[linux-cifs-client] cifs client caching wrong data

[Bjørn Tore Sund]

On 1/29/10 9:03 PM, Jeff Layton wrote:
> On Thu, 28 Jan 2010 23:05:17 +0100
> Bjørn Tore Sund<bjorn.sund at it.uib.no>  wrote:
>
>>
>> We distribute home directories to 600 linux clients using Samba and
>> cifs.  Since upgrading from Fedora 10 (mount.cifs from samba 3.2.15,
>> cifs kernel module version 1.54 on 2.6.27) to Fedora 12 (mount.cifs from
>> Samba 3.4.5, cifs kernel module version 1.60 on 2.6.31) on the clients
>> we have issues with updates on the file systems never being noticed on
>> the client site.
>>
>> The quick test we run is to do:
>>
>> mkdir a b c d e f g h i j k l m n o p q r s t u v w x; ls -a ?
>>
>> Either that, or an immediately following:
>>
>> rmdir ?; mkdir a b c d e f g h i j k l m n o p q r s t u v w x; ls -a ?
>>
>> will trigger the issue.  Normally the first one and always the second
>> one will on all or most directories fail to show them containing "." and
>> "..".  The issue is wider, we're seeing file corruptions and "ls"
>> displaying the content of a different directory than the current working
>> directory but these are not predictably reproducable.  The behaviour is
>> similar to what we saw before we started mounting with nobrl two or
>> three years ago which leads me to suspect that nobrl somehow have
>> stopped being taken into account.  Once the client have started showing
>> wrong data for a directory no amount of prodding beyond unmounting and
>> remounting will make it show correct info again.
<SNIP>
>>
>> mount.cifs is called with options nobrl and serverino.
>>
>> It is entirely possible that this is a Fedora issue but I need to start
>> somewhere and this would be the place where people know the most about
>> the cifs client and might have seen something like this before.
>>
>> -BT
>
> It sounds like a similar issue as the one reported here:
>
> http://bugzilla.kernel.org/show_bug.cgi?id=15088
>
> It might be interesting to see whether the problems go away when you
> mount with -o noserverino.

A first quick test indicates that it does.  Unfortunately, without 
serverino KDE at least used to fail miserably - but that was KDE3, I 
haven't had any reason to check whether KDE4 worked without it.  I'll do 
some thorough testing when I'm back in the office on Monday.

Thank you very much!  I'll cc myself to that bug on kernel.org and keep 
updated on progress there.

-BT
-- 
Bjørn Tore Sund       Phone: 555-84894   Email:   bjorn.sund at it.uib.no
IT department         VIP:   81724       Support: http://bs.uib.no
Univ. of Bergen

When in fear and when in doubt, run in circles, scream and shout.