Re: [PATCH resend] libxl, hotplug/Linux: default to phy backend for raw format file, take 2

[Ian Campbell <Ian.Campbell@citrix.com>]

On Mon, 2015-01-12 at 11:17 +0100, Fabio Fantoni wrote:
> Il 12/01/2015 11:06, Ian Campbell ha scritto:
> > On Mon, 2015-01-12 at 10:39 +0100, Fabio Fantoni wrote:
> >> Il 12/01/2015 10:31, Ian Campbell ha scritto:
> >>> On Mon, 2015-01-12 at 10:15 +0100, Fabio Fantoni wrote:
> >>>> In the meantime, I saw this:
> >>>> http://lists.xen.org/archives/html/xen-users/2015-01/msg00047.html
> >>>> Based on the post above seems that phy will have important risk of data
> >>>> loss if I understand good, from Ian Campbell post:
> >>>>> xl also uses qdisk for raw disk images instead of loop+blkback which
> >>>>> xend used, because there are concerns that the loop driver can lead to
> >>>>> data loss (by not implementing direct i/o the underlying device, and/or
> >>>>> not handling flushes correct, my memory is a bit fuzzy).
> >>> Stefano already corrected me on this in this very thread.
> >>>
> >>> Ian.
> >>>
> >> Thanks for your reply.
> >> I saw other posts about and if I understand good when O_DIRECT patches
> >> will be in upstream loop driver the data loss risk will be solved, right?
> > Stefano says that O_DIRECT is not needed, only correct barrier semantics
> > are and he believes those are correctly implemented.
> 
> I not understand if manual changes and/or settings are needed, about 
> this I mean:
> > only correct barrier semantics are and he believes those are correctly implemented
> If yes what exactly?

If Stefano is correct then it should all just work, no manual steps
needed (if manual steps were needed we wouldn't be taking this patch).

Ian.