Am 22.02.2011 16:02, schrieb Stefan Hajnoczi: > On Tue, Feb 22, 2011 at 10:18 AM, <jes.soren...@redhat.com> wrote: >> + else if ((laiocb->type == QEMU_AIO_WRITE) && (ret >= 0) && >> + (ret < laiocb->nbytes)) >> + ret = -ENOSPC; > > Why is write special?
I think we need the change reads, too. However not to return ENOSPC, but to return zeros instead (this is what the synchronous raw_read does, and pwrite relies on it - once we make pwrite async, we'll need this). > Why are we even allowing requests that extend beyond the end of the > device? Is the LVM volume marked growable in the QEMU block layer? Might well be a qcow2 on LVM case that Jes was debugging. Kevin