Re: [PATCH 1/2] staging: android: ion: Add __GFP_NOWARN for system contig heap

[Laura Abbott]

On 01/05/2018 11:36 AM, Chris Wilson wrote:
> Quoting Laura Abbott (2018-01-05 19:14:08)
> > syzbot reported a warning from Ion:
> >
> >    WARNING: CPU: 1 PID: 3485 at mm/page_alloc.c:3926
> >
> >    ...
> >     __alloc_pages_nodemask+0x9fb/0xd80 mm/page_alloc.c:4252
> >    alloc_pages_current+0xb6/0x1e0 mm/mempolicy.c:2036
> >    alloc_pages include/linux/gfp.h:492 [inline]
> >    ion_system_contig_heap_allocate+0x40/0x2c0
> >    drivers/staging/android/ion/ion_system_heap.c:374
> >    ion_buffer_create drivers/staging/android/ion/ion.c:93 [inline]
> >    ion_alloc+0x2c1/0x9e0 drivers/staging/android/ion/ion.c:420
> >    ion_ioctl+0x26d/0x380 drivers/staging/android/ion/ion-ioctl.c:84
> >    vfs_ioctl fs/ioctl.c:46 [inline]
> >    do_vfs_ioctl+0x1b1/0x1520 fs/ioctl.c:686
> >    SYSC_ioctl fs/ioctl.c:701 [inline]
> >    SyS_ioctl+0x8f/0xc0 fs/ioctl.c:692
> >
> > This is a warning about attempting to allocate order > MAX_ORDER. This
> > is coming from a userspace Ion allocation request. Since userspace is
> > free to request however much memory it wants (and the kernel is free to
> > deny its allocation), silence the allocation attempt with __GFP_NOWARN
> > in case it fails.
> >
> > Reported-by: syzbot+76e7efc4748495855a4d@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Reported-by: syzbot <syzkaller@xxxxxxxxxxxxxxxx>
> > Signed-off-by: Laura Abbott <labbott@xxxxxxxxxx>
> > ---
> >   drivers/staging/android/ion/ion_system_heap.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/staging/android/ion/ion_system_heap.c b/drivers/staging/android/ion/ion_system_heap.c
> > index 71c4228f8238..bc19cdd30637 100644
> > --- a/drivers/staging/android/ion/ion_system_heap.c
> > +++ b/drivers/staging/android/ion/ion_system_heap.c
> > @@ -362,7 +362,7 @@ static int ion_system_contig_heap_allocate(struct ion_heap *heap,
> >          unsigned long i;
> >          int ret;
> >   
> > -       page = alloc_pages(low_order_gfp_flags, order);
> > +       page = alloc_pages(low_order_gfp_flags | __GFP_NOWARN, order);
>
> There's both high_order_gfp and low_order_gfp. The former includes
> NOWARN and NORETRY.
>
> Interesting, ion_system_heap_create_pools() tries to mix low_order and
> high_order, but it only ever uses high_order flags. (orders[0] == 8
> forcing a permanent switch from low_order_gfp to high_order_gfp).

Good find, that got lost in a refactor back in 4.9.

> There's no good reason for low_order_gfp, high_order_gfp to be static
> rewritable variables.
>
> For this instance, I would go farther and suggest you may want
> __GFP_RETRY_MAYFAIL | __GFP_NOWARN to prevent userspace from triggering
> the lowmemkiller/oomkiller here.
>   
> (I would kill low_order_gfp_flags / high_order_gfp_flags and avoid the
> obfuscation.)
> -Chris

Yeah, I think this all needs some refactoring. The high_order/low_order
flags were originally for the system heap to allocate pages for the
page pool and I don't think they should be reused for the contig heap.
I'll see about doing a refactor.

Thanks for the review!

Laura