Subject: Re: [vserver] Possible Hashify Corruption - Update 2
From: Gordan Bobic <gordan@bobich.net>
Date: Thu, 21 Oct 2010 10:29:02 +0100

Herbert Poetzl wrote:
> On Wed, Oct 20, 2010 at 03:02:25PM +0100, Gordan Bobic wrote:
>> I have just tested this with a new kernel. And the issue doesn't arise 
>> with 2.6.35.7-vs2.3.0.36.33! So it is definitely a kernel bug, most 
>> likely in ext4.
> 
>> This really needs to be documented and stressed on the pages with 
>> download links. Hashify feature causes total data corruption with 
>> 2.6.30.10-vs2.3.0.36.14-pre8 when used on ext4 without a journal.
> 
> well, thanks for investigating this, but you are
> aware that _nobody_ is supposed to use that kernel
> or patch anymore, as it was highly experimental
> (see the pre) and superceeded by a working version

That'd be fine if there was a non-pre patch available for that kernel - 
but there isn't. And due to other patch requirements that may not have 
been ported forward to new kernels, it is plausible that that somebody 
might be stuck with it. I wasn't using such an old kernel out of choice.

If the patch shouldn't be used, it arguably shouldn't be linked on the 
download page.

> in general, I'd stay away from kernels 2.6.25-2.6.31
> especially with 'sensitive' filesystems like ext4,
> reiserfs and btrfs ...

Agreed, but as I said, sometimes you don't have a choice.

>> I haven't tried other cases (ext4 with a journal or other file systems) 
>> so I cannot speak for those, but in the mentioned use cases it is 
>> completely repeatable.
> 
> if the issue persists in the recent kernel branches,
> i.e. in 2.6.31.14 or even 2.6.33.x/2.6.35.x then
> we need to investigate, otherwise I consider this
> closed ...

As I said, 2.6.35.7 works fine. I'm not suggesting this should be 
patched/fixed in vservers (I'm suspecting the bug is actually in ext4), 
but I do think it should be documented. We are talking about the latest 
available 2.5.30 kernel here (2.6.30.10), which isn't that far off from 
2.6.31.14. Putting a note next to the link that says "there are known 
issues with this kernel that can lead to file system corruption" would 
go a long way toward saving people's data from stealthy corruption. 
Having a warning where it's due on the download page is vastly 
preferable to people writing on blogs things like "vservers sucks, it 
corrupted my data".

Gordan