On Thu, Jan 25, 2018 at 03:31:42PM +0800, Jason Wang wrote:
> We try to allocate one more entry for lockless peeking. The adding
> operation may overflow which causes zero to be passed to kmalloc().
> In this case, it returns ZERO_SIZE_PTR without any notice by ptr
> ring. Try to do producing or consuming on such ring will lead NULL
> dereference. Fix this detect and fail early.
>
> Fixes: bcecb4bbf88a ("net: ptr_ring: otherwise safe empty checks can overrun
> array bounds")
> Reported-by: syzbot+87678bcf753b44c39...@syzkaller.appspotmail.com
> Cc: John Fastabend <john.fastab...@gmail.com>
> Signed-off-by: Jason Wang <jasow...@redhat.com>
Ugh that's just way too ugly.
I'll work on dropping the extra + 1 - but calling this
function with -1 size is the real source of the bug.
Do you know how come we do that?
> ---
> include/linux/ptr_ring.h | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/include/linux/ptr_ring.h b/include/linux/ptr_ring.h
> index 9ca1726..3f99484 100644
> --- a/include/linux/ptr_ring.h
> +++ b/include/linux/ptr_ring.h
> @@ -453,6 +453,8 @@ 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)
> {
> + if (unlikely(size + 1 == 0))
> + return NULL;
> /* Allocate an extra dummy element at end of ring to avoid consumer head
> * or produce head access past the end of the array. Possible when
> * producer/consumer operations and __ptr_ring_peek operations run in
> --
> 2.7.4
