Re: [PATCH] mm: redefine the MAP_SHARED_VALIDATE to other value

[Michal Hocko <mhocko@suse.com>]

On Mon 08-07-19 21:52:53, zhong jiang wrote:
> On 2019/7/8 17:20, Michal Hocko wrote:
> > [Cc Dan]
> >
> > On Mon 08-07-19 16:05:41, zhong jiang wrote:
> >> As the mman manual says, mmap should return fails when we assign
> >> the flags to MAP_SHARED | MAP_PRIVATE.
> >>
> >> But In fact, We run the code successfully and unexpected.
> > What is the code that you are running and what is the code version.
> Just an following code, For example,
> addr = mmap(ADDR, PAGE_SIZE, PROT_WRITE|PROT_EXEC, MAP_SHARED|MAP_PRIVATE, fildes, OFFSET);

Is this a real code that relies on the failure or merely a simple test
to reflect the semantic you expect mmap to have?

> We test it and works well in linux 4.19.   As the mmap manual says,  it should fails.
> >> It is because MAP_SHARED_VALIDATE is introduced and equal to
> >> MAP_SHARED | MAP_PRIVATE.
> > This was a deliberate decision IIRC. Have a look at 1c9725974074 ("mm:
> > introduce MAP_SHARED_VALIDATE, a mechanism to safely define new mmap
> > flags").
> I  has seen the patch,  It introduce the issue.  but it only define the MAP_SHARED_VALIDATE incorrectly.
> Maybe the author miss the condition that MAP_SHARED_VALIDATE is equal to MAP_PRIVATE | MAP_SHARE.

No you are missing the point as Willy pointed out in a different email.
This is intentional. No real application could have used the combination
of two flags because it doesn't make any sense. And therefore the
combination has been chosen to chnage the mmap semantic and check for
valid mapping flags. LWN has a nice coverage[1].


[1] https://lwn.net/Articles/758594/
-- 
Michal Hocko
SUSE Labs