Re: Important: fs corruption with 2.0.33 and removable media

Eric Youngdale (eric@andante.jic.com)
Sun, 8 Mar 1998 22:51:42 -0500 (EST)


On Tue, 3 Mar 1998, Richard Waltham wrote:

> Wolfram@informatik.uni-bonn.de wrote:
> >
> > There is a very disastrous bug in 2.0.33 (untested with 2.1.x):
> > The disc change detection with removable media is not 100% working !
> > This has a destroying effect if your computer has enough RAM for caching.
> > [tested... :-(((( ]
> >
> > Please test the following with your hardware: (reproduce able for me)
> > -mount the media and unmount it
> > -eject the media
> > -insert another media
> > -wait ~ 30 min. without doing anything at your computer (important !)
> > -mount this media (better mount it ro for safety !)
> > -no disc change detection.... (look at the kernel logs)
> >
> > It looks like the disc change has been forgotten in these 30 min.
> > The disc change detection works normal if the change and mount happens within
> > less time.
> >
> > The hardware: Fujitsu M2512A2 (230MB MO Drive & 230MB Media)
> > ncr53c825 SCSI interface (with Gerard Roudier's driver)
> >
> >
> > Is this reproduce able by other people ?
> >
> > Is there any possibility to fix this bug inside the kernel ?
> > E.g. test every minute for disc change detection ?
> >
>
> No problem here - but I'm not using an M2512A. Using an Iomega jaz shows no
> problems trying the above. Also using ncr53c8xx driver;)
>
> Disk change is dependent on the drive reporting that the media has changed
> when it gets a SCSI command to do something. Its not something that the
> kernel keeps a track of.
>
> Is your drive set to automatically spin down? I have tried one of these
> drives (a really old one) on this machine but cannot remember if it's always
> set to automatically spin down or whether it is set by one of the switches.
>
> Could this be a bug in the drive firmware? eg remove media, load new, let it
> sit for sometime - maybe drive spins down and forgets disk change - mount
> new media, spins up and no disk change notification.
>
> This really sounds like a drive problem even if it is not set to spin down.
> It is the drives job to say the media has changed.

The theory is that the drive itself should be tracking when media
is changed, and notifying the kernel via the scsi layer that this has
happened. My guess is that the drive is failing to send the appropriate
signal that the media has changed.

-Eric

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu