Re: [Qemu-devel] [PATCH v6 09/11] qcow2: Clean up after refcount rebuild

[Kevin Wolf <kwolf@redhat.com>]

Am 20.10.2014 um 16:35 hat Max Reitz geschrieben:
> Because the old refcount structure will be leaked after having rebuilt
> it, we need to recalculate the refcounts and run a leak-fixing operation
> afterwards (if leaks should be fixed at all).
> 
> Signed-off-by: Max Reitz <mreitz@redhat.com>
> Reviewed-by: Benoît Canet <benoit.canet@nodalink.com>
> ---
>  block/qcow2-refcount.c | 35 +++++++++++++++++++++++++++++++++++
>  1 file changed, 35 insertions(+)
> 
> diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
> index 75e726b..3730be2 100644
> --- a/block/qcow2-refcount.c
> +++ b/block/qcow2-refcount.c
> @@ -1956,12 +1956,47 @@ int qcow2_check_refcounts(BlockDriverState *bs, BdrvCheckResult *res,
>                        nb_clusters);
>  
>      if (rebuild && (fix & BDRV_FIX_ERRORS)) {
> +        BdrvCheckResult old_res = *res;
> +
>          fprintf(stderr, "Rebuilding refcount structure\n");
>          ret = rebuild_refcount_structure(bs, res, &refcount_table,
>                                           &nb_clusters);
>          if (ret < 0) {
>              goto fail;
>          }
> +
> +        res->corruptions = 0;
> +        res->leaks = 0;
> +
> +        /* Because the old reftable has been exchanged for a new one the
> +         * references have to be recalculated */
> +        rebuild = false;
> +        memset(refcount_table, 0, nb_clusters * sizeof(uint16_t));
> +        ret = calculate_refcounts(bs, res, 0, &rebuild, &refcount_table,
> +                                  &nb_clusters);
> +        if (ret < 0) {
> +            goto fail;
> +        }
> +
> +        if (fix & BDRV_FIX_LEAKS) {
> +            /* The old refcount structures are now leaked, fix it; the result
> +             * can be ignored */
> +            pre_compare_res = *res;

I would prefer using another local variable here. At the first sight
it's not quite clear which references to pre_compare_res correspond to
which state.

> +            compare_refcounts(bs, res, BDRV_FIX_LEAKS, &rebuild,
> +                              &highest_cluster, refcount_table, nb_clusters);
> +            if (rebuild) {
> +                fprintf(stderr, "ERROR rebuilt refcount structure is still "
> +                        "broken\n");
> +            }
> +            *res = pre_compare_res;
> +        }
> +
> +        if (res->corruptions < old_res.corruptions) {
> +            res->corruptions_fixed += old_res.corruptions - res->corruptions;
> +        }
> +        if (res->leaks < old_res.leaks) {
> +            res->leaks_fixed += old_res.leaks - res->leaks;
> +        }

For these numbers to be accurate, don't we need to run
compare_refcounts() unconditionally and only make BDRV_FIX_LEAKS
conditional?

Now that I've read the rest of the series, comparing res and old_res
actually makes sense, so maybe it's not necessary to introduce a
Qcow2CheckResult.

>      } else if (fix) {
>          if (rebuild) {
>              fprintf(stderr, "ERROR need to rebuild refcount structures\n");
> -- 
> 2.1.2

Kevin