1 15.00GiB /dev/sdd1 15.00GiB Metadata,RAID1: Size:1.00GiB, Used:15.59MiB /dev/sdc1 1.00GiB /dev/sdd1 1.00GiB System,RAID1: Size:8.00MiB, Used:16.00KiB /dev/sdc1 8.00MiB /dev/sdd1 8.00MiB Unallocated: /dev/sdc1 23.99GiB /dev/sdd1 23.99GiB
This is showing the space usage for the same filesystem at the end of the previous section, with two 40 GiB devices. The "Overall" section shows aggregate information for the whole filesystem: There are 80 GiB of raw storage in total, with 32.02 GiB allocated, and 29.33 GiB of that allocation used. The "data ratio" and "metadata ratio" values are the proportion of raw disk space that is used for each byte of data written. For a RAID-1 or RAID-10 filesystem, this should be 2.0; for RAID-5 or RAID-6, this should be somewhere between 1.0 and 1.5. In the example here, the value of 2.0 for data indicates that the 29.34 GiB of used space is holding 14.67 GiB of actual data (29.34 / 2.0).
The "free" value is an estimate of the amount of data that can still be written to this FS, based on the current usage profile. The "min" value is the minimum amount of data that you can expect to be able to get onto the filesystem.
Below the "overall" section, information is shown for each of the types type of allocation, summarised and then broken down by device. In this section, you are seeing the values after the data or metadata ratio is accounted for, so we can see the 14.65 GiB of actual data used, and the 15 GiB of allocated space it lives in. (The small discrepancy with the value we worked out previously can be accounted for by rounding errors).
Why is free space so complicated?
You might think, "My whole disk is RAID-1, so why can't you just divide everything by 2 and give me a sensible value in
If everything is RAID-1 (or RAID-0, or in general all the same RAID level), then yes, we could give a sane and consistent value from
df. However, we have plans to allow per-subvolume and per-file RAID levels. In this case, it becomes impossible to give a sensible estimate as to how much space there is left.
For example, if you have one subvolume as "single", and one as RAID-1, then the first subvolume will consume raw storage at the rate of one byte for each byte of data written. The second subvolume will take two bytes of raw data for each byte of data written. So, if we have 30GiB of raw space available, we could store 30GiB of data on the first subvolume, or 15GiB of data on the second, and there is no way of knowing which it will be until the user writes that data.
So, in general, it is impossible to give an accurate estimate of the amount of free space on any btrfs filesystem. Yes, this sucks. If you have a really good idea for how to make it simple for users to understand how much space they've got left, please do let us know, but also please be aware that the finest minds in btrfs development have been thinking about this problem for at least a couple of years, and we haven't found a simple solution yet.
Why is there so much space overhead?
There are several things meant by this. One is the out-of-space issues discussed above; this is a known deficiency, which can be worked around, and will eventually be worked around properly. The other meaning is the size of the metadata block group, compared to the data block group. Note that you shouldn't compare the size of the allocations, but rather the used space in the allocations.
There are several considerations:
- The default raid level for the metadata group is dup on single drive systems, and raid1 on multi drive systems. The meaning is the same in both cases: there are two copies of everything in that group. This can be disabled at mkfs time, and it is possible to migrate raid levels online using the Balance_filters.
- There is an overhead to maintaining the checksums (approximately 0.1% – 4 bytes for each 4k block)
- Small files are also written inline into the metadata group. If you have several gigabytes of very small files, this will add up.
[incomplete; disabling features, etc]
How much space will I get with my multi-device configuration?
There is an online tool which can calculate the usable space from your drive configuration. For more details about RAID-1 mode, see the question below.
How much space do I get with unequal devices in RAID-1 mode?
For a specific configuration, you can use the online tool to see what will happen.
The general rule of thumb is if your largest device is bigger than all of the others put together, then you will get as much space as all the smaller devices added together. Otherwise, you get half of the space of all of your devices added together.
For example, if you have disks of size 3TB, 1TB, 1TB, your largest disk is 3TB and the sum of the rest is 2TB. In this case, your largest disk is bigger than the sum of the rest, and you will get 2TB of usable space.
