Re: [PATCH v2] mm/z3fold.c: Lock z3fold page before __SetPageMovable()

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 1 Jul 2019 18:16:30 -0700 Henry Burns <henryburns@google.com> wrote:

> Cc: Vitaly Wool <vitalywool@gmail.com>, Vitaly Vul <vitaly.vul@sony.com>

Are these the same person?

> Subject: Re: [PATCH v2] mm/z3fold.c: Lock z3fold page before __SetPageMovable()
> Date: Mon, 1 Jul 2019 18:16:30 -0700
> 
> On Mon, Jul 1, 2019 at 6:00 PM Shakeel Butt <shakeelb@google.com> wrote:
> >
> > On Mon, Jul 1, 2019 at 5:51 PM Henry Burns <henryburns@google.com> wrote:
> > >
> > > __SetPageMovable() expects it's page to be locked, but z3fold.c doesn't
> > > lock the page. Following zsmalloc.c's example we call trylock_page() and
> > > unlock_page(). Also makes z3fold_page_migrate() assert that newpage is
> > > passed in locked, as documentation.

The changelog still doesn't mention that this bug triggers a
VM_BUG_ON_PAGE().  It should do so.  I did this:

: __SetPageMovable() expects its page to be locked, but z3fold.c doesn't
: lock the page.  This triggers the VM_BUG_ON_PAGE(!PageLocked(page), page)
: in __SetPageMovable().
:
: Following zsmalloc.c's example we call trylock_page() and unlock_page(). 
: Also make z3fold_page_migrate() assert that newpage is passed in locked,
: as per the documentation.

I'll add a cc:stable to this fix.

> > > Signed-off-by: Henry Burns <henryburns@google.com>
> > > Suggested-by: Vitaly Wool <vitalywool@gmail.com>
> > > ---
> > >  Changelog since v1:
> > >  - Added an if statement around WARN_ON(trylock_page(page)) to avoid
> > >    unlocking a page locked by a someone else.
> > >
> > >  mm/z3fold.c | 6 +++++-
> > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/mm/z3fold.c b/mm/z3fold.c
> > > index e174d1549734..6341435b9610 100644
> > > --- a/mm/z3fold.c
> > > +++ b/mm/z3fold.c
> > > @@ -918,7 +918,10 @@ static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
> > >                 set_bit(PAGE_HEADLESS, &page->private);
> > >                 goto headless;
> > >         }
> > > -       __SetPageMovable(page, pool->inode->i_mapping);
> > > +       if (!WARN_ON(!trylock_page(page))) {
> > > +               __SetPageMovable(page, pool->inode->i_mapping);
> > > +               unlock_page(page);
> > > +       }
> >
> > Can you please comment why lock_page() is not used here?

Shakeel asked "please comment" (ie, please add a code comment), not
"please comment on".  Subtle ;)

> Since z3fold_alloc can be called in atomic or non atomic context,
> calling lock_page() could trigger a number of
> warnings about might_sleep() being called in atomic context. WARN_ON
> should avoid the problem described
> above as well, and in any weird condition where someone else has the
> page lock, we can avoid calling
> __SetPageMovable().

I think this will suffice:

--- a/mm/z3fold.c~mm-z3foldc-lock-z3fold-page-before-__setpagemovable-fix
+++ a/mm/z3fold.c
@@ -919,6 +919,9 @@ retry:
 		set_bit(PAGE_HEADLESS, &page->private);
 		goto headless;
 	}
+	/*
+	 * z3fold_alloc() can be called from atomic contexts, hence the trylock
+	 */
 	if (!WARN_ON(!trylock_page(page))) {
 		__SetPageMovable(page, pool->inode->i_mapping);
 		unlock_page(page);

However this code would be more effective if z3fold_alloc() were to be
told when it is running in non-atomic context so it can perform a
sleeping lock_page() in that case.  That's an improvement to consider
for later, please.