From: Jan Kara Subject: Re: [RFC][PATCH] Re: [BUG] ext4: cannot unfreeze a filesystem due to a deadlock Date: Mon, 2 May 2011 16:26:17 +0200 Message-ID: <20110502142617.GG4556@quack.suse.cz> References: <20110406112135.GE31057@dastard> <4DBE746F.3090707@ubuntu.com> <20110502105629.GA4556@quack.suse.cz> <4DBE9537.4050708@ubuntu.com> <20110502122055.GB5855@quack.suse.cz> <4DBEA3DF.1060306@ubuntu.com> <20110502131619.GC5855@quack.suse.cz> <4DBEB025.90704@ubuntu.com> <20110502132449.GB7708@infradead.org> <4DBEB14B.70208@ubuntu.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Christoph Hellwig , Jan Kara , Dave Chinner , Toshiyuki Okajima , Ted Ts'o , Masayoshi MIZUMA , Andreas Dilger , linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org To: Surbhi Palande Return-path: Content-Disposition: inline In-Reply-To: <4DBEB14B.70208@ubuntu.com> Sender: linux-fsdevel-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org On Mon 02-05-11 16:27:39, Surbhi Palande wrote: > On 05/02/2011 04:24 PM, Christoph Hellwig wrote: > >On Mon, May 02, 2011 at 04:22:45PM +0300, Surbhi Palande wrote: > >>This has another advantage, that it rightfully does not let you > >>update the access time when the F.S is frozen (touch_atime called > >>from a read path when the F.S is frozen) Otherwise we also need to > >>fix this path. > >In most filesystens atime updates aren't transactional. They just > >get written into inode->i_atime, and at some later point when the > >VFS tries to clean the inode it gets writtent back, either through > >a transaction or not. > > > Yes, agreed. But then when a F.S is frozen the inode should not be > dirtied? Right? So this has to be fixed? > Also, in ext4, I think that updating atime starts a transaction. Yes, it does. Any mark_inode_dirty() call causes a transaction update. Honza -- Jan Kara SUSE Labs, CR