lwIP

[中文]

ESP-IDF uses the open source lwIP lightweight TCP/IP stack. The ESP-IDF version of lwIP (esp-lwip) has some modifications and additions compared to the upstream project.

Supported APIs

ESP-IDF supports the following lwIP TCP/IP stack functions:

Adapted APIs

Warning

When using any lwIP API other than the BSD Sockets API, please make sure that the API is thread-safe. To check if a given API call is thread-safe, enable the CONFIG_LWIP_CHECK_THREAD_SAFETY configuration option and run the application. This enables lwIP to assert the correct access of the TCP/IP core functionality. If the API is not accessed or locked properly from the appropriate lwIP FreeRTOS Task, the execution will be aborted. The general recommendation is to use the ESP-NETIF component to interact with lwIP.

Some common lwIP app APIs are supported indirectly by ESP-IDF:

  • Dynamic Host Configuration Protocol (DHCP) Server & Client are supported indirectly via the ESP-NETIF functionality.

  • Domain Name System (DNS) is supported in lwIP; DNS servers could be assigned automatically when acquiring a DHCP address, or manually configured using the ESP-NETIF API.

Note

DNS server configuration in lwIP is global, not interface-specific. If you are using multiple network interfaces with distinct DNS servers, exercise caution to prevent inadvertent overwrites of one interface's DNS settings when acquiring a DHCP lease from another interface.

BSD Sockets API

The BSD Sockets API is a common cross-platform TCP/IP sockets API that originated in the Berkeley Standard Distribution of UNIX but is now standardized in a section of the POSIX specification. BSD Sockets are sometimes called POSIX Sockets or Berkeley Sockets.

As implemented in ESP-IDF, lwIP supports all of the common usages of the BSD Sockets API.

References

A wide range of BSD Sockets reference materials are available, including:

Examples

A number of ESP-IDF examples show how to use the BSD Sockets APIs:

Supported Functions

The following BSD socket API functions are supported. For full details, see lwip/lwip/src/include/lwip/sockets.h.

Non-standard functions:

Note

Some lwIP application sample code uses prefixed versions of BSD APIs, e.g., lwip_socket(), instead of the standard socket(). Both forms can be used with ESP-IDF, but using standard names is recommended.

Socket Error Handling

BSD Socket error handling code is very important for robust socket applications. Normally, socket error handling involves the following aspects:

  • Detecting the error

  • Getting the error reason code

  • Handling the error according to the reason code

In lwIP, we have two different scenarios for handling socket errors:

  • Socket API returns an error. For more information, see Socket API Errors.

  • select(int maxfdp1, fd_set *readset, fd_set *writeset, fd_set *exceptset, struct timeval *timeout) has an exception descriptor indicating that the socket has an error. For more information, see select() Errors.

Socket API Errors

Error detection

  • We can know that the socket API fails according to its return value.

Get the error reason code

  • When socket API fails, the return value does not contain the failure reason and the application can get the error reason code by accessing errno. Different values indicate different meanings. For more information, see Socket Error Reason Code.

Example:

int err;
int sockfd;

if (sockfd = socket(AF_INET,SOCK_STREAM,0) < 0) {
    // the error code is obtained from errno
    err = errno;
    return err;
}

select() Errors

Error detection

  • Socket error when select() has exception descriptor.

Get the error reason code

  • If the select() indicates that the socket fails, we can not get the error reason code by accessing errno, instead we should call getsockopt() to get the failure reason code. Since select() has exception descriptor, the error code is not given to errno.

Note

The getsockopt() function has the following prototype: int getsockopt(int s, int level, int optname, void *optval, socklen_t *optlen). Its purpose is to get the current value of the option of any type, any state socket, and store the result in optval. For example, when you get the error code on a socket, you can get it by getsockopt(sockfd, SOL_SOCKET, SO_ERROR, &err, &optlen).

