diff mbox

[v9,4/4] batman-adv: Forward IGMP/MLD reports to selected querier (only)

Message ID 1410068560-7829-5-git-send-email-linus.luessing@web.de
State Superseded, archived
Headers show

Commit Message

Linus Lüssing Sept. 7, 2014, 5:42 a.m. UTC 
With this patch IGMP or MLD reports are only forwarded to the selected
IGMP/MLD querier as RFC4541 suggests.

This is necessary to avoid multicast packet loss in bridged scenarios:
An IGMPv2/MLDv1 querier does not actively join the multicast group the
reports are sent to. Because of this, this leads to snooping
bridges/switches not being able to learn of multicast listeners in the
mesh and wrongly shutting down ports for multicast traffic to the mesh.

Signed-off-by: Linus Lüssing <linus.luessing@web.de>
---
 compat.h         |    7 +
 multicast.c      |  375 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 multicast.h      |    8 ++
 soft-interface.c |   11 ++
 types.h          |    4 +
 5 files changed, 399 insertions(+), 6 deletions(-)

Comments

Simon Wunderlich Nov. 24, 2014, 2:56 p.m. UTC | #1 
Hi Linus,

On Sunday 07 September 2014 07:42:40 Linus Lüssing wrote:
> With this patch IGMP or MLD reports are only forwarded to the selected
> IGMP/MLD querier as RFC4541 suggests.
> 
> This is necessary to avoid multicast packet loss in bridged scenarios:
> An IGMPv2/MLDv1 querier does not actively join the multicast group the
> reports are sent to. Because of this, this leads to snooping
> bridges/switches not being able to learn of multicast listeners in the
> mesh and wrongly shutting down ports for multicast traffic to the mesh.
> 
> Signed-off-by: Linus Lüssing <linus.luessing@web.de>
> ---
>  compat.h         |    7 +
>  multicast.c      |  375
> +++++++++++++++++++++++++++++++++++++++++++++++++++++- multicast.h      |  
>  8 ++
>  soft-interface.c |   11 ++
>  types.h          |    4 +
>  5 files changed, 399 insertions(+), 6 deletions(-)

As a general note: This patch adds ~400 lines for mainly processing IGMP/MLD 
stuff, right? Shouldn't this kind of IP/IGMP/MLD report detection and 
memorizing the Querier rather go in the bridge code instead of batman-adv, or 
is there any functionality there already? IMHO the preferred position would be 
the bridge code since its more general/used by more people on one hand, and 
does not bloat batman-adv on the other hand. :)

Could you give us some comments regarding who should do IGMP/MLD processing in 
the kernel so we are prepared for these kind of upstream questions?

Just some small notes regarding the code below:


