Handling modded lwIP that returns 0 instead of EINPROGRESS for connect()

[scaprile]

Apparently this modded stack blocks until TCP handshake is done and then returns... so a call to connect succeeds immediately.
See #3512

[thinker132203]

Hi @scaprile,

As promised, here are the debugging logs and the test code demonstrating the behavior on the TX82x modified lwIP stack.

By adding some custom debug prints, the logs clearly show two things:

1. Unmodified Mongoose (Silent Failure): connect() returns 0 synchronously. The MG_EV_CONNECT event is fired and the user initializes TLS. However, mg_tls_handshake() is never invoked. The connection stays silent and eventually times out.
2. Modified Mongoose (Success): By adding if (c->is_tls_hs) mg_tls_handshake(c); into the rc == 0 block of mg_connect_resolved, the TLS handshake starts immediately, and the HTTPS POST request successfully completes with a 200 OK.

Below are the detailed logs and the test snippet for your reference:

1. Original Behavior (Silent Failure & Timeout)

Modified mg_connect_resolved (added debug prints only):

    rc = connect(FD(c), &usa.sa, slen);  // Attempt to connect
    MG_DEBUG(("=== mg_connect_resolved: connect rc=%d, is_tls=%d, is_tls_hs=%d ===\n", rc, c->is_tls, c->is_tls_hs));
    if (rc == 0) {                       // Success
      setlocaddr(FD(c), &c->loc);
      mg_call(c, MG_EV_CONNECT, NULL);   // Send MG_EV_CONNECT to the user
      MG_DEBUG(("=== mg_connect_resolved: after mg_call, is_tls_hs=%d ===\n", c->is_tls_hs));
      
      // Missing mg_tls_handshake(c) here
      if (!c->is_tls_hs) c->is_tls = 0;  // user did not call mg_tls_init()
    }

Output Log:

[HTTPS] Starting local test for https://xyvision.top/api/consultationMessages/create
...
ffffffffbd4ecb98 3 mongoose.c:10639:mg_ === mg_connect_resolved: connect rc=0, is_tls=1, is_tls_hs=0 ===
[HTTPS] TCP Connected. Initializing TLS...
ffffffffbd4ecb98 3 mongoose.c:15870:mg_ 1 Setting TLS
ffffffffbd4ecb98 3 mongoose.c:15944:mg_ tls init success
[HTTPS] Sending POST request...
ffffffffbd4ecb98 3 mongoose.c:10643:mg_ === mg_connect_resolved: after mg_call, is_tls_hs=1 ===
...
// SILENCE - No TLS handshake activity
...
[106468][HTTPS] Test timed out
[106470][HTTPS] Connection closed

2. Modified Behavior (Successful TLS Handshake)

Modified mg_connect_resolved (added fix):

    if (rc == 0) {                       // Success
      setlocaddr(FD(c), &c->loc);
      mg_call(c, MG_EV_CONNECT, NULL);   // Send MG_EV_CONNECT to the user
      
      // Applied Fix:
      if (c->is_tls_hs) mg_tls_handshake(c);
      if (!c->is_tls_hs) c->is_tls = 0;  // user did not call mg_tls_init()
    }

Output Log:

[HTTPS] Starting local test for https://xyvision.top/api/consultationMessages/create
...
ffffffffbd56a368 3 mongoose.c:10639:mg_ === mg_connect_resolved: connect rc=0, is_tls=1, is_tls_hs=0 ===
[HTTPS] TCP Connected. Initializing TLS...
ffffffffbd56a368 3 mongoose.c:15944:mg_ tls init success
[HTTPS] Sending POST request...
ffffffffbd56a368 3 mongoose.c:10643:mg_ === mg_connect_resolved: after mg_call, is_tls_hs=1 ===

[26468]---------join in tls handshake-----
// TLS handshake packets exchanged successfully
[27054]---------join in tls handshake-----ffffffffbd56a368 3 mongoose.c:15824:mg_ 1 success
...
[27132][HTTPS] Received response status: 200
[27133]Full Header length: 520
[27133]JSON Body: {"created_at":"2026-04-24 ... "}
[HTTPS] Connection closed
[HTTPS] Test finished

3. Test Application Code (Context)

static const char *s_url = "https://xyvision.top/api/consultationMessages/create";
static const char *s_post_data = "{\"email\":\"\",\"message\":\"111\",\"name\":\"zxtTest\",\"phone\":\"573283282\"}";

static void official_demo_ev_handler(struct mg_connection *c, int ev, void *ev_data) {
    bool *done = (bool *) c->fn_data; 
    if (ev == MG_EV_CONNECT) {
        os_printf("[HTTPS] TCP Connected. Initializing TLS...\n");
        if (c->is_tls) {
            struct mg_tls_opts opts = {0}; 
            mg_tls_init(c, &opts);
        }
        os_printf("[HTTPS] Sending POST request...\n");
        int content_length = strlen(s_post_data);
        mg_printf(c,
                  "POST /api/consultationMessages/create HTTP/1.1\r\n"
                  "Host: xyvision.top\r\n"
                  "Content-Type: application/json\r\n"
                  "Content-Length: %d\r\n"
                  "Connection: close\r\n\r\n",
                  content_length);
        mg_send(c, s_post_data, content_length);
    } else if (ev == MG_EV_HTTP_MSG) {
        // Handle response...
        *done = true;
    }
}
// mg_mgr_init, mg_http_connect, and mg_mgr_poll loop omitted for brevity.

I hope this detailed trace provides enough context for the internal discussion. Thank you for taking the time to look into this edge case.

---

[thinker132203]

Hi @scaprile,

Thank you for the ongoing discussion. After a more thorough assessment and internal review, I’d like to share my final conclusions to help wrap up this issue:

1. Root Cause Analysis: The immediate trigger for this issue is indeed a non-standard implementation of the connect() function within the Taixin SDK. In an asynchronous context, returning 0 for a remote connection—which cannot possibly complete immediately—is improper behavior.
2. Valid Use Cases: However, according to POSIX standards, connect() is explicitly allowed to return 0 if a connection is established immediately. While niche, this scenario is entirely valid for local loopback connections. In modern Zero Trust architectures, such local paths often require TLS encryption, making this a legitimate requirement that Mongoose should ideally support.
3. Branch Consistency: Since Mongoose already maintains an explicit rc == 0 branch within mg_connect_resolved(), ensuring that this path is functionally equivalent to the asynchronous path regarding TLS initialization is highly beneficial for the library's internal consistency.
4. Feasibility: From my evaluation, the proposed fix is straightforward and introduces zero performance overhead. It simply completes a logic gap that was previously overlooked.

In conclusion: While the direct responsibility lies with the vendor's networking stack, addressing this edge case would significantly enhance Mongoose’s robustness and its adherence to POSIX semantics.

These are my personal professional views, and I leave the final architectural decision to your team's discretion.

Best regards,