[PATCH] vti: use right inner_mode for inbound inter address family policy checks

In case of inter address family tunneling (IPv6 over vti4 or IPv4 over
vti6), the inbound policy checks in vti_rcv_cb and vti6_rcv_cb are using
the wrong address family. As a result, all inbound inter address family
traffic is dropped.

Use the xfrm_ip2inner_mode helper (as done in xfrm_prepare_input and
xfrm_input) to select the inner_mode that contains the right address family
for the inbound policy checks.

Signed-off-by: Thomas Zeitlhofer <redacted>
---

Notes:
    The patch was developed by looking at the code, but without knowledge of
    the XFRM code in the kernel. It has been successfully tested, but it is
    more a guess that might be helpful for the maintainers to find a proper
    solution.

 net/ipv4/ip_vti.c  | 12 +++++++++++-
 net/ipv6/ip6_vti.c | 12 +++++++++++-
 2 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/ip_vti.c b/net/ipv4/ip_vti.c
index a917903..44d5449 100644
--- a/net/ipv4/ip_vti.c
+++ b/net/ipv4/ip_vti.c

@@ -88,6 +88,7 @@ static int vti_rcv_cb(struct sk_buff *skb, int err)
 	struct net_device *dev;
 	struct pcpu_sw_netstats *tstats;
 	struct xfrm_state *x;
+	struct xfrm_mode *inner_mode;
 	struct ip_tunnel *tunnel = XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4;
 	u32 orig_mark = skb->mark;
 	int ret;

@@ -105,7 +106,16 @@ static int vti_rcv_cb(struct sk_buff *skb, int err)
 	}
 
 	x = xfrm_input_state(skb);
-	family = x->inner_mode->afinfo->family;
+
+	inner_mode = x->inner_mode;
+
+	if (x->sel.family == AF_UNSPEC) {
+		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
+		if (inner_mode == NULL)
+			return -EPERM;
+	}
+
+	family = inner_mode->afinfo->family;
 
 	skb->mark = be32_to_cpu(tunnel->parms.i_key);
 	ret = xfrm_policy_check(NULL, XFRM_POLICY_IN, skb, family);

diff --git a/net/ipv6/ip6_vti.c b/net/ipv6/ip6_vti.c
index d90a11f..3149757 100644
--- a/net/ipv6/ip6_vti.c
+++ b/net/ipv6/ip6_vti.c

@@ -340,6 +340,7 @@ static int vti6_rcv_cb(struct sk_buff *skb, int err)
 	struct net_device *dev;
 	struct pcpu_sw_netstats *tstats;
 	struct xfrm_state *x;
+	struct xfrm_mode *inner_mode;
 	struct ip6_tnl *t = XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6;
 	u32 orig_mark = skb->mark;
 	int ret;

@@ -357,7 +358,16 @@ static int vti6_rcv_cb(struct sk_buff *skb, int err)
 	}
 
 	x = xfrm_input_state(skb);
-	family = x->inner_mode->afinfo->family;
+
+	inner_mode = x->inner_mode;
+
+	if (x->sel.family == AF_UNSPEC) {
+		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
+		if (inner_mode == NULL)
+			return -EPERM;
+	}
+
+	family = inner_mode->afinfo->family;
 
 	skb->mark = be32_to_cpu(t->parms.i_key);
 	ret = xfrm_policy_check(NULL, XFRM_POLICY_IN, skb, family);

-- 
2.1.4

On Sun, Sep 04, 2016 at 12:57:13PM +0200, Thomas Zeitlhofer wrote:

You better return -EINVAL instead of -EPERM here.
Also you should bump the LINUX_MIB_XFRMINSTATEMODEERROR
counter as we do in xfrm_input.

Same here.

Other that that it looks ok to me.

In case of inter address family tunneling (IPv6 over vti4 or IPv4 over
vti6), the inbound policy checks in vti_rcv_cb() and vti6_rcv_cb() are
using the wrong address family. As a result, all inbound inter address
family traffic is dropped.

Use the xfrm_ip2inner_mode() helper, as done in xfrm_input() (i.e., also
increment LINUX_MIB_XFRMINSTATEMODEERROR in case of error), to select the
inner_mode that contains the right address family for the inbound policy
checks.

