Am 07.12.2011 13:06, schrieb Stefan Hajnoczi: > On Wed, Dec 7, 2011 at 12:02 PM, Kevin Wolf <kw...@redhat.com> wrote: >> Am 07.12.2011 12:50, schrieb Stefan Hajnoczi: >>> On Wed, Dec 7, 2011 at 11:45 AM, Kevin Wolf <kw...@redhat.com> wrote: >>>> Backing files may be smaller than the corresponding COW file. When >>>> reading directly from the backing file, qemu-img rebase must consider >>>> this and assume zero sectors after the end of backing files. >>>> >>>> Signed-off-by: Kevin Wolf <kw...@redhat.com> >>>> --- >>>> qemu-img.c | 42 +++++++++++++++++++++++++++++++++--------- >>>> 1 files changed, 33 insertions(+), 9 deletions(-) >>> >>> Reviewed-by: Stefan Hajnoczi <stefa...@linux.vnet.ibm.com> >>> >>> It almost feels like we want a bdrv_read() variation that does the >>> zeroing beyond end of image instead of duplicating this. >> >> Almost. If we have a third user, it's probably worth it. >> >> Actually, if we make a bdrv_co_readv_backing() instead of a synchronous >> one, qcow2 would be the third user (and maybe VMDK could use it, too). >> That would require moving qemu-img into a coroutine first, which I think >> you wanted to do anyway? > > If we can make qemu-img the last user of synchronous interfaces then > it would be nice to convert it. Right now I'm happy that we're > solidifying coroutines and converting synchronous hardware emulation, > hopefully we'll get to the point where there is no real reason to keep > the synchronous API around anymore.
The point is, if I were to introduce a bdrv_co_readv_backing() function, I certainly wouldn't like to add all the bdrv_(aio_)read/write emulation code for it, too. Kevin