If you have disks of size 3TB, 2TB, 2TB, then your largest disk is 3TB and the sum of the rest of 4TB. In this case, your largest disk is smaller than the sum of the rest, and you will get (3+2+2)/2 = 3.5TB of usable space.
If the smaller disks are not the same size, the above holds true for the first case (largest device is bigger than all the others combined), but might not be true if the sum of the rest is larger. In this case, you can apply the rule multiple times.
For example, if you have disks of size 2TB, 1.5TB, 1TB, then the largest disk is 2TB and the sum is 2.5TB, but the smaller devices aren't equal, so we'll apply the rule of thumb twice. First, consider the 2TB and the 1.5TB. This set will give us 1.5TB usable and 500GB left over. Now consider the 500GB left over with the 1TB. This set will give us 500GB usable and 500GB left over. Our total set (2TB, 1.5TB, 1TB) will thus yield 2TB usable.
Another example is 3TB, 2TB, 1TB, 1TB. In this, the largest is 3TB and the sum of the rest is 4TB. Applying the rule of thumb twice, we consider the 3TB and the 2TB and get 2TB usable with 1TB left over. We then consider the 1TB left over with the 1TB and the 1TB and get 1.5TB usable with nothing left over. Our total is 3.5TB of usable space.
What does "balance" do?
btrfs filesystem balance is an operation which simply takes all of the data and metadata on the filesystem, and re-writes it in a different place on the disks, passing it through the allocator algorithm on the way. It was originally designed for multi-device filesystems, to spread data more evenly across the devices (i.e. to "balance" their usage). This is particularly useful when adding new devices to a nearly-full filesystem.
Due to the way that balance works, it also has some useful side-effects:
- If there is a lot of allocated but unused data or metadata chunks, a balance may reclaim some of that allocated space. This is the main reason for running a balance on a single-device filesystem.
- On a filesystem with damaged replication (e.g. a RAID-1 FS with a dead and removed disk), it will force the FS to rebuild the missing copy of the data on one of the currently active devices, restoring the RAID-1 capability of the filesystem.
Until at least 3.14, balance is sometimes needed to fix filesystem full issues. See Balance_Filters.
Does a balance operation make the internal B-trees better/faster?
No, balance has nothing at all to do with the B-trees used for storing all of btrfs's metadata. The B-tree implementation used in btrfs is effectively self-balancing, and won't lead to imbalanced trees. See the question above for what balance does (and why it's called "balance").
Does a balance operation recompress files?
No. Balance moves entire file extents and does not change their contents. If you want to recompress files, use btrfs filesystem defrag with the -c option.
Balance does a defragmentation, but not on a file level rather on the block group level. It can move data from less used block groups to the remaining ones, eg. using the usage balance filter.
Do I need to run a balance regularly?
In general usage, no. A full unfiltered balance typically takes a long time, and will rewrite huge amounts of data unnecessarily. You may wish to run a balance on metadata only (see Balance_Filters) if you find you have very large amounts of metadata space allocated but unused, but this should be a last resort. At some point, this kind of clean-up will be made an automatic background process.
How do I undelete a file?
There are three possible approaches to this:
- Restore from your backups (by far the easiest and most reliable method).
- Unmount the filesystem as soon as possible, and use btrfs-find-root and btrfs restore to find an earlier metadata root and restore the files.
- Unmount the filesystem as soon as possible, and use a file-carving tool (say, PhotoRec or some other carving tool) to find data that looks like the file you deleted.
There is no simple undelete tool for btrfs, and it is likely there never will be. The second and third approaches are unreliable at best. It is far better to ensure that you have good working backups in the first place.
What is the difference between mount -o ssd and mount -o ssd_spread?
Mount -o ssd_spread is more strict about finding a large unused region of the disk for new allocations, which tends to fragment the free space more over time. Mount -o ssd_spread is often faster on the less expensive SSD devices. The default for autodetected SSD devices is mount -o ssd.
Will Btrfs be in the mainline Linux Kernel?
Btrfs is already in the mainline Linux kernel. It was merged on 9th January 2009, and was available in the Linux 2.6.29 release.
