Re: [Qemu-devel] Fwd: qemu drive mirror assert fault

[Fam Zheng]

On Tue, 05/05 13:48, Kevin Wolf wrote:
> Am 05.05.2015 um 12:27 hat Fam Zheng geschrieben:
> > On Tue, 05/05 12:13, Paolo Bonzini wrote:
> > > 
> > > 
> > > On 05/05/2015 09:36, Fam Zheng wrote:
> > > I don't think this commit is relevant.  The bug is caused by
> > > bdrv_co_discard clearing the bitmap (which is not supported during
> > > iteration, and is caught by the assertion).  That bdrv_reset_dirty is
> > > not removed by the commit you pointed out.
> > 
> > You are right, I think the bug is the bdrv_reset_dirty in bdrv_co_discard. 
> > The
> > discard may zero out the src area (depending on the backend), we should
> > probably replicate this to dest (actual data or the discard operation), 
> > rather
> > than forgetting the change.
> 
> It depends. In its basic form, bdrv_discard() just means "I don't care
> about the data any more". Then clearing the dirty bitmap is correct.
> 
> The content is only important if the caller used discard to write zeros
> because can_write_zeroes_with_unmap = true. Do we have any such callers
> apart from qemu-img convert?

The guest could be taking advantage of this "can_write_zeroes_with_unmap"..

Fam