Re: KASAN, xt_TCPMSS finally found nasty use-after-free bug? 4.10.8

[Eric Dumazet]

On Sun, 2017-04-02 at 04:54 -0700, Eric Dumazet wrote:
> On Sun, 2017-04-02 at 13:45 +0200, Florian Westphal wrote:
> > Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote:
> > > -	for (i = sizeof(struct tcphdr); i <= tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) {
> > > +	for (i = sizeof(struct tcphdr); i < tcp_hdrlen - TCPOLEN_MSS; i += optlen(opt, i)) {
> > >  		if (opt[i] == TCPOPT_MSS && opt[i+1] == TCPOLEN_MSS) {
> > >  			u_int16_t oldmss;
> > 
> > maybe I am low on caffeeine but this looks fine, for tcp header with
> > only tcpmss this boils down to "20 <= 24 - 4" so we acccess offsets 20-23 which seems ok.
> 
> I am definitely low on caffeine ;)
> 
> An issue in this function is that we might add the missing MSS option,
> without checking that TCP options are already full.
> 
> But this should not cause a KASAN splat, only some malformed TCP packet
> 
> (tcph->doff would wrap)

Something like that maybe.

diff --git a/net/netfilter/xt_TCPMSS.c b/net/netfilter/xt_TCPMSS.c
index
27241a767f17b4b27d24095a31e5e9a2d3e29ce4..1465aaf0e3a15d69d105d0a50b0429b11b6439d3 100644
--- a/net/netfilter/xt_TCPMSS.c
+++ b/net/netfilter/xt_TCPMSS.c
@@ -151,7 +151,9 @@ tcpmss_mangle_packet(struct sk_buff *skb,
 	 */
 	if (len > tcp_hdrlen)
 		return 0;
-
+	/* tcph->doff is 4 bits wide, do not wrap its value to 0 */
+	if (tcp_hdrlen >= 15 * 4)
+		return 0;
 	/*
 	 * MSS Option not found ?! add it..
 	 */