Mailing List Archive

On 31 Jan 2024 08:20, Paul Leiber wrote:
> Hi Xen users,
>
> just a short info, perhaps this is relevant for you as well: I noticed
> that smartmontools/smartd silently stopped working in my Debian Bookworm
> setup a couple of months ago, presumably after an update (couldn't
> pinpoint the specific, right now, I'm on version 7.3-1+b1). Contrary to
> my expectation, I wouldn't have been notified via e-mail if there had
> been errors in the SMART logs. The cause is that there is a new check in
> the smartd.service file "ConditionVirtualization=no", which checks if
> smartd is running in a vm. If it detects a vm, smart doesn't start.
> There is a message in the logs if you look for it, but to look for it,
> you first have to have a reason to do so.
>
> Commenting out the line solves this issue.
>
> There is a discussion on GitHub on this:
>
> https://github.com/smartmontools/smartmontools/issues/62
>
> Best regards,
>
> Paul
>

Hi,

for completeness, this bug/feature affects dom0s but also any domU with
a passthrough'ed disk controller, with whatever virt platform btw.
So if you use any systemd-based distro as a file server VM and expect to
use SMART, take care.

smartmontools should have no business knowing it's running in a VM or
not, it should only care if it's handling "real" disk controllers.
To me, the systemd "ConditionVirtualization" is misused here, the real
condition for such service should be "ConditionHasDiskController".

There is also a related systemd bug, the executable
"systemd-detect-virt" (sdv) wrongly detecting dom0 as a VM.
This is the executable which sets the "ConditionVirtualization" when it
detects it's running in a VM.
Sure, dom0 -is- a VM in Xen, but for the (supposed) purpose of sdv, it
simply shouldn't be detected as such.

Bug report on systemd's github :
https://github.com/systemd/systemd/issues/28113

Debian BTS bug report :
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1038901

I'm not aware of such problems in other init systems or services.

--
++
zithro / Cyril

On Sat, Mar 16, 2024 at 7:16?PM zithro <slack@rabbit.lu> wrote:

> On 31 Jan 2024 08:20, Paul Leiber wrote:
> > Hi Xen users,
> >
> > just a short info, perhaps this is relevant for you as well: I noticed
> > that smartmontools/smartd silently stopped working in my Debian Bookworm
> > setup a couple of months ago, presumably after an update (couldn't
> > pinpoint the specific, right now, I'm on version 7.3-1+b1). Contrary to
> > my expectation, I wouldn't have been notified via e-mail if there had
> > been errors in the SMART logs. The cause is that there is a new check in
> > the smartd.service file "ConditionVirtualization=no", which checks if
> > smartd is running in a vm. If it detects a vm, smart doesn't start.
> > There is a message in the logs if you look for it, but to look for it,
> > you first have to have a reason to do so.
> >
> > Commenting out the line solves this issue.
> >
> > There is a discussion on GitHub on this:
> >
> > https://github.com/smartmontools/smartmontools/issues/62
> >
> > Best regards,
> >
> > Paul
> >
>
> Hi,
>
> for completeness, this bug/feature affects dom0s but also any domU with
> a passthrough'ed disk controller, with whatever virt platform btw.
> So if you use any systemd-based distro as a file server VM and expect to
> use SMART, take care.
>
> smartmontools should have no business knowing it's running in a VM or
> not, it should only care if it's handling "real" disk controllers.
> To me, the systemd "ConditionVirtualization" is misused here, the real
> condition for such service should be "ConditionHasDiskController".
>
> There is also a related systemd bug, the executable
> "systemd-detect-virt" (sdv) wrongly detecting dom0 as a VM.
> This is the executable which sets the "ConditionVirtualization" when it
> detects it's running in a VM.
> Sure, dom0 -is- a VM in Xen, but for the (supposed) purpose of sdv, it
> simply shouldn't be detected as such.
>
> Bug report on systemd's github :
> https://github.com/systemd/systemd/issues/28113
>
> Debian BTS bug report :
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1038901
>
> I'm not aware of such problems in other init systems or services.
>

Thinking about this a bit more -- if you'd passed through a physical disk
controller to a VM via PCI passthrough, you'd also presumably want the
SMART tools to be monitoring things. What they actually need is a
"is-real-device <dev>" function somehow; but "has-real-hardware" might be a
proxy, that would work both for dom0 systems and for systems with
pass-through; if we could get a reliable test.

-George