From mboxrd@z Thu Jan 1 00:00:00 1970 From: Chris Mason Subject: Re: anyone port data-logging patches to -rc3? Date: 23 May 2003 08:44:51 -0400 Message-ID: <1053693891.5455.51.camel@tiny.suse.com> References: <20030523003033.GC6586@zero> <20030523094633.GA1005@namesys.com> <1053692899.5455.44.camel@tiny.suse.com> <20030523123601.GB7109@namesys.com> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: list-help: list-unsubscribe: list-post: Errors-To: flx@namesys.com In-Reply-To: <20030523123601.GB7109@namesys.com> List-Id: Content-Type: text/plain; charset="us-ascii" To: Oleg Drokin Cc: Tom Vier , reiserfs-list@namesys.com On Fri, 2003-05-23 at 08:36, Oleg Drokin wrote: > Hello! > > On Fri, May 23, 2003 at 08:28:19AM -0400, Chris Mason wrote: > > > > And BTW, I think this "do O_DIRECT stuff on ordered/datalogging fs, receive error from write(2)" is counter-intuitive. > > > what do you think about just only setting reiserfs_address_space_operations->direct_IO to non-NULL > > > in case of writeback mount? > > They'd get the same error, just sooner. I'm more inclined to fix things > > Yes, they'd get it at open() time which is much better, I think. > > > so that data=ordered works with O_DIRECT, lets see what I can come up > > with. > > Still there is no chance for that to work in data=journal mode. > Right, we can null out the func pointer when we change to data=journal mode. > > > Also probably we do not want to allow remounting with different data-logging mode, > > > may be we should even add a check in reiserfs_remount() and spit out an error in such a case? > > Jeff Mahoney managed to trigger a race by switching between data=journal > > and the other modes. All we really need to do is be able to switch data > > mode when remounting from readonly to rw (as in for the root FS), so > > I'll limit the remounting code to that. > > Any chance you would also rework journaling code a bit, so that it does not occupy > ram for various tables/lists when the fs is in R/O mode? I think I would rather do that in 2.5.x first, perhaps along with some changes to more dynamic tables/lists for the journal memory structures. The data logging code started that process, but I'd like to keep the 2.4.x data logging changes smallish right now so we can get it all submitted. Aside from that raid remount bug, do you see major problems with keeping the memory around in readonly mode? -chris