Linux Ethernet Bridging - Sep 2013 - [Bridge] [RFC net] bridge: clean the nf_bridge status when forwarding the skb

From: Antonio Quartulli <antonio at open-mesh.com>

Even if enslaving a bridge interface into another bridge is
forbidden, it is still possible to create a chain of
virtual interfaces including two distinct bridges.

In this case, the skb entering the second bridge could have
the nf_bridge field already set due to a previous operation
and consequently lead to a wrong processing of the packet
itself.

To prevent this behaviour release and set to NULL the
nf_bridge field of the skb when exiting the bridge interface.

Signed-off-by: Antonio Quartulli <antonio at open-mesh.com>
---

I am not sure if this is a wanted behaviour or a real BUG. I found this
"misbehaviour" while testing batman-adv with the following
configuration:

- br0 (bridge interface) having bat0 and eth0 as slaves
- bat0 (which is a virtual interface provided by the batman-adv module and that
  works similarly to a bridge - to some extends) having br1 as slave
- br1 (second bridge interface) having eth1 as slave

Then follow these events:
- a broadcast packet arrives on eth0
- the skb enters br0 and skb->nf_bridge gets initialised and used
- the skb enters bat0 and the packet *gets encapsulated in the batman-adv packet
  which adds a batman-adv header and another Ethernet header*
- the skb enters br1 and gets ruined because nf_bridge_maybe_copy_header() (in
  br_dev_queue_push_xmit()) will try to restore an header that does not make
  sense anymore.

With this patch the nf_bridge gets de-initialised before exiting br0 and
therefore it is processed properly inside br1: nf_bridge_maybe_copy_header()
does not take place at all because nf_bridge is never initialised (the packet is
non-IP since it is a batman-adv packet)

To the developers of the bridge module I would like to ask:
1) is skb->nf_bridge allowed to be non NULL when entering br_dev_xmit() ? If
so,
   when is this supposed to happen?

2) do you think this patch is logically correct but the nf_bridge release should
   be done in batman-adv since it is the one re-encapsulating the packet?


I hope I have made the problem clear.

Best regards,


 net/bridge/br_forward.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/net/bridge/br_forward.c b/net/bridge/br_forward.c
index 4b81b14..65864bc 100644
--- a/net/bridge/br_forward.c
+++ b/net/bridge/br_forward.c
@@ -49,6 +49,11 @@ int br_dev_queue_push_xmit(struct sk_buff *skb)
 	} else {
 		skb_push(skb, ETH_HLEN);
 		br_drop_fake_rtable(skb);
+
+		/* clean the NF bridge data */
+		nf_bridge_put(skb->nf_bridge);
+		skb->nf_bridge = NULL;
+
 		dev_queue_xmit(skb);
 	}
 
