Re: delayed_refs has NO entry / btrfs_update_root:136: Aborting unused transaction(No space left).

[Christoph Anton Mitterer <calestyo@scientia.net>]

On Wed, 2019-03-20 at 10:59 +0100, Johannes Thumshirn wrote:
> First of all, have you tried a more recent kernel than the Debian
> kernels you referenced? E.g. Linus' current master or David's misc-
> next
> branch? Just so we don't try to hunt down a bug that's already fixed.

I haven't and that's a bit difficult for me unless it's packaged by the
distro (policy reasons).


Also giving out the image is a bit problematic as it's huge (8TB).

> Also if you can still reproduce the bug, please activate tracing in
> btrfs and send the trace output.

How would I do that?


In the meantime, I think I can reproduce it with fresh images so could
you try the following:

# truncate --size 1G image
# mkfs.btrfs image

# mount -o compress image /mnt
# cd /mnt

# # create some data e.g.:
# tar xaf /usr/src/linux-source-4.19.tar.xz
# cd
# umount /mnt

# losetup -r -f image
# mount -o compress /dev/loop0 /mnt

# find /mnt -type f -exec filefrag -v {} \;


And there your kernel log will explode ;-)

The culprit seems to be the device itself being read-only i.e.
losetup's -r, respectively blockdev --setro DEVICE which I've used
previously.

If you repeat the above from the losetup point, but with -r ...
everything works fine.
Haven't checked whether -o compress actually makes a difference.


Cheers,
Chris.