[Ocfs2-devel] mount point is not unique among all nodes

[Ocfs2-devel] mount point is not unique among all nodes

[Coly Li]

Hi list,

Here is a bug report on novell bugzilla (https://bugzilla.novell.com/show_bug.cgi?id=456280) that
mount point inside node A can be removed from node B.

The problem is, node B does not know an empty dir is be using as mount point on another node. Is
there any solution to return -EBUSY when a dir is be using as mount point on another node ?

Thanks in advance.
-- 
Coly Li
SuSE Labs

[Ocfs2-devel] mount point is not unique among all nodes

[Joel Becker]

On Fri, Feb 13, 2009 at 04:09:19AM +0800, Coly Li wrote:
> Hi list,
> 
> Here is a bug report on novell bugzilla (https://bugzilla.novell.com/show_bug.cgi?id=456280) that
> mount point inside node A can be removed from node B.

	Can't access the bug.

> The problem is, node B does not know an empty dir is be using as mount point on another node. Is
> there any solution to return -EBUSY when a dir is be using as mount point on another node ?

	This is a good question.  As of right now, we don't have a way
to do it.

Joel

-- 

"I always thought the hardest questions were those I could not answer.
 Now I know they are the ones I can never ask."
			- Charlie Watkins

Joel Becker
Principal Software Developer
Oracle
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127

[Ocfs2-devel] mount point is not unique among all nodes

[Sunil Mushran]

The bug refers to a heartbeat2 issue.

So your qs is: Mount a ocfs2 vol on two nodes, A & B. Then on
node A, create a directory on the ocfs2 vol and then mount a
local filesystem on that dir. Then remove that dir on node B.

While I have not tested this, it will be a problem. We are relying
on vfs_rmdir() to call d_mountpoint(dentry) to nak the unlink.
But it is obviously not cluster-aware.

File a bugzilla. We'll have to think about a solution.

Coly Li wrote:
> Hi list,
>
> Here is a bug report on novell bugzilla (https://bugzilla.novell.com/show_bug.cgi?id=456280) that
> mount point inside node A can be removed from node B.
>
> The problem is, node B does not know an empty dir is be using as mount point on another node. Is
> there any solution to return -EBUSY when a dir is be using as mount point on another node ?
>
> Thanks in advance.
>   

[Ocfs2-devel] mount point is not unique among all nodes

[Joel Becker]

On Thu, Feb 12, 2009 at 02:17:40PM -0800, Joel Becker wrote:
> On Fri, Feb 13, 2009 at 04:09:19AM +0800, Coly Li wrote:
> > Hi list,
> > 
> > Here is a bug report on novell bugzilla (https://bugzilla.novell.com/show_bug.cgi?id=456280) that
> > mount point inside node A can be removed from node B.
> 
> 	Can't access the bug.
> 
> > The problem is, node B does not know an empty dir is be using as mount point on another node. Is
> > there any solution to return -EBUSY when a dir is be using as mount point on another node ?
> 
> 	This is a good question.  As of right now, we don't have a way
> to do it.

	Sunil and I took a look.  It would seem we need to mark the
dentry lock LVB with a mountpoint count.  That way, ocfs2_unlink() could
check that right before calling ocfs2_remote_dentry_delete().
	The problem is that the VFS doesn't notify us when one of our
directories is mounted over.  Somewhere in do_add_mount() it would need
to tell ocfs2 about it, so that ocfs2 could take the EX on the dentry
lock and bump the mount count in the LVB.
	So, that's what we'll do when we get around to it.  I don't
think it's a high priority; we haven't actually run into this until now.

Joel

-- 

"The question of whether computers can think is just like the question
 of whether submarines can swim."
	- Edsger W. Dijkstra

Joel Becker
Principal Software Developer
Oracle
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127

[Ocfs2-devel] mount point is not unique among all nodes

[Coly Li]

Joel Becker Wrote:
> On Thu, Feb 12, 2009 at 02:17:40PM -0800, Joel Becker wrote:
>> On Fri, Feb 13, 2009 at 04:09:19AM +0800, Coly Li wrote:
>>> Hi list,
>>>
>>> Here is a bug report on novell bugzilla (https://bugzilla.novell.com/show_bug.cgi?id=456280) that
>>> mount point inside node A can be removed from node B.
>> 	Can't access the bug.
>>
>>> The problem is, node B does not know an empty dir is be using as mount point on another node. Is
>>> there any solution to return -EBUSY when a dir is be using as mount point on another node ?
>> 	This is a good question.  As of right now, we don't have a way
>> to do it.
> 
> 	Sunil and I took a look.  It would seem we need to mark the
> dentry lock LVB with a mountpoint count.  That way, ocfs2_unlink() could
> check that right before calling ocfs2_remote_dentry_delete().
> 	The problem is that the VFS doesn't notify us when one of our
> directories is mounted over.  Somewhere in do_add_mount() it would need
> to tell ocfs2 about it, so that ocfs2 could take the EX on the dentry
> lock and bump the mount count in the LVB.
> 	So, that's what we'll do when we get around to it.  I don't
> think it's a high priority; we haven't actually run into this until now.
> 
I filed a bug on oracle oss bugzilla, see
http://oss.oracle.com/bugzilla/show_bug.cgi?id=1075

I will continue to follow this issue. Thanks for the feedback.

-- 
Coly Li
SuSE Labs