On 08/26/2014 09:46 AM, Peter Krempa wrote:
On 08/26/14 13:21, Eric Blake wrote:
The hard part of managing the disk copy is already coded; all this had to do was convert the XML and virTypedParameters into the internal representation.
With this patch, all blockcopy operations that used the old API should also work via the new API. Additional extensions, such as supporting the granularity tunable or a network rather than file destination, will be added as later patches.
+ ret = qemuDomainBlockCopyCommon(&vm, dom->conn, disk, dest, + bandwidth, flags);
I don't think you need to do the dance with passing vm as an pointer and having it returned. Insted you can unlock/free it in qemuDomainBlockCopyCommon.
This will also clean the messy part in 5/8.
But it was the fact that there was transfer semantics that confused me. qemuDomainBlockCopyCommon cannot return early if it MUST free vm on all exit paths, and I found it easier to reason about the code if the function that allocated vm also freed it. I'll think about whether to change anything for v3.
+ + cleanup: + virStorageSourceFree(dest); + if (vm) + virObjectUnlock(vm); + return ret; +} + + static int
ACK
-- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org