From: Jonathan Corbet Subject: Re: [Ext4 Secure Delete 7/7v4] ext4/jbd2: Secure Delete: Secure delete journal blocks Date: Mon, 10 Oct 2011 17:41:15 -0600 Message-ID: <20111010174115.054c3f1f@bike.lwn.net> References: <1317971465-8517-1-git-send-email-achender@linux.vnet.ibm.com> <1317971465-8517-8-git-send-email-achender@linux.vnet.ibm.com> <20111010134736.5b721a22@bike.lwn.net> <4E93813C.1020708@linux.vnet.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 8bit Cc: linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org To: Allison Henderson Return-path: Received: from tex.lwn.net ([70.33.254.29]:58361 "EHLO vena.lwn.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751213Ab1JJXlR (ORCPT ); Mon, 10 Oct 2011 19:41:17 -0400 In-Reply-To: <4E93813C.1020708@linux.vnet.ibm.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Mon, 10 Oct 2011 16:35:24 -0700 Allison Henderson wrote: > On 10/10/2011 12:47 PM, Jonathan Corbet wrote: > > One quick question: > > > > On Fri, 7 Oct 2011 00:11:05 -0700 > > Allison Henderson wrote: > > > >> + /* Secure delete any blocks still in our range */ > >> + if (jbd2_pblk_count> 0) > >> + err = ext4_secure_delete_pblks(journal->j_inode, > >> + jbd2_pblk_start, jbd2_pblk_count); > >> + > >> +out: > >> + spin_unlock(&journal->j_pair_lock); > > > > ext4_secure_delete_pblks() appears to do its job synchronously - it has > > calls to things like sync_dirty_buffer() and such. How can you do that > > while holding ->j_pair_lock? > > > > Thanks, > > > > jon > > > > Hi Jon, > > Well j_pair_lock is a lock I added to protect the new list of vfs > -> jbd2 block pairs. It is locked by the journal commit thread to > update the list when ever a journal block is modified. The above > code here is called by the same thread that performs a punch hole or > truncate operation, not the journal commit thread. So I'm not > immediately seeing why there would be any lock problems. Is there > another case I'm missing? The problem is that ext4_secure_delete_pblks() can sleep, unless I've misunderstood things very badly. That's not something you want to do while holding a spinlock... jon