Re: [Qemu-devel] [PATCH 5/5] qcow2: merge is_zero_cluster helpers into qcow2_co_write_zeroes

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 2469 bytes --]

On 05/17/2016 03:15 AM, Denis V. Lunev wrote:
> They are used once only. This makes code more compact.
> 
> The patch also improves comments in the code.
> 
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> CC: Kevin Wolf <kwolf@redhat.com>
> ---
>  block/qcow2.c | 39 +++++++++++++++------------------------
>  1 file changed, 15 insertions(+), 24 deletions(-)
> 

> @@ -2439,20 +2418,32 @@ static coroutine_fn int qcow2_co_write_zeroes(BlockDriverState *bs,
>                                         nb_sectors);
>  
>      if (head != 0 || tail != 0) {
> -        int64_t cl_start = sector_num - head;
> +        BlockDriverState *file;
> +        uint64_t off;
> +        int nr;
> +
> +        int64_t cl_start = sector_num - head, res;
>  
>          assert(cl_start + s->cluster_sectors >= sector_num + nb_sectors);
>  
>          sector_num = cl_start;
>          nb_sectors = s->cluster_sectors;
>  
> -        if (!is_zero_cluster(bs, sector_num)) {
> +        /* check that the cluster is zeroed taking into account entire
> +           backing chain */
> +        nr = s->cluster_sectors;
> +        res = bdrv_get_block_status_above(bs, NULL, cl_start,
> +                                          s->cluster_sectors, &nr, &file);
> +        if (res < 0 || !(res & BDRV_BLOCK_ZERO)) {
>              return -ENOTSUP;
>          }

This is somewhat pessimistic in certain cases.  Suppose I have the
following cluster (borrowing from Kevin's nice notation in 154):

4k: -- -- XX --

and issue a 'write -z 2k 1k'

As written, the code will see that bdrv_get_block_status_above() does
NOT have all ZERO for the cluster, so it will return -ENOTSUP; we will
then call into the fallbacks and write explicit zeroes, so that the top
file now has an allocated cluster full of zero.  But if we were a bit
smarter, we would only check the allocation status of the backing
sectors that do not overlap with our write_zeroes request, because the
remaining sectors are about to be overwritten; in this case, we can
specifically optimize to write a zero cluster without allocation.

You'll also need to rebase this series on top of my pending work to
rewrite bdrv_co_write_zeroes to instead be bdrv_co_pwrite_zeroes with a
byte interface (patches to be posted later today), if mine gets reviewed
first.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]