On Mon, Mar 02, 2020 at 10:25:52AM +0800, Pingfan Liu wrote:
> On Fri, Feb 28, 2020 at 9:44 PM Jason Gunthorpe <jgg@xxxxxxxx> wrote:
> >
> > On Fri, Feb 28, 2020 at 07:32:29PM +0800, Pingfan Liu wrote:
> > > FOLL_LONGTERM suggests a pin which is going to be given to hardware and
> > > can't move. It would truncate CMA permanently and should be excluded.
> > >
> > > FOLL_LONGTERM has already been checked in the slow path, but not checked in
> > > the fast path, which means a possible leak of CMA page to longterm pinned
> > > requirement through this crack.
> > >
> > > Place a check in try_get_compound_head() in the fast path.
> > >
> > > Some note about the check:
> > > Huge page's subpages have the same migrate type due to either
> > > allocation from a free_list[] or alloc_contig_range() with param
> > > MIGRATE_MOVABLE. So it is enough to check on a single subpage
> > > by is_migrate_cma_page(subpage)
> > >
> > > Signed-off-by: Pingfan Liu <kernelfans@xxxxxxxxx>
> > > Cc: Ira Weiny <ira.weiny@xxxxxxxxx>
> > > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > > Cc: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> > > Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
> > > Cc: Matthew Wilcox <willy@xxxxxxxxxxxxx>
> > > Cc: John Hubbard <jhubbard@xxxxxxxxxx>
> > > Cc: "Aneesh Kumar K.V" <aneesh.kumar@xxxxxxxxxxxxx>
> > > Cc: Keith Busch <keith.busch@xxxxxxxxx>
> > > Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>
> > > Cc: Shuah Khan <shuah@xxxxxxxxxx>
> > > To: linux-mm@xxxxxxxxx
> > > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > > mm/gup.c | 26 +++++++++++++++++++-------
> > > 1 file changed, 19 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/mm/gup.c b/mm/gup.c
> > > index cd8075e..f0d6804 100644
> > > +++ b/mm/gup.c
> > > @@ -33,9 +33,21 @@ struct follow_page_context {
> > > * Return the compound head page with ref appropriately incremented,
> > > * or NULL if that failed.
> > > */
> > > -static inline struct page *try_get_compound_head(struct page *page, int refs)
> > > +static inline struct page *try_get_compound_head(struct page *page, int refs,
> > > + unsigned int flags)
> > > {
> > > - struct page *head = compound_head(page);
> > > + struct page *head;
> > > +
> > > + /*
> > > + * Huge page's subpages have the same migrate type due to either
> > > + * allocation from a free_list[] or alloc_contig_range() with param
> > > + * MIGRATE_MOVABLE. So it is enough to check on a single subpage.
> > > + */
> > > + if (unlikely(flags & FOLL_LONGTERM) &&
> > > + is_migrate_cma_page(page))
> > > + return NULL;
> >
> > This doesn't seem very good actually.
> >
> > If I understand properly, if the system has randomly decided to place,
> > say, an anonymous page in a CMA region when an application did mmap(),
> > then when the application tries to use this page with a LONGTERM pin
> > it gets an immediate failure because of the above.
> No, actually, it will fall back to slow path, which migrates and sever
> the LONGTERM pin.
>
> This patch just aims to fix the leakage in gup fast path, while in gup
> slow path, there is already logic to guard CMA against LONGTERM pin.
> >
> > This not OK - the application should not be subject to random failures
> > related to long term pins beyond its direct control.
> >
> > Essentially, failures should only originate from the application using
> > specific mmap scenarios, not randomly based on something the MM did,
> > and certainly never for anonymous memory.
> >
> > I think the correct action here is to trigger migration of the page so
> > it is not in CMA.
> In fact, it does this. The failure in gup fast path will fall back to
> slow path, where __gup_longterm_locked->check_and_migrate_cma_pages()
> does the migration.
It is probably worth revising the commit message so this flow is clear
Jason
