Re: [PATCH] x86: correct fpu emulation access to ldt

[Juergen Gross <jgross@suse.com>]

On 08/05/2015 08:24 PM, Andy Lutomirski wrote:
> On Wed, Aug 5, 2015 at 2:11 AM, Juergen Gross <jgross@suse.com> wrote:
>> On 08/04/2015 08:01 PM, Andy Lutomirski wrote:
>>>
>>> On Tue, Aug 4, 2015 at 8:02 AM, Juergen Gross <jgross@suse.com> wrote:
>>>>
>>>> Commit 14805442532c ("x86/ldt: Make modify_ldt synchronous") introduced
>>>> a new struct ldt_struct anchored at mm->context.ldt.
>>>>
>>>> Adapt the x86 fpu emulation to use that new structure.
>>>>
>>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>>
>>>
>>> Whoops!
>>>
>>> Does this need to Cc: stable?
>>
>>
>> Probably.
>>
>>> Also, want to make it slightly fancier so we can drop the dependency
>>> on CONFIG_MODIFY_LDT_SYSCALL?
>>
>>
>> Something like:
>>
>> -#define LDT_DESCRIPTOR(s) (((struct desc_struct
>> *)current->mm->context.ldt)[(s) >> 3])
>> +#ifdef CONFIG_MODIFY_LDT_SYSCALL
>> +#define LDT_DESCRIPTOR(s) (current->mm->context.ldt->entries[(s) >> 3])
>> +#else
>> +#define LDT_DESCRIPTOR(s) ((struct desc_struct){{{ .a = 0, .b = 0, }}})
>
> Careful!  I think that akpm uses some ancient gcc version that can't
> compile that.  Maybe have a global empty segment somewhere that this
> returns, or just ifdef out the two call sites.
>
> Also, I don't believe this for a second:
>
> /* s is always from a cpu register, and the cpu does bounds checking
>   * during register load --> no further bounds checks needed */
> #define LDT_DESCRIPTOR(s)    (((struct desc_struct
> *)current->mm->context.ldt)[(s) >> 3])
>
> "What the comment means is that s always came from a cpu register at
> some point in the recent past (assuming that no lazy segment save
> logic is in effect) and we cross our fingers and hope that we never
> end up accessing out of bounds if the LDT isn't the same as it was at
> the time of the fault we're handling."
>
> Sigh.
>
> Maybe the best approach would be to replace LDT_DESCRIPTOR with an
> actual function that returns a struct desc_struct.  If it's out of
> bounds or !CONFIG_MODIFY_LDT_SYSCALL, return zeros.  Otherwise return
> the descriptor.

Yeah, seems to be the better approach.

>
>> +#endif
>>
>> I'd need to specify the corresponding patch as a prerequisite for stable
>> I guess? How to do this before it is picked by Linus?
>
> Send a v2 with Cc: stable@vger.kernel.org # [commit hash you depend
> on].  Presumably Ingo will pick it up, not Linus.

I know how to specify a prerequisite. I just wasn't sure which commit
hash to use, as up to now I've only one from your tree and I guessed
that wouldn't do it.


Juergen