Re: [PATCH 1/8] powerpc/uaccess: Add unsafe_copy_from_user

Re: [PATCH 1/8] powerpc/uaccess: Add unsafe_copy_from_user

[Christoph Hellwig]

On Thu, Oct 15, 2020 at 10:01:52AM -0500, Christopher M. Riedl wrote:
> Implement raw_copy_from_user_allowed() which assumes that userspace read
> access is open. Use this new function to implement raw_copy_from_user().
> Finally, wrap the new function to follow the usual "unsafe_" convention
> of taking a label argument. The new raw_copy_from_user_allowed() calls
> __copy_tofrom_user() internally, but this is still safe to call in user
> access blocks formed with user_*_access_begin()/user_*_access_end()
> since asm functions are not instrumented for tracing.

Please also add a fallback unsafe_copy_from_user to linux/uaccess.h
so this can be used as a generic API.

Re: [PATCH 3/8] powerpc: Mark functions called inside uaccess blocks w/ 'notrace'

[Christoph Hellwig]

On Thu, Oct 15, 2020 at 10:01:54AM -0500, Christopher M. Riedl wrote:
> Functions called between user_*_access_begin() and user_*_access_end()
> should be either inlined or marked 'notrace' to prevent leaving
> userspace access exposed. Mark any such functions relevant to signal
> handling so that subsequent patches can call them inside uaccess blocks.

I don't think running this much code with uaccess enabled is a good
idea.  Please refactor the code to reduce the criticial sections with
uaccess enabled.

Btw, does powerpc already have the objtool validation that we don't
accidentally jump out of unsafe uaccess critical sections?

Re: [PATCH 3/8] powerpc: Mark functions called inside uaccess blocks w/ 'notrace'

[Peter Zijlstra]

On Fri, Oct 16, 2020 at 07:56:16AM +0100, Christoph Hellwig wrote:
> On Thu, Oct 15, 2020 at 10:01:54AM -0500, Christopher M. Riedl wrote:
> > Functions called between user_*_access_begin() and user_*_access_end()
> > should be either inlined or marked 'notrace' to prevent leaving
> > userspace access exposed. Mark any such functions relevant to signal
> > handling so that subsequent patches can call them inside uaccess blocks.
> 
> I don't think running this much code with uaccess enabled is a good
> idea.  Please refactor the code to reduce the criticial sections with
> uaccess enabled.
> 
> Btw, does powerpc already have the objtool validation that we don't
> accidentally jump out of unsafe uaccess critical sections?

It does not, there was some effort on that a while ago, but I suspect
they're waiting for the ARM64 effort to land and build on that.

Re: [PATCH 1/8] powerpc/uaccess: Add unsafe_copy_from_user

[Christophe Leroy]

Le 16/10/2020 à 08:54, Christoph Hellwig a écrit :
> On Thu, Oct 15, 2020 at 10:01:52AM -0500, Christopher M. Riedl wrote:
>> Implement raw_copy_from_user_allowed() which assumes that userspace read
>> access is open. Use this new function to implement raw_copy_from_user().
>> Finally, wrap the new function to follow the usual "unsafe_" convention
>> of taking a label argument. The new raw_copy_from_user_allowed() calls
>> __copy_tofrom_user() internally, but this is still safe to call in user
>> access blocks formed with user_*_access_begin()/user_*_access_end()
>> since asm functions are not instrumented for tracing.
> 
> Please also add a fallback unsafe_copy_from_user to linux/uaccess.h
> so this can be used as a generic API.
> 

I guess this can be done in a separate patch independant of that series ?

Christophe

Re: [PATCH 3/8] powerpc: Mark functions called inside uaccess blocks w/ 'notrace'

[Michael Ellerman]

Peter Zijlstra <peterz@infradead.org> writes:
> On Fri, Oct 16, 2020 at 07:56:16AM +0100, Christoph Hellwig wrote:
>> On Thu, Oct 15, 2020 at 10:01:54AM -0500, Christopher M. Riedl wrote:
>> > Functions called between user_*_access_begin() and user_*_access_end()
>> > should be either inlined or marked 'notrace' to prevent leaving
>> > userspace access exposed. Mark any such functions relevant to signal
>> > handling so that subsequent patches can call them inside uaccess blocks.
>> 
>> I don't think running this much code with uaccess enabled is a good
>> idea.  Please refactor the code to reduce the criticial sections with
>> uaccess enabled.
>> 
>> Btw, does powerpc already have the objtool validation that we don't
>> accidentally jump out of unsafe uaccess critical sections?
>
> It does not, there was some effort on that a while ago, but I suspect
> they're waiting for the ARM64 effort to land and build on that.

Right, we don't have objtool support.

We would definitely like objtool support at least for this uaccess
checking, I'm sure we have some escapes.

There was someone working on it in their own-time but last I heard that
was still WIP.

I didn't realise the ARM64 support was still not merged, so yeah having
that land first would probably simplify things, but we still need
someone who has time to work on it.

cheers