Example:

int err;

if (select(sockfd + 1, NULL, NULL, &exfds, &tval) <= 0) {
    err = errno;
    return err;
} else {
    if (FD_ISSET(sockfd, &exfds)) {
        // select() exception set using getsockopt()
        int optlen = sizeof(int);
        getsockopt(sockfd, SOL_SOCKET, SO_ERROR, &err, &optlen);
        return err;
    }
}

Socket Error Reason Code

Below is a list of common error codes. For a more detailed list of standard POSIX/C error codes, please see newlib errno.h and the platform-specific extensions newlib/platform_include/errno.h.

Error code

Description

ECONNREFUSED

Connection refused

EADDRINUSE

Address already in use

ECONNABORTED

Software caused connection abort

ENETUNREACH

Network is unreachable

ENETDOWN

Network interface is not configured

ETIMEDOUT

Connection timed out

EHOSTDOWN

Host is down

EHOSTUNREACH

Host is unreachable

EINPROGRESS

Connection already in progress

EALREADY

Socket already connected

EDESTADDRREQ

Destination address required

EPROTONOSUPPORT

Unknown protocol

Socket Options

The getsockopt() and setsockopt() functions allow getting and setting per-socket options respectively.

Not all standard socket options are supported by lwIP in ESP-IDF. The following socket options are supported:

Common Options

Used with level argument SOL_SOCKET.

IP Options

Used with level argument IPPROTO_IP.

For multicast UDP sockets:

  • IP_MULTICAST_IF

  • IP_MULTICAST_LOOP

  • IP_MULTICAST_TTL

  • IP_ADD_MEMBERSHIP

  • IP_DROP_MEMBERSHIP

TCP Options

TCP sockets only. Used with level argument IPPROTO_TCP.

  • TCP_NODELAY

Options relating to TCP keepalive probes:

  • TCP_KEEPALIVE: int value, TCP keepalive period in milliseconds

  • TCP_KEEPIDLE: same as TCP_KEEPALIVE, but the value is in seconds

  • TCP_KEEPINTVL: int value, the interval between keepalive probes in seconds

  • TCP_KEEPCNT: int value, number of keepalive probes before timing out

IPv6 Options

IPv6 sockets only. Used with level argument IPPROTO_IPV6.

  • IPV6_CHECKSUM

  • IPV6_V6ONLY

For multicast IPv6 UDP sockets:

  • IPV6_JOIN_GROUP / IPV6_ADD_MEMBERSHIP

  • IPV6_LEAVE_GROUP / IPV6_DROP_MEMBERSHIP

  • IPV6_MULTICAST_IF

  • IPV6_MULTICAST_HOPS

  • IPV6_MULTICAST_LOOP

fcntl()

The fcntl() function is a standard API for manipulating options related to a file descriptor. In ESP-IDF, the Virtual Filesystem Component layer is used to implement this function.

When the file descriptor is a socket, only the following fcntl() values are supported:

  • O_NONBLOCK to set or clear non-blocking I/O mode. Also supports O_NDELAY, which is identical to O_NONBLOCK.

  • O_RDONLY, O_WRONLY, O_RDWR flags for different read or write modes. These flags can only be read using F_GETFL, and cannot be set using F_SETFL. A TCP socket returns a different mode depending on whether the connection has been closed at either end or is still open at both ends. UDP sockets always return O_RDWR.

ioctl()

The ioctl() function provides a semi-standard way to access some internal features of the TCP/IP stack. In ESP-IDF, the Virtual Filesystem Component layer is used to implement this function.

When the file descriptor is a socket, only the following ioctl() values are supported:

  • FIONREAD returns the number of bytes of the pending data already received in the socket's network buffer.

  • FIONBIO is an alternative way to set/clear non-blocking I/O status for a socket, equivalent to fcntl(fd, F_SETFL, O_NONBLOCK, ...).

