Re: [PATCH v2 00/10] userns: sysctl limits for namespaces

From: Kees Cook
Date: Tue Jul 26 2016 - 12:52:53 EST

On Tue, Jul 26, 2016 at 8:06 AM, Eric W. Biederman
<ebiederm@xxxxxxxxxxxx> wrote:
> "Michael Kerrisk (man-pages)" <mtk.manpages@xxxxxxxxx> writes:
>> Hello Eric,
>> I realized I had a question after the last mail.
>> On 07/21/2016 06:39 PM, Eric W. Biederman wrote:
>>> This patchset addresses two use cases:
>>> - Implement a sane upper bound on the number of namespaces.
>>> - Provide a way for sandboxes to limit the attack surface from
>>> namespaces.
>> Can you say more about the second point? What exactly is the
>> problem that is being addressed, and how does the patch series
>> address it? (It would be good to have those details in the
>> revised commit message...)
> At some point it was reported that seccomp was not sufficient to disable
> namespace creation. I need to go back and look at that claim to see
> which set of circumstances that was referring to. Seccomp doesn't stack
> so I can see why it is an issue.

seccomp does stack. The trouble usually comes from a perception that
seccomp overhead is not trivial, so setting a system-wide policy is a
bit of a large hammer for such a limitiation. Also, at the time,
seccomp could be bypasses with ptrace, but this (as of v4.8) is no
longer true.

> The general problem is that namespaces by their nature (and especially
> in combination with the user namespaces) allow unprivileged users to use
> more of the kernel than a user would have access to without them. This
> in turn allows malicious users more kernel calls they can use in attempt
> to find an exploitable bug.
> So if you are building a sandbox/chroot jail/chromium tab or anything
> like that and you know you won't be needing a kernel feature having an
> easy way to disable the feature is useful for making the kernel
> marginally more secure, as certain attack vectors are no longer
> possible.


Kees Cook
Chrome OS & Brillo Security