Re: Inconsistency in sysfs behavior?

[Greg KH <greg@kroah.com>]

Note, Pat's email address has changed, I've changed in the CC:

On Wed, Jan 07, 2004 at 10:48:44AM -0500, Alan Stern wrote:
> The following appears to be an inconsistency in the way sysfs behaves.  
> Tell me what you think...
> 
> When a user process parks its CWD in a kobject's sysfs directory and then
> the kobject is unregistered, of course the directory is forced to remain
> in existence (albeit unlinked) because of the reference held by the
> process.  But it does not in turn hold a reference to the kobject; the
> kobject will be deleted immediately if nothing else refers to it.
> 
> On the other hand, if a user process opens a sysfs attribute file and then
> sysfs_remove_file() is called, again the file is forced to remain in
> existence (albeit unlinked) because of the reference held by the process.  
> But now it _does_ hold a reference to the kobject; if the kobject is
> unregistered it will not be deleted until the user process closes the
> attribute file.
> 
> Why this non-parallel behavior?

Because it is very difficult to determine when a user goes into a
directory because we are using the ramfs/libfs code.  It also does not
cause any errors if the kobject is removed, as the vfs cleans up
properly.

Only when a file is opened does a kobject need to be pinned, due to
possible errors that could happen.

Hope this helps,

greg k-h