Re: [PATCH 3/3] sysctl: detect overflows when converting to int

From: Kees Cook <keescook@chromium.org>
To: Heinrich Schuchardt <xypron.glpk@gmx.de>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Michal Nazarewicz <mina86@mina86.com>,
	Ingo Molnar <mingo@kernel.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Joe Perches <joe@perches.com>, Josh Hunt <johunt@akamai.com>,
	Rasmus Villemoes <linux@rasmusvillemoes.dk>,
	Rusty Russell <rusty@rustcorp.com.au>,
	Daniel Walter <dwalter@google.com>,
	David Rientjes <rientjes@google.com>,
	"David S. Miller" <davem@davemloft.net>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Aaron Tomlin <atomlin@redhat.com>,
	Prarit Bhargava <prarit@redhat.com>,
	Eric B Munson <emunson@akamai.com>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	Sam Ravnborg <sam@ravnborg.org>,
	LKML <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 3/3] sysctl: detect overflows when converting to int
Date: Mon, 8 Jun 2015 15:41:15 -0700	[thread overview]
Message-ID: <CAGXu5j+MVekAqaOm4MmrH97uOtEnsmE4zzvptjuhxhmY-86vag@mail.gmail.com> (raw)
In-Reply-To: <551C2B86.6080204@gmx.de>

On Wed, Apr 1, 2015 at 10:31 AM, Heinrich Schuchardt <xypron.glpk@gmx.de> wrote:
> On 01.04.2015 00:45, Andrew Morton wrote:
>> On Sun, 29 Mar 2015 21:28:29 +0200 Heinrich Schuchardt <xypron.glpk@gmx.de> wrote:
>>
>>> When converting unsigned long to int overflows may occur.
>>> These currently are not detected when writing to the sysctl
>>> file system.
>>>
>>> E.g. on a system where int has 32 bits and long has 64 bits
>>>   echo 0x800001234 > /proc/sys/kernel/threads-max
>>> has the same effect as
>>>   echo 0x1234 > /proc/sys/kernel/threads-max
>>>
>>> The patch adds the missing check in do_proc_dointvec_conv.
>>>
>>> With the patch an overflow will result in an error EINVAL when
>>> writing to the the sysctl file system.
>>
>> hm, why fix this?  There's a small risk of breaking
>> accidentally-working userspace, but I expect we can live with that.
>>
>> But how big a problem is this, really?  This behaviour is quite
>> expected, after all.
>>
>
> The typical user of a Linux system has never read the Kernel code and
> possibly has limited programming experience.
> Furthermore in Documentation/sysctl/kernel.txt there is no hint that
> only 32-bit integers can be used.
> So why should this typical user expect that on a 64-bit system
> +3000000000 is considered a negative number?
>
> Now that we know this is a bug why shouldn't we fix it?

I think this is worth fixing. It is, from a certain perspective,
"unexpected behavior". At the very least we could tie it to the
sysctl_writes_strict flag? Anything depending on an overflow to get
"correct" results seems extremely unlikely to me.

-Kees

-- 
Kees Cook
Chrome OS Security