Re: [2.5 patch] let COMX depend on PROC_FS

[viro@parcelfarce.linux.theplanet.co.uk]

On Tue, Jun 10, 2003 at 02:26:14PM +0100, Christoph Hellwig wrote:
> On Tue, Jun 10, 2003 at 01:55:22PM +0200, Pásztor Szilárd wrote:
> > The drivers are used by some hundreds of cards today but we tell users to
> > get the small kernelpatch from www.itc.hu and the patch, among other things,
> > exports proc_get_inode. There was a process to integrate the patch into the
> > mainstream kernel last year but, due to lack of time on my part, it was
> > suspended. I hope to be able to pick the line up again and clean things up.
> 
> So what about fixing it instead?  The usage of proc_get_inode is broken
> and so is the whole profs mess in the comx driver.  If you want to keep
> the API you need to add a ramfs-style filesystem instead of abusing
> procfs.

"broken" is a very polite way to describe that driver.  Starting with the
idea of mkdir in virtual filesystem (procfs or otherwise) creating and
populating a diretory (unmodifiable, BTW) and rmdir - removing it, even
though it's non-empty (and can't be emptied, due to above).

Guys, that's _sick_.  And that's aside of the shitload of races all over
that code (no locking whatsoever).  And kmalloc(..., GFP_KERNEL) with
interrupts disabled.  And shutting the hardware down before unregistering
netdev (yes, you check that it's down; nothing guarantees that it will
stay down while you do ->hw_exit() and friends).  And so on, and so on...

IOW, driver needs a serious rewrite, starting with its API.