Re: [PATCH] xen-block: Fix removal of backend instance via xenstore

[Paolo Bonzini <pbonzini@redhat.com>]

On 08/03/21 15:32, Anthony PERARD wrote:
> From: Anthony PERARD <anthony.perard@citrix.com>
> 
> Whenever a Xen block device is detach via xenstore, the image
> associated with it remained open by the backend QEMU and an error is
> logged:
>      qemu-system-i386: failed to destroy drive: Node xvdz-qcow2 is in use
> 
> This happened since object_unparent() doesn't immediately frees the
> object and thus keep a reference to the node we are trying to free.
> The reference is hold by the "drive" property and the call
> xen_block_drive_destroy() fails.
> 
> In order to fix that, we call drain_call_rcu() to run the callback
> setup by bus_remove_child() via object_unparent().
> 
> Fixes: 2d24a6466154 ("device-core: use RCU for list of children of a bus")
> 
> Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
> ---
> CCing people whom introduced/reviewed the change to use RCU to give
> them a chance to say if the change is fine.

If nothing else works then I guess it's okay, but why can't you do the 
xen_block_drive_destroy from e.g. an unrealize callback?

Paolo

> ---
>   hw/block/xen-block.c | 9 +++++++++
>   1 file changed, 9 insertions(+)
> 
> diff --git a/hw/block/xen-block.c b/hw/block/xen-block.c
> index a3b69e27096f..fe5f828e2d25 100644
> --- a/hw/block/xen-block.c
> +++ b/hw/block/xen-block.c
> @@ -972,6 +972,15 @@ static void xen_block_device_destroy(XenBackendInstance *backend,
>   
>       object_unparent(OBJECT(xendev));
>   
> +    /*
> +     * Drall all pending RCU callbacks as object_unparent() frees `xendev'
> +     * in a RCU callback.
> +     * And due to the property "drive" still existing in `xendev', we
> +     * cann't destroy the XenBlockDrive associated with `xendev' with
> +     * xen_block_drive_destroy() below.
> +     */
> +    drain_call_rcu();
> +
>       if (iothread) {
>           xen_block_iothread_destroy(iothread, errp);
>           if (*errp) {
>