Re: [PATCH] KEYS/DNS: Fix ____call_usermodehelper() to not lose the session keyring

[Linus Torvalds <torvalds@linux-foundation.org>]

On Thu, Jun 16, 2011 at 4:56 PM, David Howells <dhowells@redhat.com> wrote:
> ____call_usermodehelper() now erases any credentials set by the
> subprocess_inf::init() function.  The problem is that:

I absolutely puke looking at this patch.  It makes me want to dig out
my eyes with a spoon.

Yes, we had that nasty "umh" TLA before too, but it was mostly hidden
and private and kept a fairly low profile (ie only in that
wait-related enum). This makes it _truly_ barf-worthy by combining it
with a couple of new ugly function typedefs.

There really isn't any prize for the ugliest infrastructure ever.
Please just remove the pointless and ugly typedefs that are used in
just a couple of places. And in most of those cases it just makes the
code actually less readable (ie it's now totally impossible to see
what the type for the passed-in function pointers are, because it's
hidden behind that opaque type.

The only real reason to use typedefs in the kernel is if you really
have *different* types behind them, and explicitly want to make the
type more opaque. Yeah, we've occasionally broken that rule, but it's
almost always been a mistake when we do.

                                 Linus

PS. Possibly it might make more sense to just put the "struct cred *"
pointer into the "struct subprocess_info" and not change any of the
function prototypes at all?