Signed-off-by: Thomas Zeitlhofer <redacted>
---

Notes:
    v2: implement review comments from Steffen (thanks for the reply):
    
    	- return -EINVAL in case of error
    
    	- increment LINUX_MIB_XFRMINSTATEMODEERROR in case of error
    
    	  Just to point that out, in case there are arguments against it:
    	  this is done in the namespace of skb->dev and not in the
    	  t(unnel)?->net namespace.
    
    	and update the commit message accordingly.

 net/ipv4/ip_vti.c  | 15 ++++++++++++++-
 net/ipv6/ip6_vti.c | 15 ++++++++++++++-
 2 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/ip_vti.c b/net/ipv4/ip_vti.c
index a917903..c9957ed 100644
--- a/net/ipv4/ip_vti.c
+++ b/net/ipv4/ip_vti.c

@@ -88,6 +88,7 @@ static int vti_rcv_cb(struct sk_buff *skb, int err)
 	struct net_device *dev;
 	struct pcpu_sw_netstats *tstats;
 	struct xfrm_state *x;
+	struct xfrm_mode *inner_mode;
 	struct ip_tunnel *tunnel = XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4;
 	u32 orig_mark = skb->mark;
 	int ret;

@@ -105,7 +106,19 @@ static int vti_rcv_cb(struct sk_buff *skb, int err)
 	}
 
 	x = xfrm_input_state(skb);
-	family = x->inner_mode->afinfo->family;
+
+	inner_mode = x->inner_mode;
+
+	if (x->sel.family == AF_UNSPEC) {
+		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
+		if (inner_mode == NULL) {
+			XFRM_INC_STATS(dev_net(skb->dev),
+				       LINUX_MIB_XFRMINSTATEMODEERROR);
+			return -EINVAL;
+		}
+	}
+
+	family = inner_mode->afinfo->family;
 
 	skb->mark = be32_to_cpu(tunnel->parms.i_key);
 	ret = xfrm_policy_check(NULL, XFRM_POLICY_IN, skb, family);

diff --git a/net/ipv6/ip6_vti.c b/net/ipv6/ip6_vti.c
index d90a11f..52a2f73 100644
--- a/net/ipv6/ip6_vti.c
+++ b/net/ipv6/ip6_vti.c

@@ -340,6 +340,7 @@ static int vti6_rcv_cb(struct sk_buff *skb, int err)
 	struct net_device *dev;
 	struct pcpu_sw_netstats *tstats;
 	struct xfrm_state *x;
+	struct xfrm_mode *inner_mode;
 	struct ip6_tnl *t = XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6;
 	u32 orig_mark = skb->mark;
 	int ret;

@@ -357,7 +358,19 @@ static int vti6_rcv_cb(struct sk_buff *skb, int err)
 	}
 
 	x = xfrm_input_state(skb);
-	family = x->inner_mode->afinfo->family;
+
+	inner_mode = x->inner_mode;
+
+	if (x->sel.family == AF_UNSPEC) {
+		inner_mode = xfrm_ip2inner_mode(x, XFRM_MODE_SKB_CB(skb)->protocol);
+		if (inner_mode == NULL) {
+			XFRM_INC_STATS(dev_net(skb->dev),
+				       LINUX_MIB_XFRMINSTATEMODEERROR);
+			return -EINVAL;
+		}
+	}
+
+	family = inner_mode->afinfo->family;
 
 	skb->mark = be32_to_cpu(t->parms.i_key);
 	ret = xfrm_policy_check(NULL, XFRM_POLICY_IN, skb, family);

-- 
2.1.4

From: thomas.zeitlhofer+lkml@ze-it.at
Date: Wed, 7 Sep 2016 20:40:38 +0200

In case of inter address family tunneling (IPv6 over vti4 or IPv4 over
vti6), the inbound policy checks in vti_rcv_cb() and vti6_rcv_cb() are
using the wrong address family. As a result, all inbound inter address
family traffic is dropped.

Use the xfrm_ip2inner_mode() helper, as done in xfrm_input() (i.e., also
increment LINUX_MIB_XFRMINSTATEMODEERROR in case of error), to select the
inner_mode that contains the right address family for the inbound policy
checks.

Signed-off-by: Thomas Zeitlhofer <redacted>

Steffen please review this new version of the patch.

Thanks.

On Wed, Sep 07, 2016 at 08:40:38PM +0200, thomas.zeitlhofer+lkml@ze-it.at wrote:

In case of inter address family tunneling (IPv6 over vti4 or IPv4 over
vti6), the inbound policy checks in vti_rcv_cb() and vti6_rcv_cb() are
using the wrong address family. As a result, all inbound inter address
family traffic is dropped.

Use the xfrm_ip2inner_mode() helper, as done in xfrm_input() (i.e., also
increment LINUX_MIB_XFRMINSTATEMODEERROR in case of error), to select the
inner_mode that contains the right address family for the inbound policy
checks.

Signed-off-by: Thomas Zeitlhofer <redacted>
---

Notes:
    v2: implement review comments from Steffen (thanks for the reply):
    
    	- return -EINVAL in case of error
    
    	- increment LINUX_MIB_XFRMINSTATEMODEERROR in case of error
    
    	  Just to point that out, in case there are arguments against it:
    	  this is done in the namespace of skb->dev and not in the
    	  t(unnel)?->net namespace.

This is ok because the states are configured in that namespace.

I've applied this to the ipsec tree, thanks a lot!