On 09/02/2014 12:59 PM, Max Reitz wrote: >> Consider refcount_order == 0 (that is, no shared blocks, ALL blocks have >> at most refcount 1). Then, refcount_bits is (1 << 0) == 1. But 1/8 in >> integer math truncates to 0 (oops, division by zero is undefined); when >> in reality, the expression you want here is (cluster_size * 8 / >> refcount_bits). > > If it is integer division, that is. ;-) > > I'm counting on you accepting "cluster_size * 8 / refcount_bits" and not > rejecting it because of a possible integer overflow. *g*
We already document that qemu's maximum cluster size is 2M; and 2M*8 is less than 32 bits :) Maybe the qcow2 spec allows the theoretical file with a cluster size of 512M, where overflow then matters. But you are correct that I won't reject your patch, given that qemu doesn't parse all possible theoretical qcow2 files :) -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
