using the signal to exit the program may cause memory leak

[yjqhh]

Hello, Andy. Recently, I encountered the memory leak issue in 4.3.1. When in a state of no network, I use ctrl + c exit the program quickly, the issue has occurred.Then I skipped the steps of lws_service and can easily reproduce it. The issue will also be reproduced using ctrl + c during DNS retry process.

static volatile sig_atomic_t exit_flag = 0;

void handle_sigint(int signal) {
    if (signal == SIGINT) {
        exit_flag = 1;
    }
}
 
static int callback_echo(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len);

static struct lws_protocols protocols[] = {
    {"echo", callback_echo, 0, 0},
    {NULL, NULL, 0, 0 }
};

static int callback_echo(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len) {
    char msg[LWS_PRE + 13];
    char *key = "content-type:";
    char *value = "application/json";
    int length = strlen(value);
    switch (reason) {
    case LWS_CALLBACK_CLIENT_RECEIVE:
        printf("Received: %s\n", (char *)in);
        break;
    case LWS_CALLBACK_CLIENT_APPEND_HANDSHAKE_HEADER:

        char **p = (char **)in, *end = (*p) + len;
        if(lws_add_http_header_by_name(wsi, (unsigned char *)key, (unsigned char *)value, length,
        (unsigned char **)p, (unsigned char *)end)) 
            return -1;
        break;
    case LWS_CALLBACK_CLIENT_ESTABLISHED:
        printf("Connection established\n");
        int n = lws_snprintf(msg + LWS_PRE, sizeof(msg) - LWS_PRE,
                      "hello,server!");
        lws_write(wsi, (unsigned char *)msg + LWS_PRE, n, LWS_WRITE_TEXT);
        break;
    case LWS_CALLBACK_CLIENT_CLOSED:
        printf("Connection closed\n");
        break;
    case LWS_CALLBACK_CLIENT_CONNECTION_ERROR:
        printf("[Test Protocol] There was a connection error: %s\n", in ? (char*)in : "(no error information)");
        break;
    default:
        printf("default reason %d \n",reason);
        break;
    }
    return 0;
}
 
int main(int argc, char **argv)
{
    struct lws_context_creation_info info;
    memset(&info, 0, sizeof info);
    info.options = LWS_SERVER_OPTION_DO_SSL_GLOBAL_INIT; 
    info.port = CONTEXT_PORT_NO_LISTEN;
    info.protocols = protocols;
    info.gid = -1;
    info.uid = -1;
    signal(SIGINT, handle_sigint);
    
    struct lws_context *context = lws_create_context(&info);

    struct lws_client_connect_info ccinfo = {0};
    ccinfo.context = context;
    ccinfo.address = "echo.websocket.events";
    ccinfo.port = 80;  
    ccinfo.path = "/"; 
    ccinfo.host = ccinfo.address;
    ccinfo.origin = ccinfo.address;
    ccinfo.protocol = protocols[0].name;
    ccinfo.ssl_connection = 0;
    struct lws *client = lws_client_connect_via_info(&ccinfo);


    // while (!exit_flag) {
    //     lws_service(context, 50);
    // }

    lws_context_destroy(context);
    
    return 0;
}

This is the backtrace of a memory leak

I don't know if this issue has been fixed in the current version.

[lws-team]

4.3.1

There's only downsides to staying on an old tag, instead of using v4.3-stable. There are 182 patches between v4.3.1 and v4.3-stable, and new features (exactly, new api changes) are forbidden on v4.3-stable. So there are a lot of fixes you are missing.

Thanks for bringing up this problem. I pushed most of a fix on _temp branch (based on main) in this patch 85562ee and I backported it to v4.3-stable basis on v4.3_temp branch.

Tested with valgrind it fixes the problem. But I am not sure if it will break other cases yet.

[lws-team]

This passes all the ctests, I pushed it on main and v4.3-stable.

[yjqhh]

The memory leak problem has not reproduced, thanks a lot