[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <202110251706.BBEE1428@keescook>
Date: Mon, 25 Oct 2021 17:17:57 -0700
From: Kees Cook <keescook@...omium.org>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: Matthew Wilcox <willy@...radead.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Linux-MM <linux-mm@...ck.org>,
Jordy Zomer <jordy@...ing.systems>,
James Bottomley <James.Bottomley@...senpartnership.com>,
Mike Rapoport <rppt@...nel.org>,
Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH] secretmem: Prevent secretmem_users from wrapping to zero
On Mon, Oct 25, 2021 at 04:37:01PM -0700, Linus Torvalds wrote:
> If we want other semantics, it should be a new type.
Okay, that's reasonable.
> > I see that is more like a "shared resource usage count" where the shared
> > resource doesn't necessarily disappear when we reach "no users"?
>
> So I think that's really "atomic_t".
>
> And instead of saturating, people should always check such shared
> resources for limits.
Right, but people make mistakes, etc. I agree about the limit being much
more sane than saturating (though in the cases of "missed decrement"),
we get to the same place: an open-coded check for the limit that never
goes down doesn't matter if it's refcount_t nor atomic_t. :)
> > i.e. there is some resource, and it starts its life with no one using it
> > (count = 1).
>
> You are already going off into the weeds.
>
> That's not a natural thing to do. It's already confusing. Really. Read
> that sentence yourself, and read it like an outsider.
>
> "No one is using it, so count == 1" is a nonsensican statement on the
> face of it.
>
> You are thinking of a refcount_t trick, not some sane semantics.
>
> Yes, we have played off-by-one games in the kernel before. We've done
> it for various subtle reasons.
Right, sure, but it's not a rare pattern. Given that it exists, and that
it _does_ get used for allocation management (e.g. module loader), it
seems worth constructing a proper type for it so that all the open coded
stuff around these instances can be consolidated, and the API can be
defined in a way that will behave sanely.
> I really don't see what's wrong with 'atomic_t', and just checking for limits.
It's that last part. :) If we go through atomic_dec() see a zero and do
something, okay, fine. But these places need to check for insane
conditions too ("we got a -1 back -- this means there's a bug but what
do we do?"). Same for atomic_inc(): "oh, we're at our limit, do
something", but what above discovering ourselves above the limit?
There's nothing about using the atomic_t primitives that enforces these
kinds of checks. (And there likely shouldn't be for atomic_t -- it's a
plain type.) But we likely need something that fills in this API gap
between atomic_t and refcount_t.
> So if a user can ever trigger a saturating counter, that's a big big
> problem in itself.
Yes! It is. :) But they don't get to gain control over a Use-after-Free.
The risk to the system is DoS instead of loss of execution control.
That's a meaningful risk downgrade.
So, what's the right semantics for an atomic type that could be used in
the module loader, that would catch kernel counting bugs in a safe manner?
The "refcount_t but 1-based" is close, but clearly not the right name. :)
--
Kees Cook
Powered by blists - more mailing lists