lwip accept => error 23 (Too many open files in system)

[slompf]

Hi,

I'm using lwip to receive data from a remote host. This works fine, but sometimes the connection goes down, so lwip_read returns -1 (error 113, Software caused connection abort). If this happens, I'm closing the socket I read.

After this happens sometimes, accept returns -1 (error 23, Too many open files in system).

Now I'm wondering, what to do to avoid the error of accept. What else but closing the socket can I do?

In order to make clearer what I'm doing, I striped down the logic into this pseudo code:

Code: Select all

for (;;)
{
    fd_set read_fds;
    memcpy(&read_fds, &current_fds, sizeof(read_fds));
    int triggered_nfds = select(highest_socket + 1, &read_fds, NULL, NULL, &timeout);
    if (triggered_nfds > 0) // more than zero sockets have data
    {
        if (FD_ISSET(listen_socket, &read_fds)) // check if listener socket has data
        {
            int socket = accept(listen_socket, (struct sockaddr *)NULL, (socklen_t *)NULL);

            if (socket < 0)
            {
                // log error 
                break;
            }
            else
            {
                // set:
                /* 10 second timeout */
                /* enable sending keepalive probes for socket */
                /* 180 sec idle before start sending probes */
                /* 30 sec between probes */
                /* Drop connection after 4 probes without response */
                FD_SET(socket, &current_fds);

                // save connection
            }

        triggered_nfds--;
    }

    for (each accepted connection)
    {
        if (FD_ISSET(connection_socket, &read_fds))
        {
            int recv_size = lwip_read(connection_socket, buffer, buffer_size);
            if(recv_size < 0)
            {
                // log error
                FD_CLR(connection_socket, &current_fds); // clear socket in buffer
                lwip_close(connection_socket);
            }else{
                // do sth with buffer
            }
            triggered_nfds--;
        }
    }
}

Thanks for your help,
Torsten

[ESP_Angus]

Hi Torsten,

LWIP has a limit of the maximum number of sockets in the system. ENFILE (23) will be returned if this limit is reached:
https://docs.espressif.com/projects/esp ... ax-sockets

The solution may be to either increase this limit in the project config, or fix a socket leak if you're not closing all the accept()-ed sockets - it's hard to see from the pseudocode if they will always be closed.

[slompf]

Hi

but in general, calling lwip_close after recv returned -1 should be enough to close the socket?

Thanks for your answer,
Torsten

[ESP_Angus]

but in general, calling lwip_close after recv returned -1 should be enough to close the socket?

Yes, this will close the socket at the ESP32 end. Of course, it only works if recv returns -1 first (ie the other end has closed the socket and isn't holding it open, or some packet loss meant the RST packet was not received yet, etc.)

[ESP_Mahavir]

In case you are using ESP-IDF v3.3 (or prior) release then `lwip_close` won't actually free up socket pcb, you have to use thread safe counterpart API `lwip_close_r` assuming you are using default configuration (https://github.com/espressif/esp-lwip/b ... ets.c#L877).

Ideally you can modify application to use POSIX APIs (i.e. `close` for `lwip_close` and so on), VFS layer in ESP-IDF will do correct translation as provided at https://github.com/espressif/esp-idf/bl ... lwip.c#L61 and thus ensuring appropriate behaviour.

[slompf]

Thanks for the response. Found an additional explanation here: https://github.com/espressif/esp-idf/pull/537 .