Re: [RFC PATCH 0/3] move the secure_computing call

[Andrew Lutomirski]

On Thu, May 24, 2012 at 4:56 PM, H. Peter Anvin <hpa@xxxxxxxxx> wrote:
> On 05/24/2012 04:43 PM, Andrew Lutomirski wrote:
>>
>> IMO the behavior should change.  Alternatively, a post-ptrace syscall
>> should have to pass the *tracer's* seccomp filter, but that seems
>> overcomplicated and confusing.
>>
>> OTOH, allowing ptrace in a seccomp filter is asking for trouble anyway
>> -- if you can ptrace something outside the sandbox, you've escaped.
>>
>
> This is my suggestion: if there is demand, make it possible to install a
> *second* seccomp filter program which is run on the result of the
> ptrace.  I.e.:
>
> Untraced:       process -> seccomp1 -> kernel
>
> Traced:         process -> seccomp1 -> ptrace -> seccomp2 -> kernel

Just to clarify: are you suggesting that, for now, the traced behavior
should be:

process -> seccomp -> ptrace -> kernel?

If so, I think the man page or something should have a big fat warning
that seccomp filters should *never* allow ptrace (even PTRACE_TRACEME)
unless they fully understand the issue.

In any case, I think that the UML interaction is missing the point.
UML will *emulate* the seccomp filter.  If it chooses to use host
seccomp filters for some business, that's its business.

--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/