-- 
1.8.1.5

On Thu, 26 Sep 2013 22:19:50 +0200
Antonio Quartulli <antonio@meshcoding.com> wrote:
> From: Antonio Quartulli <antonio@open-mesh.com>
> 
> Even if enslaving a bridge interface into another bridge is
> forbidden, it is still possible to create a chain of
> virtual interfaces including two distinct bridges.
> 
> In this case, the skb entering the second bridge could have
> the nf_bridge field already set due to a previous operation
> and consequently lead to a wrong processing of the packet
> itself.
> 
> To prevent this behaviour release and set to NULL the
> nf_bridge field of the skb when exiting the bridge interface.
> 
> Signed-off-by: Antonio Quartulli <antonio@open-mesh.com>
> ---
> 
> I am not sure if this is a wanted behaviour or a real BUG. I found this
> "misbehaviour" while testing batman-adv with the following
configuration:
> 
> - br0 (bridge interface) having bat0 and eth0 as slaves
> - bat0 (which is a virtual interface provided by the batman-adv module and
that
>   works similarly to a bridge - to some extends) having br1 as slave
> - br1 (second bridge interface) having eth1 as slave
> 
> Then follow these events:
> - a broadcast packet arrives on eth0
> - the skb enters br0 and skb->nf_bridge gets initialised and used
> - the skb enters bat0 and the packet *gets encapsulated in the batman-adv
packet
>   which adds a batman-adv header and another Ethernet header*
> - the skb enters br1 and gets ruined because nf_bridge_maybe_copy_header()
(in
>   br_dev_queue_push_xmit()) will try to restore an header that does not
make
>   sense anymore.
> 
> With this patch the nf_bridge gets de-initialised before exiting br0 and
> therefore it is processed properly inside br1:
nf_bridge_maybe_copy_header()
> does not take place at all because nf_bridge is never initialised (the
packet is
> non-IP since it is a batman-adv packet)
> 
> To the developers of the bridge module I would like to ask:
> 1) is skb->nf_bridge allowed to be non NULL when entering br_dev_xmit()
? If so,
>    when is this supposed to happen?
> 
> 2) do you think this patch is logically correct but the nf_bridge release
should
>    be done in batman-adv since it is the one re-encapsulating the packet?
> 
> 
> I hope I have made the problem clear.
> 
> Best regards,
> 
> 
>  net/bridge/br_forward.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/net/bridge/br_forward.c b/net/bridge/br_forward.c
> index 4b81b14..65864bc 100644
> --- a/net/bridge/br_forward.c
> +++ b/net/bridge/br_forward.c
> @@ -49,6 +49,11 @@ int br_dev_queue_push_xmit(struct sk_buff *skb)
>  	} else {
>  		skb_push(skb, ETH_HLEN);
>  		br_drop_fake_rtable(skb);
> +
> +		/* clean the NF bridge data */
> +		nf_bridge_put(skb->nf_bridge);
> +		skb->nf_bridge = NULL;
> +
>  		dev_queue_xmit(skb);
>  	}
>  
You need to bracket this with CONFIG_BRIDGE_NETFILTER
since nf_bridge is only in skbuff if that option is enabled.

On Thu, 26 Sep 2013 22:19:50 +0200
Antonio Quartulli <antonio at meshcoding.com> wrote:
> From: Antonio Quartulli <antonio at open-mesh.com>
> 
> Even if enslaving a bridge interface into another bridge is
> forbidden, it is still possible to create a chain of
> virtual interfaces including two distinct bridges.
> 
> In this case, the skb entering the second bridge could have
> the nf_bridge field already set due to a previous operation
> and consequently lead to a wrong processing of the packet
> itself.
> 
> To prevent this behaviour release and set to NULL the
> nf_bridge field of the skb when exiting the bridge interface.
> 
> Signed-off-by: Antonio Quartulli <antonio at open-mesh.com>
> ---
> 
> I am not sure if this is a wanted behaviour or a real BUG. I found this
> "misbehaviour" while testing batman-adv with the following
configuration:
> 
> - br0 (bridge interface) having bat0 and eth0 as slaves
> - bat0 (which is a virtual interface provided by the batman-adv module and
that
>   works similarly to a bridge - to some extends) having br1 as slave
> - br1 (second bridge interface) having eth1 as slave
> 
> Then follow these events:
> - a broadcast packet arrives on eth0
> - the skb enters br0 and skb->nf_bridge gets initialised and used
> - the skb enters bat0 and the packet *gets encapsulated in the batman-adv
packet
>   which adds a batman-adv header and another Ethernet header*
> - the skb enters br1 and gets ruined because nf_bridge_maybe_copy_header()
(in
>   br_dev_queue_push_xmit()) will try to restore an header that does not
make
>   sense anymore.
> 
> With this patch the nf_bridge gets de-initialised before exiting br0 and
> therefore it is processed properly inside br1:
nf_bridge_maybe_copy_header()
> does not take place at all because nf_bridge is never initialised (the
packet is
> non-IP since it is a batman-adv packet)
> 
> To the developers of the bridge module I would like to ask:
> 1) is skb->nf_bridge allowed to be non NULL when entering br_dev_xmit()
? If so,
>    when is this supposed to happen?
> 
> 2) do you think this patch is logically correct but the nf_bridge release
should
>    be done in batman-adv since it is the one re-encapsulating the packet?
> 
> 
> I hope I have made the problem clear.
> 
> Best regards,
> 
> 
>  net/bridge/br_forward.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/net/bridge/br_forward.c b/net/bridge/br_forward.c
> index 4b81b14..65864bc 100644
> --- a/net/bridge/br_forward.c
> +++ b/net/bridge/br_forward.c
> @@ -49,6 +49,11 @@ int br_dev_queue_push_xmit(struct sk_buff *skb)
>  	} else {
>  		skb_push(skb, ETH_HLEN);
>  		br_drop_fake_rtable(skb);
> +
> +		/* clean the NF bridge data */
> +		nf_bridge_put(skb->nf_bridge);
> +		skb->nf_bridge = NULL;
> +
>  		dev_queue_xmit(skb);
>  	}
>  
You need to bracket this with CONFIG_BRIDGE_NETFILTER
since nf_bridge is only in skbuff if that option is enabled.

