On 17/11/2015 11:20, Fam Zheng wrote:
> When a request R is absorbed by request M, it is appended to the
> "mr_next" queue led by M, and is completed together with the completion
> of M, in virtio_blk_rw_complete.
>
> During DMA restart in virtio_blk_dma_restart_bh, requests in s->rq are
> parsed and submitted again, possibly with a stale req->mr_next. It could
> be a problem if the request merging in virtio_blk_handle_request hasn't
> refreshed every mr_next pointer, in which case, virtio_blk_rw_complete
> could walk through unexpected requests following the stale pointers.
>
> Fix this by unsetting the pointer in virtio_blk_rw_complete. It is safe
> because this req is either completed and freed right away, or it will be
> restarted and parsed from scratch out of the vq later.
>
> Signed-off-by: Fam Zheng <f...@redhat.com>
>
> ---
>
> v3: Fix as Stefan suggested.
> ---
> hw/block/virtio-blk.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
> index e70fccf..848f3fe 100644
> --- a/hw/block/virtio-blk.c
> +++ b/hw/block/virtio-blk.c
> @@ -112,6 +112,10 @@ static void virtio_blk_rw_complete(void *opaque, int ret)
> * happen on the other side of the migration).
> */
> if (virtio_blk_handle_rw_error(req, -ret, is_read)) {
> + /* Break the link in case the next request is added to the
> + * restart queue and is going to be parsed from the ring
> again.
> + */
> + req->mr_next = NULL;
> continue;
> }
> }
>
This is now a write-after-free for rerror/werror=stop. The right place
to set req->mr_next is inside virtio_blk_handle_rw_error, I think.
Paolo
