Re: [Qemu-block] does blk_commit_all need blk_all_next?

[Paolo Bonzini]

On Monday, October 24, 2016 8:12:48 PM, "Max Reitz" <address@hidden> wrote:
> On 24.10.2016 10:45, Paolo Bonzini wrote:
> > blk_commit_all is in block/block-backend.c only because it uses
> > blk_all_next.  This is also the only reason why it needs a stub
> > (stubs/blk-commit-all.c).
> > 
> > The only difference between blk_next and blk_all_next is that the latter
> > "iterates over all BlockBackends, even the ones which are hidden (i.e.
> > are not referenced by the monitor)".  Should blk_commit_all really
> > iterate over BlockBackends such as the NBD server or the block jobs'?
> 
> I guess bdrv_commit_all() did, so blk_commit_all() does now, too.
> 
> The issue is, though, that currently "hidden BB" is equivalent to
> "anonymous BB". And I think that Kevin is working towards basically all
> BBs being anonymous, i.e. hidden.
> 
> When using -drive without id but with node-name, then you'll get a
> hidden BB. So we can't get away without blk_all_next().  But maybe
> it would make sense to skip all BBs that are not attached to a
> device?

Yeah, it sounds like it shouldn't be blk_all_next() but another public
API.

Paolo