Re: [Qemu-block] [PATCH v2 5/8] qcow: Switch to a byte-based driver

[Eric Blake]

On 06/04/2018 04:33 PM, Jeff Cody wrote:
> On Thu, May 31, 2018 at 03:50:43PM -0500, Eric Blake wrote:
> > We are gradually moving away from sector-based interfaces, towards
> > byte-based.  The qcow driver is now ready to fully utilize the
> > byte-based callback interface, as long as we override the default
> > alignment to still be 512 (needed at least for asserts present
> > because of encryption, but easier to do everywhere than to audit
> > which sub-sector requests are handled correctly, especially since
> > we no longer recommend qcow for new disk images).
> >
> > Signed-off-by: Eric Blake <address@hidden>

> > -static coroutine_fn int qcow_co_readv(BlockDriverState *bs, int64_t sector_num,
> > -                         int nb_sectors, QEMUIOVector *qiov)
> > +static void qcow_refresh_limits(BlockDriverState *bs, Error **errp)
> > +{
> > +    /* At least encrypted images require 512-byte alignment. Apply the
> > +     * limit universally, rather than just on encrypted images, as
> > +     * it's easier to let the block layer handle rounding than to
> > +     * audit this code further. */
> > +    bs->bl.request_alignment = BDRV_SECTOR_SIZE;
> > +}
> > +
> > +static coroutine_fn int qcow_co_preadv(BlockDriverState *bs, uint64_t offset,
> > +                                       uint64_t bytes, QEMUIOVector *qiov,
> > +                                       int flags)
> >   {
> >       BDRVQcowState *s = bs->opaque;
> >       int offset_in_cluster;
> > @@ -624,9 +632,8 @@ static coroutine_fn int qcow_co_readv(BlockDriverState *bs, 
> > int64_t sector_num,
> >       QEMUIOVector hd_qiov;
> >       uint8_t *buf;
> >       void *orig_buf;
> > -    int64_t offset = sector_num * BDRV_SECTOR_SIZE;
> > -    int64_t bytes = nb_sectors * BDRV_SECTOR_SIZE;
> >
> > +    assert(!flags);
>
> Why this assert here and in the _pwritev()?

We're changing from an interface that didn't have flags to one that 
does, but we are not prepared to handle any flags, so the assert proves 
the block layer doesn't hand us any flags we aren't expecting (there are 
no block layer flags for pread at the moment; and no flags for pwrite 
because we didn't set bs->supported_write_flags).

--
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org