Re: Rethinking sigcontext's xfeatures slightly for PKRU's benefit?

[Andy Lutomirski]

On Fri, Dec 18, 2015 at 1:45 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Fri, Dec 18, 2015 at 1:12 PM, Dave Hansen
> <dave.hansen@xxxxxxxxxxxxxxx> wrote:
>>
>> But, if we are picking out an execute-only pkey more dynamically, we've
>> got to keep the default value for the entire process somewhere.
>
> How dynamic do we want to make this, though?
>
> I haven't looked at the details, and perhaps more importantly, I don't
> know what exactly are the requirements you've gotten from the people
> who are expected to actually use this.
>
> I think we might want to hardcode a couple of keys as "kernel
> reserved". And I'd rather reserve them up-front than have some user
> program be unhappy later when we want to use them.
>
> I guess we want to leave key #0 for "normal page", so my suggesting to
> use that for the execute-only was probably misguided.
>
> But I do think we might want to have that "no read access" as a real
> fixed key too, because I think the kernel itself would want to use it:
>
>  (a) to make sure that it gets the right fault when user space passes
> in a execute-only address to a system call.
>
>  (b) for much more efficient PAGEALLOC_DEBUG for kernel mappings.
>
> so I do think that we'd want to reserve two of the 16 keys up front.
>
> Would it be ok for the expected users to have those keys simply be
> fixed? With key 0 being used for all default pages, and key 1 being
> used for all execute-only pages? And then defaulting PKRU to 4,
> disallowing access to that key #1?
>
> I could imagine that some kernel person would want to use even more
> keys, but I think two fixed keys are kind of the minimal we'd want to
> use.

I imagine we'd reserve key 0 for normal page and key 1 for deny-read.
Let me be a bit more concrete about what I'm suggesting:

We'd have thread_struct.baseline_pkru.  It would start with key 0
allowing all access and key 1 denying reads.

We'd have a syscall like set_protection_key that could allocate unused
keys and change the values of keys that have been allocated.  Those
changes would be reflected in baseline_pkru.  Changes to keys 0 and 1
in baseline_pkru would not be allowed.

Signal delivery would load baseline_pkru into the PKRU register.
Signal restore would restore PKRU to its previous value.

WRPKRU would, of course, override baseline_pkru, but it wouldn't
change baseline_pkru.  The set_protection_key syscall would modify
*both* real PKRU and baseline_pkru.

Apps that don't want to use the baseline_pkru mechanism could use
syscalls to claim ownership of protection keys but then manage them
purely with WRPKRU directly.  We could optionally disallow
mprotect_key on keys that weren't allocated in advance.

Does that seem sane?

--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/