[dm-devel] Re: dm: bind new table before destroying old

[Mike Anderson]

Milan Broz <mbroz at redhat.com> wrote:
> On 11/11/2009 04:11 PM, Alasdair G Kergon wrote:
> > On Wed, Nov 11, 2009 at 03:56:05PM +0900, Kiyoshi Ueda wrote:
> >> I believe dm-mpath needs to flush such workqueues in postsuspend.
> > 
> > Yes - should be an easy change.
> 
> similar problem is probably in crypt target, I'll check it later.
> 
> >> Also, we need something to block message ioctl to suspended device.
> >  
> >> As for the message ioctl, I don't have any good idea, but...
> >>   - Reject message ioctl to suspended device in dm-ioctl
> > 
> > I think the crypt target expects to be able to do this to manipulate
> > the in-core encryption key.
> 
> yes, crypt target must be able to process messages when in suspended state.
> 

I hit a issue like Kiyoshi described of the target_message
(multipath_message) generating new work while dev_remove is trying
complete in parallel with calling fail_path / reinstate_path.

I added two accessor functions to dm-table.c (dm_table_md_suspended and
dm_table_md_freeing). I am calling both functions from multipath_message
to return early if we are in one of these states. Depending on the usage
model of the other targets message functions dm_table_md_freeing (or a new
function dm_table_md_deleting) could be called in target_message instead
of each targets message function.

The test runs using a mutex and the alternate option of using spin_lock
are currently passing I will post tomorrow the series for comments.

I also tested suspend / resume in parallel with calling fail_path /
reinstate_path. 

-andmike
--
Michael Anderson
andmike at linux.vnet.ibm.com