Re: [vserver] Possible Hashify Corruption - Update 2

From: Gordan Bobic <gordan_at_bobich.net>
Date: Wed 20 Oct 2010 - 23:25:33 BST
Message-ID: <4CBF6C5D.7050209@bobich.net>

On 20/10/2010 21:41, Martin Fick wrote:
> --- On Wed, 10/20/10, Gordan Bobic<gordan@bobich.net> wrote:
>> You're most welcome. It'd be interesting to find out at
>> what kernel version this issue went away. Everything before that should
>> be marked as dangerous, at least with ext4.
>
> I think that generally older kernels with ext4 are considered dangerous. First google link for ext4 data corruption:
>
> http://wiki.archlinux.org/index.php/Ext4#Data_Corruption
>
> Of course, going back in time and marking old kernels dangerous would be kinda hard. :)
>
> From the article:
>
> "For kernel versions earlier than 2.6.30, consider adding bootflags=data=ordered"

Except it's not as simple as that. data=ordered is only meaningful when
there is a journal. When there is no journal, data=writeback is
implicit. You cannot set data=ordered without a journal.

Unfortunately, there seem to be a number of edge cases around not using
a journal that have been missed for a disappointingly long time in ext4
(the hard-link corruption case mentioned here, and until very recently,
discard option being ignored if there is no journal, even though it is
specifically people with SSDs that are likely to run without a journal).
But that's a whole different rant.

Gordan
Received on Wed Oct 20 23:25:41 2010

[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Wed 20 Oct 2010 - 23:25:42 BST by hypermail 2.1.8