Re: [Qemu-devel] [PULL 39/42] block: look for zero blocks in bs->file

[Paolo Bonzini]

Il 13/09/2013 11:14, Peter Lieven ha scritto:
> On 13.09.2013 11:10, Paolo Bonzini wrote:
>> Il 13/09/2013 11:08, Peter Lieven ha scritto:
>>> On 13.09.2013 10:00, Peter Lieven wrote:
>>>> On 06.09.2013 17:39, Stefan Hajnoczi wrote:
>>>>> From: Paolo Bonzini <address@hidden>
>>>>>
>>>>> Reviewed-by: Eric Blake <address@hidden>
>>>>> Signed-off-by: Paolo Bonzini <address@hidden>
>>>>> Signed-off-by: Stefan Hajnoczi <address@hidden>
>>>>> ---
>>>>>    block.c | 16 +++++++++++++++-
>>>>>    1 file changed, 15 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/block.c b/block.c
>>>>> index 3c39769..a325efc 100644
>>>>> --- a/block.c
>>>>> +++ b/block.c
>>>>> @@ -3075,7 +3075,7 @@ static int64_t coroutine_fn
>>>>> bdrv_co_get_block_status(BlockDriverState *bs,
>>>>>    {
>>>>>        int64_t length;
>>>>>        int64_t n;
>>>>> -    int64_t ret;
>>>>> +    int64_t ret, ret2;
>>>>>          length = bdrv_getlength(bs);
>>>>>        if (length < 0) {
>>>>> @@ -3117,6 +3117,20 @@ static int64_t coroutine_fn
>>>>> bdrv_co_get_block_status(BlockDriverState *bs,
>>>>>                }
>>>>>            }
>>>>>        }
>>>>> +
>>>>> +    if (bs->file &&
>>>>> +        (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
>>>>> +        (ret & BDRV_BLOCK_OFFSET_VALID)) {
>>>>> +        ret2 = bdrv_co_get_block_status(bs->file, ret >>
>>>>> BDRV_SECTOR_BITS,
>>>>> +                                        *pnum, pnum);
>>>>> +        if (ret2 >= 0) {
>>>>> +            /* Ignore errors.  This is just providing extra
>>>>> information, it
>>>>> +             * is useful but not necessary.
>>>>> +             */
>>>>> +            ret |= (ret2 & BDRV_BLOCK_ZERO);
>>>>> +        }
>>>>> +    }
>>>>> +
>>>>>        return ret;
>>>>>    }
>>>> for allocated blocks in iscsi this actually leads to 2 call outs to
>>>> iscsi_get_block_status.
>>>>
>>>> because the raw driver passes the get_block_status request to bs->file.
>>>>
>>> correct patch?
>> Correct but a bit hackish... Stefan, Kevin, any ideas?
> add a BDRV_BLOCK_RAW flag to the get_block_status flags?

Perhaps, but then hide it so it's never returned by the block.c
bdrv_get_block_status API.

Paolo