On 10/01/2013 09:50 AM, Kevin Wolf wrote: > Am 01.10.2013 um 17:41 hat Eric Blake geschrieben: >> On 10/01/2013 07:20 AM, Kevin Wolf wrote: >>> This series adds a blockdev-add QMP command and a row of cleanup patches to >>> separate legacy -drive/drive_add behaviour from blockdev_init() to the >>> wrapper >>> function drive_init(). >>> >>> v2: >>> >>> - Removed I/O throttling and copy-on-read options from the schema: Both are >>> candidates for becoming block filters instead. For the time being, you can >>> use the existing QMP commands in order to enable I/O throttling on a >>> device >>> created with blockdev-add in a second step. >> >> Tolerable as a stop-gap; but doesn't that mean that there is a window >> where throttling is not active? Anything that can restrict block >> operations must ultimately be specified atomically up front to avoid a >> non-deterministic burst during the race window. But in the interest of >> incremental improvements, I can live with the approach used here, while >> still waiting for throttling to be implemented as a true filter device. > > Immediately after blockdev-add, the block device is still completely > unused. You can set the throttling options before you do the > corresponding device-add or block job command or whatever you're > planning to do with the device.
Ah. So there's no race window after all. Good to know. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
