于 2013-4-3 17:03, Kevin Wolf 写道:
Am 03.04.2013 um 07:35 hat Wenchao Xia geschrieben:
于 2013-4-2 21:59, Kevin Wolf 写道:
Am 01.04.2013 um 12:01 hat Wenchao Xia geschrieben:
   Last operaton should be cancelled first.

Signed-off-by: Wenchao Xia <xiaw...@linux.vnet.ibm.com>

Should it? This commit message does little to convince me.

Kevin

   I think so, if two request are "snapshot_blkdev ide0 newimage.qcow2",
"snapshot_blkdev_internal ide0 snap0", then in rollback delete of
snap0 would be wrong.

I think this problem exists only because you didn't properly separate
the prepare and the commit stage. If any changes only took effect in a
non-failing commit stage, then you don't ever have to delete a snapshot.
Instead, you just abort the creation of the snapshot.

Kevin

  OK, I got your point, will split bdrv_snapshot_create() before adding
it in qmp_transaction.

--
Best Regards

Wenchao Xia


Reply via email to