Re: [Qemu-devel] [PATCH v2 14/17] block: Resize bitmaps on bdrv_truncate

[Max Reitz <mreitz@redhat.com>]

On 2015-03-03 at 16:24, John Snow wrote:
>
>
> On 03/03/2015 11:02 AM, Max Reitz wrote:
>> On 2015-03-02 at 18:20, John Snow wrote:
>>> Signed-off-by: John Snow <jsnow@redhat.com>
>>> ---
>>>   block.c                | 22 ++++++++++++++++++++
>>>   include/block/block.h  |  1 +
>>>   include/qemu/hbitmap.h | 10 +++++++++
>>>   util/hbitmap.c         | 55
>>> ++++++++++++++++++++++++++++++++++++++++++++++++++
>>>   4 files changed, 88 insertions(+)
>>>
>>> diff --git a/block.c b/block.c
>>> index e6b2696..5eaa874 100644
>>> --- a/block.c
>>> +++ b/block.c
>>> @@ -3543,6 +3543,7 @@ int bdrv_truncate(BlockDriverState *bs, int64_t
>>> offset)
>>>       ret = drv->bdrv_truncate(bs, offset);
>>>       if (ret == 0) {
>>>           ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
>>> +        bdrv_dirty_bitmap_truncate(bs);
>>>           if (bs->blk) {
>>>               blk_dev_resize_cb(bs->blk);
>>>           }
>>> @@ -5562,6 +5563,27 @@ BdrvDirtyBitmap
>>> *bdrv_reclaim_dirty_bitmap(BlockDriverState *bs,
>>>       return parent;
>>>   }
>>> +static void dirty_bitmap_truncate(BdrvDirtyBitmap *bitmap, uint64_t
>>> size)
>>> +{
>>> +    /* Should only be frozen during a block backup job, which should
>>> have
>>> +     * blocked any resize actions. */
>>> +    assert(!bdrv_dirty_bitmap_frozen(bitmap));
>>> +    hbitmap_truncate(bitmap->bitmap, size);
>>> +}
>>> +
>>> +void bdrv_dirty_bitmap_truncate(BlockDriverState *bs)
>>> +{
>>> +    BdrvDirtyBitmap *bitmap;
>>> +    uint64_t size = bdrv_nb_sectors(bs);
>>> +
>>> +    QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
>>> +        if (bdrv_dirty_bitmap_frozen(bitmap)) {
>>> +            continue;
>>> +        }
>>
>> Hm!
>>
>> The assert() above doesn't do anything because this condition will just
>> skip over frozen bitmaps. Maybe we should drop it?
>>
>> I guess my R-b stands since bdrv_dirty_bitmap_frozen(bitmap) will always
>> be false anyway (hence the assertion), but it would stand with this
>> block dropped, too.
>>
>> Max
>>
>
> It's there for parity with the other bdrv helpers that perform a 
> similar check. I am guarding against, in the future, any potential 
> users from deciding to call this function directly for whichever reason.
>
> Requesting an action on a drive that contains bitmaps that are not 
> ready for the action is OK. Requesting an action on a bitmap that is 
> not ready for the action should never happen, hence the 'documentation 
> assert.'
>
> My default action will be to leave it in.

Technically that's fine (which is why I said my R-b stands), my only 
problem with that is that if something really strange happens and we 
actually end up with a frozen bitmap here, the action we're taking is 
wrong. The bitmap will have the wrong size after the BDS was truncated, 
and that's not right.

I'd find just calling dirty_bitmap_truncate() regardless of the frozen 
state more correct, the idea being "We want to resize all bitmaps, 
because all bitmaps attached to this BDS need to be resized", and then, 
in dirty_bitmap_truncate(), we notice that we cannot resize a frozen 
bitmap. But we need to. So the assertion fails.

Max

> --js