[fedora-virt] I/O errors in guest when cache=none, qcow2 on Btrfs

[Chris Murphy]

Could someone check out this bug, and see if it needs upstream
attention? It's currently set to kernel; but I have no idea if it
would be libvirt or qemu upstreams to make aware of this.

The gist is that on Fedora 21 and 22, when virtio blkc + cache=none +
qcow2 on Btrfs, the guest OS (regardless of the file system it uses)
starts to experience many I/O errors. If the qcow2 is on XFS, or if
cache=writeback or writethrough, the problem doesn't happen.

Regression testing shows the problem does not happen on Fedora 20's
versions of libvirt and qemu, even with newer kernels. So maybe it's
not a kernel problem, or maybe it's a collision of kernel and libvirt
or qemu problem, hence the inquiry. Upstream Btrfs is aware of this
bug and are looking into it.

The fallout of the bug is that gnome-boxes experiences problems since
they're currently using cache=none by default (and there's no way to
change this in the GUI) when qcow2 is on Btrfs.

blk_update_request: I/O error, dev vda, sector XXXXXXXX when qcow2 is on Btrfs
https://bugzilla.redhat.com/show_bug.cgi?id=1204569

-- 
Chris Murphy