Re: [Qemu-devel] [PATCH] docs: Define refcount_bits value

[Stefan Hajnoczi]

On Fri, May 23, 2014 at 07:56:16AM -0600, Eric Blake wrote:
> On 05/23/2014 07:41 AM, Maria Kustova wrote:
> 
> >           96 -  99:  refcount_order
> >                      Describes the width of a reference count block entry 
> > (width
> > -                    in bits = 1 << refcount_order). For version 2 images, 
> > the
> > -                    order is always assumed to be 4 (i.e. the width is 16 
> > bits).
> > +                    in bits: refcount_bits = 1 << refcount_order). For 
> > version 2
> > +                    images, the order is always assumed to be 4
> > +                    (i.e. refcount_bits = 16).
> 
> In light of all the recent CVE fixes (and possibly a separate patch if
> any code is broken), I wonder if we need more work to ensure that
> refcount_order is capped to a worthwhile maximum rather than causing
> undefined behavior.  That is, a refcount_order of 0x10004 should be an
> error, and not a synonym of refcount_order of 4, since '1 << 0x10004' is
> undefined.
> 
> Furthermore, this raises some questions in my mind. Later on, we document:
> 
>     refcount_block_entries = (cluster_size / sizeof(uint16_t))
> 
> which implies a hard cap of refcount_bits=16 as the maximum, which in
> turn implies a hard cap of refcount_order of 4 as the maximum.  Or is it
> possible to specify a larger refcount_order, in which case
> refcount_block_entries is dynamically sized to uint32_t, and in which
> case the rest of the docs need to be fixed to accommodate that?
> 
> Also,
> 
> Refcount block entry (x = refcount_bits - 1):
> 
>     Bit  0 -  x:    Reference count of the cluster. If refcount_bits
> implies a
>                     sub-byte width, note that bit 0 means the least
> significant
>                     bit in this context.
> 
> but nothing is said about bits x+1 - 15 (which only exist when
> refcount_order < 4, but which presumably must be all 0 bits for the file
> to be valid).

Only refcount_order = 4 is supported by QEMU at the moment.

I agree the spec could be made a bit clearer though.  Maybe Kevin wants
to send a patch to explain the details of refcount entry sizing.

Stefan