[Devel] [PATCH vz9 v2] dm-qcow2: support mixed compression images

[Pavel Tikhomirov]

On 2/19/25 12:36, Alexander Atanasov wrote:
> On 19.02.25 4:27, Pavel Tikhomirov wrote:
>>
>>
>> On 2/18/25 22:06, Alexander Atanasov wrote:
>>> Currently code uses only top qcow2 delta to handle decompression.
>>> Images have compression type in the qcow2 header already set when 
>>> loaded.
>>> To support mixed compression switch to use qio->qcow2 and respective
>>> qcow2 header so the decompression is performed with the correct type.
>>> Remove qcow2 argument in functions called from decompression where
>>> possible and use qio->qcow2 which is the respective qcow2 image for
>>> the qio, a qio references only one image.
>>>
>>> https://virtuozzo.atlassian.net/browse/VSTOR-97155
>>> Signed-off-by: Alexander Atanasov <alexander.atanasov at virtuozzo.com>
>>> ---
>>>   drivers/md/dm-qcow2-map.c | 70 +++++++++++++++++++++++----------------
>>>   1 file changed, 42 insertions(+), 28 deletions(-)
>>>
>>>
>>> v1->v2:
>>>   - drop extra stream reset for zstd
>>>
>>> diff --git a/drivers/md/dm-qcow2-map.c b/drivers/md/dm-qcow2-map.c
>>> index ccc8d4484049..568bf68981b4 100644
>>> --- a/drivers/md/dm-qcow2-map.c
>>> +++ b/drivers/md/dm-qcow2-map.c
>>> @@ -3009,8 +3009,9 @@ static int copy_buf_to_bvec_iter(const struct 
>>> bio_vec *bvec,
>>>       return ret;
>>>   }
>>> -static int copy_clu_part_to_qio(struct qcow2 *qcow2, const void 
>>> *buf, struct qio *qio)
>>> +static int copy_clu_part_to_qio(const void *buf, struct qio *qio)
>>>   {
>>> +    struct qcow2 *qcow2 = qio->qcow2;
>>>       u32 max, seek, clu_size = qcow2->clu_size;
>>>       seek = bytes_off_in_cluster(qcow2, qio);
>>> @@ -3047,7 +3048,7 @@ static int copy_zcow_slice(loff_t start, loff_t 
>>> end, void *qio_p,
>>>       return copy_buf_to_bvec_iter(bvec, &iter, buf + off, clu_size - 
>>> off);
>>>   }
>>> -static int prepare_zcow_slices(struct qcow2 *qcow2, void *buf, 
>>> struct qio *qio)
>>> +static int prepare_zcow_slices(void *buf, struct qio *qio)
>>>   {
>>>       loff_t consumed = 0;
>>>       /* Place required slices in that pages like further COW expects */
>>> @@ -3655,46 +3656,59 @@ static int complete_metadata_writeback(struct 
>>> qcow2 *qcow2)
>>>   }
>>>   /* Process completed compressed READs */
>>> -static void process_compressed_read(struct qcow2 *qcow2, struct 
>>> list_head *read_list,
>>> +static void process_compressed_read(struct list_head *read_list,
>>>                       struct list_head *cow_list)
>>>   {
>>> +    struct qcow2 *qcow2, *qcow2_prev = NULL;
>>>       struct qcow2_bvec *qvec;
>>>       struct qio_ext *ext;
>>>       int ret;
>>>       void *buf = NULL, *arg;
>>>       struct qio *qio;
>>>       bool for_cow;
>>> +    size_t dctxlen_alloced = 0;
>>
>> Let's fix spelling in variable name: s/alloced/allocated/
> 
> there is nothing wrong with the spelling here.
> we do *alloc, and not *allocate - that is why to form a past participle 
> is to add an ed suffix. this form is widely used and recognised in 
> various projects both in source code and documentation. you will find it 
> in some dictionaries too. And if you check, you will find it in kernel 
> source code too, where it is used more than 500 times.

Ok. I just didn't see any "alloced" in the dictionaries:

https://www.collinsdictionary.com/conjugation/english/allocate
Past Participle
allocated

https://www.merriam-webster.com/dictionary/allocate

And my spell checker highlights it.

But it seems ok, as we already have it all over 
drivers/md/dm-qcow2-map.c, and codespell also allows it.

> 

-- 
Best regards, Tikhomirov Pavel
Senior Software Developer, Virtuozzo.