Re: [PATCH v4 2/3] mm: introduce MAP_VALIDATE a mechanism for adding new mmap flags

[Dan Williams]

On Mon, Sep 18, 2017 at 2:31 AM, Jan Kara <jack@xxxxxxx> wrote:
> On Sun 17-09-17 19:39:45, Christoph Hellwig wrote:
>> On Sat, Sep 16, 2017 at 08:44:14PM -0700, Dan Williams wrote:
>> > So it wasn't all that easy, and Linus declined to take it. I think we
>> > should add a new ->mmap_validate() file operation and save the
>> > tree-wide cleanup until later.
>>
>> Note that we already have a mmap_capabilities callout for nommu,
>> I wonder if we could generalize that.
>
> So if I understood Dan right, Linus refused to merge the patch which adds
> 'flags' argument to ->mmap callback. That is actually logically independent
> change from validating flags passed to mmap(2) syscall. Dan did it just to
> save himself from adding a VMA flag for MAP_DIRECT.
>
> For validating flags passed to mmap(2), I agree we could use
> ->mmap_capabilities() instead of mmap_supported_mask Dan has added. But I
> don't have a strong opinion there.

The drawback I see with mmap_capabilities is that it requires all mmap
flags to have a corresponding vm_flag. After the cold reaction the
VM_DAX flag received I'd want to be sure they were on board with this
direction.