Pranjal Kumar Dutta wrote: > Hi, > I have some problems in understanding some parts of the > (802.1q)vlan module in linux kernel and its interaction with linux > bridging code. > > I have the following problems in understanding the codes. > > 1. In linux 802.1q code a vlan is characterised as a generic > net_device. Every real net_device (Ethernet NICs e.g) has a vlan_group > attached to it to identify the port's membership of vlans. A > vlan_group contains an array of pointers to all vlan_devices (in the > form of net_devices) to which the real device is member of. As per my > understanding goes here whenever a vlan is attached to an interface > the following is called. > > static struct net_device *register_vlan_device(const char *eth_IF_name, > > unsigned short VLAN_ID) > Inside this function I could see that a new net_device is created for > the corresponding vlan attched to it. So that means a VLAN domain in > the box is NOT represented by a single net_device, rather it is > interface specific. For the same vlan X every interface will be having > its own net_device for X in its vlan_group. Then how the vlan layer > glues all the vlan_devices > (in the form of net_devices) for a particular vlan X into a domain? > Because theoretically speaking while bridging ethernet flows MAC > Learning and forwarding should be based on vlan domains. > I'm not sure I understand your question fully, but I think this info may be helpful: When 8021q.o is isnmodded it calls dev_add_pack(&vlan_packet_type); which registers the function vlan_skb_recv() as a handler for the vlan dix type, namely 0x8100. The code in net/core/dev.c which dequeues frames - queued by device drivers using netif_rx() - uses a hash table to find the handler for each frame by dix type. After 8021q.o is insmodded net/core/dev.c will send all frames with dix type 0x8100 to vlan_skb_recv(). vlan_skb_recv() checks to see if the originating device (e.g. eth0) has a vlan device with the correct vid on top of it (e.g. eth0.10). If it has then it passes the frame up through this device (i.e. changes skb->dev, rearranges the header if neccessary, and calls netif_rx()). If not, it frees the skb and returns a failure code. Alex
