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

[Max Reitz <mreitz@redhat.com>]

On 2014-10-21 at 17:11, Kevin Wolf wrote:
> Am 21.10.2014 um 16:55 hat Max Reitz geschrieben:
>> On 2014-10-21 at 12:16, Max Reitz wrote:
>>> On 2014-10-21 at 11:59, Kevin Wolf wrote:
>>>> 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.
>>> Why not.
>>>
>>>>> +            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?
>>> Actually, there is no difference, because at the point of this
>>> patch, you cannot use BDRV_FIX_ERRORS without BDRV_FIX_LEAKS. But
>>> it'd be more correct, right.
>> Wait, it would not be more correct. The result of the
>> compare_refcounts() call inside of the "if (fix & BDRV_FIX_LEAKS)"
>> conditional block is ignored, its only purpose is to fix leaks
>> resulting from rebuild_refcount_structure().
>>
>> So the question is whether we should discard the result of that
>> compare_refcounts() call. I think we should. Its sole purpose is to
>> fix leaks due to the rebuilt refcount structures, and qemu-img will
>> double check anyway.
> Right, the other leaks should have been fixed by rebuilding the refcount
> structures. So what you're saying is that we could do this:
>
>      if (res->corruptions < old_res.corruptions) {
>          res->corruptions_fixed += old_res.corruptions - res->corruptions;
>      }
>      assert(res->leaks == 0);
>      res->leaks_fixed = old_res.leaks;
>
> If this weren't true, we'd ignore leaked clusters even with
> BDRV_FIX_LEAKS set.

Okay, so for obvious reasons this is not nice. Well, I don't know what 
will be worse. Writing the code which takes the leak fix result into 
account or having to review it. I think I'll just see how many new leaks 
appeared due to the rebuild and how many of those could not be fixed, 
and then add that result to res->leaks. That should work out without 
being too complicated.

Max