[SSI] Re: /initrd fails to unmount

[David B. Z. <dav...@hp...>]

Last time I checked process 2 was not preventing /initrd from being 
unmounted.  One of my test clusters umounts /initrd just fine.  Another 
with Fiber-Channel disks does not, because the FC modules creates a 
daemon while loading from linuxrc and it doesn't daemonize itself 
properly.  Another of our SSI group has a cluster which also doesn't 
umount /initrd, but I haven't looked at that one yet.

In your cluster I would do an "fuser -m /initrd" after booting and look 
at the process(es) that are still using the filesystem.  I assume it 
will be daemons or processes created during linuxrc.  If you can fix any 
hardware specific modules/drivers, the problem should go away.  If not I 
can try to find out what is still wrong.

If you are using the Compaq Fiber Channel modules then apply the patch 
below:

--- sandbox.orig/drivers/scsi/cpqfcTSworker.c   Thu Oct 25 13:53:50 2001
+++ sandbox/drivers/scsi/cpqfcTSworker.c        Thu Aug  1 14:14:04 2002
@@ -194,6 +194,10 @@
      current->fs = fs;
      atomic_inc(&fs->count);
    }
+  // From routine daemonize() to close all files
+  exit_files(current);
+  current->files = init_task.files;
+  atomic_inc(&current->files->count);

    siginitsetinv(&current->blocked, SHUTDOWN_SIGS);



Bertrand Sirodot wrote:
> Hi,
> 
> I would like to know if anybody has made any progress 
> with this bug. From what I can see nothing is holding up 
> anything in /initrd, but ,because at the end of 
> the /initrd/linuxrc script, it does a chroot, which it 
> means that eventhough we can't see anything 
> using /initrd, it still keeps a reference to /initrd.
> 
> The only way, I can see, to fix it would be to maybe to a 
> cd somewhere else before doing the chroot.
> 
> What do you think?
> Regards,
> Bertrand.
> 


-- 
David B. Zafman                            | Hewlett-Packard Company
Linux Kernel Developer                  | Open SSI Clustering Project
mailto:dav...@hp...        | http://www.hp.com
"Thus spake the master programmer:  When you have learned to snatch
the error code from the trap frame, it will be time for you to leave."