[sheepdog] Call to sd_truncate()

[Liu Yuan]

On Thu, Dec 12, 2013 at 12:25:52PM +0100, Hadrien KOHL wrote:
> I just sent the question to the qemu developer mailing list. But as I did,
> I remembered this sd_truncate is also called
> from  sd_co_writev(BlockDriverState *bs, int64_t sector_num, int
> nb_sectors, QEMUIOVector *qiov):
> 
> static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t
> sector_num,
>                         int nb_sectors, QEMUIOVector *qiov)
> {
>     SheepdogAIOCB *acb;
>      int ret;
> 
>     if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
>         ret = sd_truncate(bs, (sector_num + nb_sectors) * BDRV_SECTOR_SIZE);
>         if (ret < 0) {
>             return ret;
>         }

>From the code, so if growable is true, seems that sheepdog support automatic
grow of volume. So my last email about 'can't resize live VM' is wrong.

By a simply glance of the code, seems that bs->growable is set by qcow2.c, which
indicate that this might be a feature from qcow2 format.

>         bs->total_sectors = sector_num + nb_sectors;
>     }
> 
>     acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors);
>     acb->aio_done_func = sd_write_done;
>     acb->aiocb_type = AIOCB_WRITE_UDATA;
> 
>     ret = sd_co_rw_vector(acb);
> ...
> }
> 
> This function is mapped to bdrv_co_writev in the qemu driver.
> Looking at the test in this function, the only way a shrink could happen -
> I guess - is if the units in sector_num and nb_sectors are inconsistent
> (esp since there's a conversion to "BDRV_SECTOR_SIZE") in relation to
> bs->total_sectors.
> I am using qcow2 file format and created the image with qemu-img -t qcow2
> create sheepdog:name. Could it be related?

I think this might be a bug about running qcow2 in sheepdog, but I can't figure
out what is the problem by statically reviewing the code. Is there a reliable
way to reproduce it?

Thanks
Yuan