> [...]
> --- a/multicast.c
> +++ b/multicast.c
> [...]
> @@ -512,10 +515,247 @@ out:
>  }
> 
>  /**
> + * batadv_mcast_is_valid_igmp - check for an IGMP packet
> + * @skb: the IPv4 packet to check
> + *
> + * Checks whether a given IPv4 packet is a valid IGMP packet and if so
> + * sets the skb transport header accordingly.
> + *
> + * The caller needs to ensure the correct setup of the skb network header.
> + * This call might reallocate skb data.
> + */
> +static bool batadv_mcast_is_valid_igmp(struct sk_buff *skb)
> +{
> +	struct iphdr *iphdr;
> +	unsigned int len = skb_network_offset(skb) + sizeof(*iphdr);
> +
> +	if (!pskb_may_pull(skb, len))
> +		return false;
> +
> +	iphdr = ip_hdr(skb);
> +
> +	if (iphdr->ihl < 5 || iphdr->version != 4)
> +		return false;
> +
> +	if (ntohs(iphdr->tot_len) < ip_hdrlen(skb))
> +		return false;
> +
> +	len += ip_hdrlen(skb) - sizeof(*iphdr);
> +	skb_set_transport_header(skb, len);
> +
> +	if (iphdr->protocol != IPPROTO_IGMP)
> +		return false;
> +
> +	/* TODO: verify checksums */
> +
> +	if (!pskb_may_pull(skb, len + sizeof(struct igmphdr)))
> +		return false;
> +
> +	/* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
> +	 * all-systems destination addresses (224.0.0.1) for general queries
> +	 */
> +	if (igmp_hdr(skb)->type == IGMP_HOST_MEMBERSHIP_QUERY &&
> +	    !igmp_hdr(skb)->group &&
> +	    ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
> +		return false;
> +
> +	return true;
> +}
> +
> +/**
> + * batadv_mcast_is_report_ipv4 - check for IGMP reports (and queries)
> + * @bat_priv: the bat priv with all the soft interface information
> + * @skb: the ethernet frame destined for the mesh
> + * @orig: if IGMP report: to be set to the querier to forward the skb to
> + *
> + * Checks whether the given frame is an IGMP report and if so sets the
> + * orig pointer to the originator of the selected IGMP querier if one
> exists + * and returns true. Otherwise returns false.
> + *
> + * If the packet is a general IGMP query instead then we delete the
> memorized, + * foreign IGMP querier (if one exists): We became the selected
> querier and + * therefore do not need to forward reports into the mesh.
> + *
> + * This call might reallocate skb data.
> + */
> +static bool batadv_mcast_is_report_ipv4(struct batadv_priv *bat_priv,
> +					struct sk_buff *skb,
> +					struct batadv_orig_node **orig)
> +{
> +	struct batadv_mcast_querier_state *querier;
> +	struct batadv_orig_node *orig_node;
> +
> +	if (!batadv_mcast_is_valid_igmp(skb))
> +		return false;
> +
> +	querier = &bat_priv->mcast.querier_ipv4;
> +
> +	switch (igmp_hdr(skb)->type) {
> +	case IGMP_HOST_MEMBERSHIP_REPORT:
> +	case IGMPV2_HOST_MEMBERSHIP_REPORT:
> +	case IGMPV3_HOST_MEMBERSHIP_REPORT:
> +		rcu_read_lock();
> +		orig_node = rcu_dereference(querier->orig);
> +		if (orig_node && atomic_inc_not_zero(&orig_node->refcount)) {
> +			/* TODO: include multicast routers via MRD (RFC4286) */
> +			batadv_dbg(BATADV_DBG_MCAST, bat_priv,
> +				   "Redirecting IGMP Report to %pM\n",
> +				   orig_node->orig);
> +			*orig = orig_node;
> +		}
> +		rcu_read_unlock();
> +
> +		return true;
> +	case IGMP_HOST_MEMBERSHIP_QUERY:
> +		/* RFC4541, section 2.1.1.1.b) says:
> +		 * ignore general queries from 0.0.0.0
> +		 */
> +		if (!ip_hdr(skb)->saddr || igmp_hdr(skb)->group)
> +			break;

Why do we break here if group != 0? This looks wrong, maybe you meant && or 
group != 0?

> [...]
> diff --git a/soft-interface.c b/soft-interface.c
> index d4d892c..0f5fb30 100644
> --- a/soft-interface.c
> +++ b/soft-interface.c
> @@ -183,6 +183,7 @@ static int batadv_interface_tx(struct sk_buff *skb,
>  	if (atomic_read(&bat_priv->mesh_state) != BATADV_MESH_ACTIVE)
>  		goto dropped;
> 
> +	skb_set_network_header(skb, ETH_HLEN);
>  	soft_iface->trans_start = jiffies;
>  	vid = batadv_get_vid(skb, 0);
>  	ethhdr = eth_hdr(skb);
> @@ -397,7 +398,9 @@ void batadv_interface_rx(struct net_device *soft_iface,
>  	/* skb->dev & skb->pkt_type are set here */
>  	if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
>  		goto dropped;
> +
>  	skb->protocol = eth_type_trans(skb, soft_iface);
> +	skb_reset_network_header(skb);

Why do we need that header stuff here? And there is an extra newline too ... If 
you need that for the multicast stuff better move it there, otherwise it is set 
for every packet ...

Cheers,
   Simon
Linus Lüssing Nov. 26, 2014, 4:37 p.m. UTC | #2 
On Mon, Nov 24, 2014 at 03:56:06PM +0100, Simon Wunderlich wrote:
> As a general note: This patch adds ~400 lines for mainly processing IGMP/MLD 
> stuff, right? Shouldn't this kind of IP/IGMP/MLD report detection and 
> memorizing the Querier rather go in the bridge code instead of batman-adv, or 
> is there any functionality there already? IMHO the preferred position would be 
> the bridge code since its more general/used by more people on one hand, and 
> does not bloat batman-adv on the other hand. :)
> 
> Could you give us some comments regarding who should do IGMP/MLD processing in 
> the kernel so we are prepared for these kind of upstream questions?

Currently, with this patch, the decisions about where to forward IGMP/MLD
reports in batman-adv is done on the multicast listener / report sender side.

With this approach, as seen in the picture, even a node without
the bridge (the blue node 'L') needs to snoop reports. It can't
rely on a bridge.


On the other hand, the current upstream implementation without
bridge support does not (necessarily) need it (will add an
explanation to the wiki page).

Though there is no (does not seem to be) any conceptual issue with
either the current upstream implementation or this proposition as is,
you just made me realize that there are issues with mixed setups...
backwards compatibility issues... For instance if node 'L' were of
a post-multicast but pre-multicast-bridge optimizations version.
diff mbox

Patch

diff --git a/compat.h b/compat.h
index 9f997b3..f3e7032 100644
--- a/compat.h
+++ b/compat.h
@@ -255,6 +255,13 @@  static inline void skb_reset_mac_len(struct sk_buff *skb)
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
 
+static inline int batadv_ipv6_skip_exthdr(const struct sk_buff *skb, int start,
+					  u8 *nexthdrp, __be16 *frag_offp)
+{
+	return ipv6_skip_exthdr(skb, start, nexthdrp);
+}
+#define ipv6_skip_exthdr batadv_ipv6_skip_exthdr
+
 #define batadv_interface_add_vid(x, y, z) \
 __batadv_interface_add_vid(struct net_device *dev, __be16 proto,\
                           unsigned short vid);\
diff --git a/multicast.c b/multicast.c
index 09b3e6f..24cbf07 100644
--- a/multicast.c
+++ b/multicast.c
@@ -21,6 +21,9 @@ 
 #include "hard-interface.h"
 #include "translation-table.h"
 
+#include <linux/igmp.h>
+#include <net/mld.h>
+
 /**
  * batadv_mcast_get_bridge - get the bridge on top of the softif if it exists
  * @soft_iface: netdev struct of the mesh interface
@@ -512,10 +515,247 @@  out:
 }
 
