Re: [PATCH net V3 1/2] ptr_ring: try vmalloc() when kmalloc() fails

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Feb 08, 2018 at 02:58:40PM +0800, Jason Wang wrote:
> 
> 
> On 2018年02月08日 12:45, Michael S. Tsirkin wrote:
> > On Thu, Feb 08, 2018 at 11:59:24AM +0800, Jason Wang wrote:
> > > This patch switch to use kvmalloc_array() for using a vmalloc()
> > > fallback to help in case kmalloc() fails.
> > > 
> > > Reported-by: syzbot+e4d4f9ddd4295539735d@syzkaller.appspotmail.com
> > > Fixes: 2e0ab8ca83c12 ("ptr_ring: array based FIFO for pointers")
> > I guess the actual patch is the one that switches tun to ptr_ring.
> 
> I think not, since the issue was large allocation.
> 
> > 
> > In fact, I think the actual bugfix is patch 2/2. This specific one
> > just makes kmalloc less likely to fail but that's
> > not what syzbot reported.
> 
> Agree.
> 
> > 
> > Then I would add this patch on top to make kmalloc less likely to fail.
> 
> Ok.
> 
> > > Signed-off-by: Jason Wang <jasowang@redhat.com>
> > 
> > 
> > > ---
> > >   include/linux/ptr_ring.h | 10 +++++-----
> > >   1 file changed, 5 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/include/linux/ptr_ring.h b/include/linux/ptr_ring.h
> > > index 1883d61..2af71a7 100644
> > > --- a/include/linux/ptr_ring.h
> > > +++ b/include/linux/ptr_ring.h
> > > @@ -466,7 +466,7 @@ static inline int ptr_ring_consume_batched_bh(struct ptr_ring *r,
> > >   static inline void **__ptr_ring_init_queue_alloc(unsigned int size, gfp_t gfp)
> > >   {
> > > -	return kcalloc(size, sizeof(void *), gfp);
> > > +	return kvmalloc_array(size, sizeof(void *), gfp | __GFP_ZERO);
> > >   }
> > >   static inline void __ptr_ring_set_size(struct ptr_ring *r, int size)
> > This implies a bunch of limitations on the flags. From kvmalloc_node
> > docs:
> > 
> >   * Reclaim modifiers - __GFP_NORETRY and __GFP_NOFAIL are not supported.
> >   * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
> >   * preferable to the vmalloc fallback, due to visible performance drawbacks.
> > 
> > Fine with all the current users, but if we go this way, please add
> > documentation so future users don't misuse this API.
> 
> I suspect this is somehow a overkill since this means we need sync with
> mm/vmalloc changes in the future to keep it synced.
> 
> > Alternatively, test flags and call kvmalloc or kcalloc?
> 
> Similar to the above issue, I would rather leave it as is.
> 
> Thanks

How do we prevent someone from inevitably trying to use this with
GFP_ATOMIC?

> > 
> > 
> > > @@ -601,7 +601,7 @@ static inline int ptr_ring_resize(struct ptr_ring *r, int size, gfp_t gfp,
> > >   	spin_unlock(&(r)->producer_lock);
> > >   	spin_unlock_irqrestore(&(r)->consumer_lock, flags);
> > > -	kfree(old);
> > > +	kvfree(old);
> > >   	return 0;
> > >   }
> > > @@ -641,7 +641,7 @@ static inline int ptr_ring_resize_multiple(struct ptr_ring **rings,
> > >   	}
> > >   	for (i = 0; i < nrings; ++i)
> > > -		kfree(queues[i]);
> > > +		kvfree(queues[i]);
> > >   	kfree(queues);
> > > @@ -649,7 +649,7 @@ static inline int ptr_ring_resize_multiple(struct ptr_ring **rings,
> > >   nomem:
> > >   	while (--i >= 0)
> > > -		kfree(queues[i]);
> > > +		kvfree(queues[i]);
> > >   	kfree(queues);
> > > @@ -664,7 +664,7 @@ static inline void ptr_ring_cleanup(struct ptr_ring *r, void (*destroy)(void *))
> > >   	if (destroy)
> > >   		while ((ptr = ptr_ring_consume(r)))
> > >   			destroy(ptr);
> > > -	kfree(r->queue);
> > > +	kvfree(r->queue);
> > >   }
> > >   #endif /* _LINUX_PTR_RING_H  */
> > > -- 
> > > 2.7.4