On Tue, Jan 12, 2021 at 1:54 AM Oscar Salvador <osalvador@xxxxxxx> wrote:
>
> On Tue, Jan 12, 2021 at 01:34:58AM -0800, Dan Williams wrote:
> > The conversion to move pfn_to_online_page() internal to
> > soft_offline_page() missed that the get_user_pages() reference needs to
> > be dropped when pfn_to_online_page() fails.
>
> I would be more specific here wrt. get_user_pages (madvise).
> soft_offline_page gets called from more places besides madvise_*.
Sure.
>
> > When soft_offline_page() is handed a pfn_valid() &&
> > !pfn_to_online_page() pfn the kernel hangs at dax-device shutdown due to
> > a leaked reference.
> >
> > Fixes: feec24a6139d ("mm, soft-offline: convert parameter to pfn")
> > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Naoya Horiguchi <nao.horiguchi@xxxxxxxxx>
> > Cc: David Hildenbrand <david@xxxxxxxxxx>
> > Cc: Michal Hocko <mhocko@xxxxxxxxxx>
> > Cc: Oscar Salvador <osalvador@xxxxxxx>
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
>
> LGTM, thanks for catching this:
>
> Reviewed-by: Oscar Salvador <osalvador@xxxxxxx>
>
> A nit below.
>
> > ---
> > mm/memory-failure.c | 20 ++++++++++++++++----
> > 1 file changed, 16 insertions(+), 4 deletions(-)
> >
> > diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> > index 5a38e9eade94..78b173c7190c 100644
> > --- a/mm/memory-failure.c
> > +++ b/mm/memory-failure.c
> > @@ -1885,6 +1885,12 @@ static int soft_offline_free_page(struct page *page)
> > return rc;
> > }
> >
> > +static void put_ref_page(struct page *page)
> > +{
> > + if (page)
> > + put_page(page);
> > +}
>
> I am not sure this warrants a function.
> I would probably go with "if (ref_page).." in the two corresponding places,
> but not feeling strong here.
I'll take another look, it felt cluttered...
>
> > +
> > /**
> > * soft_offline_page - Soft offline a page.
> > * @pfn: pfn to soft-offline
> > @@ -1910,20 +1916,26 @@ static int soft_offline_free_page(struct page *page)
> > int soft_offline_page(unsigned long pfn, int flags)
> > {
> > int ret;
> > - struct page *page;
> > bool try_again = true;
> > + struct page *page, *ref_page = NULL;
> > +
> > + WARN_ON_ONCE(!pfn_valid(pfn) && (flags & MF_COUNT_INCREASED));
>
> Did you see any scenario where this could happen? I understand that you are
> adding this because we will leak a reference in case pfn is not valid anymore.
>
I did not, more future proofing / documenting against refactoring that
fails to consider that case.
