[Pkg-xen-devel] Bug#919460: Bug#919460: xen: disk I/O problems on stretch PV guest restores after security update

[Sergio Gelato]

* Hans van Kranenburg [2019-01-17 00:41:39 +0100]:
> > After "reboot -f"ing some of the affected domUs (which made them
> > functional again), I rebooted the dom0. This time all domUs were
> > restored normally. (Of course those that still had their filesystems
> > mounted read-only stayed that way.)
> > 
> > Is anyone else seeing this?
> 
> The usual questions here would be like "can you reproduce the issue"
> etc... Because if you consistently can cause the problem to happen,
> you're in a positition to start trying things.

Since this only happened on the reboot immediately following the Xen
upgrade, in order to reproduce it I would need to either try it on
another system or downgrade Xen on my test system and upgrade it again.
The latter doesn't look like a good use of my limited time.

I will eventually want to upgrade my production dom0's. Any domU's that
I care about on these systems will be shut down prior to the reboot
(that was the main lesson for me from this test), but I can leave a
canary behind and see what happens to it. So eventually I should have
an idea whether this is reproducible across systems. This will take
a few weeks, though: plenty of time for others to run into the same
problem if it's reproducible.

I've been wondering whether the observed behaviour might be explained in
terms of the specific changes made by the latest security patches. I
don't see any such obvious explanation myself, but maybe to an expert
it would leap out; that's partly why I reported this. If this is the
case, then maybe there is no fix other than documenting the issue in
release notes.

It's also conceivable that the bug is in that domU kernel rather than in
Xen. I might set up canary domU's with 4.9.144 (stretch-proposed-updates)
and/or 4.19.x.