linux-ext4 - Re: inline_data,metadata_csum a lot of bad csum for directories

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:	Thu, 4 Dec 2014 16:10:01 -0800
From:	"Darrick J. Wong" <darrick.wong@...cle.com>
To:	"Theodore Ts'o" <tytso@....edu>
Cc:	Andreas Dilger <adilger@...ger.ca>,
	Dmitry Monakhov <dmonakhov@...nvz.org>,
	ext4 development <linux-ext4@...r.kernel.org>
Subject: Re: inline_data,metadata_csum  a lot of bad csum for directories

On Mon, Dec 01, 2014 at 09:08:32PM -0800, Darrick J. Wong wrote:
> On Mon, Dec 01, 2014 at 10:58:13PM -0500, Theodore Ts'o wrote:
> > On Mon, Dec 01, 2014 at 04:32:02PM -0700, Andreas Dilger wrote:
> > > On Dec 1, 2014, at 7:09 AM, Dmitry Monakhov <dmonakhov@...nvz.org> wrote:
> > > > 
> > > > Ted can you please clarify what is status of inline_data patchset.
> > > > I've enabled inline_data testing in my test environment
> > > 
> > > I suspect that since inline_data and metadata_csum were developed
> > > in isolation, they don't work well together yet.  I believe that
> > > Darrick is working on fixing some issues in this area.
> 
> Fixing metadata_csum bugs, yes.  There could be plenty more in inline_data --
> there are a couple of spots (setting extra_isize in particular) where I'm not
> entirely sure it's setting the inode checksum before writing the raw inode out
> to disk.
> 
> On the plus side, from running generic/011 in a loop I think I've found some
> sort of strange bug where unlinked inodes fail csum validation, causing e2fsck
> to complain.  A cheap solution is not to bother revalidating an inode if its
> link count is zero, but I want to know why the kernel isn't doing the right
> thing with the unlinked inodes.

Ahah, I introduced a race condition with the crypto API rework in ext4_chksum.
I failed to notice that _setkey() modifies s_chksum_driver without any sort of
concurrency protection.  Soo... I'll send a patch tonight that fixes it.

Sorry about that,

--D

> 
> --D
> 
> > 
> > The standard xfstests-bld test configuration is with inline_data and
> > metadata_csum in combination:
> > 
> > export FS=ext4
> > export TEST_DEV=$SM_TST_DEV
> > export TEST_DIR=$SM_TST_MNT
> > export MKFS_OPTIONS="-q -I 256 -O inline_data,metadata_csum,64bit"
> > export EXT_MOUNT_OPTIONS="-o block_validity"
> > TESTNAME="Ext4 4k block w/inline"
> > 
> > Are you using the very latest version of e2fsprogs.git#next branch for
> > your tests?  If not, that could be why you are getting some test
> > failures.  It is absolutely required for any tests involving
> > metdata_csum.
> > 
> > I have a xfstests-build 32-bit rootfs.img I can upload for your
> > convenience if that's what you are using.  Getting a formally upgraded
> > xfstests-bld rootfs.img with the latest xfstests and some other bug
> > fixes and the latest e2fsprogs was on my todo list; sorry, the
> > metadata csum v3 patches that went in in 3.18-rc3 required an updated
> > e2fsprogs that's not in the released rootfs.img.
> > 
> > > > Please post your test matrix
> > > > 
> > > > My config:
> > > > e2fsprogs git commit: g210baa5
> > 
> > OK, that should be sufficiently up to date.  As far as my test matrix,
> > please see:
> > 
> > http://git.kernel.org/cgit/fs/ext2/xfstests-bld.git/tree/kvm-xfstests/test-appliance/files/root/conf
> > 
> > with these test versions:
> > 
> > e2fsprogs	v1.42.12-415-gf0a7814 (Sun, 23 Nov 2014 20:32:08 -0500)
> > fio		fio-2.1.13-77-g0f9940a (Wed, 15 Oct 2014 14:16:35 -0600)
> > quota		f215e3f (Mon, 2 Dec 2013 23:06:23 -0300)
> > xfsprogs	v3.2.1-6-gba24eb7 (Fri, 18 Jul 2014 10:47:11 +1000)
> > xfstests-bld	2a20a9d (Fri, 31 Oct 2014 10:08:53 -0400)
> > xfstests	linux-v3.8-523-g481c28f (Tue, 14 Oct 2014 22:59:39 +1100)
> > 
> > > > ext4.git commit: recent tytso/dev ca83cc75a16f67
> > > > ./kvm-xfstests -c inline generic/011  generic/013
> > > > generic/013       [13:51:44]
> > > > [  368.507266] EXT4-fs error (device vdd): ext4_iget:3918: inode #294: comm rm: checksum invalid
> > > > [  368.511098] EXT4-fs error (device vdd): ext4_iget:3918: inode #294: comm rm: checksum invalid
> > > > [  369.098863] EXT4-fs error (device vdd): ext4_iget:3918: inode #2108:comm rm: checksum invalid
> > 
> > Hmm, works for me:
> > 
> > ./kvm-xfstests -c inline generic/011  generic/013
> > 
> > BEGIN TEST: Ext4 4k block w/inline Mon Dec  1 22:56:21 EST 2014
> > Device: /dev/vdd
> > mk2fs options: -q -I 256 -O inline_data,metadata_csum,64bit
> > mount options: -o block_validity
> > FSTYP         -- ext4
> > PLATFORM      -- Linux/i686 kvm-xfstests 3.18.0-rc3-00530-g32f3869
> > MKFS_OPTIONS  -- -q -I 256 -O inline_data,metadata_csum,64bit /dev/vdc
> > MOUNT_OPTIONS -- -o acl,user_xattr -o block_validity /dev/vdc /vdc
> > 
> > generic/011 2s ...  [22:56:24] [22:56:26] 2s
> > generic/013 47s ... [22:56:26] [22:57:21] 55s
> > Ran: generic/011 generic/013
> > Passed all 2 tests
> > 
> > 						- Ted
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> > the body of a message to majordomo@...r.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@...r.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html