From mboxrd@z Thu Jan  1 00:00:00 1970
From: Philipp Hachtmann <phacht@linux.vnet.ibm.com>
Subject: VFS: Parallel unmounting of filesystems produces system call error
Date: Tue, 18 Nov 2014 16:49:35 +0100
Message-ID: <20141118164935.19771a88@lilie>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Cc: linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
To: viro@zeniv.linux.org.uk
Return-path: <linux-kernel-owner@vger.kernel.org>
Sender: linux-kernel-owner@vger.kernel.org
List-Id: linux-fsdevel.vger.kernel.org

Hi Al,

I just ran some tests involving many parallel mount and unmount
operations. I sometimes get EINVAL and the umount program says "not
mounted". Which is definitively wrong. It feels like a race condition
somewhere in the VFS code.

I use todays master branch (commit fc14f9c1272f62c3) on s390 with 1024
DASD devices.

In fs/namespace.c, sys_umount:

> 	retval = user_path_mountpoint_at(AT_FDCWD, name, 
> lookup_flags, &path);
...
> 	retval = -EINVAL;
> 	if (path.dentry != path.mnt->mnt_root)
> 		goto dput_and_out;
This seems to be erroneously true from time to time.


The script does something equivalent to:
> for n in $DEVNODES; do
>	umount $n &
> done

Do you have any idea what's wrong?


Kind regards

Philipp