[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Denis Cheng]

ocfs2 uses a subsystem under sysfs: /sys/o2cb, which is not good,

the canonical use of sysfs is one kset under /sys/fs, like the approach
gfs2 does.

if someone will agree my opinion, I can improve this code.

--
Denis Cheng

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Joel Becker]

On Fri, Dec 07, 2007 at 12:22:17PM +0800, Denis Cheng wrote:
> ocfs2 uses a subsystem under sysfs: /sys/o2cb, which is not good,
> 
> the canonical use of sysfs is one kset under /sys/fs, like the approach
> gfs2 does.
> 
> if someone will agree my opinion, I can improve this code.

o2cb is, in theory, independent of the filesystem.  Thus, it doesn't
necessarily belong under /sys/fs.

The bigger problem is that there are tools expecting /sys/o2cb to be
there.  We can't just drop it.

Joel

-- 

"Drake!  We're LEAVING!"

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

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[rae l]

On Dec 7, 2007 1:12 PM, Joel Becker <Joel.Becker@oracle.com> wrote:
> On Fri, Dec 07, 2007 at 12:22:17PM +0800, Denis Cheng wrote:
> > ocfs2 uses a subsystem under sysfs: /sys/o2cb, which is not good,
> >
> > the canonical use of sysfs is one kset under /sys/fs, like the approach
> > gfs2 does.
> >
> o2cb is, in theory, independent of the filesystem.  Thus, it doesn't
> necessarily belong under /sys/fs.
I agree it's independent, but /sys/oc2b just don't conform the uniform
of the linux kernel sysfs usage.

> The bigger problem is that there are tools expecting /sys/o2cb to be
> there.  We can't just drop it.
Do you mean the ocfs2-tools? We could also change it. Or any other tools?

The kernel community doesn't promise a stable kernel API, so does the sysfs,
Except If it would break many userspace tools, we should consider with
caution or keep the binary compatibility,

As time goes by, the wrong way shouldn't be continued, we are to make
changes sooner or later.

-- 
Denis Cheng

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Mark Fasheh]

On Fri, Dec 07, 2007 at 03:21:20PM +0800, rae l wrote:
> > The bigger problem is that there are tools expecting /sys/o2cb to be
> > there.  We can't just drop it.
> Do you mean the ocfs2-tools? We could also change it. Or any other tools?
> 
> The kernel community doesn't promise a stable kernel API, so does the sysfs,
> Except If it would break many userspace tools, we should consider with
> caution or keep the binary compatibility,

It would break ocfs2-tools, majorly. mount.ocfs2, fsck.ocfs2, tunefs.ocfs2,
mkfs.ocfs2, o2cb_ctl are just the programs which I can think of off the top
of my head which look for o2cb/interface_revision. Basically anything that
needs to interact with the cluster uses that file at least once.

So sorry, but shuffling it around would make many Ocfs2 users very, very
unhappy.
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh@oracle.com

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Joel Becker]

On Fri, Dec 07, 2007 at 03:21:20PM +0800, rae l wrote:
> On Dec 7, 2007 1:12 PM, Joel Becker <Joel.Becker@oracle.com> wrote:
> > o2cb is, in theory, independent of the filesystem.  Thus, it doesn't
> > necessarily belong under /sys/fs.
> I agree it's independent, but /sys/oc2b just don't conform the uniform
> of the linux kernel sysfs usage.

	By your measure.  By others, it does.  In fact, we asked around
before we put it there.

> > The bigger problem is that there are tools expecting /sys/o2cb to be
> > there.  We can't just drop it.
> Do you mean the ocfs2-tools? We could also change it. Or any other tools?

	I mean ocfs2-tools.  We have to deal with existing
installations, not just new tools.

> The kernel community doesn't promise a stable kernel API, so does the sysfs,
> Except If it would break many userspace tools, we should consider with
> caution or keep the binary compatibility,

	sysfs is a userspace ABI.  The kernel *does* try to keep
userspace ABI stable or backwards compatible.

Joel

-- 

"What do you take me for, an idiot?"  
        - General Charles de Gaulle, when a journalist asked him
          if he was happy.

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

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[rae l]

On Dec 12, 2007 3:06 AM, Greg KH <greg@kroah.com> wrote:
> On Fri, Dec 07, 2007 at 03:21:20PM +0800, rae l wrote:
> > On Dec 7, 2007 1:12 PM, Joel Becker <Joel.Becker@oracle.com> wrote:
> > > On Fri, Dec 07, 2007 at 12:22:17PM +0800, Denis Cheng wrote:
> > > > ocfs2 uses a subsystem under sysfs: /sys/o2cb, which is not good,
> > > >
> > > > the canonical use of sysfs is one kset under /sys/fs, like the approach
> > > > gfs2 does.
> > > >
> > > o2cb is, in theory, independent of the filesystem.  Thus, it doesn't
> > > necessarily belong under /sys/fs.
> > I agree it's independent, but /sys/oc2b just don't conform the uniform
> > of the linux kernel sysfs usage.
>
> And what would that be?  :)
The sysfs entries of ocfs2 are under /sys/oc2b, but I think they
generally should be under /sys/fs/oc2b, aren't they?

But as Joel said, this change will break ocfs2-tools, so the decision
was not made.

