From: Jan Kara Subject: Re: [PATCH 5/4] exofs: Handle error from d_splice_alias() Date: Tue, 12 Jun 2012 11:04:12 +0200 Message-ID: <20120612090412.GB6021@quack.suse.cz> References: <1338322067-17566-1-git-send-email-jack@suse.cz> <4FC5F8CC.20400@panasas.com> <20120608215950.GQ30000@ZenIV.linux.org.uk> <4FD611AA.5010006@panasas.com> <20120611190108.GF16086@thunk.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Boaz Harrosh , Al Viro , Jan Kara , linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org To: Ted Ts'o Return-path: Received: from cantor2.suse.de ([195.135.220.15]:36673 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751847Ab2FLJEP (ORCPT ); Tue, 12 Jun 2012 05:04:15 -0400 Content-Disposition: inline In-Reply-To: <20120611190108.GF16086@thunk.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Mon 11-06-12 15:01:08, Ted Tso wrote: > On Mon, Jun 11, 2012 at 06:41:30PM +0300, Boaz Harrosh wrote: > > > > My point being that please any changes made to ext2, in this area please also > > apply to exofs, since it is just another copy/paste of ext2. I'll ACK any > > which way you guys decide to properly go with, as part of the VFS changes. > > Well, I already have this quick and dirty fix to address the problem > in ext4. See commit 7e936b7372. If we need to make changes to all of > the file systems to accomodate some new VFS abstraction, it might be > worth considering whether it's easier/simpler to just put in a quick > check like I did for ext4 (just so I could plug the security hole[1] > quickly). > > [1] It's a denial of service attack for kiosks that do automounts of > USB sticks; granted, it's not that big a of a security deal, but some > people care about such things. > > Of course, if the new/changed VFS abstraction solves other problems, > that's cool, but if not, sometimes a simple brute force check is > better than something complicated if elegant. :-) I think that fix in ext4 is fine. Just you don't catch the situation when the directory entry points e.g. to a parent and that's deadlockable trivially as well. Even if it points to some unrelated directory, you can easily deadlock rename which tries to lock both directories. So I attempted for a fix in VFS because that's the only place having enough information to be able to tell whether you are creating directory hardlink or not. Honza -- Jan Kara SUSE Labs, CR