Re: [Qemu-block] [Qemu-devel] [PATCH v10 08/10] Implement new driver for block replication

[Fam Zheng]

On Tue, 10/13 17:46, Wen Congyang wrote:
> On 10/13/2015 05:41 PM, Fam Zheng wrote:
> > On Tue, 10/13 16:59, Wen Congyang wrote:
> >> On 10/13/2015 12:25 AM, Stefan Hajnoczi wrote:
> >>> On Fri, Sep 25, 2015 at 02:17:36PM +0800, Wen Congyang wrote:
> >>>> +static void backup_job_completed(void *opaque, int ret)
> >>>> +{
> >>>> +    BDRVReplicationState *s = opaque;
> >>>> +
> >>>> +    if (s->replication_state != BLOCK_REPLICATION_DONE) {
> >>>> +        /* The backup job is cancelled unexpectedly */
> >>>> +        s->error = -EIO;
> >>>> +    }
> >>>> +
> >>>> +    bdrv_op_block(s->hidden_disk, BLOCK_OP_TYPE_BACKUP_TARGET,
> >>>> +                  s->active_disk->backing_blocker);
> >>>> +    bdrv_op_block(s->secondary_disk, BLOCK_OP_TYPE_BACKUP_SOURCE,
> >>>> +                  s->hidden_disk->backing_blocker);
> >>>> +
> >>>> +    bdrv_put_ref_bh_schedule(s->secondary_disk);
> >>>
> >>> Why is bdrv_put_ref_bh_schedule() necessary?
> >>
> >> It is copied from block_job_cb(). According to the comments in 
> >> bdrv_put_ref_bh_schedule():
> >> /*
> >>  * Release a BDS reference in a BH
> >>  *
> >>  * It is not safe to use bdrv_unref() from a callback function when the 
> >> callers
> >>  * still need the BlockDriverState.  In such cases we schedule a BH to 
> >> release
> >>  * the reference.
> >>  */
> >>
> >> If the comment is right, I think it is necessary to call 
> >> bdrv_put_ref_bh_schedule() here.
> >> Because the job is created on the BDS s->secondary disk, 
> >> backup_job_completed() is
> >> called in block_job_completed(), and we will still use s->secondary_disk 
> >> in block_job_release().
> > 
> > Where is the matching bdrv_ref called?
> 
> It is in block_job_create()....
> 
> source: we call in bdrv_ref() in block_job_create(), and the user should 
> unref it.
> target: the user call bdrv_ref(), and we will unref it in the job
> 
> I don't know why we design it like this...
> 

Maybe it's better to unref it in block_job_release. Then we can simply drop
bdrv_put_ref_bh_schedule.

Fam