Does Btrfs run with older kernels?
v0.16 of (out-of-tree) Btrfs maintains compatibility with kernels back to 2.6.18. Kernels older than that will not work.
btrfs made it into mainline in 2.6.29, and development and bugfixes since then have gone directly into the main kernel. Backporting btrfs from a newer kernel to an earlier one may be a difficult process due to changes in the VFS or block layer APIs; there are no known projects or people doing this on a regular basis.
We strongly recommend that you keep up-to-date with the latest released kernels from kernel.org -- we try to maintain a list of sources that make that task easier for most major distributions.
How long will the Btrfs disk format keep changing?
The Btrfs disk format is not immutable. However, any changes that are made will not invalidate existing filesystems. If a new feature results in a change to the filesystem format, it will be implemented in a way which is both safe and compatible: filesystems made with the new feature will safely refuse to mount on older kernels which do not support the feature, and existing filesystems will not have the new feature enabled without explicit manual intervention to add the feature to the filesystem.
How do I upgrade to the 2.6.31 format?
The 2.6.31 kernel can read and write Btrfs filesystems created by older kernels, but it writes a slightly different format for the extent allocation trees. Once you have mounted with 2.6.31, the stock Btrfs in 2.6.30 and older kernels will not be able to mount your filesystem.
We don't want to force people into 2.6.31 only, and so the newformat code is available against 2.6.30 as well. All fixes will also be maintained against 2.6.30. For details on downloading, see the Btrfs source repositories.
Can I find out compression ratio of a file?
compsize takes a list of files on a btrfs filesystem and measures used compression types and effective compression ratio: https://github.com/kilobyte/compsize
There's a patchset http://thread.gmane.org/gmane.comp.file-systems.btrfs/37312 that extends the FIEMAP interface to return the physical length of an extent (ie. the compressed size).
The size obtained is not exact and is rounded up to block size (4KB). The real amount of compressed bytes is not reported and recorded by the filesystem (only the block count) in it's structures. It is saved in the disk blocks but solely processed by the compression code.
Can I change metadata block size without recreating the filesytem?
No, the value passed to mkfs.btrfs -n SIZE cannot be changed once the filesystem is created. A backup/restore is needed.
Note, that this will likely never be implemented because it would require major updates to the core functionality.
Why do I have "single" chunks in my RAID filesystem?
After mkfs with an old version of mkfs.btrfs, and writing some data to the filesystem,
btrfs filesystem df /mountpoint will show something like this:
Data, RAID1: total=3.08TiB, used=3.02TiB Data, single: total=8.00MiB, used=0.00B System, RAID1: total=3.88MiB, used=336.00KiB System, single: total=4.00MiB, used=0.00B Metadata, RAID1: total=4.19GiB, used=3.56GiB Metadata, single: total=8.00MiB, used=0.00B GlobalReserve, single: total=512.00MiB, used=0.00B
single chunks are perfectly normal, and are a result of the way that mkfs works. They are small, harmless, and will remain unused as the FS grows, so you won't risk any unreplicated data. You can get rid of them with a balance:
$ btrfs balance start -dusage=0 -musage=0 /mountpoint
(but don't do this immediately after running mkfs, as you may end up losing your RAID configuration and reverting to single mode -- put at least some data on the FS first).
This issue has been fixed in mkfs.btrfs since v4.2.
What is the GlobalReserve and why does 'btrfs fi df' show it as single even on RAID filesystems?
The global block reserve is last-resort space for filesystem operations that may require allocating workspace even on a full filesystem. An example is removing a file, subvolume or truncating a file. This is mandated by the COW model, even removing data blocks requires to allocate some metadata blocks first (and free them once the change is persistent).
The block reserve is only virtual and is not stored on the devices. It's an internal notion of Metadata but normally unreachable for the user actions (besides the ones mentioned above). For ease it's displayed as single.
The size of global reserve is determined dynamically according to the filesystem size but is capped at 512MiB. The value used greater than 0 means that it is in use.
What is the difference between -c and -p in send?
It's easiest to understand if you look at what receive does. Receive takes a stream of instructions, creates a new subvolume, and uses the instructions to modify that subvolume until it looks like the one being sent.
