Re: Vendor-sec hosting and future of closed lists

[Dan Rosenberg <dan.j.rosenberg () gmail com>]

> > Rather than requiring individuals to perform substantial amounts of
> > digging through patches, which I agree is infeasible, perhaps it would
> > be more reasonable to establish a general policy that bug reporters
> > and maintainers can use to work with distro security teams and the
> > rest of the security community.
> >
> > For example, a public or private list could be established for all
> > *potential* kernel security issues, and just as is the case with
> > CC'ing stable, a policy could be developed where maintainers are
> > expected to CC this list for fixes that might possibly have security
> > relevance, with a tendency towards erring on the safe side if security
> > impact is unclear.
>
> This proposal just fell down right there, as it has been rightly pointed
> out that numerous bug fixes in the kernel in the past have later been
> deemed "security fixes".  So what you are asking for is for _all_
> bugfixes to be sent to such a list.
>
> Well, we have that already, we have mailing lists that get every single
> patch that is merged into the kernel, and there's the big lkml list as
> well with hundreds of fixes posted every week.

Of course failing to anticipate security impact is bound to happen in
the kernel; it frequently happens in userland too, and is unavoidable.
 That doesn't mean we can't try, and it doesn't mean we should be
overly paranoid and have security folks manually audit every patch.
Currently, maintainers and bug reporters are expected to ask
themselves a simple question when deciding whether or not to CC
stable: "does this fix a bug or security issue, or is it a new
feature?".  Similarly, I don't think it's too much to ask for people
to consider the question of "does this bug it allow an unprivileged
user to crash the system, gain additional access, or otherwise cross
privilege boundaries?"  And if the answer is "I don't know, maybe?",
then they should CC this list to be safe.  I think this would result
in not nearly as much volume as you're anticipating.

> > I think security communication needs to be
> > improved at the commit level (as opposed to the reporting), since
> > maintainers are often much more knowledgeable and better able to
> > understand security impact than the users who are often presenting
> > issues.
>
> I don't think you understand the rate of change in the kernel and how
> trying to do this for every commit is unfeasable and unworkable.  You do
> know how fast it goes, right?

Why is CC'ing a security list any more difficult than CC'ing stable?

> > Criteria could be set up for what kinds of issues would be
> > candidates for being sent to this list.  I don't think this would
> > require substantially more work on anyone's part, but by creating a
> > culture where potential security issues are treated seriously, it
> > would at least stop some of the silent patching that's been going on.
> >
> > Once potential security issues have been submitted to such a list, I'm
> > sure there would be no shortage of people willing and able to analyze
> > security impact for each issue, including assigning CVEs.  While
> > digging through every kernel patch might be too much work, with the
> > cooperation of maintainers this can be reduced to a much smaller
> > subset that would be easily dealt with.
>
> I would be happy if someone could just document the patches that _are_
> applied to stable kernel releases.  I bet you can't keep up with that,
> they are moving so fast.
>
> Sorry, I don't think this is workable as you are proposing, but feel
> free to prove me wrong :)

Perhaps you're right, but maybe we can generate some discussion to
come up with a solution that improves upstream security communication
and IS workable.

Thanks,
Dan