Re: [PATCH] proc: restrict kernel stack dumps to root

[Kees Cook <keescook@chromium.org>]

On Thu, Sep 13, 2018 at 4:55 AM, Jann Horn <jannh@google.com> wrote:
> On Thu, Sep 13, 2018 at 12:28 AM Kees Cook <keescook@chromium.org> wrote:
>>
>> On Wed, Sep 12, 2018 at 8:29 AM, Jann Horn <jannh@google.com> wrote:
>> > +linux-api, I guess
>> >
>> > On Tue, Sep 11, 2018 at 8:39 PM Jann Horn <jannh@google.com> wrote:
>> >>
>> >> Restrict the ability to inspect kernel stacks of arbitrary tasks to root
>> >> in order to prevent a local attacker from exploiting racy stack unwinding
>> >> to leak kernel task stack contents.
>> >> See the added comment for a longer rationale.
>> >>
>> >> There don't seem to be any users of this userspace API that can't
>> >> gracefully bail out if reading from the file fails. Therefore, I believe
>> >> that this change is unlikely to break things.
>> >> In the case that this patch does end up needing a revert, the next-best
>> >> solution might be to fake a single-entry stack based on wchan.
>> >>
>> >> Fixes: 2ec220e27f50 ("proc: add /proc/*/stack")
>> >> Cc: stable@vger.kernel.org
>> >> Signed-off-by: Jann Horn <jannh@google.com>
>> >> ---
>> >>  fs/proc/base.c | 14 ++++++++++++++
>> >>  1 file changed, 14 insertions(+)
>> >>
>> >> diff --git a/fs/proc/base.c b/fs/proc/base.c
>> >> index ccf86f16d9f0..7e9f07bf260d 100644
>> >> --- a/fs/proc/base.c
>> >> +++ b/fs/proc/base.c
>> >> @@ -407,6 +407,20 @@ static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
>> >>         unsigned long *entries;
>> >>         int err;
>> >>
>> >> +       /*
>> >> +        * The ability to racily run the kernel stack unwinder on a running task
>> >> +        * and then observe the unwinder output is scary; while it is useful for
>> >> +        * debugging kernel issues, it can also allow an attacker to leak kernel
>> >> +        * stack contents.
>> >> +        * Doing this in a manner that is at least safe from races would require
>> >> +        * some work to ensure that the remote task can not be scheduled; and
>> >> +        * even then, this would still expose the unwinder as local attack
>> >> +        * surface.
>> >> +        * Therefore, this interface is restricted to root.
>> >> +        */
>> >> +       if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
>> >> +               return -EACCES;
>>
>> In the past, we've avoided hard errors like this in favor of just
>> censoring the output. Do we want to be more cautious here? (i.e.
>> return 0 or a fuller seq_printf(m, "[<0>] privileged\n"); return 0;)
>
> In my mind, this is different because it's a place where we don't have
> to selectively censor output while preserving parts of it, and it's a
> place where, as Laura said, it's useful to make lack of privileges
> clearly visible because that informs users that they may have to retry
> with more privileges.
>
> Of course, if you have an example of software that actually breaks due
> to this, I'll change it. But I looked at the three things in Debian
> codesearch that seem to use it, and from what I can tell, they all
> bail out cleanly when the read fails.

I prefer -EACCESS too, but I thought I'd mention the alternative. So, I guess:

Reviewed-by: Kees Cook <keescook@chromium.org>

:)

-Kees

-- 
Kees Cook
Pixel Security