Re: [Qemu-devel] [PATCH 1/3] qcow2: Check bs->drv in copy_sectors()

[Laszlo Ersek <lersek@redhat.com>]

On 03/10/14 23:44, Max Reitz wrote:
> Before dereferencing bs->drv for a call to its member bdrv_co_readv(),
> copy_sectors() should check whether that pointer is indeed valid, since
> it may have been set to NULL by e.g. a concurrent write triggering the
> corruption prevention mechanism.
> 
> Signed-off-by: Max Reitz <mreitz@redhat.com>
> ---
> To be precise, this still is a race condition. If bs->drv is set to NULL
> after the check and before the call to bdrv_co_readv(), QEMU will
> obviously still crash. However, in order to circumvent this behavior, we
> would probably have to re-lock s->lock, check bs->drv, take the function
> pointer to bdrv_co_readv() and then unlock s->lock before the function
> is called. I found this rather ugly and therefore this still has a very
> small chance of running into a race condition.
> Therefore, I'm asking for your opinion on this, whether we can really
> take this chance or should rather "do it right". In fact, if I were a
> reviewer, I'd probably reject this patch and request the solution with
> the function pointer (if there is no better solution), but I was afraid
> to send such an ugly patch.
> ---
>  block/qcow2-cluster.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
> index 36c1bed..9499df9 100644
> --- a/block/qcow2-cluster.c
> +++ b/block/qcow2-cluster.c
> @@ -380,6 +380,10 @@ static int coroutine_fn copy_sectors(BlockDriverState *bs,
>  
>      BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
>  
> +    if (!bs->drv) {
> +        return -ENOMEDIUM;
> +    }
> +
>      /* Call .bdrv_co_readv() directly instead of using the public block-layer
>       * interface.  This avoids double I/O throttling and request tracking,
>       * which can lead to deadlock when block layer copy-on-read is enabled.
> 

I can't answer your question nor review this patch -- instead, I have a
question of my own: when you say "set to NULL by [...] the corruption
prevention mechanism", do you mean qcow2_pre_write_overlap_check():

        bs->drv = NULL; /* make BDS unusable */

If so: I thought that it was quite a bold move, but also that we'd find
the SIGSEGVs sooner or later... :)

Thanks
Laszlo