Revision as of 23:24, 9 February 2017

General topics

How do I report bugs and issues?

Please report bugs on Bugzilla on kernel.org (requires registration) setting the component to Btrfs, and report bugs and issues to the mailing list (linux-btrfs@vger.kernel.org; you are not required to subscribe). For quick questions you may want to join the IRC #btrfs channel on freenode (and stay around for some time in case you do not get the answer right away).

Please use btrfs-progs somewhere in the bug subject if you're reporting a bug for the userspace tools.

Never use the Bugzilla on "original" Btrfs project page at Oracle.

If you include kernel log backtraces in bug reports sent to the mailing list, please disable word wrapping in your mail user agent to keep the kernel log in a readable format.

Please attach files (like logs or dumps) directly to the bug and don't use pastebin-like services.

• Report a bug
• List open Bugzilla issues
• List all Bugzilla issues
• Josef's request to use bugzilla.kernel.org

Mount/unmount problems

I can't mount my filesystem, and I get a kernel oops!

First, update your kernel to the latest one available and try mounting again. If you have your kernel on a btrfs filesystem, then you will probably have to find a recovery disk with a recent kernel on it.

Second, try mounting with options -o recovery or -o ro or -o recovery,ro (using the new kernel). One of these may work successfully.

Finally, if and only if the kernel oops in your logs has something like this in the middle of it,

? replay_one_dir_item+0xb5/0xb5 [btrfs]
? walk_log_tree+0x9c/0x19d [btrfs]
? btrfs_read_fs_root_no_radix+0x169/0x1a1 [btrfs]
? btrfs_recover_log_trees+0x195/0x29c [btrfs]
? replay_one_dir_item+0xb5/0xb5 [btrfs]
? btree_read_extent_buffer_pages+0x76/0xbc [btrfs]
? open_ctree+0xff6/0x132c [btrfs]

then you should try using Btrfs-zero-log.

Filesystem can't be mounted by label

See the next section.

Only one disk of a multi-volume filesystem will mount

If you have labelled your filesystem and put it in /etc/fstab, but you get:

# mount LABEL=foo
mount: wrong fs type, bad option, bad superblock on /dev/sdd2,
      missing codepage or helper program, or other error
      In some cases useful info is found in syslog - try
      dmesg | tail  or so

or if one volume of a multi-volume filesystem fails when mounting, but the other succeeds:

# mount /dev/sda1 /mnt/fs
mount: wrong fs type, bad option, bad superblock on /dev/sdd2,
      missing codepage or helper program, or other error
      In some cases useful info is found in syslog - try
      dmesg | tail  or so
# mount /dev/sdb1 /mnt/fs
#

Then you need to ensure that you run a btrfs device scan first:

# btrfs device scan

This should be in many distributions' startup scripts (and initrd images, if your root filesystem is btrfs), but you may have to add it yourself.

My filesystem won't mount and none of the above helped. Is there any hope for my data?

Maybe. Any number of things might be wrong. The restore tool is a non-destructive way to dump data to a backup drive and may be able to recover some or all of your data, even if we can't save the existing filesystem.

Unsorted/uncategorized

Defragmenting a directory doesn't work

Running this:

# btrfs filesystem defragment ~/stuff

does not defragment the contents of the directory.

This is by design. btrfs fi defrag operates on the single filesystem object passed to it, e.g. a (regular) file. When ran on a directory, it defragments the metadata held by the subvolume containing the directory, and not the contents of the directory. If you want to defragment the contents of a directory, you have to use the recursive mode with the -r flag (see recursive defragmentation).

Compression doesn't work / poor compression ratios

First of all make sure you have passed "compress" mount option in fstab or mount command. If yes, and ratios are unsatisfactory, then you might try "compress-force" option. This way you make the btrfs to compress everything. The reason why "compress" ratios are so low is because btrfs very easily backs out of compress decision. (Probably not to waste too much CPU time on bad compressing data).

Copy-on-write doesn't work

You've just copied a large file, but still it consumed free space. Try:

# cp --reflink=always file1 file2

I get the message "failed to open /dev/btrfs-control skipping device registration" from "btrfs dev scan"

You are missing the /dev/btrfs-control device node. This is usually set up by udev. However, if you are not using udev, you will need to create it yourself:

# mknod /dev/btrfs-control c 10 234

You might also want to report to your distribution that their configuration without udev is missing this device.

How to clean up old superblock ?

The preferred way is to use the wipefs utility that is part of the util-linux package. Running the command with the device will not destroy the data, just list the detected filesystems:

# wipefs /dev/sda
offset               type
----------------------------------------------------------------
0x10040              btrfs   [filesystem]
                     UUID:  7760469b-1704-487e-9b96-7d7a57d218a5

To actually remove the filesystem use:

# wipefs -o 0x10040 /dev/sda
8 bytes [5f 42 48 52 66 53 5f 4d] erased at offset 0x10040 (btrfs)

ie. copy the offset number to the commandline parameter.

Related problem:

Long time ago I created btrfs on /dev/sda. After some changes btrfs moved to /dev/sda1.

Use wipefs as well, it deletes only a small portion of sda that will not interfere with the next partition data.

What if I don't have wipefs at hand?

