From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Date: Tue, 3 Dec 2013 02:48:54 -0800 Subject: [Cluster-devel] [PATCH 12/18] ocfs2: use generic posix ACL infrastructure In-Reply-To: <20131202230007.GK12253@quack.suse.cz> References: <20131201115903.910559036@bombadil.infradead.org> <20131201120655.852590677@bombadil.infradead.org> <20131202230007.GK12253@quack.suse.cz> Message-ID: <20131203104854.GA3223@infradead.org> List-Id: To: cluster-devel.redhat.com MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit On Tue, Dec 03, 2013 at 12:00:07AM +0100, Jan Kara wrote: > Hum, this changes the cluster locking. Previously ocfs2_acl_get() used > from ocfs2_acl_chmod() grabbed cluster wide inode lock. Now getting of ACL > isn't protected by the inode lock. That being said the cluster locking > around setattr looks fishy anyway - if two processes on different > nodes are changing attributes of the same file, changing ACLs post fact > after dropping inode lock could cause interesting effects. Also I'm > wondering how inode_change_ok() can ever be safe without holding inode > lock... Until we grab that other node is free to change e.g. owner of the > inode thus leading even to security implications. But maybe I'm missing > something. Mark, Joel? Hmm, indeed. How does ocfs2_iop_get_acl get away without that lock? Btw, ocfs2 changes will need careful testing as I couldn't find any easy way to run xfstests on ocfs2 out of the box.