From: Ted Ts'o Subject: Re: Collapsing the number of feature flags (was Re: [PATCH v2.3 00/23] ext4: Add metadata checksumming) Date: Sat, 11 Feb 2012 21:55:18 -0500 Message-ID: <20120212025518.GA22492@thunk.org> References: <20120107082751.21970.84856.stgit@elm3c44.beaverton.ibm.com> <20120208180847.GA3654@thunk.org> <3EBD659A-0A10-4DAE-9EA9-E736CE187574@dilger.ca> <20120210211131.GB5381@thunk.org> <28776282-021D-473A-99A1-F5A3DDC08966@dilger.ca> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "Darrick J. Wong" , linux-ext4@vger.kernel.org To: Andreas Dilger Return-path: Received: from li9-11.members.linode.com ([67.18.176.11]:45425 "EHLO test.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755411Ab2BLCzY (ORCPT ); Sat, 11 Feb 2012 21:55:24 -0500 Content-Disposition: inline In-Reply-To: <28776282-021D-473A-99A1-F5A3DDC08966@dilger.ca> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Sat, Feb 11, 2012 at 06:38:03PM -0700, Andreas Dilger wrote: > > I think that making it difficult for users to use anything but the default > options in mke2fs/tune2fs/debugfs is equivalent to what you suggest. > Unless you prevent the use of the old feature flags entirely, then it > would still be possible to create filesystems with mixed up combinations > of features. I think the important outcome is that users are clearly > made aware that they are stepping off the well-trodden path, and if > things break they get to keep both pieces. Having two feature flags > that mean the same thing doesn't really help this, IMHO. The problem is that sometimes people upgrade the kernel before they upgrade userspace, and sometimes they upgrade e2fsprogs but are still running an older kernel --- and there times when those are good and proper things to do (i.e., they have to use some ancient obsolete RHEL kernel becuase it's they only way they can talk to the EMC storage device, but they want to use a newer e2fsprogs to fix bugs that have since been fixed in the latest version of e2fsprogs). So we have to support both, and so yes, we can't remove any code for quite a while. What we *can* do is say that if you enable metadata checksums, you also must get all of these other features along for the ride. It's true that means users can still create mixed up feature set --- but not if they want the new metadata checksum feature. It's an eventual solution, but it's not magic pixie dust, yes. Realistically most of the time we haven't had *that* many problems caused by random feature combinations. There are certain combinations that are problematic, but we know what they are up front. If someone really wants to add that screen for certain feature combinations and causes mke2fs/e2fsck to complain, or maybe even outright refuse to function, that's a different thing, that's going to be fairly specific for the distro. I don't think that's something we want to add upstream.... - Ted