Netconn API

lwIP supports two lower-level APIs as well as the BSD Sockets API: the Netconn API and the Raw API.

The lwIP Raw API is designed for single-threaded devices and is not supported in ESP-IDF.

The Netconn API is used to implement the BSD Sockets API inside lwIP, and it can also be called directly from ESP-IDF apps. This API has lower resource usage than the BSD Sockets API. In particular, it can send and receive data without firstly copying it into internal lwIP buffers.

Important

Espressif does not test the Netconn API in ESP-IDF. As such, this functionality is enabled but not supported. Some functionality may only work correctly when used from the BSD Sockets API.

For more information about the Netconn API, consult lwip/lwip/src/include/lwip/api.h and part of the **unofficial** lwIP Application Developers Manual.

lwIP FreeRTOS Task

lwIP creates a dedicated TCP/IP FreeRTOS task to handle socket API requests from other tasks.

A number of configuration items are available to modify the task and the queues (mailboxes) used to send data to/from the TCP/IP task:

IPv6 Support

Both IPv4 and IPv6 are supported in a dual-stack configuration and are enabled by default. Both IPv6 and IPv4 may be disabled separately if they are not needed, see Minimum RAM Usage.

IPv6 support is limited to Stateless Autoconfiguration only. Stateful configuration is not supported in ESP-IDF, nor in upstream lwIP.

IPv6 Address configuration is defined by means of these protocols or services:

  • SLAAC IPv6 Stateless Address Autoconfiguration (RFC-2462)

  • DHCPv6 Dynamic Host Configuration Protocol for IPv6 (RFC-8415)

None of these two types of address configuration is enabled by default, so the device uses only Link Local addresses or statically-defined addresses.

Stateless Autoconfiguration Process

To enable address autoconfiguration using the Router Advertisement protocol, please enable:

This configuration option enables IPv6 autoconfiguration for all network interfaces, which differs from the upstream lwIP behavior, where the autoconfiguration needs to be explicitly enabled for each netif with netif->ip6_autoconfig_enabled=1.

DHCPv6

DHCPv6 in lwIP is very simple and supports only stateless configuration. It could be enabled using:

Since the DHCPv6 works only in its stateless configuration, the Stateless Autoconfiguration Process has to be enabled as well via CONFIG_LWIP_IPV6_AUTOCONFIG.

Moreover, the DHCPv6 needs to be explicitly enabled from the application code using:

dhcp6_enable_stateless(netif);

DNS Servers in IPv6 Autoconfiguration

In order to autoconfigure DNS server(s), especially in IPv6-only networks, we have these two options:

  • Recursive Domain Name System (DNS): this belongs to the Neighbor Discovery Protocol (NDP) and uses Stateless Autoconfiguration Process.

    The number of servers must be set CONFIG_LWIP_IPV6_RDNSS_MAX_DNS_SERVERS, this option is disabled by default, i.e., set to 0.

  • DHCPv6 stateless configuration, uses DHCPv6 to configure DNS servers. Note that this configuration assumes IPv6 Router Advertisement Flags (RFC-5175) to be set to

    • Managed Address Configuration Flag = 0

    • Other Configuration Flag = 1

ESP-lwIP Custom Modifications

Additions

The following code is added, which is not present in the upstream lwIP release:

Thread-Safe Sockets

It is possible to close() a socket from a different thread than the one that created it. The close() call blocks, until any function calls currently using that socket from other tasks have returned.

It is, however, not possible to delete a task while it is actively waiting on select() or poll() APIs. It is always necessary that these APIs exit before destroying the task, as this might corrupt internal structures and cause subsequent crashes of the lwIP. These APIs allocate globally referenced callback pointers on the stack so that when the task gets destroyed before unrolling the stack, the lwIP could still hold pointers to the deleted stack.

On-Demand Timers

lwIP IGMP and MLD6 feature both initialize a timer in order to trigger timeout events at certain times.

The default lwIP implementation is to have these timers enabled all the time, even if no timeout events are active. This increases CPU usage and power consumption when using automatic Light-sleep mode. ESP-lwIP default behavior is to set each timer on demand, so it is only enabled when an event is pending.

To return to the default lwIP behavior, which is always-on timers, disable CONFIG_LWIP_TIMERS_ONDEMAND.

lwIP Timers API

When not using Wi-Fi, the lwIP timer can be turned off via the API to reduce power consumption.

The following API functions are supported. For full details, see lwip/lwip/src/include/lwip/timeouts.h.

  • sys_timeouts_init()

  • sys_timeouts_deinit()

Additional Socket Options

  • Some standard IPV4 and IPV6 multicast socket options are implemented, see Socket Options.

  • Possible to set IPV6-only UDP and TCP sockets with IPV6_V6ONLY socket option, while normal lwIP is TCP-only.

IP Layer Features

  • IPV4-source-based routing implementation is different

  • IPV4-mapped IPV6 addresses are supported

Customized lwIP Hooks

The original lwIP supports implementing custom compile-time modifications via LWIP_HOOK_FILENAME. This file is already used by the ESP-IDF port layer, but ESP-IDF users could still include and implement any custom additions via a header file defined by the macro ESP_IDF_LWIP_HOOK_FILENAME. Here is an example of adding a custom hook file to the build process, and the hook is called my_hook.h, located in the project's main folder:

idf_component_get_property(lwip lwip COMPONENT_LIB)
target_compile_options(${lwip} PRIVATE "-I${PROJECT_DIR}/main")
target_compile_definitions(${lwip} PRIVATE "-DESP_IDF_LWIP_HOOK_FILENAME=\"my_hook.h\"")

Customized lwIP Options From ESP-IDF Build System

The most common lwIP options are configurable through the component configuration menu. However, certain definitions need to be injected from the command line. The CMake function target_compile_definitions() can be employed to define macros, as illustrated below:

idf_component_get_property(lwip lwip COMPONENT_LIB)
target_compile_definitions(${lwip} PRIVATE "-DETHARP_SUPPORT_VLAN=1")

This approach may not work for function-like macros, as there is no guarantee that the definition will be accepted by all compilers, although it is supported in GCC. To address this limitation, the add_definitions() function can be utilized to define the macro for the entire project, for example: add_definitions("-DFALLBACK_DNS_SERVER_ADDRESS(addr)=\"IP_ADDR4((addr), 8,8,8,8)\"").

Alternatively, you can define your function-like macro in a header file which will be pre-included as an lwIP hook file, see Customized lwIP Hooks.

Limitations

ESP-IDF additions to lwIP still suffer from the global DNS limitation, described in Adapted APIs. To address this limitation from application code, the FALLBACK_DNS_SERVER_ADDRESS() macro can be utilized to define a global DNS fallback server accessible from all interfaces. Alternatively, you have the option to maintain per-interface DNS servers and reconfigure them whenever the default interface changes.

The number of IP addresses returned by network database APIs such as getaddrinfo() and gethostbyname() is restricted by the macro DNS_MAX_HOST_IP. By default, the value of this macro is set to 1.

In the implementation of getaddrinfo(), the canonical name is not available. Therefore, the ai_canonname field of the first returned addrinfo structure will always refer to the nodename argument or a string with the same contents.

Calling send() or sendto() repeatedly on a UDP socket may eventually fail with errno equal to ENOMEM. This failure occurs due to the limitations of buffer sizes in the lower-layer network interface drivers. If all driver transmit buffers are full, the UDP transmission will fail. For applications that transmit a high volume of UDP datagrams and aim to avoid any dropped datagrams by the sender, it is advisable to implement error code checking and employ a retransmission mechanism with a short delay.

Increasing the number of TX buffers in the Wi-Fi project configuration may also help.

