Re: [PATCH 1/1] kernel/sysctl.c: avoid overflow

From: Dave Young <dyoung@redhat.com>
To: Willy Tarreau <w@1wt.eu>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Heinrich Schuchardt <xypron.glpk@gmx.de>,
	Arnaldo Carvalho de Melo <acme@redhat.com>,
	Kees Cook <keescook@chromium.org>,
	Don Zickus <dzickus@redhat.com>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Hugh Dickins <hughd@google.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Daniel Cashman <dcashman@google.com>,
	Alexei Starovoitov <ast@plumgrid.com>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	Ilya Dryomov <idryomov@gmail.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/1] kernel/sysctl.c: avoid overflow
Date: Wed, 15 Jun 2016 16:33:32 +0800	[thread overview]
Message-ID: <20160615083332.GA6466@dhcp-128-65.nay.redhat.com> (raw)
In-Reply-To: <20160614204149.GA2723@1wt.eu>

On 06/14/16 at 10:41pm, Willy Tarreau wrote:
> On Tue, Jun 14, 2016 at 01:19:06PM -0700, Andrew Morton wrote:
> > On Sat, 11 Jun 2016 03:33:08 +0200 Heinrich Schuchardt <xypron.glpk@gmx.de> wrote:
> > 
> > > An undetected overflow may occur in do_proc_dointvec_minmax_conv_param.
> > > 
> > > ...
> > >
> > > --- a/kernel/sysctl.c
> > > +++ b/kernel/sysctl.c
> > > @@ -2313,7 +2313,17 @@ static int do_proc_dointvec_minmax_conv(bool *negp, unsigned long *lvalp,
> > >  {
> > >  	struct do_proc_dointvec_minmax_conv_param *param = data;
> > >  	if (write) {
> > > -		int val = *negp ? -*lvalp : *lvalp;
> > > +		int val;
> > > +
> > > +		if (*negp) {
> > > +			if (*lvalp > (unsigned long) INT_MAX + 1)
> > > +				return -EINVAL;
> > > +			val = -*lvalp;
> > > +		} else {
> > > +			if (*lvalp > (unsigned long) INT_MAX)
> > > +				return -EINVAL;
> > > +			val = *lvalp;
> > > +		}
> > >  		if ((param->min && *param->min > val) ||
> > >  		    (param->max && *param->max < val))
> > >  			return -EINVAL;
> > 
> > hm.
> > 
> > What happens if someone does
> > 
> > 	echo -1 > /proc/foo
> > 
> > expecting to get 0xffffffff?  That's a reasonable shorthand, and if we
> > change that to spit out EINVAL then people's stuff may break.
> 
> I'd go even further, I don't see anymore how it becomes possible
> to actually *write* 0xffffffff at all! This function is used by
> proc_dointvec_minmax() which is used with extra1=&zero and extra2
> not set with some unsigned ints to allow the full range to be
> configured (eg: dirty_expire_interval is the first I found by a
> quick random look).

sysctl_writes_strict use extra1 = -1 and extra2 = 1

But I do not get why -1 does not work, 1 < (unsigned long) INT_MAX + 1
so val = -1, it is still right?

> 
> So for me this change is bogus.
> 
> Willy
> 

Thanks
Dave