When you use -p, the receiver will take a snapshot of the corresponding subvol, and then use the send stream to modify it. So, if you do this:
btrfs sub snap -r /@A /snaps/@A.1 btrfs send /snaps/@A.1 | btrfs receive /backups btrfs sub snap -r /@A /snaps/@A.2 btrfs send -p /snaps/@A.1 /snaps/@A.2 | btrfs receive /backups
the second receive will first snapshot /backups/@A.1 to /backups/@A.2, and then use the information in the send stream to make it look like /snaps/@A.2.
When you use -c, then receiver will simply use the clone range ioctl (i.e. reflink copies) from data held in the corresponding subvol. So, doing this:
btrfs sub snap -r /@A /snaps/@A.1 btrfs send /snaps/@A.1 | btrfs receive /backups cp --reflink=always /@A/bigfile /@B/bigfile.copy btrfs sub snap -r /@B /snaps/@B.1 btrfs send -c /snaps/@A.1 /snaps/@B.1 | btrfs receive /backups
the second receive will start with a blank subvolume, and build it up from scratch. The send will send all of the metadata of @B.1, but will leave out the data for @B.1/bigfile, because it's already in the backups filesystem, and can be reflinked from there.
So if you use -c on its own (without -p), all of the metadata of the subvol is sent, but some of the data is assumed to exist on the far side. If you use -p, then some of the metadata is assumed to exist on the other side as well, in the form of an earlier subvol that can be snapshotted.
How do I recover from a "parent transid verify failed" error?
parent transid verify failed on 29360128 wanted 1486656 found 1486662
If the second two numbers (wanted 1486656 and found 1486662) are close together (within about 20 of each other), then mounting with
may help. If it's successful with a read-only mount, then try again without the
ro option, for a read-write mount.
If the usebackuproot doesn't work, then the FS is basically unrecoverable in its current state with current tools. You should use
btrfs restore to refresh your backups, and then restore from them.
What does "parent transid verify failed" mean?
"Parent transid verify failed" is the result of a failed internal consistency check of the filesystem's metadata.
The way that btrfs works -- CoW, or Copy-on-Write -- means that when any item of data or metadata is modified, it's not overwritten in place, but instead is written to some other location, and then the metadata which referenced it is updated to point at the new location. The changed metadata is also updated in the same way: a new copy is written elsewhere, and the things that referenced that item are updated. This process ultimately bubbles up all the btrfs trees, until only the superblock is left to be updated. The superblock(s) are the only things which are overwritten in place, and each one can be done atomically. This process ensures that if you follow the filesystem structures downwards from the superblock, you will always see a completely consistent structure for the filesystem.
Now, if this cascade of writes happened for every single update, btrfs would be orders of magnitude slower than it is. So, for performance reasons, the updates are cached in memory, and every so often the in-memory data is written to disk in a consistent way, and the superblocks on the disk are then updated to point at the new structures. This update of the superblocks is called a transaction in btrfs. Each transaction is given a number, called the transid (or sometimes generation). These transids increase by one on every transaction.
Now, when writing metadata, every metadata block has the current transid embedded in it. The upwards-cascading effect of the CoW process means that in the metadata trees, the transid of every block in the metadata must be lower than or equal to the transid of the block above it in the tree. This is a fundamental property of the design of the filesystem -- if it's not true, then something has broken very badly, and the FS rightly complains (and may prevent the FS from being mounted at all).
How does this happen?
The fundamental problem identified by transid failures is that some metadata block is referenced by an older metadata block. This can happen in a couple of ways.
- The superblock got written out before it should have been, before the trees under it were completely consistent (i.e. before all the cached metadata in RAM was flushed to permanent storage), and then the FS was interrupted in writing out the remainder of the metadata. To ensure that this is the case, btrfs uses a feature called a barrier to impose an ordering on the writes to the hardware. The barrier should ensure that the superblock can't get written out before all the other metadata has reached the permanent storage. However, (a) the hardware may not implement barriers correctly, or at all -- most common on USB storage devices, but has also been seen directly on some drives and controllers, or (b) there might be a bug in btrfs (unlikely, because it's tested quite thoroughly using tools written specifically to identify this problem).
