From: Djalal Harouni Subject: Re: [PATCH] fs/ext{3,4}: fix potential race when setversion ioctl updates inode Date: Wed, 4 Jan 2012 00:14:32 +0100 Message-ID: <20120103231432.GA23522@dztty> References: <20120103013152.GA26455@dztty> <20120103124624.GB31457@quack.suse.cz> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andrew Morton , Andreas Dilger , Theodore Ts'o , Yongqiang Yang , linux-ext4@vger.kernel.org, linux-kernel@vger.kernel.org, Al Viro To: Jan Kara Return-path: Received: from numidia.opendz.org ([98.142.220.152]:56679 "EHLO numidia.opendz.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754399Ab2ACXL2 (ORCPT ); Tue, 3 Jan 2012 18:11:28 -0500 Content-Disposition: inline In-Reply-To: <20120103124624.GB31457@quack.suse.cz> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Tue, Jan 03, 2012 at 01:46:24PM +0100, Jan Kara wrote: > Hello, > > On Tue 03-01-12 02:31:52, Djalal Harouni wrote: > > > > The EXT{3,4}_IOC_SETVERSION ioctl() updates the inode without i_mutex, > > this can lead to a race with the other operations that update the same > > inode. > > > > Patch tested. > Thanks for the patch but I don't quite understand the problem. > i_generation is set when: > a) inode is loaded from disk > b) inode is allocated > c) in SETVERSION ioctl > > The only thing that can race here seems to be c) against c) and that is > racy with i_mutex as well. So what problems do you exactly observe without > the patch? Right, but what about the related i_ctime change ? (i_ctime is updated in other places...) The i_ctime update must reflect the _appropriate_ inode modification operation. This is why IMHO we should protect them to avoid a lost update. BTW the i_generation which is used by NFS and fuse filesystems is updated even if the inode is marked immutable, is this the intended behaviour? > Honza Thanks for your response. -- tixxdz http://opendz.org