From: Tyler Hicks Subject: Re: ecryptfs lock loop Date: Wed, 15 Apr 2009 10:40:03 -0500 Message-ID: <49E5FFD3.6090309@linux.vnet.ibm.com> References: <49E4E5A4.80109@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-2 Content-Transfer-Encoding: 7bit Cc: kirkland@canonical.com, ecryptfs-devel@lists.launchpad.net, linux-fsdevel@vger.kernel.org, Linux kernel mailing list , xatrix101@gmail.com To: Jiri Slaby Return-path: In-Reply-To: <49E4E5A4.80109@gmail.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org Jiri Slaby wrote: > Hi, > > one student (in CC) found out a suspicioous locking dependence in > ecryptfs code while debugging/running a static lockdep analyzer. > > I'm unable to say whether it is only a theoretical issue or a real bug, > any ideas? Hi Jiri - This looks to be a real bug. Thanks for pointing it out! > > Here it comes: > > function ecryptfs_send_message: > ------------------------------ > ecryptfs_daemon_hash_mux <- ecryptfs_msg_ctx_lists_mux > (in function ecryptfs_send_message_locked) It would be nice if we could keep from having to hold the ecryptfs_daemon_hash_mux for the entire execution of ecryptfs_send_message_locked(). Maybe we could just hold it in ecryptfs_send_message(), while it calls ecryptfs_find_daemon_by_euid(). I think we would need to grab the daemon->mux before releasing the ecryptfs_daemon_hash_mux to make sure that a QUIT message wasn't received and the daemon freed before we add the msg_ctx to the daemon->msg_ctx_out_queue in ecryptfs_send_miscdev(). It would remove the first lock dependency you pointed out, but I'll need some more time to determine if it creates another bad lock loop. Thanks again! Tyler > > function ecryptfs_wait_for_response: > ----------------------------------- > cryptfs_msg_ctx_lists_mux <- msg_ctx->mux > > function ecryptfs_process_response: > ---------------------------------- > msg_ctx->mux <- ecryptfs_daemon_hash_mux > > > <- means lock dependency