From: Andreas Dilger Subject: Re: e2fsprogs-1.39-tyt3 Date: Tue, 1 May 2007 01:56:41 -0600 Message-ID: <20070501075641.GB5967@schatzie.adilger.int> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-ext4@vger.kernel.org To: Theodore Ts'o Return-path: Received: from mail.clusterfs.com ([206.168.112.78]:35543 "EHLO mail.clusterfs.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S965635AbXEAH4n (ORCPT ); Tue, 1 May 2007 03:56:43 -0400 Content-Disposition: inline In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org On Apr 30, 2007 11:22 -0400, Theodore Ts'o wrote: > One concern I still have is the fact that we're exposing a lot of > interfaces in libext2fs.so which are very specifically tied to the > current 48-bit physical/32-bit logical on-disk extent data structure. > If/when we add support for the 64/64-bit extent data structure, and some > kind of compressed/bit-packed extent format, those interfaces will have > to be extended. Some other comments: - it appears you have all of extents.c copied into both "block.c" and in "extent.c" - it appears you are missing a couple of extent sanity checks for headers and such that were added more recently, but those might also have been removed by you - do the patches as they stand actually fix the duplicate block test cases yet? It seems all of the BLOCK_CHANGED handling was removed from block_iterate_extents(). - it would be easier to read if the patches were generated with "diff -up" (can be set in .quiltrc via 'export QUILT_DIFF_OPTS="-up"'. I also like "export QUILT_NO_DIFF_TIMESTAMPS=1" to avoid tons of gratuitous changes to patches when they are refreshed. > Another problem is that while external extent blocks are getting byte > swapped, there is no byte swapping going on for the extents stored in > the inode. I haven't tried it on a big endian system, such as Power > machine, but I'm pretty sure it's going to blow up spectacularly. Ah, interesting. We have 64-bit machines for testing, but no big-endian ones. > The patches can be found at: > > ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs-interim/e2fsprogs-1.39-tyt3 What else I think is important to get into this patch series is a change to "mkfs.ext4" so that the default inode size is increased to 256. I'm not sure how/if this can be done via mke2fs.conf. I also wouldn't object to changing the default inode_ratio on large filesystems to be a lot fewer than 1/8192 bytes. At a minimum, allowing an inode_ratio > 1/4MB should be allowed. Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.