[Gllug] Remounting readonly fails

Mark Williams mark.666 at blueyonder.co.uk
Fri Apr 7 07:18:13 UTC 2006


Tet wrote:

>On 4/4/06, Nix <nix at esperi.org.uk> wrote:
>
>  
>
>>>Sure it isnt something silly like a shell having a directory on /usr
>>>as its 'curent working directory'   (hwoever this SHOULD show up in 'fuser').
>>>      
>>>
>>That's most likely, yes.
>>    
>>
>
>Nope, for two reasons:
>
>1. The filesystem had been mounted readonly beforehand. It was
>remounted rw, a yum update was run, then the remount readonly failed.
>In that brief period of time, it's unlikely that any process on the
>system would have a cwd in /usr without me explicitly starting one
>that did. Furthermore, the chances of a process being in the state you
>described, such that it can hide from fuser are vanishingly small
>IMHO.
>
>2. A process with a cwd on a given filesystem doesn't cause a readonly
>remount to fail anyway. Only an open file or a writable memory map can
>do that.
>
>I once wrote a patch to fuser that showed if maps and open files were
>writable, but the maintainer chose to ignore it. I suppose I should
>dig it out and compile it up again and see if it sheds any light on
>the problem...
>
>Tet
>  
>
Is this timing related? ie, will it re-mount after a while, or would 
inserting a  sync  before the final remount help?

Mark

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list