On 10/12, Eyal Edri wrote:
>
>
> ----- Original Message -----
> > From: "Sandro Bonazzola" <sbonazzo(a)redhat.com>
> > To: secalert(a)redhat.com
> > Cc: security(a)ovirt.org, infra(a)ovirt.org
> > Sent: Thursday, October 9, 2014 9:09:20 AM
> > Subject: Re: [
engineering.redhat.com #319333] Re: [Security]
System job to deploy rpms
> >
> > Il 08/10/2014 18:18, Red Hat Product Security ha scritto:
> > > On Wed Oct 08 08:35:15 2014, sbonazzo(a)redhat.com wrote:
> > >> Il 08/10/2014 12:02, Ohad Basan ha scritto:
> > >>> Hello everyone.
> > >>>
> > >>> I've created a small job (not yet enabled)
> > >>> that gets an rpm and then deploys it to the static repo at
> > >>
resources.ovirt.org
> > >>> for this I've sent this patch
http://gerrit.ovirt.org/#/c/33863/
> > >>> that will add the "resources" user. it will have
permissions
only
> > >> for the static rpms directory and will scp the files to there.
> > >>> is it acceptable by everybody security-wise?
> > >>>
> > >>
> > >> Adding security list to the loop.
> > >
> > > Hi, thanks for this. I'm a bit confused though. Is this
pertaining to the
> > > infrastructure for the oVirt project, or is this code going into
the oVirt
> > > code itself that is then consumed by downstream users? I only
ask because
> > > of the reference to
resources.ovirt.org so I'm unsure whether
this is a
> > > code question or an infrastructure question.
> > >
> > > Can you please advise?
> >
> > It's infrastructure question
>
> let me try to clarify.
> today our continuous delivery process is async partially:
> -
jenkins.ovirt.org builds and publish the rpms into
resources.ovirt.org under jenkins home (unprivileged user).
> - a cron job scans the target dir and checks if new rpms are there
(via flag used by the script) and updates the repos accordingly.
>
> the idea behind this is not to allow direct access from jenkins to
resources.ovirt.org via ssh.
>
> now what ohad is suggesting is to change the process and ALLOW
direct access to certain repositories under
resources.ovirt.org, with
the following changes:
> - new user will be used - resources
> - the user will have limited sudo access only to read/write to the
relevant repository (static repos)
> - no cron job will run async to update it.
>
> so the question is are we comfortable with this change? is it safe
or has the same security level as the current async one?
> if its safe we might consider changing the original flow as well to
be synced and not use a cron job.
>
> your input is appreciated,
As far as I know, the only reason we had to use the cron (kiril might
know better, but he does not work with us anymore) was to avoid
exposing the signing key for the packages to the jenkins ssh user, but
that part was never automated anyhow and the nightly-static packages
are not signed.
So I think that creating a new user without privileges to mess
anything up (only access to the nightly repos) is as secure as the
cron approach, but more convenient.
In the future I'm thinking on using a simple web service to deploy the
rpms, that way the clients do not need ssh access to the machine at
all. But that's not ready yet.
This seems reasonable to me. If you're at all unsure of uncomfortable I do suggest
reaching out to infosec, but nothing described in this or the prior mail have any red
flags leaping out at me.
--
Vincent Danen / Red Hat Product Security