Re: NetBSD syzbot broken?

["'Aleksandr Nogikh' via syzkaller-netbsd-bugs" <syzkaller-netbsd-bugs%googlegroups.com@localhost>]

On Thu, Apr 11, 2024 at 3:36 AM Taylor R Campbell <riastradh%netbsd.org@localhost> wrote:
>
> > Date: Wed, 3 Apr 2024 20:05:54 +0200
> > From: Aleksandr Nogikh <nogikh%google.com@localhost>
> >
> > That is indeed due to some problem with the host qemu, but that makes it
> > even more mysterious. I didn't update anything on the host and I think
> > nobody else did that either. And it's now unfortunately impossible to
> > pinpoint the exact moment when it began to fail with "failed to create the
> > VM Instance:" -- syzbot cleans up old logs and in some of these logs there
> > are actually ordinary NetBSD compilation errors. Could a NetBSD host have
> > updated the qemu package itself?
>
> Not sure what you mean by this -- is syzbot running NetBSD guests
> under a NetBSD host?
>
> If so, this message is pretty weird:
>
>    Could not access KVM kernel module: No such file or directory
>    qemu-system-x86_64: failed to initialize kvm: No such file or directory
>
> I would expect to see something about nvmm, not kvm, if that were the
> case; qemu under a NetBSD host has never supported `-accel kvm' or
> anything like that -- only `-accel nvmm'.  If there haven't been any
> recent changes about this, surely this would have failed all along
> because NetBSD as a host has never had kvm.
>
> But maybe I misunderstood your question?  I'm fuzzy on how syzbot
> operates at a high level -- I've only dug into the issues it reports.
> Certainly I'd be astonished if a NetBSD _guest_ changed any kind of
> package installation on the _host_.


Ah, I'm sorry, I have confused it with the FreeBSD setup on syzbot.
NetBSD fuzzing indeed runs on a Linux host.

>
> > In the syzkaller code, we do have an option to distinguish whether we need
> > to explicitly specify the raw format or not:
> > https://github.com/google/syzkaller/blob/51c4dcff83b0574620c280cc5130ef59cc4a2e32/vm/qemu/qemu.go#L459
> >
> > We can set UseNewQemuImageOptions=true for NetBSD and it should hopefully
> > be working again. But I still don't understand why it has popped up..
>
> It seems like if a file is supposed to be interpreted as a raw image,
> it would be prudent to say `format=raw'.  How do I find how the image
> (/syzkaller/managers/ci2-netbsd/latest.tmp/image) is created and what
> format it is supposed to be in?

I think it's easier to just use the new arguments format here. I've
sent a PR: https://github.com/google/syzkaller/pull/4672

But the main problem is that it tries to use kvm acceleration, but
fails to: `qemu-system-x86_64: failed to initialize kvm: No such file
or directory`.

It's still unclear whether it didn't use kvm for netbsd builds before,
but somehow began to, or for some reason we lost the nested
virtualization support on our GCE instance.

>
> Conceivably if something changed about that somehow (e.g., maybe
> NetBSD newfs or mkimage started doing putting in different formatting
> that might confuse qemu's format detection) that could break this.
> But it seems unlikely.

-- 
You received this message because you are subscribed to the Google Groups "syzkaller-netbsd-bugs" group.
To unsubscribe from this group and stop receiving emails from it, send an email to syzkaller-netbsd-bugs+unsubscribe%googlegroups.com@localhost.
To view this discussion on the web visit https://groups.google.com/d/msgid/syzkaller-netbsd-bugs/CANp29Y6JpzTTJ%3DUSmV39hLdB6%2BrwDs8ehTH_dm7raxmNuy76pw%40mail.gmail.com.