Re: [Qemu-devel] [PATCH 3/3] block: mirror - zero unallocated target sectors when zero init not present

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1641 bytes --]

On 09/28/2015 03:32 PM, Jeff Cody wrote:

> I guess that makes sense.  What about  the case when the target is a raw
> device without zero init?  There is no backing file... Of course, perhaps
> in the raw case the user should be using sync==full anyways.
> 
>>
>> 2) even with mode == "existing" you expect the data to be consistent at
>> the end of the mirroring
>>
> 
> The reason I added the "existing" exception was so the user could avoid the
> time penalty of zeroing out the data if they knew the target had already
> explicitly been zeroed.  Do you think it is fair to assume that if the user
> specified existing, that they take responsibility for setting up the target
> image how they like (including data initialization)?  Or should we add
> another option for mirror, to allow the user to bypass the zero fill?

mode == 'existing' puts the burden on the caller to ensure that the file
they are passing in starts with known contents (either contents don't
matter because we are doing sync == 'full' to write every sector, or
contents MUST initially match what the guest would see looking at the
backing image when doing a shallow clone).  But if there is a way for a
user to pass in an existing file which they have pre-zeroed, even though
the file would normally be treated as though it did not have zero fill,
then the option to bypass a redundant zero fill might be useful.  I'm
not sure it's worth implementing without a known user, though, and I
don't know that libvirt would use it.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]