linux-ext4 - Re: [PATCH] ext4: fix handling mount -o remount,nolazytime

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] [thread-next>] [day] [month] [year] [list]

Message-ID: <2326451b-faf2-72a5-cb55-89cb6d8ce9ed@yandex-team.ru>
Date:   Thu, 20 Feb 2020 18:11:04 +0300
From:   Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
To:     "Theodore Y. Ts'o" <tytso@....edu>
Cc:     Andreas Dilger <adilger.kernel@...ger.ca>,
        linux-ext4@...r.kernel.org, Karel Zak <kzak@...hat.com>,
        Dmitry Monakhov <dmtrmonakhov@...dex-team.ru>
Subject: Re: [PATCH] ext4: fix handling mount -o remount,nolazytime

On 19/02/2020 19.22, Theodore Y. Ts'o wrote:
> On Wed, Feb 19, 2020 at 12:19:52PM +0300, Konstantin Khlebnikov wrote:
>> Tool "mount" from util-linux >= 2.27 knows about flag MS_LAZYTIME and
>> handles options "lazytime" and "nolazytime" as fs-independent.
>>
>> For ext4 it works for enabling lazytime: mount(MS_REMOUNT | MS_LAZYTIME),
>> but does not work for disabling: mount(MS_REMOUNT).
>>
>> Currently ext4 has performance issue in lazytime implementation caused by
>> contention around inode_hash_lock in ext4_update_other_inodes_time().
>>
>> Fortunately lazytime still could be disabled without unmounting by passing
>> "nolazytime" as fs-specific mount option: mount(MS_REMOUNT, "nolazytime").
>> But modern versions of tool "mount" cannot do that.
>>
>> This patch fixes remount for modern tool and keeps backward compatibility.
> 
> Actually, if you are using ancient versions of mount that don't know
> about MS_LAZYTIME, then when you do something like mount -o
> remount,usrquota /dev/sdb" with your patch, it will disable
> MS_LAZYTIME, which would be a backwards incompatible change.
> 
> So if we make this change, and there is someone who wants to use
> lazytime on some ancient enterprise linux system which is still using
> an old version of util-linux, and then take a kernel with this change,
> then it will result in a change in the behavior they will see.  The
> good news is that RHEL 8 is using util-linux 2.32, but RHEL 7 is still
> using util-linux 2.23.
> 
> Lazytime is not enabled by default, so this issue is really only a
> problem for someone which explicitly enables lazytime using a newer
> version of util-linux, and then disables lazytime with a newer version
> of util-linux.  So the behaviour of a2fd66d069d8 ("ext4: set lazytime
> on remount if MS_LAZYTIME is set by mount") was in fact an explicit
> decision to do things in that way.
> 
> So maybe we might want to change things, assuming that it's unlikely
> users will try to be running new kernels on ancient distros.  But I
> really wouldn't want to add a Fixes tag, and I would want to make sure
> this doesn't get backported to older kernels, since the change does
> *not* keep backwards compatibility.
> 
> Unfortunately, it's not possible to do this without breaking
> compatibility for at least some systems.  The question is whether or
> not we think systems running util-linux less than 2.27 is something we
> care about for new kernels.  Times may have changed since
> a2fd66d069d8.
> 
> So I might be willing to take this patch (I invite comments from
> others), but there will need to be a DO NOT BACKPORT warning in the
> commit description.

Usually all these options are saved in /etc/fstab and
mount -o remount,... includes them into line passed into syscall.
In this case remounting any other option will not disable lazytime.

But there might be implementations of /bin/mount which doesn't do that.

> 
> Cheers,
> 
> 						- Ted
>