On Fri, Jun 01, 2018 at 05:49:24PM +0200, Ján Tomko wrote:
> On Fri, Jun 01, 2018 at 05:05:30PM +0200, Kashyap Chamarthy wrote:
> > On Fri, Jun 01, 2018 at 02:11:12PM +0100, Daniel P. Berrangé wrote:
> > > On Fri, Jun 01, 2018 at 02:46:40PM +0200, Peter Krempa wrote:
> > > > On Fri, Jun 01, 2018 at 13:32:20 +0100, Daniel Berrange wrote:
> >
> > [...]
> >
> > > > > The reason the config file documents 'root' is because that is what
> > > > > configure defaults to. If you pass --with-qemu-user to configure,
> > > > > we don't update the config file example though, and perhaps we should.
> >
> > Thanks for that 'configure' context.
> >
> > > > > Alternatively should we make configure defualt to 'qemu' instead of
> > > > > 'root', since it is generally considered insane to run QEMU as root.
> > > >
> > > > But user 'qemu' is not by default present on all systems. Even the
> > > > libvirt.spec file creates the account.
> > >
> > > Yes, that's the reason configure defaults to 'root', but I really hate
> > > the fact that we default to a config that no one should ever run in
> > > practice.
> > >
>
> Nobody should be using ./configure with the defaults in practice anyway.
> For real world usage, people should go through their distribution's
> packages which should have specified --with-qemu-user.
>
> > > We could check for existance of 'qemu' in configure and complain if
> > > it is missing, but that's painful in itself as it is valid to build
> > > on a host without the user, as long as it exists at runtime.
> > >
>
> We probe for many things that are only relevant at runtime (paths to
> various binaries come to mind). For many things we try to be smart and
> try to guess what the user wants and turn off configure options
> when dependencies for drivers and features are missing, instead of
> building everything by default and letting the user disable stuff
> manually if they don't want the dependencies.
>
> Which brings me to the question: who are the ./configure defaults for?
> Distributions usually specify all possible configure options when
> building the packages and end users IMO should not be building from
> the source. Should we try to build everything and error out when
> it's not present? Or try to build as much as we can with the libaries
> and dependencies currently present in the system?
The general view of "configure" defaults is that they should
"do the right thing".
Distros could build with the defaults - we just use the explicit
--with/--without args as a safety net, so features don't accidentally
go missing/appear if the build root contents changes in surprising
ways.
> > > I tend to think we should just blindly use qemu/qemu by default and
> > > document that creating these accounts is a requirement. Users will
> > > quickly see if they're missing when they try to start a guest.
> >
> > I'll try to audit what user all the different distributions (that
> > matter) use to launch QEMU. If they are all are using 'qemu' anyway,
> > then probably we can just go with 'qemu:qemu', and document the
> > requirement as such.
>
> Just by defaulting to qemu:qemu if they are present in the system
> you can eliminate many setups that would default to root.
Yeah, that would be a decent step forward if we didn't want to
unconditionally use qemu:qemu.
> But if we start requiring things we consider sensible at configure time,
> I'd happily clean up more of our automagic and let the users frugal
> on system libraries supply --without- arguments for stuff they don't
> want.
If libvirt code considers it optional, then configure should always
automatically probe for it, requiring --without- args would be a
step backwards to how things worked in old days, which was a constant
source of pain for people to figure out how to which arg needed
disabling next.
> > > > As a second thought, we generally use commented-out bits that are the
> > > > non-default configuration. So this fits the pattern in the extent that
> > > > any sane distro specified it's own user/group using the configure
> > > > options and if for any reason the user wants to run this as root it's
> > > > done just by uncommenting it.
> > >
> > > Most commented out bits are not a security flaw if uncommented though.
> > > The fact that we show 'user=root' in the config file though puts across
> > > the misleading idea that it is a reasonable thing todo, when in fact it
> > > is a horribly insecure thing todo.
> >
>
> In that case,
> #allow_disk_format_probing = 1
Yes, I think there's actually a good case to be made for that to
go. We had to have it as a get out of jail free card when we disabled
format probing by default, so we had some compat with existing legacy
tools/deployments. After all this time, I think we could reasonably
justify dropping this though.
> should also probably go. Maybe {vnc,tls}_listen = "0.0.0.0" as well.
listening on 0.0.0.0 is only bad if you've not enabled authentication,
otherwise it is an entirely normal/sensible config choice.
Regards,
Daniel
--
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org -o- https://fstop138.berrange.com :|
|: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list
