Re: [PATCH] dm: also check for conflicting table type in dm_table_set_type

[Kiyoshi Ueda <k-ueda@ct.jp.nec.com>]

Hi Mike,

On 05/21/2010 07:26 AM +0900, Mike Snitzer wrote:
> Factorize live table type compatibility check into
> dm_table_type_matches_live_table().  Catching a conflicting table type
> early, during table load, avoids failing during resume -- which always
> leaves the DM device suspended.
> 
> So also check type during dm_table_set_type() -- which happens as part
> of a table load.  But preserve the existing check during resume, in
> dm_swap_table(), because a racing table load could stage an inactive
> table that conflicts with the table type that dm_swap_table() is about
> to make live:

As you noted, the additional check in dm_table_set_type() is racy and
has no guarantee to detect problematic table loading in some cases.
Although I'm not sure such a duplicated racy check is really needed,
it will detect most cases and users may be happy with that.

So I have no objection, but at least, please put some comments that
it's racy in the code below.

> +finish:
> +	if (!dm_table_type_matches_live_table(t))
> +		return -EINVAL;
> +

Thanks,
Kiyoshi Ueda