[Pkg-virtualbox-devel] Bug#874105: virtualbox: Virtualbox seems to conflict with BTRFS when trying to install ReactOS 0.4.6

Gianfranco Costamagna locutusofborg at debian.org
Mon Sep 4 15:34:05 UTC 2017


control: tags -1 unreproducible
control: severity -1 important

On Sun, 03 Sep 2017 11:05:39 +0200 Svein Engelsgjerd <waxhead at dirtcellar.net> wrote:
> Package: virtualbox
> Version: 5.1.26-dfsg-2
> Severity: grave
> Justification: causes non-serious data loss

to be confirmed, and not sure if it is a vbox issue

> Dear Maintainer,
> 
>    * What led up to the situation?
> I wanted to try to install ReactOS 0.4.6. My rootfs is btrfs with metadata=raid10 and data=raid1
> While VirtualBox was running dmesg showed a constant stream of csum failed root 5 ino XXX off XXX csum XXX expected csum XXX mirror 1 messages.
> Note that the mirror was always 1.
> 
>    * What exactly did you do (or not do) that was effective (or
>      ineffective)?
> I stopped the VM that was running the ReactOS installer. All the errors stopped
> I ran a scrub on my btrfs filesystem afterwards. No errors detected.
> I checked smart data for my SSD's No errors detected, no sign of wear.
> I tried various file operations (md5sum, sha1sum, cp, hexeditor etc...) on the .vdi file - no corruption messages
> 
>    * What was the outcome of this action?
> While this could just as easily be a btrfs bug I suspect that  VirtualBox's vboxdrv is the cause here.
> I may be wrong about this, so my apologies if that is the case.
> Btrfs fixes the corruptions and appear to be in a good state as long as the VirutalBox VM is not running.
> qemu/kvm works fine. Using that instead.
> 
>    * What outcome did you expect instead?
> I expected no errors while running VirtualBox on top of my BTRFS rootfilessytem.
> 
>    * Other information:
> While I did not experience any dataloss I fear that this is fragile and *might* cause dataloss for some.
> this is the reason I classify this bug as grave. This is also causing me to avoid testing for a while
> until I get some feedback on this one.
> 
> I am aware that BTRFS have it's share of bugs, but I run with a minimal configuration that have not caused
> any issues for years (it has actually fixed a few silent data corruptions here and there).
> 
> So since my hardware and data checked out fine and the data is instact I think this could be a conflict
> between the btrfs and/or vboxdrv module. I am not skilled enoguh to find that out for myself.
> 
> BTW: The VirtualBox vdi file contains a VFAT filesystem (reactos uses that). This does NOT appear to be
> the issues as I have tested other stuff with VFAT without any issues.
> 

quoting upstream on vbox-dev irc channel:
[16:54:47] <aeichner> LocutusOfBorg: Michael just made me aware of https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=874105 . Very very unlikely that this is some conflict with VBox, I think this a bug in the direct I/O (O_DIRECT) code path of BTRFS leading to the corruption. The I/O pattern created  by the ReactOS installer inside VBox merely triggers it.
[16:55:15] <aeichner> LocutusOfBorg: The user might try to enable the "Use host I/O cache" checkbox in the VM config to check whether the issue goes away.

can you please check this and set back to serious the severity if you are still sure this is a vbox issue?
I admit, this is the first time people open such a bug, so I'm inclined to blame something else
(specially whit btrfs that is far from perfect)

G.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: OpenPGP digital signature
URL: <http://lists.alioth.debian.org/pipermail/pkg-virtualbox-devel/attachments/20170904/423ef599/attachment.sig>


More information about the Pkg-virtualbox-devel mailing list