From mboxrd@z Thu Jan 1 00:00:00 1970 From: Theodore Ts'o Subject: Re: [PATCH] Fix up workarounds for dietlibc breakage Date: Thu, 9 Jan 2014 10:25:55 -0500 Message-ID: <20140109152555.GA6192@thunk.org> References: <1389238019-18921-1-git-send-email-tytso@mit.edu> <18FF1403-0799-4877-88F2-31FC6D960F03@dilger.ca> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Ext4 Developers List To: Andreas Dilger Return-path: Received: from imap.thunk.org ([74.207.234.97]:47093 "EHLO imap.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751248AbaAIPZ5 (ORCPT ); Thu, 9 Jan 2014 10:25:57 -0500 Content-Disposition: inline In-Reply-To: <18FF1403-0799-4877-88F2-31FC6D960F03@dilger.ca> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Wed, Jan 08, 2014 at 08:59:54PM -0700, Andreas Dilger wrote: > Before e2fsprogs used posix_memalign() the MMP code just allocated a > 2x sized buffer, and then picked the region in the middle that was > page aligned. I'd be fine going back to this to reduce the dependency > on this interface. Yeah, I've considered doing that. It's only a single block, so we'd only be wasting 4k of memory. OTOH, there are patches floating around that add posix_memalign() support to dietlibc. IIRC, they are included in Fedora at the very least. What I don't know is why they aren't in the upstream distribution. Did someone not bother to send them upstream, or did Felix nix them for some reason? I have on my low priority todo list to try to follow up on this, and that's why I haven't tried doing the 2x sized buffer thing. BTW, if we are going to do the 2x buffer thing, we should be doing that in unix_io.c, which already has direct I/O support. There's a lot of duplicate code ext2fs_mmp_read() and ext2fs_mmp_write(); my bad for noticing this when I accepted the MMP patches. - Ted P.S. BTS, there is an e-mail trail of Felix refusing to fix the "dietlibc ignores the TZ environment variable" complaint. I think he thought the timezone code that they were using was too hairy already, and the primary use case of dietlibc was one where it was unlikely that the TZ variable would be used.