Re: Question about pseudo filesystems

[Daniel Phillips <phillips@arcor.de>]

On Wednesday 04 September 2002 20:02, Jamie Lokier wrote:
> I have a small problem with a module I'm writing.  It uses a
> pseudo-filesystem, rather like futexes and pipes, so that it can hand
> out special file descriptors on request.
> 
> Following the examples of pipe.c and futex.c, but specifically for a 2.4
> kernel, I'm doing this:
> 
> 	static DECLARE_FSTYPE (mymod_fs_type, "mymod_fs",
> 			       mymod_read_super, FS_NOMOUNT);
> 
> 
> 	static int __init mymod_init (void)
> 	{
> 		int err = register_filesystem (&mymod_fs_type);
> 		if (err)
> 			return err;
> 		mymod_mnt = kern_mount (&mymod_fs_type);
> 		if (IS_ERR (mymod_mnt)) {
> 			unregister_filesystem (&mymod_fs_type);
> 			return PTR_ERR (mymod_mnt);
> 		}
> 	}
> 
> 	static void __exit mymod_exit (void)
> 	{
> 		mntput (mymod_mnt);
> 		unregister_filesystem (&mymod_fs_type);
> 	}
> 
> Unfortunately, when I come to _unload_ the module, it can't be unloaded
> because the kern_mount increments the module reference count.
> 
> (pipe.c in 2.4 appears to have the same problem, but of course nobody
> can ever unload it anyway so it doesn't matter).
> 
> I'm handing out file descriptors rather like futexes from 2.5: they all
> share the same dentry, which is the root of the filesystem.  In my
> code's case, that dentry is created in `mymod_read_super' (just the same
> way as 2.4 pipe.c).
> 
> Somehow, it looks like I need to mount the filesystem when it first
> generates an fd, and unmount it when the last fd is destroyed -- but is
> it safe to unmount the filesystem _within_ a release function of an
> inode on the filesystem?
> 
> Either that, or I need something else.

It's a good example of why the module interface is stupidly wrong, and
__exit needs to be called by the module unloader, returning 0 if it's
ok to unload.  Then your __exit can whatever condition it's interested
in and, if all is well, do the kern_umount.

-- 
Daniel