There are three superblocks: the first one is located at 64K, the second one at 64M, the third one at 256GB. The following lines reset the magic string on all the three superblocks

# dd if=/dev/zero bs=1 count=8 of=/dev/sda seek=$((64*1024+64))
# dd if=/dev/zero bs=1 count=8 of=/dev/sda seek=$((64*1024*1024+64))
# dd if=/dev/zero bs=1 count=8 of=/dev/sda seek=$((256*1024*1024*1024+64))

If you want to restore the superblocks magic string,

# echo "_BHRfS_M" | dd bs=1 count=8 of=/dev/sda seek=$((64*1024+64))
# echo "_BHRfS_M" | dd bs=1 count=8 of=/dev/sda seek=$((64*1024*1024+64))
# echo "_BHRfS_M" | dd bs=1 count=8 of=/dev/sda seek=$((256*1024*1024*1024+64))

I get "No space left on device" errors, but df says I've got lots of space

First, check how much space has been allocated on your filesystem:

$ sudo btrfs fi show
Label: 'media'  uuid: 3993e50e-a926-48a4-867f-36b53d924c35
	Total devices 1 FS bytes used 61.61GB
	devid    1 size 133.04GB used 133.04GB path /dev/sdf

Note that in this case, all of the devices (the only device) in the filesystem are fully utilised. This is your first clue.

Next, check how much of your metadata allocation has been used up:

$ sudo btrfs fi df /mount/point
Data: total=127.01GB, used=56.97GB
System, DUP: total=8.00MB, used=20.00KB
System: total=4.00MB, used=0.00
Metadata, DUP: total=3.00GB, used=2.32GB
Metadata: total=8.00MB, used=0.00

Note that the Metadata used value is fairly close (75% or more) to the Metadata total value, but there's lots of Data space left. What has happened is that the filesystem has allocated all of the available space to either data or metadata, and then one of those has filled up (usually, it's the metadata space that does this). For now, a workaround is to run a partial balance:

$ sudo btrfs fi balance start -dusage=5 /mount/point

Note that there should be no space between the -d and the usage. This command will attempt to relocate data in empty or near-empty data chunks (at most 5% used, in this example), allowing the space to be reclaimed and reassigned to metadata.

If the balance command ends with "Done, had to relocate 0 out of XX chunks", then you need to increase the "dusage" percentage parameter till at least one chunk is relocated. More information is available elsewhere in this wiki, if you want to know what a balance does, or what options are available for the balance command.

btrfs scrub will not start nor cancel

As per Marc's blog on how to fix btrfs cancel: Problem:

gargamel:~# btrfs scrub start -d /dev/mapper/dshelf1
ERROR: scrub is already running.
To cancel use 'btrfs scrub cancel /dev/mapper/dshelf1'.
gargamel:~# btrfs scrub status  /dev/mapper/dshelf1
scrub status for 6358304a-2234-4243-b02d-4944c9af47d7
        scrub started at Tue Apr  8 08:36:18 2014, running for 46347 seconds
        total bytes scrubbed: 5.70TiB with 0 errors
gargamel:~# btrfs scrub cancel  /dev/mapper/dshelf1
ERROR: scrub cancel failed on /dev/mapper/dshelf1: not running

Fix:

gargamel:~# perl -pi -e 's/finished:0/finished:1/' /var/lib/btrfs/*

Verification:

gargamel:~# btrfs scrub status  /dev/mapper/dshelf1
scrub status for 6358304a-2234-4243-b02d-4944c9af47d7
        scrub started at Tue Apr  8 08:36:18 2014 and finished after 46347 seconds
        total bytes scrubbed: 5.70TiB with 0 errors
gargamel:~# btrfs scrub start -d /dev/mapper/dshelf1
scrub started on /dev/mapper/dshelf1, fsid 6358304a-2234-4243-b02d-4944c9af47d7 (pid=24196)

I cannot delete an empty directory

First case, if you get:

• rmdir: failed to remove ‘emptydir’: Operation not permitted

then this is probably because "emptydir" is actually a subvolume.

You can check whether this is the case with:

# btrfs subvolume list -a /mountpoint

To delete the subvolume you'll have to run:

# btrfs subvolume delete emptydir

Second case, if you get:

• rmdir: failed to remove ‘emptydir’: Directory not empty

then you may have an empty directory with a non-zero i_size.

You can check whether this is the case with:

# stat -c %s emptydir
3196         <-- unexpected non-zero size

Running "btrfs check" on that (unmounted) filesystem will confirm the issue and list other problematic directories (if any).

You will get a similar output (excerpt):

checking fs roots
root 5 inode 557772 errors 200, dir isize wrong
root 266 inode 24021 errors 200, dir isize wrong
...

Such errors should be fixable with "btrfs check --repair" provided you run a recent enough version of btrfs-progs.

Note that "btrfs check --repair" should not be used lightly as in some cases it can make a problem worse instead of fixing anything.

Deciphering error messages from syslog

balance will reduce metadata integrity, use force if you want this

This means that conversion will remove a degree of metadata redundancy, for example when going from profile raid1 or dup to single

unable to start balance with target metadata profile 32

This means that a conversion has been attempted from profile raid1 to dup with btrfs-progs earlier than version 4.7.

parent transid verify failed

Example:

parent transid verify failed on 4316004352 wanted 289 found 283

• 4316004352 is the byte offset of the metadata block
• 289 expected generation of the block
• 283 generation found in the block

Under normal circumstances the generation numbers must match. A mismatch can be caused by a lost write after a crash (ie. a dangling block "pointer"; software bug, hardware bug), misdirected write (the block was never written to that location; software bug, hardware bug).