On 23.01.19 09:20, Vladimir Sementsov-Ogievskiy wrote:
> 16.01.2019 16:05, Max Reitz wrote:
>> On 14.01.19 15:48, Vladimir Sementsov-Ogievskiy wrote:
>>> 14.01.2019 17:13, Max Reitz wrote:
>>>> On 14.01.19 15:01, Vladimir Sementsov-Ogievskiy wrote:
>>>>> 14.01.2019 16:10, Max Reitz wrote:
>>>>>> On 29.12.18 13:20, Vladimir Sementsov-Ogievskiy wrote:
>>>>>>> Simplify backup_incremental_init_copy_bitmap using the function
>>>>>>> bdrv_dirty_bitmap_next_dirty_area.
>>>>>>>
>>>>>>> Note: move to job->len instead of bitmap size: it should not matter but
>>>>>>> less code.
>>>>>>>
>>>>>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
>>>>>>> ---
>>>>>>>     block/backup.c | 40 ++++++++++++----------------------------
>>>>>>>     1 file changed, 12 insertions(+), 28 deletions(-)
>>>>>>
>>>>>> Overall: What is this function even supposed to do?  To me, it looks
>>>>>> like it marks all areas in job->copy_bitmap dirty that are dirty in
>>>>>> job->sync_bitmap.
>>>>>>
>>>>>> If so, wouldn't just replacing this by hbitmap_merge() simplify things
>>>>>> further?
>>>>>>
>>>>>>> diff --git a/block/backup.c b/block/backup.c
>>>>>>> index 435414e964..fbe7ce19e1 100644
>>>>>>> --- a/block/backup.c
>>>>>>> +++ b/block/backup.c
>>>>>>> @@ -406,43 +406,27 @@ static int coroutine_fn backup_run_incremental(BackupBlockJob *job)
>>>>>>
>>>>>> [...]
>>>>>>
>>>>>>> +    while (bdrv_dirty_bitmap_next_dirty_area(job->sync_bitmap,
>>>>>>> +                                             &offset, &bytes))
>>>>>>> +    {
>>>>>>> +        uint64_t cluster = offset / job->cluster_size;
>>>>>>> +        uint64_t last_cluster = (offset + bytes) / job->cluster_size;
>>>>>>>     
>>>>>>> -        next_cluster = DIV_ROUND_UP(offset, job->cluster_size);
>>>>>>> -        hbitmap_set(job->copy_bitmap, cluster, next_cluster - cluster);
>>>>>>> -        if (next_cluster >= end) {
>>>>>>> +        hbitmap_set(job->copy_bitmap, cluster, last_cluster - cluster + 1);
>>>>>>
>>>>>> Why the +1?  Shouldn't the division for last_cluster round up instead?
>>>>>>
>>>>>>> +
>>>>>>> +        offset = (last_cluster + 1) * job->cluster_size;
>>>>>>
>>>>>> Same here.
>>>>>
>>>>> last cluster is not "end", but it's last dirty cluster. so number of dirty clusters is last_cluster - cluster + 1, and next offset is calculated through +1 too.
>>>>>
>>>>> If I round up division result, I'll get last for most cases, but "end" (next after the last), for the case when offset % job->cluster_size == 0, so, how to use it?
>>>>
>>>> Doesn't bdrv_dirty_bitmap_next_dirty_area() return a range [offset,
>>>> offset + bytes), i.e. where "offset + bytes" is the first clean offset?
>>>
>>> oops, you are right. then I need
>>> uint64_t last_cluster = (offset + bytes - 1) / job->cluster_size;
>>
>> That, or you just use a rounding up division and rename it from
>> last_cluster to end_cluster or first_clean_cluster or something (and
>> subsequently drop the +1s).
> 
> This will not work, as ((offset + bytes) / job->cluster_size) is not first clean cluster
> or end cluster. It's a cluster, where is first clean bit located, but it may have dirty
> bits too (or, may not).
> 
> So, to rewrite based on end_cluster, it should be calculated as
> 
> (offset + bytes - 1) / job->cluster_size + 1

That's why I asked "Shouldn't the division for last_cluster round up
instead?"

Max

> and, I'm going to do so, one "+1" instead of two, and, may be, a bit more understandable.
>