Re: [PATCH v3 0/7] File Sealing & memfd_create()

[David Herrmann <dh.herrmann@gmail.com>]

Hi

On Fri, Jun 13, 2014 at 5:17 PM, Andy Lutomirski <luto@amacapital.net> wrote:
> On Fri, Jun 13, 2014 at 8:15 AM, David Herrmann <dh.herrmann@gmail.com> wrote:
>> Hi
>>
>> On Fri, Jun 13, 2014 at 5:10 PM, Andy Lutomirski <luto@amacapital.net> wrote:
>>> On Fri, Jun 13, 2014 at 3:36 AM, David Herrmann <dh.herrmann@gmail.com> wrote:
>>>> Hi
>>>>
>>>> This is v3 of the File-Sealing and memfd_create() patches. You can find v1 with
>>>> a longer introduction at gmane:
>>>>   http://thread.gmane.org/gmane.comp.video.dri.devel/102241
>>>> An LWN article about memfd+sealing is available, too:
>>>>   https://lwn.net/Articles/593918/
>>>> v2 with some more discussions can be found here:
>>>>   http://thread.gmane.org/gmane.linux.kernel.mm/115713
>>>>
>>>> This series introduces two new APIs:
>>>>   memfd_create(): Think of this syscall as malloc() but it returns a
>>>>                   file-descriptor instead of a pointer. That file-descriptor is
>>>>                   backed by anon-memory and can be memory-mapped for access.
>>>>   sealing: The sealing API can be used to prevent a specific set of operations
>>>>            on a file-descriptor. You 'seal' the file and give thus the
>>>>            guarantee, that it cannot be modified in the specific ways.
>>>>
>>>> A short high-level introduction is also available here:
>>>>   http://dvdhrm.wordpress.com/2014/06/10/memfd_create2/
>>>
>>> Potentially silly question: is it guaranteed that mmapping and reading
>>> a SEAL_SHRINKed fd within size bounds will not SIGBUS?  If so, should
>>> this be documented?  (The particular issue here would be reading
>>> holes.  It should work by using the zero page, but, if so, we should
>>> probably make it a real documented guarantee.)
>>
>> No, this is not guaranteed. See the previous discussion in v2 on Patch
>> 2/4 between Hugh and me.
>>
>> Summary is: If you want mmap-reads to not fail, use mlock(). There are
>> many situations where a fault might fail (think: OOM) and sealing is
>> not meant to protect against that. Btw., holes are automatically
>> filled with fresh pages by shmem. So a read only fails in OOM
>> situations (or memcg limits, etc.).
>>
>
> Isn't the point of SEAL_SHRINK to allow servers to mmap and read
> safely without worrying about SIGBUS?

No, I don't think so.
The point of SEAL_SHRINK is to prevent a file from shrinking. SIGBUS
is an effect, not a cause. It's only a coincidence that "OOM during
reads" and "reading beyond file-boundaries" has the same effect:
SIGBUS.
We only protect against reading beyond file-boundaries due to
shrinking. Therefore, OOM-SIGBUS is unrelated to SEAL_SHRINK.

Anyone dealing with mmap() _has_ to use mlock() to protect against
OOM-SIGBUS. Making SEAL_SHRINK protect against OOM-SIGBUS would be
redundant, because you can achieve the same with SEAL_SHRINK+mlock().

Thanks
David