From: "Darrick J. Wong" Subject: Re: [PATCH] Remove VLAIS usage from JBD2 code Date: Fri, 9 Nov 2012 15:24:51 -0800 Message-ID: <20121109232451.GA11827@blackbox.djwong.org> References: <1351622404-18214-1-git-send-email-behanw@converseincode.com> <1351622404-18214-2-git-send-email-behanw@converseincode.com> <20121030191307.GC19576@blackbox.djwong.org> <20121108155307.GB766@thunk.org> <20121109231912.GA6444@blackbox.djwong.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Behan Webster , linux-ext4@vger.kernel.org, Mark Charlebois To: "Theodore Ts'o" Return-path: Received: from userp1040.oracle.com ([156.151.31.81]:21051 "EHLO userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751659Ab2KIXY5 (ORCPT ); Fri, 9 Nov 2012 18:24:57 -0500 Content-Disposition: inline In-Reply-To: <20121109231912.GA6444@blackbox.djwong.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Fri, Nov 09, 2012 at 03:19:12PM -0800, Darrick J. Wong wrote: > On Thu, Nov 08, 2012 at 10:53:07AM -0500, Theodore Ts'o wrote: > > Hey Darrick, > > > > I was looking at this code a bit more closely while applying the > > revised version of this patch --- and this in particular raised a red > > flag for me: > > > > *(u32 *)desc.ctx = crc; > > ... > > return *(u32 *)desc.ctx; > > > > Does this raise any byte swapping issues? Looking at how the crc32 > > code in crypto/ works, I'm almost certain this is broken on big-endian > > systems, and we need to add le32_to_cup() and cpu_to_le32() calls here. > > > > Am I missing something? > > crc32_[bl]e_generic() in lib/crc32.c contains the appropriate be/le-to-cpu > conversion calls to ensure that the results are always returned in cpu-endian > format. The crypto/crc32.c file is basically just a crypto-hash wrapper around > the lib/crc32.c code, and don't touch the endianness at all. The ext4 code > receives the results from the crypto code in cpu-endian format and always > converts it to (or from) little-endian format when accessing disk structures. > > As far as e2fsprogs, it leaves the checksums in le format, only converting it > to cpu-endian format when verifying or setting the checksum. I forgot to mention -- I tried creating a metadata_csum filesystem on an x86 box. Then I bounced it between that machine and my one remaining ppc32 box, running a mount/read/write/unmount/fsck script on each bounce. I didn't see any errors. Sadly, I no longer have anything weirder (sparc64/s390x) to test. --D > > --D > > > > - Ted > -- > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html