On Dec 7, 2007 5:28 PM, Joel Becker <Joel.Becker@oracle.com> wrote:
>         I mean ocfs2-tools.  We have to deal with existing
> installations, not just new tools.
To Joel: I mean this change will be done in new kernels, and new
ocfs2-tools, old kernel require old ocfs2-tools, that does not break
existing installations, is that OK?

-- 
Denis Cheng

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Greg KH]

On Fri, Dec 07, 2007 at 03:21:20PM +0800, rae l wrote:
> On Dec 7, 2007 1:12 PM, Joel Becker <Joel.Becker@oracle.com> wrote:
> > On Fri, Dec 07, 2007 at 12:22:17PM +0800, Denis Cheng wrote:
> > > ocfs2 uses a subsystem under sysfs: /sys/o2cb, which is not good,
> > >
> > > the canonical use of sysfs is one kset under /sys/fs, like the approach
> > > gfs2 does.
> > >
> > o2cb is, in theory, independent of the filesystem.  Thus, it doesn't
> > necessarily belong under /sys/fs.
> I agree it's independent, but /sys/oc2b just don't conform the uniform
> of the linux kernel sysfs usage.

And what would that be?  :)

confused,

greg k-h

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Mark Fasheh]

On Wed, Dec 12, 2007 at 02:28:57PM -0800, Greg KH wrote:
> > The sysfs entries of ocfs2 are under /sys/oc2b, but I think they
> > generally should be under /sys/fs/oc2b, aren't they?
> 
> Yes, they should be, it looks like someone messed up putting them in the
> sysfs root directory.

'o2cb' refers to the stuff in ocfs2_nodemanager, so we wanted to keep it out
of the fs namespace (think more "cluster" than "file system").

We felt at the time (and still do) that it was better than polluting
/sys/fs/ with non file system related things.

Also, I feel that for perspective, I should mention that this layout isn't
anything new - we've been doing this for a while.


> > But as Joel said, this change will break ocfs2-tools, so the decision
> > was not made.
> 
> It should be pretty simple to change the tools to work for both places,
> right?

Changing a path (or just testing two of them) isn't a big deal, the problem
is that existing versions of ocfs2-tools will break very badly. That's
clearly unnaceptable for a production file system.
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh@oracle.com

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Greg KH]

On Wed, Dec 12, 2007 at 09:02:41AM +0800, rae l wrote:
> On Dec 12, 2007 3:06 AM, Greg KH <greg@kroah.com> wrote:
> > On Fri, Dec 07, 2007 at 03:21:20PM +0800, rae l wrote:
> > > On Dec 7, 2007 1:12 PM, Joel Becker <Joel.Becker@oracle.com> wrote:
> > > > On Fri, Dec 07, 2007 at 12:22:17PM +0800, Denis Cheng wrote:
> > > > > ocfs2 uses a subsystem under sysfs: /sys/o2cb, which is not good,
> > > > >
> > > > > the canonical use of sysfs is one kset under /sys/fs, like the approach
> > > > > gfs2 does.
> > > > >
> > > > o2cb is, in theory, independent of the filesystem.  Thus, it doesn't
> > > > necessarily belong under /sys/fs.
> > > I agree it's independent, but /sys/oc2b just don't conform the uniform
> > > of the linux kernel sysfs usage.
> >
> > And what would that be?  :)
> The sysfs entries of ocfs2 are under /sys/oc2b, but I think they
> generally should be under /sys/fs/oc2b, aren't they?

Yes, they should be, it looks like someone messed up putting them in the
sysfs root directory.

> But as Joel said, this change will break ocfs2-tools, so the decision
> was not made.

It should be pretty simple to change the tools to work for both places,
right?

thanks,

greg k-h

[Ocfs2-devel] [RFC] to make files of ocfs2 under sysfs canonical

[Greg KH]

On Wed, Dec 12, 2007 at 02:52:50PM -0800, Mark Fasheh wrote:
> On Wed, Dec 12, 2007 at 02:28:57PM -0800, Greg KH wrote:
> > > The sysfs entries of ocfs2 are under /sys/oc2b, but I think they
> > > generally should be under /sys/fs/oc2b, aren't they?
> > 
> > Yes, they should be, it looks like someone messed up putting them in the
> > sysfs root directory.
> 
> 'o2cb' refers to the stuff in ocfs2_nodemanager, so we wanted to keep it out
> of the fs namespace (think more "cluster" than "file system").
> 
> We felt at the time (and still do) that it was better than polluting
> /sys/fs/ with non file system related things.
> 
> Also, I feel that for perspective, I should mention that this layout isn't
> anything new - we've been doing this for a while.

Then you all have been ignoring the patches that are in -mm that move
this directory under /sys/fs/ for quite some time now :)

> > > But as Joel said, this change will break ocfs2-tools, so the decision
> > > was not made.
> > 
> > It should be pretty simple to change the tools to work for both places,
> > right?
> 
> Changing a path (or just testing two of them) isn't a big deal, the problem
> is that existing versions of ocfs2-tools will break very badly. That's
> clearly unnaceptable for a production file system.

Update the tools today, and then let the kernel change happen in a few
months.  Not that big of a deal.

thanks,

greg k-h