The nlmsg_len() helper returned "int" from a u32 calculation that could
possible go negative. WARN() if this calculation ever goes negative
(instead returning 0), or if the result would be larger than INT_MAX
(instead returning INT_MAX).

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Jozsef Kadlecsik <kadlec@netfilter.org>
Cc: Florian Westphal <fw@strlen.de>
Cc: syzbot <syzkaller@googlegroups.com>
Cc: Yajun Deng <yajun.deng@linux.dev>
Cc: netdev@vger.kernel.org
Cc: netfilter-devel@vger.kernel.org
Cc: coreteam@netfilter.org
Signed-off-by: Kees Cook <keescook@chromium.org>
---
 include/net/netlink.h | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/include/net/netlink.h b/include/net/netlink.h
index 7a2a9d3144ba..f8cb0543635e 100644
--- a/include/net/netlink.h
+++ b/include/net/netlink.h
@@ -576,7 +576,15 @@ static inline void *nlmsg_data(const struct nlmsghdr *nlh)
  */
 static inline int nlmsg_len(const struct nlmsghdr *nlh)
 {
-	return nlh->nlmsg_len - NLMSG_HDRLEN;
+	u32 nlmsg_contents_len;
+
+	if (WARN_ON_ONCE(check_sub_overflow(nlh->nlmsg_len,
+					    (u32)NLMSG_HDRLEN,
+					    &nlmsg_contents_len)))
+		return 0;
+	if (WARN_ON_ONCE(nlmsg_contents_len > INT_MAX))
+		return INT_MAX;
+	return nlmsg_contents_len;
 }
 
 /**
-- 
2.34.1

On Wed, 31 Aug 2022 20:06:09 -0700 Kees Cook wrote:
>  static inline int nlmsg_len(const struct nlmsghdr *nlh)
>  {
> -	return nlh->nlmsg_len - NLMSG_HDRLEN;
> +	u32 nlmsg_contents_len;
> +
> +	if (WARN_ON_ONCE(check_sub_overflow(nlh->nlmsg_len,
> +					    (u32)NLMSG_HDRLEN,
> +					    &nlmsg_contents_len)))
> +		return 0;
> +	if (WARN_ON_ONCE(nlmsg_contents_len > INT_MAX))
> +		return INT_MAX;
> +	return nlmsg_contents_len;

We check the messages on input, making sure the length is valid wrt
skb->len, and sane, ie > NLMSG_HDRLEN. See netlink_rcv_skb().

Can we not, pretty please? :(

On Wed, Aug 31, 2022 at 08:18:25PM -0700, Jakub Kicinski wrote:
> On Wed, 31 Aug 2022 20:06:09 -0700 Kees Cook wrote:
> >  static inline int nlmsg_len(const struct nlmsghdr *nlh)
> >  {
> > -	return nlh->nlmsg_len - NLMSG_HDRLEN;
> > +	u32 nlmsg_contents_len;
> > +
> > +	if (WARN_ON_ONCE(check_sub_overflow(nlh->nlmsg_len,
> > +					    (u32)NLMSG_HDRLEN,
> > +					    &nlmsg_contents_len)))
> > +		return 0;
> > +	if (WARN_ON_ONCE(nlmsg_contents_len > INT_MAX))
> > +		return INT_MAX;
> > +	return nlmsg_contents_len;
> 
> We check the messages on input, making sure the length is valid wrt
> skb->len, and sane, ie > NLMSG_HDRLEN. See netlink_rcv_skb().
> 
> Can we not, pretty please? :(

This would catch corrupted values...

Is the concern the growth in image size? The check_sub_overflow() isn't
large at all -- it's just adding a single overflow bit test. The WARNs
are heavier, but they're all out-of-line.

-- 
Kees Cook

On Wed, 31 Aug 2022 23:27:08 -0700 Kees Cook wrote:
> This would catch corrupted values...
> 
> Is the concern the growth in image size? The check_sub_overflow() isn't
> large at all -- it's just adding a single overflow bit test. The WARNs
> are heavier, but they're all out-of-line.

It turns the most obvious function into a noodle bar :(

Looking at this function in particular is quite useful, because 
it clearly indicates that the nlmsg_len includes the header.

How about we throw in a

	WARN_ON_ONCE(nlh->nlmsg_len < NLMSG_HDRLEN ||
		     nlh->nlmsg_len > INT_MAX);

but leave the actual calculation human readable C?

On Thu, Sep 1, 2022 at 12:49 PM Jakub Kicinski <kuba@kernel.org> wrote:
>
> On Wed, 31 Aug 2022 23:27:08 -0700 Kees Cook wrote:
> > This would catch corrupted values...
> >
> > Is the concern the growth in image size? The check_sub_overflow() isn't
> > large at all -- it's just adding a single overflow bit test. The WARNs
> > are heavier, but they're all out-of-line.
>
> It turns the most obvious function into a noodle bar :(
>
> Looking at this function in particular is quite useful, because
> it clearly indicates that the nlmsg_len includes the header.
>
> How about we throw in a
>
>         WARN_ON_ONCE(nlh->nlmsg_len < NLMSG_HDRLEN ||
>                      nlh->nlmsg_len > INT_MAX);
>
> but leave the actual calculation human readable C?

This is inlined, and will add a lot of extra code. We are making the
kernel slower at each release.

What about letting fuzzers like syzbot find the potential issues ?

DEBUG_NET_WARN_ON_ONCE(...);