On Thu, 26 Sep 2013 22:19:50 +0200
Antonio Quartulli <antonio at meshcoding.com> wrote:
> From: Antonio Quartulli <antonio at open-mesh.com>
> 
> Even if enslaving a bridge interface into another bridge is
> forbidden, it is still possible to create a chain of
> virtual interfaces including two distinct bridges.
> 
> In this case, the skb entering the second bridge could have
> the nf_bridge field already set due to a previous operation
> and consequently lead to a wrong processing of the packet
> itself.
> 
> To prevent this behaviour release and set to NULL the
> nf_bridge field of the skb when exiting the bridge interface.
> 
> Signed-off-by: Antonio Quartulli <antonio at open-mesh.com>
> ---
> 
> I am not sure if this is a wanted behaviour or a real BUG. I found this
> "misbehaviour" while testing batman-adv with the following
configuration:
> 
> - br0 (bridge interface) having bat0 and eth0 as slaves
> - bat0 (which is a virtual interface provided by the batman-adv module and
that
>   works similarly to a bridge - to some extends) having br1 as slave
> - br1 (second bridge interface) having eth1 as slave
> 
> Then follow these events:
> - a broadcast packet arrives on eth0
> - the skb enters br0 and skb->nf_bridge gets initialised and used
> - the skb enters bat0 and the packet *gets encapsulated in the batman-adv
packet
>   which adds a batman-adv header and another Ethernet header*
> - the skb enters br1 and gets ruined because nf_bridge_maybe_copy_header()
(in
>   br_dev_queue_push_xmit()) will try to restore an header that does not
make
>   sense anymore.
> 
> With this patch the nf_bridge gets de-initialised before exiting br0 and
> therefore it is processed properly inside br1:
nf_bridge_maybe_copy_header()
> does not take place at all because nf_bridge is never initialised (the
packet is
> non-IP since it is a batman-adv packet)
> 
> To the developers of the bridge module I would like to ask:
> 1) is skb->nf_bridge allowed to be non NULL when entering br_dev_xmit()
? If so,
>    when is this supposed to happen?
> 
> 2) do you think this patch is logically correct but the nf_bridge release
should
>    be done in batman-adv since it is the one re-encapsulating the packet?
> 
> 
> I hope I have made the problem clear.
> 
> Best regards,
> 
> 
>  net/bridge/br_forward.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/net/bridge/br_forward.c b/net/bridge/br_forward.c
> index 4b81b14..65864bc 100644
> --- a/net/bridge/br_forward.c
> +++ b/net/bridge/br_forward.c
> @@ -49,6 +49,11 @@ int br_dev_queue_push_xmit(struct sk_buff *skb)
>  	} else {
>  		skb_push(skb, ETH_HLEN);
>  		br_drop_fake_rtable(skb);
> +
> +		/* clean the NF bridge data */
> +		nf_bridge_put(skb->nf_bridge);
> +		skb->nf_bridge = NULL;
> +
>  		dev_queue_xmit(skb);
>  	}
>  
I think the header will also be garbage if bridge on bridge with netfilter is
used.
See nf_bridge_save_header.