Re: [PATCH 1/4] block: Add bdrv_make_empty()

[Kevin Wolf]

Am 28.04.2020 um 15:26 hat Max Reitz geschrieben:
> Right now, all users of bdrv_make_empty() call the BlockDriver method
> directly.  That is not only bad style, it is also wrong, unless the
> caller has a BdrvChild with a WRITE permission.
> 
> Introduce bdrv_make_empty() that verifies that it does.
> 
> Signed-off-by: Max Reitz <address@hidden>
> ---
>  include/block/block.h |  1 +
>  block.c               | 23 +++++++++++++++++++++++
>  2 files changed, 24 insertions(+)
> 
> diff --git a/include/block/block.h b/include/block/block.h
> index b05995fe9c..d947fb4080 100644
> --- a/include/block/block.h
> +++ b/include/block/block.h
> @@ -351,6 +351,7 @@ BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts 
> *opts,
>  void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr);
>  void bdrv_refresh_limits(BlockDriverState *bs, Error **errp);
>  int bdrv_commit(BlockDriverState *bs);
> +int bdrv_make_empty(BdrvChild *c, Error **errp);
>  int bdrv_change_backing_file(BlockDriverState *bs,
>      const char *backing_file, const char *backing_fmt);
>  void bdrv_register(BlockDriver *bdrv);
> diff --git a/block.c b/block.c
> index 2e3905c99e..b0d5b98617 100644
> --- a/block.c
> +++ b/block.c
> @@ -6791,3 +6791,26 @@ void bdrv_del_child(BlockDriverState *parent_bs, 
> BdrvChild *child, Error **errp)
>  
>      parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
>  }
> +
> +int bdrv_make_empty(BdrvChild *c, Error **errp)
> +{
> +    BlockDriver *drv = c->bs->drv;
> +    int ret;
> +
> +    assert(c->perm & BLK_PERM_WRITE);

If I understand correctly, bdrv_make_empty() is called to drop an
overlay whose content is identical to what it would read from its
backing file (in particular after a commit operation). This means that
the caller promises that the visible content doesn't change.

So should we check BLK_PERM_WRITE_UNCHANGED instead?

Kevin