Re: [Qemu-stable] [PATCH for-2.10 v3] block: Skip implicit nodes in query-block/blockstats

[Kevin Wolf]

Am 20.07.2017 um 16:06 hat Max Reitz geschrieben:
> On 2017-07-20 15:11, Kevin Wolf wrote:
> > @@ -133,13 +133,21 @@ BlockDeviceInfo *bdrv_block_device_info(BlockBackend 
> > *blk,
> >              qapi_free_BlockDeviceInfo(info);
> >              return NULL;
> >          }
> > +
> >          if (bs0->drv && bs0->backing) {
> > +            info->backing_file_depth++;
> >              bs0 = bs0->backing->bs;
> >              (*p_image_info)->has_backing_image = true;
> >              p_image_info = &((*p_image_info)->backing_image);
> >          } else {
> >              break;
> >          }
> > +
> > +        /* Skip automatically inserted nodes that the user isn't aware of 
> > for
> > +         * query-block (blk != NULL), but not for query-named-block-nodes 
> > */
> > +        while (blk && bs0 && bs0->drv && bs0->implicit) {
> > +            bs0 = backing_bs(bs0);
> > +        }
> 
> If the bottom-most backing file is implicit, this will leave bs0 to be
> NULL. The surrounding loop does not look like it could cope well with
> that. (And even if it could, we would have to reset has_backing_image to
> false, wouldn't we?)
> 
> Not necessarily an issue here because all of our implicit nodes are
> somewhere in between, but still...
> 
> (And I just saw you're even asserting this in another place...)
> 
> >      }
> >  
> >      return info;
> 
> [...]
> 
> > @@ -446,6 +459,14 @@ static BlockStats *bdrv_query_bds_stats(const 
> > BlockDriverState *bs,
> >          return s;
> >      }
> >  
> > +    /* Skip automatically inserted nodes that the user isn't aware of in
> > +     * a BlockBackend-level command. Stay at the exact node for a 
> > node-level
> > +     * command. */
> > +    while (blk_level && bs->drv && bs->implicit) {
> > +        bs = backing_bs(bs);
> > +        assert(bs);
> 
> (...namely here.)
> 
> I extremely doubt this assertion will hold in the future, but well, it's
> good enough for now.
> 
> So, I've always wanted "file" to be the default child for block filters.
> This patch very much establishes that "backing" is the default child, right?

Manos is going to address this when he uses an I/O throttling filter
node to implement the legacy options. The plan there is to replace it
with some function that asserts that there is only one child (because
only filter nodes are supposed to be implicit) and then uses that child,
no matter whether it's bs->file, bs->backing or something else.

I just tried to keep things simple for 2.10.

Kevin

pgpy8HtpouUj2.pgp
Description: PGP signature