Re: [PATCH] fs/userfaultfd.c: simplify the calculation of new_flags

[Andrea Arcangeli]

On Sat, Oct 05, 2019 at 06:46:40AM +0800, Wei Yang wrote:
> On Wed, Oct 02, 2019 at 08:45:05PM -0400, Andrea Arcangeli wrote:
> >Hello,
> >
> >On Tue, Aug 06, 2019 at 01:38:59PM +0800, Wei Yang wrote:
> >> Finally new_flags equals old vm_flags *OR* vm_flags.
> >> 
> >> It is not necessary to mask them first.
> >> 
> >> Signed-off-by: Wei Yang <richardw.yang@xxxxxxxxxxxxxxx>
> >> ---
> >>  fs/userfaultfd.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >> 
> >> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> >> index ccbdbd62f0d8..653d8f7c453c 100644
> >> --- a/fs/userfaultfd.c
> >> +++ b/fs/userfaultfd.c
> >> @@ -1457,7 +1457,7 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
> >>  			start = vma->vm_start;
> >>  		vma_end = min(end, vma->vm_end);
> >>  
> >> -		new_flags = (vma->vm_flags & ~vm_flags) | vm_flags;
> >> +		new_flags = vma->vm_flags | vm_flags;
> >>  		prev = vma_merge(mm, prev, start, vma_end, new_flags,
> >>  				 vma->anon_vma, vma->vm_file, vma->vm_pgoff,
> >>  				 vma_policy(vma),
> >
> >And then how do you clear the flags after the above?
> >
> >It must be possible to clear the flags (from
> >UFFDIO_REGISTER_MODE_MISSING|UFFDIO_REGISTER_MODE_WP to only one set
> >or invert).
> >
> >We have no WP support upstream yet, so maybe that's why it looks
> >superfluous in practice, but in theory it isn't because it would then
> >need to be reversed by Peter's (CC'ed) -wp patchset.
> >
> >The register code has already the right placeholder to support -wp and
> >so it's better not to break them.
> >
> >I would recommend reviewing the uffd-wp support and working on testing
> >the uffd-wp code instead of changing the above.
> >
> 
> Sorry, I don't get your point. This change is valid to me even from arithmetic
> point of view.
> 
>     vm_flags == VM_UFFD_MISSING | VM_UFFD_WP
> 
> The effect of current code is clear these two bits then add them. This equals
> to just add these two bits.
> 
> I am not sure which part I lost.

The cleaned removed the "& ~" and that was enough to quickly tell the
cleaned up version was wrong.

What I should have noticed right away as well is that the code was
already wrong, sorry. That code doesn't require a noop code cleanup,
it requires a fix and the "& ~" needs to stay.

This isn't going to make any difference upstream until the uffd-wp
support is merged so it is enough to queue it in Peter's queue, or you
can merge it independently.

Thanks,
Andrea