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]
Message-ID: <20230825115410.rvjmreiac6hclucm@quack3>
Date:   Fri, 25 Aug 2023 13:54:10 +0200
From:   Jan Kara <jack@...e.cz>
To:     Amir Goldstein <amir73il@...il.com>
Cc:     Theodore Ts'o <tytso@....edu>, Jan Kara <jack@...e.cz>,
        linux-ext4@...r.kernel.org
Subject: Re: [PATCH 0/11] ext4: Cleanup read-only and fs aborted checks

On Fri 25-08-23 10:24:52, Amir Goldstein wrote:
> On Fri, Aug 25, 2023 at 10:15 AM Amir Goldstein <amir73il@...il.com> wrote:
> >
> > On Thu, Aug 3, 2023 at 6:23 PM Theodore Ts'o <tytso@....edu> wrote:
> > >
> > >
> > > On Fri, 16 Jun 2023 18:50:46 +0200, Jan Kara wrote:
> > > > This series arised from me trying to fix races when the ext4 filesystem gets
> > > > remounted read-write and users can race in writes before quota subsystem is
> > > > prepared to take them. This particular problem got fixed in VFS in the end
> > > > but the cleanups are still good in my opinion so I'm submitting them. They
> > > > get rid of EXT4_MF_ABORTED flag and cleanup some sb_rdonly() checks.
> > > >
> > > > Honza
> > > >
> > > > [...]
> > >
> > > Applied, thanks!
> > >
> > > [01/11] ext4: Remove pointless sb_rdonly() checks from freezing code
> > >         commit: 98175720c9ed3bac857b0364321517cc2d695a3f
> > > [02/11] ext4: Use sb_rdonly() helper for checking read-only flag
> > >         commit: d5d020b3294b69eaf3b8985e7a37ba237849c390
> > > [03/11] ext4: Make ext4_forced_shutdown() take struct super_block
> > >         commit: eb8ab4443aec5ffe923a471b337568a8158cd32b
> > > [04/11] ext4: Make 'abort' mount option handling standard
> > >         commit: 22b8d707b07e6e06f50fe1d9ca8756e1f894eb0d
> > > [05/11] ext4: Drop EXT4_MF_FS_ABORTED flag
> > >         commit: 95257987a6387f02970eda707e55a06cce734e18
> > > [06/11] ext4: Avoid starting transaction on read-only fs in ext4_quota_off()
> > >         commit: e0e985f3f8941438a66ab8abb94cb011b9fb39a7
> > > [07/11] ext4: Warn on read-only filesystem in ext4_journal_check_start()
> > >         commit: e7fc2b31e04c46c9e2098bba710c9951c6b968af
> > > [08/11] ext4: Drop read-only check in ext4_init_inode_table()
> > >         commit: ffb6844e28ef6b9d76bee378774d7afbc3db6da9
> > > [09/11] ext4: Drop read-only check in ext4_write_inode()
> > >         commit: f1128084b40e520bea8bb32b3ff4d03745ab7e64
> > > [10/11] ext4: Drop read-only check from ext4_force_commit()
> > >         commit: 889860e452d7436ca72018b8a03cbd89c38d6384
> > > [11/11] ext4: Replace read-only check for shutdown check in mmp code
> > >         commit: 1e1566b9c85fbd6150657ea17f50fd42b9166d31
> > >
> > > Best regards,
> > > --
> > > Theodore Ts'o <tytso@....edu>
> >
> > Hi Jan,
> >
> > Yesterday I ran fanotify LTP tests on linux-next and noticed a regression
> > with fanotify22 which tests the FAN_FS_ERROR event on ext4.
> > It's 100% reproducible on my machine (see below).
> >
> > I've bisected the regression down to this series.
> 
> Forgot to say that the good baseline for the test is Christian's vfs.all
> branch merged into Linus' master and the regression is after merging
> commit 1e1566b9c85 from Ted's tree.

Thanks for report! I had a look and it is the LTP test that is problematic.

1) It has four testcases, each of which ends up triggering more or less
fatal error on this filesystem. However the filesystem is not unmounted &
mounted again between testcases so it assumes that we continue reporting
further errors after fatal filesystem shutdown. This is a wrong assumption
as after such fatal error it isn't really defined what succeeds and what
not.

2) The patchset in ext4 tree slightly changed the behavior of the 'abort'
mount option by unifying it with the filesystem shutdown functionality
because having two different ways to abort a filesystem led to places
checking one but not the other. As a result once the filesystem is
shutdown using the 'abort' mount option, we don't report any more errors
because it's kind of pointless noise - things are expected to fail on
shutdown filesystem. And this upsets the test.

I'll fix the test.

								Honza
-- 
Jan Kara <jack@...e.com>
SUSE Labs, CR

Powered by blists - more mailing lists