As shown in the :ref:`esp-netif structure` diagram, ESP-NETIF is in fact an intermediary between the I/O driver and the TCP/IP stack. This manual describes customization of these two sides, that is if you need to implement :ref:`esp_netif_custom_driver` or if you need to employ :ref:`esp_netif_tcpip_stack`.
This section outlines implementing a new I/O driver with ESP-NETIF connection capabilities.
By convention, the I/O driver has to register itself as an ESP-NETIF driver, and thus holds a dependency on ESP-NETIF component and is responsible for providing data path functions, post-attach callback and in most cases, also default event handlers to define network interface actions based on driver's lifecycle transitions.
According to the diagram shown in the :ref:`esp-netif structure` part, the following three API functions for the packet data path must be defined for connecting with ESP-NETIF:
The first two functions for transmitting and freeing the rx buffer are provided as callbacks, i.e., they get called from ESP-NETIF (and its underlying TCP/IP stack) and I/O driver provides their implementation.
The receiving function on the other hand gets called from the I/O driver, so that the driver's code simply calls :cpp:func:`esp_netif_receive()` on a new data received event.
A final part of the network interface initialization consists of attaching the ESP-NETIF instance to the I/O driver, by means of calling the following API:
It is assumed that the ``esp_netif_iodriver_handle`` is a pointer to driver's object, a struct derived from ``struct esp_netif_driver_base_s``, so that the first member of I/O driver structure must be this base structure with pointers to:
So when the :cpp:func:`esp_netif_attach()` gets called from the initialization code, the post-attach callback from I/O driver's code gets executed to mutually register callbacks between ESP-NETIF and I/O driver instances. Typically the driver is started as well in the post-attach callback. An example of a simple post-attach callback is outlined below:
I/O drivers also typically provide default definitions of lifecycle behavior of related network interfaces based on state transitions of I/O drivers. For example *driver start*``->``*network start*, etc.
The packet data path functions for transmitting and freeing the rx buffer (defined in the I/O driver) are called from the ESP-NETIF, specifically from its TCP/IP stack connecting layer.
Note that ESP-IDF provides several network stack configurations for the most common network interfaces, such as for the Wi-Fi station or Ethernet. These configurations are defined in :component_file:`esp_netif/include/esp_netif_defaults.h` and should be sufficient for most network drivers.
In some cases, you might want to define a custom lwIP based interface, for example if you need to update :component_file:`esp_netif/lwip/netif/wlanif.c` with a specific packet pool. In that case, you would have to define an explicit dependency to lwIP and include :component_file:`esp_netif/include/lwip/esp_netif_net_stack.h` for the relevant lwIP configuration structures.
.._esp_netif_tcpip_stack:
ESP-NETIF Custom TCP/IP Stack
-----------------------------
It is possible to use a custom TCP/IP stack with ESP-IDF, provided it implements BSD API. You can add support for your own TCP/IP stack, while using the generic ESP-NETIF functionality, so the application code can stay the same as with the lwIP.
In this case, please choose ``ESP_NETIF_PROVIDE_CUSTOM_IMPLEMENTATION`` in the ESP-NETIF component configuration menu. This option will disable lwIP implementation of the ESP-NETIF functions and provide only header files with declarations of types and API. You will have to supply the necessary implementation in your custom component. You can refer to the :component_file:`esp_netif/loopback/esp_netif_loopback.c` for example of dummy implementations of these functions.
It is also possible to build ESP-IDF without lwIP, please refer to :idf_file:`components/esp_netif_stack/README.md`.