Hi, On Thu, Mar 20, 2014 at 02:38:32PM +0100, Kevin Wolf wrote: > Am 17.03.2014 um 07:53 hat Hu Tao geschrieben: > > This adds a preallocation=full mode to qcow2 image creation, which > > creates a non-sparse image file. > > > > Signed-off-by: Hu Tao <hu...@cn.fujitsu.com> > > I see that you changed the implementation from bdrv_preallocate() to > bdrv_create() options in v4 after Stefan had asked whether > bdrv_preallocate() was really necessary or just complicating things. > > In hindsight, do you really think the new code is simpler? I certainly > think this one is rather adventurous and the formulas to precalculate > the image file size are easy to get wrong.
When I tried to implement the bdrv_preallocate() way, I found that it was inevitable to precalculate the size. For example, in qcow2::bdrv_preallcate() we have to call low level raw::bdrv_preallocate() first, but we have to pass in the actual image file size which we have to precalculate before qcow2::bdrv_preallocate() can create any metadata. Another problem I don't understand about bdrv_preallocate() is that how it ensures existing data not being corrupted for every call to bdrv_preallocate(). Regards, Hu Tao
