On Wed, May 16, 2007 at 12:17:53AM +0200, Hauke Fath wrote:

> But since the kernel actually knows the USB device has been disconnected,
> it could either (nfs hard-mount-like) put any processes to sleep that had
> files open there, or (nfs soft-mount-like) fail the accesses, and clean up.
> Instead, it spills the beans.

It's even worse. The device driver is freed, but the filesystem on top
of it doesn't know about it and will use invalid pointers to the driver.

Most of the time this just stalls a userland process forever. Sometimes
the kernel crashes.

> Holes like that in USB support are one of the major reasons why NetBSD is
> on the way out in our labs. A machine locking up because you try to use a
> USB stick with it - that doesn't fly.

Indeed.

-- 
                                Michael van Elst
Internet: mlelstv@serpens.de
                                "A potential Snark may lurk in every tree."