 /**
+ * batadv_mcast_is_valid_igmp - check for an IGMP packet
+ * @skb: the IPv4 packet to check
+ *
+ * Checks whether a given IPv4 packet is a valid IGMP packet and if so
+ * sets the skb transport header accordingly.
+ *
+ * The caller needs to ensure the correct setup of the skb network header.
+ * This call might reallocate skb data.
+ */
+static bool batadv_mcast_is_valid_igmp(struct sk_buff *skb)
+{
+	struct iphdr *iphdr;
+	unsigned int len = skb_network_offset(skb) + sizeof(*iphdr);
+
+	if (!pskb_may_pull(skb, len))
+		return false;
+
+	iphdr = ip_hdr(skb);
+
+	if (iphdr->ihl < 5 || iphdr->version != 4)
+		return false;
+
+	if (ntohs(iphdr->tot_len) < ip_hdrlen(skb))
+		return false;
+
+	len += ip_hdrlen(skb) - sizeof(*iphdr);
+	skb_set_transport_header(skb, len);
+
+	if (iphdr->protocol != IPPROTO_IGMP)
+		return false;
+
+	/* TODO: verify checksums */
+
+	if (!pskb_may_pull(skb, len + sizeof(struct igmphdr)))
+		return false;
+
+	/* RFC2236+RFC3376 (IGMPv2+IGMPv3) require the multicast link layer
+	 * all-systems destination addresses (224.0.0.1) for general queries
+	 */
+	if (igmp_hdr(skb)->type == IGMP_HOST_MEMBERSHIP_QUERY &&
+	    !igmp_hdr(skb)->group &&
+	    ip_hdr(skb)->daddr != htonl(INADDR_ALLHOSTS_GROUP))
+		return false;
+
+	return true;
+}
+
+/**
+ * batadv_mcast_is_report_ipv4 - check for IGMP reports (and queries)
+ * @bat_priv: the bat priv with all the soft interface information
+ * @skb: the ethernet frame destined for the mesh
+ * @orig: if IGMP report: to be set to the querier to forward the skb to
+ *
+ * Checks whether the given frame is an IGMP report and if so sets the
+ * orig pointer to the originator of the selected IGMP querier if one exists
+ * and returns true. Otherwise returns false.
+ *
+ * If the packet is a general IGMP query instead then we delete the memorized,
+ * foreign IGMP querier (if one exists): We became the selected querier and
+ * therefore do not need to forward reports into the mesh.
+ *
+ * This call might reallocate skb data.
+ */
+static bool batadv_mcast_is_report_ipv4(struct batadv_priv *bat_priv,
+					struct sk_buff *skb,
+					struct batadv_orig_node **orig)
+{
+	struct batadv_mcast_querier_state *querier;
+	struct batadv_orig_node *orig_node;
+
+	if (!batadv_mcast_is_valid_igmp(skb))
+		return false;
+
+	querier = &bat_priv->mcast.querier_ipv4;
+
+	switch (igmp_hdr(skb)->type) {
+	case IGMP_HOST_MEMBERSHIP_REPORT:
+	case IGMPV2_HOST_MEMBERSHIP_REPORT:
+	case IGMPV3_HOST_MEMBERSHIP_REPORT:
+		rcu_read_lock();
+		orig_node = rcu_dereference(querier->orig);
+		if (orig_node && atomic_inc_not_zero(&orig_node->refcount)) {
+			/* TODO: include multicast routers via MRD (RFC4286) */
+			batadv_dbg(BATADV_DBG_MCAST, bat_priv,
+				   "Redirecting IGMP Report to %pM\n",
+				   orig_node->orig);
+			*orig = orig_node;
+		}
+		rcu_read_unlock();
+
+		return true;
+	case IGMP_HOST_MEMBERSHIP_QUERY:
+		/* RFC4541, section 2.1.1.1.b) says:
+		 * ignore general queries from 0.0.0.0
+		 */
+		if (!ip_hdr(skb)->saddr || igmp_hdr(skb)->group)
+			break;
+
+		spin_lock_bh(&querier->orig_lock);
+		orig_node = rcu_dereference(querier->orig);
+		if (orig_node)
+			rcu_assign_pointer(querier->orig, NULL);
+		spin_unlock_bh(&querier->orig_lock);
+
+		batadv_dbg(BATADV_DBG_MCAST, bat_priv,
+			   "Snooped own IGMP Query\n");
+		break;
+	}
+
+	return false;
+}
+
+/**
+ * batadv_mcast_is_valid_mld - check for an MLD packet
+ * @skb: the IPv6 packet to check
+ *
+ * Checks whether a given IPv6 packet is a valid MLD packet and if so
+ * sets the skb transport header accordingly.
+ *
+ * The caller needs to ensure the correct setup of the skb network header.
+ * This call might reallocate skb data.
+ */
+static bool batadv_mcast_is_valid_mld(struct sk_buff *skb)
+{
+	const struct ipv6hdr *ip6hdr;
+	struct mld_msg *mld;
+	__be16 frag_off;
+	u8 nexthdr;
+	unsigned int len = skb_network_offset(skb) + sizeof(*ip6hdr);
+
+	if (!pskb_may_pull(skb, len))
+		return false;
+
+	ip6hdr = ipv6_hdr(skb);
+
+	if (ip6hdr->version != 6 || ip6hdr->payload_len == 0)
+		return false;
+
+	if (skb->len < len + ntohs(ip6hdr->payload_len))
+		return false;
+
+	nexthdr = ip6hdr->nexthdr;
+	len = ipv6_skip_exthdr(skb, len, &nexthdr, &frag_off);
+
+	if (len < 0)
+		return false;
+
+	skb_set_transport_header(skb, len);
+
+	/* TODO: verify checksums */
+
+	if (nexthdr != IPPROTO_ICMPV6)
+		return false;
+
+	if (!pskb_may_pull(skb, len + sizeof(*mld)))
+		return false;
+
+	mld = (struct mld_msg *)icmp6_hdr(skb);
+
+	if (mld->mld_type == ICMPV6_MGM_QUERY) {
+		/* RFC2710+RFC3810 (MLDv1+MLDv2): link-local source addresses */
+		if (!(ipv6_addr_type(&ip6hdr->saddr) & IPV6_ADDR_LINKLOCAL))
+			return false;
+
+		/* RFC2710+RFC3810 (MLDv1+MLDv2): multicast link layer all-nodes
+		 * destination address (ff02::1) for general queries
+		 */
+		if (ipv6_addr_any(&mld->mld_mca) &&
+		    !ipv6_addr_is_ll_all_nodes(&ip6hdr->daddr))
+			return false;
+	}
+
+	return true;
+}
+
+/**
+ * batadv_mcast_is_report_ipv6 - check for MLD reports (and queries)
+ * @bat_priv: the bat priv with all the soft interface information
+ * @skb: the ethernet frame destined for the mesh
+ * @orig: if MLD report: to be set to the querier to forward the skb to
+ *
+ * Checks whether the given frame is an MLD report and if so sets the
+ * orig pointer to the originator of the selected MLD querier if one exists
+ * and returns true. Otherwise returns false.
+ *
+ * If the packet is a general MLD query instead then we delete the memorized,
+ * foreign MLD querier (if one exists): We became the selected querier and
+ * therefore do not need to forward reports into the mesh.
+ *
+ * This call might reallocate skb data.
+ */
+static bool batadv_mcast_is_report_ipv6(struct batadv_priv *bat_priv,
+					struct sk_buff *skb,
+					struct batadv_orig_node **orig)
+{
+	struct mld_msg *mld;
+	struct batadv_mcast_querier_state *querier;
+	struct batadv_orig_node *orig_node;
+
+	if (!batadv_mcast_is_valid_mld(skb))
+		return false;
+
+	querier = &bat_priv->mcast.querier_ipv6;
+	mld = (struct mld_msg *)icmp6_hdr(skb);
+
+	switch (mld->mld_type) {
+	case ICMPV6_MGM_REPORT:
+	case ICMPV6_MLD2_REPORT:
+		rcu_read_lock();
+		orig_node = rcu_dereference(querier->orig);
+		if (orig_node && atomic_inc_not_zero(&orig_node->refcount)) {
+			/* TODO: include multicast routers via MRD (RFC4286) */
+			batadv_dbg(BATADV_DBG_MCAST, bat_priv,
+				   "Redirecting MLD Report to %pM\n",
+				   orig_node->orig);
+			*orig = orig_node;
+		}
+		rcu_read_unlock();
+
+		return true;
+	case ICMPV6_MGM_QUERY:
+		spin_lock_bh(&querier->orig_lock);
+		orig_node = rcu_dereference(querier->orig);
+		if (orig_node)
+			rcu_assign_pointer(querier->orig, NULL);
+		spin_unlock_bh(&querier->orig_lock);
+
+		batadv_dbg(BATADV_DBG_MCAST, bat_priv,
+			   "Snooped own MLD Query\n");
+		break;
+	}
+
+	return false;
+}
+
+/**
  * batadv_mcast_forw_mode_check_ipv4 - check for optimized forwarding potential
  * @bat_priv: the bat priv with all the soft interface information
  * @skb: the IPv4 packet to check
  * @is_unsnoopable: stores whether the destination is snoopable
+ * @orig: for IGMP reports: to be set to the querier to forward the skb to
  *
  * Checks whether the given IPv4 packet has the potential to be forwarded with a
  * mode more optimal than classic flooding.
@@ -525,7 +765,8 @@  out:
  */
 static int batadv_mcast_forw_mode_check_ipv4(struct batadv_priv *bat_priv,
 					     struct sk_buff *skb,
-					     bool *is_unsnoopable)
+					     bool *is_unsnoopable,
+					     struct batadv_orig_node **orig)
 {
 	struct iphdr *iphdr;
 
@@ -533,6 +774,9 @@  static int batadv_mcast_forw_mode_check_ipv4(struct batadv_priv *bat_priv,
 	if (!pskb_may_pull(skb, sizeof(struct ethhdr) + sizeof(*iphdr)))
 		return -ENOMEM;
 
+	if (batadv_mcast_is_report_ipv4(bat_priv, skb, orig))
+		return 0;
+
 	iphdr = ip_hdr(skb);
 
 	/* TODO: Implement Multicast Router Discovery (RFC4286),
@@ -554,6 +798,7 @@  static int batadv_mcast_forw_mode_check_ipv4(struct batadv_priv *bat_priv,
  * @bat_priv: the bat priv with all the soft interface information
  * @skb: the IPv6 packet to check
  * @is_unsnoopable: stores whether the destination is snoopable
+ * @orig: for MLD reports: to be set to the querier to forward the skb to
  *
  * Checks whether the given IPv6 packet has the potential to be forwarded with a
  * mode more optimal than classic flooding.
@@ -563,7 +808,8 @@  static int batadv_mcast_forw_mode_check_ipv4(struct batadv_priv *bat_priv,
  */
 static int batadv_mcast_forw_mode_check_ipv6(struct batadv_priv *bat_priv,
 					     struct sk_buff *skb,
-					     bool *is_unsnoopable)
+					     bool *is_unsnoopable,
+					     struct batadv_orig_node **orig)
 {
 	struct ipv6hdr *ip6hdr;
 
@@ -571,6 +817,9 @@  static int batadv_mcast_forw_mode_check_ipv6(struct batadv_priv *bat_priv,
 	if (!pskb_may_pull(skb, sizeof(struct ethhdr) + sizeof(*ip6hdr)))
 		return -ENOMEM;
 
+	if (batadv_mcast_is_report_ipv6(bat_priv, skb, orig))
+		return 0;
+
 	ip6hdr = ipv6_hdr(skb);
 
 	/* TODO: Implement Multicast Router Discovery (RFC4286),
@@ -593,6 +842,7 @@  static int batadv_mcast_forw_mode_check_ipv6(struct batadv_priv *bat_priv,
  * @bat_priv: the bat priv with all the soft interface information
  * @skb: the multicast frame to check
  * @is_unsnoopable: stores whether the destination is snoopable
+ * @orig: for IGMP/MLD reports: to be set to the querier to forward the skb to
  *
  * Checks whether the given multicast ethernet frame has the potential to be
  * forwarded with a mode more optimal than classic flooding.
@@ -602,7 +852,8 @@  static int batadv_mcast_forw_mode_check_ipv6(struct batadv_priv *bat_priv,
  */
 static int batadv_mcast_forw_mode_check(struct batadv_priv *bat_priv,
 					struct sk_buff *skb,
-					bool *is_unsnoopable)
+					bool *is_unsnoopable,
+					struct batadv_orig_node **orig)
 {
 	struct ethhdr *ethhdr = eth_hdr(skb);
 
@@ -615,10 +866,10 @@  static int batadv_mcast_forw_mode_check(struct batadv_priv *bat_priv,
 	switch (ntohs(ethhdr->h_proto)) {
 	case ETH_P_IP:
 		return batadv_mcast_forw_mode_check_ipv4(bat_priv, skb,
-							 is_unsnoopable);
+							 is_unsnoopable, orig);
 	case ETH_P_IPV6:
 		return batadv_mcast_forw_mode_check_ipv6(bat_priv, skb,
-							 is_unsnoopable);
+							 is_unsnoopable, orig);
 	default:
 		return -EINVAL;
 	}
@@ -786,12 +1037,16 @@  batadv_mcast_forw_mode(struct batadv_priv *bat_priv, struct sk_buff *skb,
 	bool is_unsnoopable = false;
 	struct ethhdr *ethhdr;
 
-	ret = batadv_mcast_forw_mode_check(bat_priv, skb, &is_unsnoopable);
+	ret = batadv_mcast_forw_mode_check(bat_priv, skb, &is_unsnoopable,
+					   orig);
 	if (ret == -ENOMEM)
 		return BATADV_FORW_NONE;
 	else if (ret < 0)
 		return BATADV_FORW_ALL;
 
+	if (*orig)
+		return BATADV_FORW_SINGLE;
+
 	ethhdr = eth_hdr(skb);
 
 	tt_count = batadv_tt_global_hash_count(bat_priv, ethhdr->h_dest,
@@ -823,6 +1078,101 @@  batadv_mcast_forw_mode(struct batadv_priv *bat_priv, struct sk_buff *skb,
 }
 
 /**
+ * batadv_mcast_snoop_query_ipv4 - snoop for the selected MLD querier
+ * @skb: the unencapsulated ethernet frame coming from the mesh
+ * @orig_node: the originator this frame came from
+ *
+ * Checks whether the given frame is a general IGMP query from the selected
+ * querier and if so memorizes the originator this frame came from.
+ */
+static void batadv_mcast_snoop_query_ipv4(struct sk_buff *skb,
+					  struct batadv_orig_node *orig_node)
+{
+	struct batadv_mcast_querier_state *querier;
+
+	if (!batadv_mcast_is_valid_igmp(skb))
+		return;
+
+	/* we are only interested in general queries (group == 0.0.0.0) */
+	if (igmp_hdr(skb)->type != IGMP_HOST_MEMBERSHIP_QUERY ||
+	    igmp_hdr(skb)->group)
+		return;
+
+	/* RFC4541, section 2.1.1.1.b) says: ignore queries from 0.0.0.0 */
+	if (!ip_hdr(skb)->saddr)
+		return;
+
+	querier = &orig_node->bat_priv->mcast.querier_ipv4;
+
+	spin_lock_bh(&querier->orig_lock);
+	rcu_assign_pointer(querier->orig, orig_node);
+	spin_unlock_bh(&querier->orig_lock);
+
+	batadv_dbg(BATADV_DBG_MCAST, orig_node->bat_priv,
+		   "Snooped IGMP Query from originator %pM\n", orig_node->orig);
+}
+
+/**
+ * batadv_mcast_snoop_query_ipv6 - snoop for the selected MLD querier
+ * @skb: the unencapsulated ethernet frame coming from the mesh
+ * @orig_node: the originator this frame came from
+ *
+ * Checks whether the given frame is a general MLD query from the selected
+ * querier and if so memorizes the originator this frame came from.
+ */
+static void batadv_mcast_snoop_query_ipv6(struct sk_buff *skb,
+					  struct batadv_orig_node *orig_node)
+{
+	struct mld_msg *mld;
+	struct batadv_mcast_querier_state *querier;
+
+	if (!batadv_mcast_is_valid_mld(skb))
+		return;
+
+	mld = (struct mld_msg *)icmp6_hdr(skb);
+
+	/* we are only interested in general queries (mca == ::) */
+	if (mld->mld_type != ICMPV6_MGM_QUERY ||
+	    !ipv6_addr_any(&mld->mld_mca))
+		return;
+
+	querier = &orig_node->bat_priv->mcast.querier_ipv6;
+
+	spin_lock_bh(&querier->orig_lock);
+	rcu_assign_pointer(querier->orig, orig_node);
+	spin_unlock_bh(&querier->orig_lock);
+
+	batadv_dbg(BATADV_DBG_MCAST, orig_node->bat_priv,
+		   "Snooped MLD Query from originator %pM\n", orig_node->orig);
+}
+
+/**
+ * batadv_mcast_snoop_query - snoop the selected IGMP/MLD querier
+ * @skb: the unencapsulated ethernet frame coming from the mesh
+ * @orig_node: the originator this frame came from
+ *
+ * Checks whether the given frame is a general IGMP or MLD query
+ * from the selected querier and if so memorizes the originator
+ * this frame came from.
+ *
+ * This call might reallocate skb data.
+ */
+void batadv_mcast_snoop_query(struct sk_buff *skb,
+			      struct batadv_orig_node *orig_node)
+{
+	struct ethhdr *ethhdr = eth_hdr(skb);
+
+	switch (ntohs(ethhdr->h_proto)) {
+	case ETH_P_IP:
+		batadv_mcast_snoop_query_ipv4(skb, orig_node);
+		break;
+	case ETH_P_IPV6:
+		batadv_mcast_snoop_query_ipv6(skb, orig_node);
+		break;
+	}
+}
+
+/**
  * batadv_mcast_want_unsnoop_update - update unsnoop counter and list
  * @bat_priv: the bat priv with all the soft interface information
  * @orig: the orig_node which multicast state might have changed of
@@ -1101,6 +1451,7 @@  void batadv_mcast_free(struct batadv_priv *bat_priv)
 void batadv_mcast_purge_orig(struct batadv_orig_node *orig)
 {
 	struct batadv_priv *bat_priv = orig->bat_priv;
+	struct batadv_mcast_querier_state *querier;
 
 	if (!(orig->capabilities & BATADV_ORIG_CAPA_HAS_MCAST))
 		atomic_dec(&bat_priv->mcast.num_disabled);
@@ -1108,4 +1459,16 @@  void batadv_mcast_purge_orig(struct batadv_orig_node *orig)
 	batadv_mcast_want_unsnoop_update(bat_priv, orig, BATADV_NO_FLAGS);
 	batadv_mcast_want_ipv4_update(bat_priv, orig, BATADV_NO_FLAGS);
 	batadv_mcast_want_ipv6_update(bat_priv, orig, BATADV_NO_FLAGS);
+
+	querier = &bat_priv->mcast.querier_ipv4;
+	spin_lock_bh(&querier->orig_lock);
+	if (rcu_dereference(querier->orig) == orig)
+		rcu_assign_pointer(querier->orig, NULL);
+	spin_unlock_bh(&querier->orig_lock);
+
+	querier = &bat_priv->mcast.querier_ipv6;
+	spin_lock_bh(&querier->orig_lock);
+	if (rcu_dereference(querier->orig) == orig)
+		rcu_assign_pointer(querier->orig, NULL);
+	spin_unlock_bh(&querier->orig_lock);
 }
diff --git a/multicast.h b/multicast.h
index e7bc9bf..fa8e7be 100644
--- a/multicast.h
+++ b/multicast.h
@@ -40,6 +40,9 @@  enum batadv_forw_mode
 batadv_mcast_forw_mode(struct batadv_priv *bat_priv, struct sk_buff *skb,
 		       struct batadv_orig_node **mcast_single_orig);
 
+void batadv_mcast_snoop_query(struct sk_buff *skb,
+			      struct batadv_orig_node *orig_node);
+
 void batadv_mcast_init(struct batadv_priv *bat_priv);
 
 int batadv_mcast_flags_seq_print_text(struct seq_file *seq, void *offset);
@@ -61,6 +64,11 @@  batadv_mcast_forw_mode(struct batadv_priv *bat_priv, struct sk_buff *skb,
 	return BATADV_FORW_ALL;
 }
 
+static inline void batadv_mcast_snoop_query(struct sk_buff *skb,
+					    struct batadv_orig_node *orig_node)
+{
+}
+
 static inline int batadv_mcast_init(struct batadv_priv *bat_priv)
 {
 	return 0;
diff --git a/soft-interface.c b/soft-interface.c
index d4d892c..0f5fb30 100644
--- a/soft-interface.c
+++ b/soft-interface.c
@@ -183,6 +183,7 @@  static int batadv_interface_tx(struct sk_buff *skb,
 	if (atomic_read(&bat_priv->mesh_state) != BATADV_MESH_ACTIVE)
 		goto dropped;
 
+	skb_set_network_header(skb, ETH_HLEN);
 	soft_iface->trans_start = jiffies;
 	vid = batadv_get_vid(skb, 0);
 	ethhdr = eth_hdr(skb);
@@ -397,7 +398,9 @@  void batadv_interface_rx(struct net_device *soft_iface,
 	/* skb->dev & skb->pkt_type are set here */
 	if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
 		goto dropped;
+
 	skb->protocol = eth_type_trans(skb, soft_iface);
+	skb_reset_network_header(skb);
 
 	/* should not be necessary anymore as we use skb_pull_rcsum()
 	 * TODO: please verify this and remove this TODO
@@ -435,6 +438,9 @@  void batadv_interface_rx(struct net_device *soft_iface,
 			skb->mark &= ~bat_priv->isolation_mark_mask;
 			skb->mark |= bat_priv->isolation_mark;
 		}
+
+		if (orig_node)
+			batadv_mcast_snoop_query(skb, orig_node);
 	} else if (batadv_is_ap_isolated(bat_priv, ethhdr->h_source,
 					 ethhdr->h_dest, vid)) {
 		goto dropped;
@@ -747,8 +753,13 @@  static int batadv_softif_init_late(struct net_device *dev)
 #ifdef CONFIG_BATMAN_ADV_MCAST
 	bat_priv->mcast.querier_ipv4.exists = false;
 	bat_priv->mcast.querier_ipv4.shadowing = false;
+	rcu_assign_pointer(bat_priv->mcast.querier_ipv4.orig, NULL);
+	spin_lock_init(&bat_priv->mcast.querier_ipv4.orig_lock);
 	bat_priv->mcast.querier_ipv6.exists = false;
 	bat_priv->mcast.querier_ipv6.shadowing = false;
+	rcu_assign_pointer(bat_priv->mcast.querier_ipv6.orig, NULL);
+	spin_lock_init(&bat_priv->mcast.querier_ipv6.orig_lock);
+
 	bat_priv->mcast.flags = BATADV_NO_FLAGS;
 	bat_priv->mcast.bridged = false;
 	atomic_set(&bat_priv->multicast_mode, 1);
diff --git a/types.h b/types.h
index bd56fc1..0a26fe9 100644
--- a/types.h
+++ b/types.h
@@ -627,10 +627,14 @@  struct batadv_priv_dat {
  * @exists: whether a querier exists in the mesh
  * @shadowing: if a querier exists, whether it is potentially shadowing
  *  multicast listeners (i.e. querier is behind our own bridge segment)
+ * @orig: node on which the selected querier resides
+ * @orig_lock: protects updates of the selected querier in 'orig'
  */
 struct batadv_mcast_querier_state {
 	bool exists;
 	bool shadowing;
+	struct batadv_orig_node __rcu *orig; /* rcu protected pointer */
+	spinlock_t orig_lock; /* protects updates of orig */
 };
 
 /**