AW: [CFT] 2.6.0-test7: Fix use-after-free bug in ds.

[Pavel Roskin]

On Wed, 15 Oct 2003, David Hinds wrote:

> On Wed, Oct 15, 2003 at 03:55:19PM -0400, Pavel Roskin wrote:
> >
> > It's a great simplification in the 2.6 kernels that ds doesn't need
> > to be loaded after all socket drivers.  This allows loading more
> > drivers after ds.
>
> I don't see why this is a particularly useful thing.

Simplicity of administration.  If a new bridge is found, there should be no
need to kill cardmgr, unload ds, load the socket driver, load ds and
restart cardmgr.  I believe the PCMCIA subsystem is too complex compared
to other subsystems, where it's usually enough to load another module.

I'm subscribed to several mailing lists for wireless devices.  Hardly a
week passes without a post from somebody confused by the design of the
PCMCIA support in Linux.

I know an embedded system that has exactly two executables - busybox and
cardmgr.  Everything is compiled into the kernel, yet a userspace program
is needed to bind kernel drivers to kernel sockets.

> > But the next logical step is to remove the separation between
> > pcmcia_core and ds, which is now unnecessary.
>
> The original reason for separation between ds and pcmcia_core was a
> conceptual one.  "cs" implemented the core Card Services API's.  The
> "ds" module is (or I guess I should say, "was") itself an ordinary Card
> Services client driver; it is special only in that it talks to cardmgr
> and exports some services to other clients.  It is not required that a
> PCMCIA driver use the "ds" API and be managed by cardmgr.

I have no problem with implementing ds as a special client module.  I have
a problem with the requirement that it has to be loaded after all other
socket drivers.  I'm glad that it has been lifted, and I'll be
disappointed if this requirement is reinstated.

-- 
Regards,
Pavel Roskin