Performance Optimization

TCP/IP performance is a complex subject, and performance can be optimized toward multiple goals. The default settings of ESP-IDF are tuned for a compromise between throughput, latency, and moderate memory usage.

Maximum Throughput

Espressif tests ESP-IDF TCP/IP throughput using the wifi/iperf example in an RF-sealed enclosure.

The wifi/iperf/sdkconfig.defaults file for the iperf example contains settings known to maximize TCP/IP throughput, usually at the expense of higher RAM usage. To get maximum TCP/IP throughput in an application at the expense of other factors, it is suggested to apply settings from this file into the project sdkconfig.

Important

Suggest applying changes a few at a time and checking the performance each time with a particular application workload.

If using a Wi-Fi network interface, please also refer to Wi-Fi Buffer Usage.

Minimum Latency

Except for increasing buffer sizes, most changes that increase throughput also decrease latency by reducing the amount of CPU time spent in lwIP functions.

  • For TCP sockets, lwIP supports setting the standard TCP_NODELAY flag to disable Nagle's algorithm.

Minimum RAM Usage

Most lwIP RAM usage is on-demand, as RAM is allocated from the heap as needed. Therefore, changing lwIP settings to reduce RAM usage may not change RAM usage at idle, but can change it at peak.

  • Reducing CONFIG_LWIP_MAX_SOCKETS reduces the maximum number of sockets in the system. This also causes TCP sockets in the WAIT_CLOSE state to be closed and recycled more rapidly when needed to open a new socket, further reducing peak RAM usage.

  • Reducing CONFIG_LWIP_TCPIP_RECVMBOX_SIZE, CONFIG_LWIP_TCP_RECVMBOX_SIZE and CONFIG_LWIP_UDP_RECVMBOX_SIZE reduce RAM usage at the expense of throughput, depending on usage.

  • Reducing CONFIG_LWIP_TCP_MSL and CONFIG_LWIP_TCP_FIN_WAIT_TIMEOUT reduces the maximum segment lifetime in the system. This also causes TCP sockets in the TIME_WAIT and FIN_WAIT_2 states to be closed and recycled more rapidly.

  • Disabling CONFIG_LWIP_IPV6 can save about 39 KB for firmware size and 2 KB RAM when the system is powered up and 7 KB RAM when the TCP/IP stack is running. If there is no requirement for supporting IPV6, it can be disabled to save flash and RAM footprint.

  • Disabling CONFIG_LWIP_IPV4 can save about 26 KB of firmware size and 600 B RAM on power up and 6 KB RAM when the TCP/IP stack is running. If the local network supports IPv6-only configuration, IPv4 can be disabled to save flash and RAM footprint.

If using Wi-Fi, please also refer to Wi-Fi Buffer Usage.

Peak Buffer Usage

The peak heap memory that lwIP consumes is the theoretically-maximum memory that the lwIP driver consumes. Generally, the peak heap memory that lwIP consumes depends on:

  • the memory required to create a UDP connection: lwip_udp_conn

  • the memory required to create a TCP connection: lwip_tcp_conn

  • the number of UDP connections that the application has: lwip_udp_con_num

  • the number of TCP connections that the application has: lwip_tcp_con_num

  • the TCP TX window size: lwip_tcp_tx_win_size

  • the TCP RX window size: lwip_tcp_rx_win_size

So, the peak heap memory that the lwIP consumes can be calculated with the following formula:

lwip_dynamic_peek_memory = (lwip_udp_con_num * lwip_udp_conn) + (lwip_tcp_con_num * (lwip_tcp_tx_win_size + lwip_tcp_rx_win_size + lwip_tcp_conn))

Some TCP-based applications need only one TCP connection. However, they may choose to close this TCP connection and create a new one when an error occurs (e.g., a sending failure). This may result in multiple TCP connections existing in the system simultaneously, because it may take a long time for a TCP connection to close, according to the TCP state machine, refer to RFC793.

Provide feedback about this document