L2 Layer Management¶
The L2 stack is designed to hide the whole networking link-layer part
and the related device drivers from the upper network stack. This is made
struct net_if declared in
The upper layers are unaware of implementation details beyond the net_if
object and the generic API provided by the L2 layer in
Only the L2 layer can talk to the device driver, linked to the net_if object. The L2 layer dictates the API provided by the device driver, specific for that device, and optimized for working together.
In order to create an L2 layer, or a driver for a specific L2 layer, one needs to understand how the L3 layer interacts with it and how the L2 layer is supposed to behave. See also network stack architecture for more details. The generic L2 API has these functions:
recv(): All device drivers, once they receive a packet which they put into a
struct net_pkt, will push this buffer to the network stack via
net_recv_data(). At this point, the network stack does not know what to do with it. Instead, it passes the buffer along to the L2 stack’s
recv()function for handling. The L2 stack does what it needs to do with the packet, for example, parsing the link layer header, or handling link-layer only packets. The
recv()function will return
NET_DROPin case of an erroneous packet,
NET_OKif the packet was fully consumed by the L2, or
NET_CONTINUEif the network stack should then handle it.
send(): Similar to receive function, the network stack will call this function to actually send a network packet. All relevant link-layer content will be generated and added by this function. The
send()function returns the number of bytes sent, or a negative error code if there was a failure sending the network packet.
enable(): This function is used to enable/disable traffic over a network interface. The function returns
<0if error and
>=0if no error.
get_flags(): This function will return the capabilities of an L2 driver, for example whether the L2 supports multicast or promiscuous mode.
Network device drivers fully follows Zephyr device driver model as a basis. Please refer to Device Driver Model.
There are, however, two differences:
- The driver_api pointer must point to a valid
- The network device driver must use
ETH_NET_DEVICE_INIT()for Ethernet devices. These macros will call the
DEVICE_AND_API_INIT()macro, and also instantiate a unique
struct net_ifrelated to the created device driver instance.
Implementing a network device driver depends on the L2 stack it belongs to: Ethernet, IEEE 802.15.4, etc. In the next section, we will describe how a device driver should behave when receiving or sending a network packet. The rest is hardware dependent and is not detailed here.
On reception, it is up to the device driver to fill-in the network packet with
as many data buffers as required. The network packet itself is a
struct net_pkt and should be allocated through
net_pkt_rx_alloc_with_buffer(). Then all data buffers will be
automatically allocated and filled by
After all the network data has been received, the device driver needs to
net_recv_data(). If that call fails, it will be up to the
device driver to unreference the buffer via
On sending, the device driver send function will be called, and it is up to the device driver to send the network packet all at once, with all the buffers.
Each Ethernet device driver will need, in the end, to call
ETH_NET_DEVICE_INIT() like this:
ETH_NET_DEVICE_INIT(..., CONFIG_ETH_INIT_PRIORITY, &the_valid_net_if_api_instance, 1500);
Device drivers for IEEE 802.15.4 L2 work basically the same as for
Ethernet. What has been described above, especially for
here as well. There are two specific differences however:
- It requires a dedicated device driver API:
struct ieee802154_radio_api, which overloads
struct net_if_api. This is because 802.15.4 L2 needs more from the device driver than just
recv()functions. This dedicated API is declared in include/net/ieee802154_radio.h. Each and every IEEE 802.15.4 device driver must provide a valid pointer on such relevantly filled-in API structure.
- Sending a packet is slightly different than in Ethernet. IEEE 802.15.4 sends
relatively small frames, 127 bytes all inclusive: frame header,
payload and frame checksum. Buffers are meant to fit such
frame size limitation. But a buffer containing an IPv6/UDP packet
might have more than one fragment. IEEE 802.15.4 drivers
handle only one buffer at a time. This is why the
struct ieee802154_radio_apirequires a tx function pointer which differs from the
struct net_if_apisend function pointer. Instead, the IEEE 802.15.4 L2, provides a generic
ieee802154_radio_send()meant to be given as
struct net_ifsend function. It turn, the implementation of
ieee802154_radio_send()will ensure the same behavior: sending one buffer at a time through
struct ieee802154_radio_apitx function, and unreferencing the network packet only when all the transmission were successful.
Each IEEE 802.15.4 device driver, in the end, will need to call
NET_DEVICE_INIT_INSTANCE() that way:
NET_DEVICE_INIT_INSTANCE(..., the_device_init_prio, &the_valid_ieee802154_radio_api_instance, IEEE802154_L2, NET_L2_GET_CTX_TYPE(IEEE802154_L2), 125);
Network Layer 2 abstraction layer.
- #include <net_l2.h>
Network L2 structure.
Used to provide an interface to lower network stack.