Rusty Bird

Oh right, so the problem also affects the Fedora 41 templates on R4.2. It's just not visible, because none of the preinstalled packages currently have an update from RPM Fusion....

Sorry for chiming in late. I had assumed that this functionality is intended to be optional (opt-in or opt-out) but it looks like it's always-on. Isn't the effect a bit...

My understanding is that Btrfs used to be more prone to checksum errors with direct I/O mainly because direct I/O exercises substantially different parts of the Btrfs codebase that had...

>> ["misbehaving (...) application modifies a buffer in-flight when doing an O_DIRECT write"](https://lore.kernel.org/linux-btrfs/c3cc0815c5756d07201c57063f3759250f662c77.1600961206.git.josef@toxicpanda.com/). In our case the application is xen-blkback on a loop device > it turned out that using...

> > Looks like the untrusted qube side can cause checksum errors on the trusted dom0 side. I was hoping that the dom0 backend prevents this. > Is this a...

IMHO we should indeed disable direct I/O in the not-script at the moment. @DemiMarie @marmarek It's difficult to figure out who's technically wrong here in this stack of blkfront-on-Windows, blkback,...

Do we know that it is always safe for XFS and ext4 though? That Btrfs commit said it's simply application misbehavior to modify the buffer during a direct I/O write....

The change was cherry-picked for the [R4.2 branch](https://github.com/QubesOS/qubes-linux-utils/commits/release4.2/) three weeks ago, but the qubes-utils package version hasn't been bumped yet. @marmarek > I think I just experienced disk corruption (recoverable,...

> > > I think I just experienced disk corruption [...] > > > > This happened with Windows VMs both times? > > Linux AppVMs. You might want to...

Not sure what's going on with SELinux here, but: > sudo mount /dev/qubes_dom0/vm-MYQUBE-private /mnt [Never mount a VM volume in dom0.](https://forum.qubes-os.org/t/possible-to-run-locate-mlocate-in-dom0-to-index-appvms/9299/6) Do it in a DisposableVM instead (ideally based on...