- The data written to disk was modified (corrupted) in RAM before it was written out to disk. If the transid in a metadata block was modified, or a pointer to another block was changed in a way that happened to hit another metadata block (but an older one, or just the wrong one), then you can see transid failures. This could happen with a faulty computing device (bad RAM, CPU, PSU or power regulation), or it could happen as a result of a bug in some part of the kernel which allowed another part of the kernel to write on data it shouldn't have written to.
If you see transids close together, then it's probably the first case above, and mounting with
-o usebackuproot may be able to step backwards and use an earlier superblock with a slightly older, non-broken set of trees in it.
If you see wildly different transids, or a transid which is hugely out of proportion, then you probably have the second case, and there's little hope for in-place recovery (transids don't increase very fast -- once every 30s is the normal behaviour, although it might be somewhat faster than that with lots of snapshotting).
What is a subvolume?
A subvolume is like a directory - it has a name, there's nothing on it when it is created, and it can hold files and other directories. There's at least one subvolume in every Btrfs filesystem, the top-level subvolume.
As well as being like directories, subvolumes can be mounted independently of the rest of the filesystem. They are also the unit of snapshotting: you can make an atomic snapshot of a single subvolume, but not a whole tree of them; you can't make an atomic snapshot of anything smaller than a subvolume (like, say, a single directory).
Subvolumes can be given capacity limits, through the qgroups/quota facility, but otherwise share the single storage pool of the whole btrfs filesystem. They may even share data between themselves (through deduplication or snapshotting).
How do I find out which subvolume is mounted?
A specific subvolume can be mounted by -o subvol=/path/to/subvol option, but currently it's not implemented to read that path directly from /proc/mounts. If the filesystem is mounted via a /etc/fstab entry, then output of mount command will show the subvol path, as it reads it from /etc/mtab.
Generally working way to read the path, like for bind mounts, is from /proc/self/mountinfo
27 21 0:19 /subv1 /mnt/ker rw,relatime - btrfs /dev/loop0 rw,space_cache ^^^^^^
What is a snapshot?
A snapshot is a frozen image of all the files and directories of a subvolume. For example, if you have two files ("a" and "b") in a subvolume, you take a snapshot and you delete "b", the file you just deleted is still available in the snapshot you took. The great thing about Btrfs snapshots is you can operate on any files or directories vs lvm when it is the whole logical volume.
Note that a snapshot is not a backup: Snapshots work by use of btrfs's copy-on-write behaviour. A snapshot and the original it was taken from initially share all of the same data blocks. If that data is damaged in some way (cosmic rays, bad disk sector, accident with dd to the disk), then the snapshot and the original will both be damaged. Snapshots are useful to have local online "copies" of the filesystem that can be referred back to, or to implement a form of deduplication, or to fix the state of a filesystem for making a full backup without anything changing underneath it. They do not in themselves make your data any safer.
Since backup from tape are a pain here is the thoughts of a lazy sysadm that create a home directory as a Btrfs file system for their users, lets try some fancy net attached storage ideas.
- Then there could be a snaphot every 6 hours via cron
- Then there could be a snaphot every 6 hours via cron
The logic would look something like this for rolling 3 day rotation that would use cron @ midnight
- rename /home_today_00, /home_backday_1
- create a symbolic link for /home_backDay_00 that points to real dir of /home_backday_1
- rename /home_today_06, /home_backDay_06 , Need to do this for all hours (06..18)
- delete the /home_backday_3
- rename /home_backday_2 to /home_backday_3 day
- rename /home_backday_1 to /home_backday_2 day
Automated rolling snapshots are easily done with a script like Marc MERLIN's btrfs-snaps script
Can I mount subvolumes with different mount options?
The generic mount options can be different for each subvolume, see the list below. Btrfs-specific mount options cannot be specified per-subvolume, but this will be possible in the future (a work in progress).
Generic mount options:
- nodev, nosuid, ro, rw, and probably more. See section FILESYSTEM INDEPENDENT MOUNT OPTIONS of man page mount(8).
Yes for btrfs-specific options:
- subvol or subvolid
- compress/compress-force, autodefrag, inode_cache, ...
- the options affecting the whole filesystem like space_cache, discard, ssd, ...
