Am Wed, 23 Mar 2011 13:47:34 +0100
schrieb Lars Marowsky-Bree <l...@novell.com>:

> On 2011-03-23T11:50:02, Lars Ellenberg <lars.ellenb...@linbit.com> wrote:
> 
> > >  # Take advantage of /etc/mtab if present, use portable mount command
> > >  # otherwise. Normalize format to "dev mountpoint fstype".
> 
> I wonder if we shouldn't just always rely on "mount" and insist on that
> providing proper data.
> 
> Re-implementing mount seems like a bad idea. If more extensive checks
> than a binary "mounted xor not" are needed, this would be noticed by the
> deeper monitor levels or the application-level monitor, no?
> 
> 
> Regards,
>     Lars
> 

I had a weird case in which a filesystem (GFS2) had gotten unmounted
without /etc/mtab being modified.

It is possible to enforce that behavour by calling the umount() system
call directly from any program that has sufficient privileges.
(For example by executing `umount <mountpoint> -n` )

It can apparently also happen if a hardware (or in this case network)
caused event leads to a unmount by the kernel.

mount without arguments as well as the the filesystem resource agent
both still list the filesystem as mounted, simply because BOTH just
look at /etc/mtab. The application level resource agent of course
noticed the errors, but were unable to recover (the application on
course depends on its filesystem being mounted) pacemaker did not
remount the filesystem because the resource agents monitor() function
did not notice that it wasn't.

I agree, the "correct" way of doing this would be to use a system tool
that lists all mounted filesystems for us, regardless of the actual
system used. However `mount` is the wrong tool, since `mount` without
parameters only prints out thew contents of /etc/mtab (even if given an
additional "-n" flag this doesn't change) and as such is not suitable
as a `monitor()` IMHO.

Currently neither /etc/mtab nor `mount` are reliable to monitor the
mount state of a device or mountpoint. /proc/mounts is accurate with
some limitations.

regards

Corvus
_______________________________________________________
Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/

Reply via email to