[PATCH V3] netfilter: h323: avoid potential attack

I think hackers chould build a malicious h323 packet to overflow
the pointer p which will panic during the memcpy(addr, p, len)
For example, he may fabricate a very large taddr->ipAddress.ip;
As suggested by Eric, this module is protected by a lock (nf_h323_lock)
so adding a variable h323_buffer_valid_bytes that would contain
the number of valid bytes would not require to change prototypes of
get_h2x5_addr.

Signed-off-by: Zhouyi Zhou <redacted>
Signed-off-by: Eric Dumazet <redacted>
Reviewed-by: Sergei Shtylyov <redacted> 

---
 net/netfilter/nf_conntrack_h323_main.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
index 9511af0..65d84bc 100644
--- a/net/netfilter/nf_conntrack_h323_main.c
+++ b/net/netfilter/nf_conntrack_h323_main.c

@@ -110,6 +110,11 @@ int (*nat_q931_hook) (struct sk_buff *skb,
 
 static DEFINE_SPINLOCK(nf_h323_lock);
 static char *h323_buffer;
+static unsigned int h323_buffer_valid_bytes;
+/* check offset overflow and out of range data reference */
+#define CHECK_BOUND(p, n) ((n) > h323_buffer_valid_bytes ||		\
+			   ((void *)(p) + (n) - (void *)h323_buffer	\
+			    > h323_buffer_valid_bytes))
 
 static struct nf_conntrack_helper nf_conntrack_helper_h245;
 static struct nf_conntrack_helper nf_conntrack_helper_q931[];

@@ -145,6 +150,7 @@ static int get_tpkt_data(struct sk_buff *skb, unsigned int protoff,
 
 	if (*data == NULL) {	/* first TPKT */
 		/* Get first TPKT pointer */
+		h323_buffer_valid_bytes = tcpdatalen;
 		tpkt = skb_header_pointer(skb, tcpdataoff, tcpdatalen,
 					  h323_buffer);
 		BUG_ON(tpkt == NULL);

@@ -247,6 +253,9 @@ static int get_h245_addr(struct nf_conn *ct, const unsigned char *data,
 		return 0;
 	}
 
+	if (CHECK_BOUND(p, len + sizeof(__be16)))
+		return 0;
+
 	memcpy(addr, p, len);
 	memset((void *)addr + len, 0, sizeof(*addr) - len);
 	memcpy(port, p + len, sizeof(__be16));

@@ -669,6 +678,9 @@ int get_h225_addr(struct nf_conn *ct, unsigned char *data,
 		return 0;
 	}
 
+	if (CHECK_BOUND(p, len + sizeof(__be16)))
+		return 0;
+
 	memcpy(addr, p, len);
 	memset((void *)addr + len, 0, sizeof(*addr) - len);
 	memcpy(port, p + len, sizeof(__be16));

@@ -1248,6 +1260,7 @@ static unsigned char *get_udp_data(struct sk_buff *skb, unsigned int protoff,
 	if (dataoff >= skb->len)
 		return NULL;
 	*datalen = skb->len - dataoff;
+	h323_buffer_valid_bytes = *datalen;
 	return skb_header_pointer(skb, dataoff, *datalen, h323_buffer);
 }
 

-- 
1.9.1

On Fri, Jan 29, 2016 at 11:25:35AM +0800, Zhouyi Zhou wrote:

We don't want obscure macros. You add a function for this, the
compiler will likely inline it.

On Mon, 2016-02-01 at 18:58 +0100, Pablo Neira Ayuso wrote:

On Fri, Jan 29, 2016 at 11:25:35AM +0800, Zhouyi Zhou wrote:

quoted

I think hackers chould build a malicious h323 packet to overflow
the pointer p which will panic during the memcpy(addr, p, len)
For example, he may fabricate a very large taddr->ipAddress.ip;
As suggested by Eric, this module is protected by a lock (nf_h323_lock)
so adding a variable h323_buffer_valid_bytes that would contain
the number of valid bytes would not require to change prototypes of
get_h2x5_addr.

Signed-off-by: Zhouyi Zhou <redacted>
Signed-off-by: Eric Dumazet <redacted>
Reviewed-by: Sergei Shtylyov <redacted> 

---
 net/netfilter/nf_conntrack_h323_main.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/net/netfilter/nf_conntrack_h323_main.c b/net/netfilter/nf_conntrack_h323_main.c
index 9511af0..65d84bc 100644
--- a/net/netfilter/nf_conntrack_h323_main.c
+++ b/net/netfilter/nf_conntrack_h323_main.c

@@ -110,6 +110,11 @@ int (*nat_q931_hook) (struct sk_buff *skb,
 
 static DEFINE_SPINLOCK(nf_h323_lock);
 static char *h323_buffer;
+static unsigned int h323_buffer_valid_bytes;
+/* check offset overflow and out of range data reference */
+#define CHECK_BOUND(p, n) ((n) > h323_buffer_valid_bytes ||		\
+			   ((void *)(p) + (n) - (void *)h323_buffer	\
+			    > h323_buffer_valid_bytes))

We don't want obscure macros. You add a function for this, the
compiler will likely inline it.

BTW, I did not signed-off this patch.

Zhouyi Zho, just add your own signature, let people add their own.

Thanks.