From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tejun Heo Subject: Re: [PATCH 05/11] sysfs: sysfs_chmod_file handle multiple superblocks Date: Tue, 24 Jun 2008 13:45:27 +0900 Message-ID: <48607BE7.2060203@gmail.com> References: <20080618170729.808539948@theryb.frec.bull.fr> <20080618170730.853353875@theryb.frec.bull.fr> <485DD93A.6020600@gmail.com> <486018CB.80005@fr.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <486018CB.80005@fr.ibm.com> Sender: linux-kernel-owner@vger.kernel.org To: Daniel Lezcano Cc: Benjamin Thery , Greg Kroah-Hartman , Andrew Morton , Eric Biederman , Serge Hallyn , linux-kernel@vger.kernel.org, Al Viro , Linux Containers List-Id: containers.vger.kernel.org Hello, Daniel Lezcano wrote: >> I think it would be great if sysfs_chmod_file can do all-or-nothing >> instead of failing half way through but given the interface of >> notify_change(), it could be difficult to implement. Any ideas? > > Is it acceptable to queue the notifications in a list until we are in > the loop and loop again to notify when exiting the first loop without > error ? Can you please take a look at the following patch? http://article.gmane.org/gmane.linux.file-systems/24484 Which replaces notify_change() call to two calls to sysfs_setattr() and fsnotify_change(). The latter never fails and the former should always succeed if inode_change_ok() succeeds (inode_setattr() never fails unless the size is changing), so I think the correct thing to do is... * Separate out sysfs_do_setattr() which doesn't do inode_change_ok() and just sets the attributes. Making it a void function which triggers WARN_ON() when inode_setattr() fails would be a good idea. * Implement sysfs_chmod_file() in similar way rename/move are implemented - allocate all resources and check conditions and then iff everything looks okay commit the operation by calling sysfs_do_setattr(). How does that sound? Thanks. -- tejun