Interaction with partitions, device managers and logical volumes
Btrfs has subvolumes, does this mean I don't need a logical volume manager and I can create a big Btrfs filesystem on a raw partition?
There is not a single answer to this question. Here are the issues to think about when you choose raw partitions or LVM:
- raw partitions are slightly faster than logical volumes
- btrfs does write optimisation (sequential writes) across a filesystem
- subvolume write performance will benefit from this algorithm
- creating multiple btrfs filesystems, each on a different LV, means that the algorithm can be ineffective (although the kernel will still perform some optimization at the block device level)
- Online resizing and relocating the filesystem across devices:
- the pvmove command from LVM allows filesystems to move between devices while online
- raw partitions can only be moved to a different starting cylinder while offline
- raw partitions can only be made bigger if there is free space after the partition, while LVM can expand an LV onto free space anywhere in the volume group - and it can do the resize online
- subvolume/logical volume size constraints
- LVM is convenient for creating fixed size logical volumes (e.g. 10MB for each user, 20GB for each virtual machine image, etc)
- per-subvolume quota can be set using the qgroups quota feature
Based on the above, all of the following are valid strategies, depending upon whether your priority is performance or flexibility:
- create a raw partition on each device, and create btrfs on top of the partition (or combine several such partitions into btrfs raid1)
- create subvolumes within btrfs (e.g. for /home/user1, /home/user2, /home/media, /home/software)
- in this case, any one subvolume could grow to use up all the space, leaving none for other subvolumes
- create a single volume group, with two logical volumes (LVs), each backed by separate devices
- create a btrfs raid1 across the two LVs
- create subvolumes within btrfs (e.g. for /home/user1, /home/user2, /home/media, /home/software)
- in this case, any one subvolume could grow to use up all the space, leaving none for other subvolumes
- however, it performs well and is convenient
- create a single volume group, create several pairs of logical volumes
- create several btrfs raid1 filesystems, each spanning a pair of LVs
- mount each filesystem on a distinct mount point (e.g. for /home/user1, /home/user2, /home/media, /home/software)
- in this case, each mount point has a fixed size, so one user can't use up all the space
Does the Btrfs multi-device support make it a "rampant layering violation"?
Yes and no. Device management is a complex subject, and there are many different opinions about the best way to do it. Internally, the Btrfs code separates out components that deal with device management and maintains its own layers for them. The vast majority of filesystem metadata has no idea there are multiple devices involved.
Many advanced features such as checking alternate mirrors for good copies of a corrupted block are meant to be used with RAID implementations below the FS.
What are the differences among MD-RAID / device mapper / btrfs raid?
Note: device here means a block device -- often a partition, but it might also be something like a full disk, or a DRBD network device. It is possible with all of the descriptions below, to construct a RAID-1 array from two or more devices, and have those devices live on the same physical drive. This configuration does not offer any form of redundancy for your data.
MD-RAID supports RAID-0, RAID-1, RAID-10, RAID-5, and RAID-6.
MD-RAID operates directly on the devices. RAID-1 is defined as "data duplicated to all devices", so a raid with three 1 TB devices will have 1TB of usable space but there will be 3 copies of the data.
Likewise, RAID-0 is defined as "data striped across all devices", so a raid with three 1 TB devices will have 3 TB usable space, but to read/write a stripe all 3 devices must be written to or read from, as part of the stripe will be on each disk. This offers additional speed on slow devices, but no additional redundancy benefits at all.
RAID-10 requires at least 4 devices, and is constructed as a stripe across 2 mirrors. So a raid with four 1 TB devices yields 2 TB usable and 2 copies of the data. A raid with 6 × 1 TB devices yields 3 TB usable data with 2 copies of all the data (3 mirrors of 1 TB each, striped)
btrfs supports RAID-0, RAID-1, and RAID-10. As of Linux 3.9, btrfs also supports RAID-5 and RAID-6 although that code is still experimental.
btrfs combines all the devices into a storage pool first, and then duplicates the chunks as file data is created. RAID-1 is defined currently as "2 copies of all the data on different devices". This differs from MD-RAID and dmraid, in that those make exactly n copies for n devices. In a btrfs RAID-1 on three 1 TB devices we get 1.5 TB of usable data. Because each block is only copied to 2 devices, writing a given block only requires exactly 2 devices to be written to; reading can be made from only one.
