lwIP

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

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

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 material is 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.

  • socket()

  • bind()

  • accept()

  • shutdown()

  • getpeername()

  • getsockopt() & setsockopt() (see Socket Options)

  • close() (via 虚拟文件系统组件)

  • read(), readv(), write(), writev() (via 虚拟文件系统组件)

  • recv(), recvmsg(), recvfrom()

  • send(), sendmsg(), sendto()

  • select() (via 虚拟文件系统组件)

  • poll() (Note: on ESP-IDF, poll() is implemented by calling select internally, so using select() directly is recommended if a choice of methods is available.)

  • fcntl() (see fcntl)

Non-standard functions:

注解

Some lwIP application sample code uses prefixed versions of BSD APIs, for example 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 the socket error handling involves the following aspects:

  • Detecting the error.

  • Geting the error reason code.

  • Handle the error according to the reason code.

In lwIP, we have two different scenarios of 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 exception descriptor indicating that the socket has an error. For more information, see select() Errors.

Socket API Errors

The 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 doesn’t 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

The error detection
  • Socket error when select() has exception descriptor

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

注解

getsockopt function prototype int getsockopt(int s, int level, int optname, void *optval, socklen_t *optlen). Its function 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 more detailed list of standard POSIX/C error codes, please see newlib errno.h <https://github.com/espressif/newlib-esp32/blob/master/newlib/libc/include/sys/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/setting per-socket options.

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, 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 虚拟文件系统组件 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/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/write modes. These can read via F_GETFL only, they cannot be set using F_SETFL. A TCP socket will return 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.

ioctls

The ioctl() function provides a semi-standard way to access some internal features of the TCP/IP stack. In ESP-IDF, the 虚拟文件系统组件 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 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 needing to first copy it into internal lwIP buffers.

重要

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 this wiki page which is 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 as dual stack and enabled by default (IPv6 may be disabled if it’s not needed, see Minimum RAM usage). IPv6 support is limited to Stateless Autoconfiguration only, Stateful configuration is not supported in ESP-IDF (not 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 Router Advertisement protocol please enable:

This configuration option enables IPv6 autoconfiguration for all network interfaces (in contrast to the upstream lwIP, 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 support 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, too, by means of CONFIG_LWIP_IPV6_AUTOCONFIG. Moreover, the DHCPv6 needs to be explicitly enabled form 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 – this belongs to the Neighbor Discovery Protocol (NDP), uses Stateless Autoconfiguration Process. Number of servers must be set CONFIG_LWIP_IPV6_RDNSS_MAX_DNS_SERVERS, this is option is disabled (set to 0) by default.

  • DHCPv6 stateless configuration – uses DHCPv6 to configure DNS servers. Note that the 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 to the one that created it. The close() call will block 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 stack, so that when the task gets destroyed before unrolling the stack, the lwIP would still hold pointers to the deleted stack)

On demand timers

lwIP IGMP and MLD6 features 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 behaviour is to set each timer “on demand” so it is only enabled when an event is pending.

To return to the default lwIP behaviour (always-on timers), disable CONFIG_LWIP_TIMERS_ONDEMAND.

Lwip timers API

When users are not using WiFi, these APIs provide users with the ability to turn off LwIP timer 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()

Abort TCP connections when IP changes

CONFIG_LWIP_TCP_KEEP_CONNECTION_WHEN_IP_CHANGES is disabled by default. This disables the default lwIP behaviour of keeping TCP connections open if an interface IP changes, in case the interface IP changes back (for example, if an interface connection goes down and comes back up). Enable this option to keep TCP connections open in this case, until they time out normally. This may increase the number of sockets in use if a network interface goes down temporarily.

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 (normal lwIP is TCP only).

IP layer features

  • IPV4 source based routing implementation is different.

  • IPV4 mapped IPV6 addresses are supported.

Limitations

Calling send() or sendto() repeatedly on a UDP socket may eventually fail with errno equal to ENOMEM. This is a limitation of buffer sizes in the lower layer network interface drivers. If all driver transmit buffers are full then UDP transmission will fail. Applications sending a high volume of UDP datagrams who don’t wish for any to be dropped by the sender should check for this error code and re-send the datagram after a short delay.

Increasing the number of TX buffers in the Wi-Fi or Ethernet project configuration (as applicable) may also help.

Performance Optimization

TCP/IP performance is a complex subject, and performance can be optimized towards 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 then suggest applying settings from this file into the project sdkconfig.

重要

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 缓冲区使用情况.

Minimum latency

Except for increasing buffer sizes, most changes which increase throughput will 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 will also cause TCP sockets in the WAIT_CLOSE state to be closed and recycled more rapidly (if 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 memory usage at the expense of throughput, depending on usage.

  • Disable CONFIG_LWIP_IPV6 can save about 39 KB for firmware size and 2KB RAM when system power up and 7KB RAM when TCPIP stack running. If there is no requirement for supporting IPV6 then it can be disabled to save flash and RAM footprint.

If using Wi-Fi, please also refer to Wi-Fi 缓冲区使用情况.

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 (such as a sending failure) occurs. 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).