Re: Inconsistency in sysfs behavior?

[Alan Stern <stern@rowland.harvard.edu>]

On Wed, 7 Jan 2004, Greg KH wrote:

> On Wed, Jan 07, 2004 at 04:50:24PM -0500, Alan Stern wrote:
> > 
> > I had in mind approaching this the opposite way.  Instead of trying to 
> > make open directories also pin a kobject, why not make open attribute 
> > files not pin them?
> > 
> > It shouldn't be hard to avoid any errors; in fact I had a patch from some
> > time ago that would do the trick (although in a hacked-up kind of way).  
> > The main idea is to return -ENXIO instead of calling the show()/store()
> > routines once the attribute has been removed.
> 
> And you can do this without adding another lock, race free?

I used dentry->d_inode->i_sem.  While I'm not very familiar with the ins
and outs of the filesystem code, that ought to be safe enough.

The real problem was finding a way to indicate that the file was
disconnected from its kobject.  I did that by setting
dentry->d_inode->i_mode to 0.  (I didn't want to erase dentry->d_fsdata
for fear that it might be needed somewhere else.)  That's definitely not a
good way; it was intended only for my proof-of-principle.  No doubt
someone else could do a much better job.

Alan Stern