Skip to contentSkip to conversation

/bin/df showing btrfs filesystem full

DISCUSSION Kenneth Topp 2 participants 2 messages Jan 6, 2020 Btrfs file system

lore
Share & tools
lore
#0 lore raw

Hi.

I have an issue were periodically a btrfs filesystem shows 100% utilized 0% available

# df -h /home
Filesystem           Size  Used Avail Use% Mounted on
/dev/mapper/cprt-50   30T   20T     0 100% /home

Then it goes back to normal as follows:

# df -h /home
Filesystem           Size  Used Avail Use% Mounted on
/dev/mapper/cprt-50   30T   20T  9.7T  67% /home

This filesystem was created on this kernel 5.4.6 that it's currently running. This filesystem went from 0tb used to 20tb used and would show this problem periodically as i was filling up the drive. There was no ENOSPACE issues, so I thought it was just related to the heavy writing, but now that the system is in regular service, it's still periodically "filling up". but again, the only symptom I can see is gnome and df showing the drive being full. nothing else indicates that the drive is full.

I have some other btrfs filesystems that didn't show any issues. They were created under earlier kernels, but with the same options. the other difference is this new filesystem is on top 4kn drives, where the others are all 512e.

Any advice would be welcome, for now I'm just ignoring the problem, and making sure my backups are good.

filesystem creation commands:

mkfs.btrfs -f -O no-holes -d single -m raid1 -L tm /dev/mapper/cprt-50 /dev/mapper/cprt-53

first time mounted was with this:
mount -o clear_cache,space_cache=v2 LABEL=tm /mnt

diagnostics commands:

# uname -a
Linux static.bllue.org 5.4.6-301.fc31.x86_64 #1 SMP Tue Dec 24 15:09:19 EST 2019 x86_64 x86_64 x86_64 GNU/Linux
# btrfs --version
btrfs-progs v5.4
# btrfs fi show
Label: 't2' uuid: ce50d21c-7727-4a53-b804-d02480643dfa

        Total devices 2 FS bytes used 640.00KiB
        devid    1 size 447.13GiB used 2.01GiB path /dev/mapper/cprt-30
        devid    2 size 447.13GiB used 2.01GiB path /dev/mapper/cprt-31

Label: 'btm' uuid: 0a5b42a7-0e39-48fa-be1f-4aa29bc323f2

        Total devices 2 FS bytes used 19.45TiB
        devid    1 size 14.55TiB used 9.75TiB path /dev/mapper/cprt-50
        devid    2 size 14.55TiB used 9.75TiB path /dev/mapper/cprt-53

# btrfs fi df /home # Replace /home with the mount point of your btrfs-filesystem
Data, single: total=19.45TiB, used=19.43TiB System, RAID1: total=32.00MiB, used=2.05MiB Metadata, RAID1: total=26.00GiB, used=25.73GiB GlobalReserve, single: total=512.00MiB, used=0.00B

#1 lore raw
Quoted 11 lines · Kenneth Topp

Hi.

I have an issue were periodically a btrfs filesystem shows 100% utilized 0% available

# df -h /home
Filesystem           Size  Used Avail Use% Mounted on
/dev/mapper/cprt-50   30T   20T     0 100% /home
Quoted 5 lines

Then it goes back to normal as follows:

# df -h /home
Filesystem           Size  Used Avail Use% Mounted on
/dev/mapper/cprt-50   30T   20T  9.7T  67% /home

A known bug if the metadata reservation reaches a threshold, only for v5.4 kernel.

The latest patchset trying to address this bug is here:
https://patchwork.kernel.org/project/linux-btrfs/list/?series=223921

Quoted 11 lines

This filesystem was created on this kernel 5.4.6 that it's currently running.  This filesystem went from 0tb used to 20tb used and would show this problem periodically as i was filling up the drive.   There was no ENOSPACE issues, so I thought it was just related to the heavy writing, but now that the system is in regular service, it's still periodically "filling up".  but again, the only symptom I can see is gnome and df showing the drive being full.  nothing else indicates that the drive is full.

Quoted 7 lines

I have some other btrfs filesystems that didn't show any issues.  They were created under earlier kernels, but with the same options.  the other difference is this new filesystem is on top 4kn drives, where the others are all 512e.

Any advice would be welcome, for now I'm just ignoring the problem, and making sure my backups are good.

Only statfs() call is affected, so your data is fine.

But quite some programs would use statfs() to determine if the fs is full, so there would be quite some inconvenience, but should be no data corruption.

Thanks,
Qu

Quoted 37 lines

filesystem creation commands:

mkfs.btrfs -f  -O no-holes -d single -m raid1 -L tm /dev/mapper/cprt-50 /dev/mapper/cprt-53

first time mounted was with this:
mount -o clear_cache,space_cache=v2 LABEL=tm /mnt

diagnostics commands:

#   uname -a
Linux static.bllue.org 5.4.6-301.fc31.x86_64 #1 SMP Tue Dec 24 15:09:19 EST 2019 x86_64 x86_64 x86_64 GNU/Linux
#   btrfs --version
btrfs-progs v5.4
#   btrfs fi show
Label: 't2'  uuid: ce50d21c-7727-4a53-b804-d02480643dfa Total devices 2 FS bytes used 640.00KiB
        devid    1 size 447.13GiB used 2.01GiB path /dev/mapper/cprt-30 devid    2 size 447.13GiB used 2.01GiB path /dev/mapper/cprt-31

Label: 'btm'  uuid: 0a5b42a7-0e39-48fa-be1f-4aa29bc323f2 Total devices 2 FS bytes used 19.45TiB
        devid    1 size 14.55TiB used 9.75TiB path /dev/mapper/cprt-50 devid    2 size 14.55TiB used 9.75TiB path /dev/mapper/cprt-53

#   btrfs fi df /home # Replace /home with the mount point of your btrfs-filesystem
Data, single: total=19.45TiB, used=19.43TiB
System, RAID1: total=32.00MiB, used=2.05MiB
Metadata, RAID1: total=26.00GiB, used=25.73GiB GlobalReserve, single: total=512.00MiB, used=0.00B