Re: [PATCH 2.6.10-rd1-bk12] user-level lookup handler for tmpfs

[Mike Waychison <Michael.Waychison@Sun.COM>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Adam J. Richter wrote:
> 	This patch eliminates the user level race condition that
> I mentioned in my original trapfs announcement and which Michael
> Waychison also complained out.  Now, if an attempt is made to
> open or stat a file name that is already blocking on a user
> level helper, the new attempt will also invoke a user level
> helper and block.
> 
> 	A side effect of this is that, if a user level program
> wants to create a plain text file, the helper program has to create
> it with some file name that the program knows to ignore and then
> rename them to the correct file names.  This is because there is
> nothing special about the user level helper program's attempts to
> open the nonexistant files, so that open() invokes another instance
> instance of a user level helper program, which needs to know to ignore
> the file name.  This should be pretty easy to do, and I've added an
> untested hypothetical example of it to
> Documentation/filesystems/lookup-trap.txt.

So the ignoring is done by the userspace agent?

> 
> 	If you want to have a directory tree that has no such file
> names, then make it a "mount --bind" copy of a subdirectory of a
> hidden tmpfs file system, something like this:
> 
> 	mount -t tmpfs /hidden
> 	mkdir /hidden/mirror /hidden/tmp-files
> 	mount --bind /hidden/mirror /public
> 
> 	...and then have your user level helper program create
> files in /hidden/tmp-files and move them to /hidden/mirror or /public
> (makes no difference, although using /hidden/mirror would make it clearer
> that the move is within the same file system).

Stuff like moving a directory with a mountpoint on it can't be done
(from userspace).  This scheme would require doing an explicit mkdir and
a mount --move.

> 
> 	Serialization is up to the user level helper programs to deal
> with if necessary.  This approach not only keeps the kernel code
> small, but also allows the user level helper programs to avoid
> unnecessary serialization.
> 
> 	Anyhow, here is the updated patch.  Please let me know what
> you think.  I will probably post this and the devfs patch to
> lkml soon for comments from a wider audience.
> 

- --
Mike Waychison
Sun Microsystems, Inc.
1 (650) 352-5299 voice
1 (416) 202-8336 voice

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
NOTICE:  The opinions expressed in this email are held by me,
and may not represent the views of Sun Microsystems, Inc.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBiQmQdQs4kOxk3/MRAo6rAJ907iugzAz6h/m9zvzO8dpvG+a3UACghljw
aRWbEwiNizzvjat4r+1dYmg=
=lwzL
-----END PGP SIGNATURE-----