Re: address@hidden: Bug#136501: Parted segfaults on startup with Minix-created partition table]

[Andrew Clausen]

On Sun, Mar 03, 2002 at 09:41:38PM -0700, Andreas Dilger wrote:
> On Mar 04, 2002  08:11 +1100, Andrew Clausen wrote:
> > Parted does funky stuff with caches, etc., so I wouldn't be
> > surprised if we're causing some corner case.
> 
> Actually, the parted BLKFLSBUF tickles a bug that was in ext3 2.2 for
> a while (hit that on an internal server).  Didn't we discuss removing
> this call entirely?

Haven't gotten around to fixing it yet.  But removing it entirely
is Wrong...

> Basically, if a device is in-use in the kernel
> calling BLKFLSBUF is not sufficient to force all buffers to disk (it
> will refuse to flush busy buffers), and we won't be able to change a
> partition that is in-use anyways.

The problem is: partition devices vs. master devices.  libparted
does IO through master devices, so partition devices and master
devices must be flushed (otherwise you get very nasty cache-coherency
problems).  However, it shouldn't flush partition devices that are
being used (which never causes problems, because we aren't
modifying data in that part of the master device anyway)

Andrew