From: Jan Kara Subject: Re: [PATCH v1 0/9] do not use s_dirt in ext4 Date: Wed, 28 Mar 2012 12:15:49 +0200 Message-ID: <20120328101549.GB18751@quack.suse.cz> References: <1332254489-2300-1-git-send-email-dedekind1@gmail.com> <20120322095342.GC14485@quack.suse.cz> <1332410747.18717.12.camel@sauron.fi.intel.com> <20120322103309.GA14484@quack.suse.cz> <1332854998.31549.40.camel@sauron.fi.intel.com> <20120327201401.GF5020@quack.suse.cz> <1332924282.31549.51.camel@sauron.fi.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Jan Kara , Ted Tso , Ext4 Mailing List , Linux FS Maling List , Linux Kernel Maling List To: Artem Bityutskiy Return-path: Received: from cantor2.suse.de ([195.135.220.15]:54590 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757731Ab2C1KQF (ORCPT ); Wed, 28 Mar 2012 06:16:05 -0400 Content-Disposition: inline In-Reply-To: <1332924282.31549.51.camel@sauron.fi.intel.com> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Wed 28-03-12 11:44:42, Artem Bityutskiy wrote: > On Tue, 2012-03-27 at 22:14 +0200, Jan Kara wrote: > > @@ -181,9 +181,22 @@ static int ext4_file_open(struct inode * inode, struct file * filp) > > path.dentry = mnt->mnt_root; > > cp = d_path(&path, buf, sizeof(buf)); > > if (!IS_ERR(cp)) { > > + handle_t *handle; > > + int err; > > + > > + handle = ext4_journal_start_sb(sb, 1); > > + if (IS_ERR(handle)) > > + return PTR_ERR(handle); > > + err = ext4_journal_get_write_access(handle, > > + EXT4_SB(sb)->s_sbh); > > Why do we need to bother with journal in this case - AFAIU, we update a > single field, not critical, and we do not really need the journal for > this - we can just call 'mark_buffer_dirty(sbi->s_sbh)' and let the SB > be written out directly and asynchronously. Well, except that if someone is trying to modify the superblock via the journal in parallel, the journalling layer will loudly complain about buffer becoming dirty under it's hands... So there are only two ways you can reasonably do this in ext3/ext4 - just modify the data (which is a bit ugly but works), or do full journalling. Honza -- Jan Kara SUSE Labs, CR