Re: [Qemu-devel] [PATCH v2 05/12] block: Introduce "drained begin/end" API

[Kevin Wolf]

Am 12.10.2015 um 13:50 hat Fam Zheng geschrieben:
> The semantics is that after bdrv_drained_begin(bs), bs will not get new 
> external
> requests until the matching bdrv_drained_end(bs).
> 
> Signed-off-by: Fam Zheng <address@hidden>
> ---
>  block.c                   |  2 ++
>  block/io.c                | 18 ++++++++++++++++++
>  include/block/block.h     | 19 +++++++++++++++++++
>  include/block/block_int.h |  2 ++
>  4 files changed, 41 insertions(+)
> 
> diff --git a/block.c b/block.c
> index 1f90b47..9b28a07 100644
> --- a/block.c
> +++ b/block.c
> @@ -2058,6 +2058,8 @@ static void bdrv_move_feature_fields(BlockDriverState 
> *bs_dest,
>      bs_dest->device_list = bs_src->device_list;
>      bs_dest->blk = bs_src->blk;
>  
> +    bs_dest->quiesce_counter = bs_src->quiesce_counter;
> +
>      memcpy(bs_dest->op_blockers, bs_src->op_blockers,
>             sizeof(bs_dest->op_blockers));
>  }

This feels wrong. As I understand it, bdrv_drained_begin/end works on
specific nodes and not on trees. Including the field in
bdrv_move_feature_fields() means that it moves to the top of the tree
(i.e. it stays at in the same C object, which however belongs to a
different logical node now).

What I could imagine is that you did this so you can use
bdrv_draind_end() on the same BDS as you called bdrv_drained_start() on.
However, that's not the interface of bdrv_swap(), which really means
that the BDSes are swapped. So with this hunk you just end up having a
bug that cancels out the weirdness of the bdrv_swap() interface.

If you rebase on my bdrv_swap() removal series, things become a bit more
obvious. If you don't, you should drop this hunk and change some
bdrv_drained_end() calls, e.g. in the next patch, you'd have to call
bdrv_drained_begin(state->old_bs), but bdrv_drained_end(state->new_bs).

The rest of this patch looks good.

Kevin