Re: [Qemu-devel] [PATCH v6 for 2.1 00/10] Modify block jobs to use node-names

[Jeff Cody]

On Tue, Jun 24, 2014 at 10:48:52AM +0800, Fam Zheng wrote:
> On Mon, 06/23 21:08, Stefan Hajnoczi wrote:
> > On Thu, Jun 19, 2014 at 12:26:00PM -0400, Jeff Cody wrote:
> > > On Thu, Jun 19, 2014 at 05:17:16PM +0800, Stefan Hajnoczi wrote:
> > > > On Tue, Jun 17, 2014 at 05:53:48PM -0400, Jeff Cody wrote:
> > > > Let's discuss this topic in a sub-thread and figure out what to do for
> > > > QEMU 2.1.  This is an important issue to solve before the release
> > > > because we can't change QMP command semantics easily later.
> > > > 
> > > > My questions are:
> > > > a. How do we fix resize, snapshot-sync, etc?  It seems like we need to
> > > >    propagate child op blockers.
> > > > 
> > > > b. Is it a good idea to perform op blocker checks on the root node?
> > > >    It's inconsistent with resize, snapshot-sync, etc.  Permissions in
> > > >    BDS graphs with multiple root nodes (e.g. guest device and NBD
> > > >    run-time server) will be different depending on which root you
> > > >    specify.
> > > 
> > > I don't think (b) is the ultimate solution.  It is used as a stop-gap
> > > because op blockers in the current implementation is essentially
> > > analogous to the in-use flag.  But is it good enough for 2.1?  If
> > > *everything* checks the topmost node in 2.1, then I think we are OK in
> > > all cases except where images files share a common BDS.
> > 
> > Checking op blockers on the root node as a stop-gap is a good idea.
> > Let's apply it across all commands (e.g. snapshot-sync, resize).
> > 
> > Fam pointed out that this approach is vulnerable to blockdev-add, where
> > blockers could be set/checked on an incomplete BDS graph (since you can
> > add new nodes on top).  Do we need to move the blockers up the graph if
> > a new root node is inserted?
> 
> My concern is if we allow adding new root on top, it's not easy to know the
> real root then.
> 
> To give an example:
> 
> If we have
> 
>     [base id=""] <- [active id="drive0" blockers=...]
> 
> When user does
> 
>     (QMP) block-commit device="drive0" ...
> 
> We should check drive0, which is OK.
> 
> Then, assume user adds a new root on top, we would take care of moving the
> blockers:
> 
>     [base id=""] <- [active id="drive0"] <- [active id="drive1" blockers=]
> 
> At this point, what if user does something on drive0 again?
> 
>     (QMP) block-commit device="drive0" ...
> 
> The right thing to do is to check blockers on "drive1", since it's the real
> root now.  But how do we know? Do we need to add a back reference pointer
> ->overlap_hd in BDS, or do we maintain a look up table, or do we search all 
> BDS
> graphs to figure out?
> 
> None is easier than if we put the blockers in the bottom BDS, in the first
> place:
> 
>     [base id="" blockers=...] <- [active id="drive0"]
>                 ^^^^^^^^^^^^
> 

I think you are right.  If we place the blocker at the bottom-most
BDS, then that would be a more restrictive blocker.  This may end up
being more restrictive than needed, but more importantly it should
make everything safe.

Also, it is an easy change for 2.1 - just call bdrv_find_base(bs), and
set/check/clear blockers on the returned BDS.

> Even if user adds a new root, we don't need to worry about moving blockers,
> because the bottom is not changed.
> 
>     [base id="" blockers=...] <- [active id="drive0"] <- [active id="drive1"]
> 
> Checking the blockers are easy, either for drive0 or drive1: just follow the
> backing chain until getting to the end.
> 
> Fam