Subject: Re: DMAPI test suite.
From: Ram Pai <linuxram@us.ibm.com>
To: Steve Lord <lord@xfs.org>
Cc: Dean Roehrich <roehrich@clink.americas.sgi.com>, linux-xfs@oss.sgi.com
In-Reply-To: <40360FC8.4060405@xfs.org>
References: <200402181519.i1IFJS9i11055016@clink.americas.sgi.com>
	 <1077265286.7503.37.camel@dyn319009bld.beaverton.ibm.com>
	 <40360FC8.4060405@xfs.org>
Content-Type: text/plain
Organization: 
Message-Id: <1077473812.2822.32.camel@localhost.localdomain>
Mime-Version: 1.0
Date: 22 Feb 2004 10:16:52 -0800
Content-Transfer-Encoding: 7bit
Sender: linux-xfs-bounce@oss.sgi.com
Errors-to: linux-xfs-bounce@oss.sgi.com
Precedence: bulk

On Fri, 2004-02-20 at 05:46, Steve Lord wrote:
> Ram Pai wrote:
> > On Wed, 2004-02-18 at 07:19, Dean Roehrich wrote:
> > 
> >>It probably needs to be upgraded to the latest autotools.  It's a bit old.
> >>
> >>I don't have an rpm for the binaries.
> > 
> > 
> > Ok. I am not able to run this test suite. In any case, I want to
> > validate the behavior of XFS for pre-unmount event.
> > 
> > As per the spec the pre-unmount event should be generated before the
> > operation is about to check for open files.  However given the current
> > VFS code I am not sure, how could ever any filesystem support this
> > feature correctly.
> 
> You are probably correct here, the vfs keeps track of the busy count
> and apart from the unmount_begin operation which is called in the
> MNT_FORCE case, there is not a lot of opportunity to insert this
> call. If the pre-unmount event is allowed to fail the unmount
> then this will not work either since it does not stop the unmount.

> > 
> > The VFS code may not even inform the filesystem specific unmount handler
> > in some cases. As a result no filesystem can ever support this feature
> > correctly.  Is my understanding correct? Does XFS provide the
> > pre-unmount event correctly?
> >
> 
> There is a pre_unmount event in xfs, it is after the vfs has cleaned
> up all the inodes (in the put_super code), 

Looking through the xfs dmapi code, 2 questions pop up:
1. How can the pre_unmount event, when generated through the put_super()
code provide the DM_UNMOUNT_FORCE information.  There is now way for the
put_super() code to know if the unmount is a forceful one or not. 

2. And finally I find that all the DMAPI event generators are stubbed
to   fs_nosys or fs_noerr or fs_noval . So does XFS ever generate DM
events ?

RP

> if it returns an error
> then the vfs will still finish the unmount, and you may well get one
> of those self destruct in 5 seconds messages.


> 
> > Any ideas?
> > RP
> 
> Al Viro is distinctly anti dmapi, so chances are slim to none I would
> say.

hmm...that makes it difficult. Is he against Dmapi implemented in VFS or
against the overall idea of supporting DMAPI anywhere in linux?  I hope
he is just against DMAPI implemented at the VFS level and is ok with 
supporting filesystems to implement it. 

RP