On Fri 31-03-17 20:58:16, Ben Hutchings wrote:
> On Tue, 2017-03-28 at 15:11 +0200, Michal Hocko wrote:
> > On Wed 22-03-17 10:09:43, Jiri Slaby wrote:
> > [...]
> > > @@ -1245,6 +1254,10 @@ struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
> > >
> > > page = pmd_page(*pmd);
> > > VM_BUG_ON(!PageHead(page));
> > > +
> > > + if (flags & FOLL_WRITE && !can_follow_write_pmd(*pmd, page, flags))
> > > + goto out;
> > > +
> > > if (flags & FOLL_TOUCH) {
> > > pmd_t _pmd;
> > > /*
> >
> > I have just noticed that this patch is not correct fo 3.12 because we
> > should return NULL rather than the page in this case. 3.2 is wrong as
> > well AFAICS.
>
> Thanks for reporting this. This is the same mistake I made initially
> with follow_page() in 3.2. But I had a test case which caught that
> before release, and I don't have a test case for this.
I believe Miroslav has used the test case embeded in the patch
description to catch the bug.
--
Michal Hocko
SUSE Labs