RAID-0 is similarly defined, with the stripe split across as many devices as possible. 3 × 1 TB devices yield 3 TB usable space, but offers no redundancy at all.
RAID-10 is built on top of these definitions. Every stripe is split across to exactly 2 RAID-1 sets and those RAID-1 sets are written to exactly 2 devices (hence 4 devices minimum). A btrfs RAID-10 volume with 6 × 1 TB devices will yield 3 TB usable space with 2 copies of all data.
An archive of the btrfs mailing list describes how RAID-5/6 is implemented in btrfs.
Case study: btrfs-raid 5/6 versus MD-RAID 5/6
(content comes from )
The advantage in btrfs-raid 5/6 is that unlike MD-RAID, btrfs knows what blocks are actually used by data/metadata, and can use that information in a rebuild/recovery situation to only sync/rebuild the actually used blocks on a re-added or replacement device, skipping blocks that were entirely unused/empty in the first place.
MD-RAID can't do that, because it tries to be a filesystem agnostic layer that doesn't know nor care what blocks on the layers above it were actually used or empty. For it to try to track that would be a layering violation and would seriously complicate the code and/or limit usage to only those filesystems or other layers above that it supported/understood/could-properly-track.
A comparable relationship exists between a ramdisk (comparable to MD-RAID) and tmpfs (comparable to btrfs) -- the first is transparent and allows the flexibility of putting whatever filesystem or other upper layer on top, while the latter is the filesystem layer itself, allowing nothing else above it. But the ramdisk/tmpfs case deals with memory emulating block device storage, while the MD-RAID/btrfs case deals with multiple block devices emulating a single device. In both cases each has its purpose, with the strengths of one being the limitations of the other, and you choose the one that best matches your use case.
To learn more about using Raid 5 and Raid 6 with btrfs, see the RAID56 page.
About the project
What is your affiliation with WinBtrfs
[CRFS] is a network file system protocol. It was designed at around the same time as Btrfs. Its wire format uses some Btrfs disk formats and crfsd, a CRFS server implementation, uses Btrfs to store data on disk. More information can be found at http://oss.oracle.com/projects/crfs/ and http://en.wikipedia.org/wiki/CRFS.
Will Btrfs become a clustered file system
No. Btrfs's main goal right now is to be the best non-cluster file system.
If you want a cluster file system, there are many production choices that can be found in the Distributed file systems section on Wikipedia. Keep in mind that each file system has their own benefits or limitations, so find the best fit for your environment.
The closest cluster file system that uses Btrfs as its underlying file system is Ceph
I've heard that ...
... files cannot be deleted once the filesystem is full
Not true. Even though the deletion requires some extra space (because of the copy-on-write design), there is a global reserve that is used for this type of operation.
Historically (before about 3.16 or so), it was indeed possible to reach a situation where files could not be deleted. This was due to bugs in the handling of space. However, many of these bugs have been fixed, and current evidence (May 2015) suggests that if there are remaining bugs in this area, they are very, very hard to hit.
... btrfs is broken by design (aka. Edward Shishkin's "Unbound(?) internal fragmentation in Btrfs")
Mailing list thread Unbound(?) internal fragmentation in Btrfs (2010)
Short answer: not a design problem but a bug (see 1) that was fixed
The core of the argument is that the b-tree implementation (from year 2010) does not follow the assumptions found in classic literature about b-trees and can lead to underutilization of the space. There are two issues raised: one about variable record length (in particular the file data inlined in the leaves), and a second issue about an inefficient b-tree balancing algorithm (see 3).
Inline extents are not split across leaves and are limited in size of one leaf. This was a decision made based on experience with reiserfs, where inline items were split, causing increased code complexity (and bugs) without much gain (see 2).
To completely avoid the theoretical worst case, you can turn off the file inlining by mounting with max_inline=0 (see 2).
In practice, the worst case means that an inline file extent may occupy one leaf block (which is more limited to the memory page size, usually 4k). This leads to the same space efficiency as in other filesystems, where a small file occupies a full block (see 4).
Discussed on LWN: