Differences

This shows you the differences between two versions of the page.

--- igmp_snooping [2009/04/26 18:00] – alex
+++ igmp_snooping [2012/10/17 11:40] (current) – claudiu
@@ Line 28: / Line 28: @@
   * if igmp snooping is disabled, multicast traffic is flooded to all ports in vlan, regardless of frame nature (i.e. it is igmp or not); static multicast fdb entries are ignored;
   * mrouters can be statically configured while igmp snooping is disabled on vlan, but they are ignored until igmp snooping is enabled; when igmp snooping is enabled, the rules listed above apply;
+  * mrouters are not activated until: interface is up AND interface has access to mrouter vlan (either access mode vlan or trunk allowed vlan)
+When changing access vlan, with mrouter already configured (in new vlan):
+<code>
+w0d: IGMPSN: mgt: Vlan 15, port Fa0/5 port_id 7 in non-fwd mode
+w0d: IGMPSN: mgt: received pmvlan_linkchange down event on vlan 15
+w0d: IGMPSN: Adding router port Fa0/5 to all GCEs in Vlan 18
+w0d: L2MM: added rport Fa0/5 on Vlan 18
+w0d: %SYS-5-CONFIG_I: Configured from console by vty0 (10.0.0.2)
+</code>
+====== Implementation Approach ======
+Analyzing the forwarding cases, it turns out that when incoming packets are forwarded to more than one port at the same time, they are forwarded to either all ports in a specific igmp group or all mrouter ports (but not a mix of the two). In other words, depending on the igmp snooping logic, either an igmp group or a list of mrouter ports is sufficient to determine which ports to forward to.
+===== Forwarding to ports in an igmp group =====
+  * igmp group membership is stored as special fdb entry
+  * is_static field in struct net_switch_fdb_entry is change to a bitwise field:
+    * b0 tells if entry is dynamic or static
+    * b1 tells if entry is regular fdb entry or igmp membership entry
+    * for igmp membership entries, a b0 value corresponding to "static" means user-configured group membership
+  * all igmp membership entries have a virtual MAC address formed as 01-00 plus the group address (4 bytes)
+  * when flooding to all ports in a group, the following actions are taken:
+    * first the virtual mac corresponding to the packet destination is build;
+    * the fdb bucket corresponding to the virtual mac is determined;
+    * the fdb bucket is walked to determine the ports to forward to:
+      * virtual mac must be equal (between fdb entry and packet)
+      * vlan must be equal (between fdb entry and packet)
+  * igmp group listing is done only on the slow path (upon user request) by walking the entire fdb and building the igmp group list:
+    * the kernel module is asked for a list of all igmp membership fdb entries (with the ability to filter by a specific group and/or vlan);
+    * userspace code walks the fdb entry list and builds a (temporary) list of groups and, for each group, a list of member ports
+The following changes need to be made to existing code:
+  * fdb queries from userspace need to filter out igmp membership entries;
+  * standard (non-multicast) packet fdb lookup needs to ignore igmp membership entries;
+===== Forwarding to mrouter ports =====
+Whether a port is mrouter or not is a function of the port and vlan. In other words, mrouter can be stored as a flag (bit) depending on the port and vlan. Since vlan number is limited and relatively small (4096), mrouter information can be stored as a per-port vlan-indexed bitmap. The data structure would be similar to vlan membership for trunk ports.
+Flooding to all mrouter ports in a specific vlan becomes very simple:
+  * use the VDB to iterate through all ports in the required vlan;
+  * for each port check if it's mrouter in the required vlan;
+Algorithm can be easily cloned from sw_flood().
 ====== CLI Commands ======