fix: Recovery-Rettungsweg härten (Watchdog-Cap, Konsole, Softap-Bit, Scan-Retry)

CMakeLists.txt

@@ -44,9 +44,16 @@ endif()
 if(NOT DEFINED NINLIGHT_RECOVERY_PASSWORD)
   set(NINLIGHT_RECOVERY_PASSWORD "ninlight-recovery")
 endif()
+# Unlike NINLIGHT_WIFI_SSID, the recovery SSID is compared byte-for-byte against scan
+# results (nl_recovery.c) and must fit struct tls_bss_info_t::ssid[32]; empty would match
+# any hidden/empty-SSID AP in the scan, so it must not be empty either.
+string(LENGTH "${NINLIGHT_RECOVERY_SSID}" _nl_rec_ssid_len)
+if(_nl_rec_ssid_len EQUAL 0 OR _nl_rec_ssid_len GREATER 32)
+  message(FATAL_ERROR "NINLIGHT_RECOVERY_SSID must be 1..32 characters")
+endif()
 string(LENGTH "${NINLIGHT_RECOVERY_PASSWORD}" _nl_rec_len)
-if(_nl_rec_len LESS 8)
-  message(FATAL_ERROR "NINLIGHT_RECOVERY_PASSWORD needs at least 8 characters (WPA2)")
+if(_nl_rec_len LESS 8 OR _nl_rec_len GREATER 63)
+  message(FATAL_ERROR "NINLIGHT_RECOVERY_PASSWORD must be 8..63 characters (WPA2)")
 endif()
 foreach(_v NINLIGHT_WIFI_SSID NINLIGHT_WIFI_PASSWORD NINLIGHT_AP_PASSWORD NINLIGHT_TOKEN
            NINLIGHT_RECOVERY_SSID NINLIGHT_RECOVERY_PASSWORD)

config.local.cmake.example

@@ -4,5 +4,8 @@
 # set(NINLIGHT_WIFI_SSID "MyWifi")
 # set(NINLIGHT_WIFI_PASSWORD "secret")
 # set(NINLIGHT_AP_PASSWORD "ninlight")
+# Boot recovery station (plan 3): SSID 1..32 chars, password 8..63 chars (WPA2)
+# set(NINLIGHT_RECOVERY_SSID "ninlight-recovery")
+# set(NINLIGHT_RECOVERY_PASSWORD "ninlight-recovery")
 # Used from plan 3/4 on (path to the old RSA signing key):
 # set(NINLIGHT_MIGRATION_KEY "/path/to/fw_sign.pem")

src/app_main.c

@@ -197,8 +197,8 @@ static void app_task(void *arg)
 
     tls_watchdog_init(20u * 1000u * 1000u); /* 20 s */
     nl_ota_init();
+    nl_console_start(); /* keep the serial fallback alive during nl_recovery_try() below */
     nl_recovery_try(); /* may not return: a successful recovery resets the chip */
-    nl_console_start();
 
     {
         static nl_settings_t boot; /* copy: the network task may change s_settings later */

src/net/nl_recovery.c

@@ -12,6 +12,15 @@
 #define NL_RECOVERY_SCAN_MS 3000u
 #define NL_RECOVERY_JOIN_MS 15000u
 #define NL_RECOVERY_BUF     2048u
+/* Generous cap on how long the download itself may run. wm_http_fwup.c's own resume loop
+   can already take up to ~20 minutes on a slow/aborted transfer (200 rounds x up to 5 s
+   receive timeout / 2 s retry delay), and a real hang - e.g. its unbounded
+   "while (!p) { tls_os_time_delay(1); pbuf_alloc(...) }" retry on allocation failure
+   (wm_http_fwup.c:178) - never returns from t_http_fwup() at all. Past this cap we stop
+   feeding the watchdog (see below) instead of waiting indefinitely: a watchdog reset is
+   strictly better than a silent hang with no Wi-Fi and no console, and a fresh boot simply
+   retries recovery from scratch. */
+#define NL_RECOVERY_DOWNLOAD_MAX_MS (10u * 60u * 1000u)
 
 static volatile bool s_scan_done;
 
@@ -32,24 +41,50 @@ static bool recovery_ap_present(void)
     uint32_t max_entries, n;
     bool found = false;
     u32 i;
+    int scan_rc;
 
     if (buf == NULL)
         return false;
     s_scan_done = false;
     tls_wifi_scan_result_cb_register(scan_done_cb);
-    if (tls_wifi_scan() != WM_SUCCESS) {
+
+    /* wm_wifi.h: "If not SUCCESS, user needs to call this function again to trigger scan" -
+       retry a transient failure (e.g. WM_WIFI_SCANNING_BUSY) instead of silently skipping
+       the whole recovery attempt, but stay inside the shared 3 s budget. */
+    for (;;) {
+        scan_rc = tls_wifi_scan();
+        if (scan_rc == WM_SUCCESS)
+            break;
+        if (waited >= NL_RECOVERY_SCAN_MS) {
+            printf("recovery: tls_wifi_scan() failed (rc=%d), giving up\r\n", scan_rc);
             tls_wifi_scan_result_cb_register(NULL);
             tls_mem_free(buf);
             return false;
         }
-    while (!s_scan_done && waited < NL_RECOVERY_SCAN_MS) {
         tls_watchdog_clr(); /* boot phase: only this path runs, the light_task heartbeat
                                 gate in app_task() does not watch us yet */
         tls_os_time_delay(HZ / 10);
         waited += 100;
     }
-    if (s_scan_done && tls_wifi_get_scan_rslt((u8 *)buf, NL_RECOVERY_BUF) == WM_SUCCESS) {
+
+    while (!s_scan_done && waited < NL_RECOVERY_SCAN_MS) {
+        tls_watchdog_clr();
+        tls_os_time_delay(HZ / 10);
+        waited += 100;
+    }
+
+    if (!s_scan_done) {
+        /* Without this line, a scan whose callback simply never fired in time is
+           indistinguishable in the log from "no AP found" - the EVK verification needs
+           exactly that distinction. */
+        printf("recovery: scan result not ready after %u ms, giving up\r\n", (unsigned)waited);
+    } else if (tls_wifi_get_scan_rslt((u8 *)buf, NL_RECOVERY_BUF) != WM_SUCCESS) {
+        printf("recovery: scan done after %u ms, but tls_wifi_get_scan_rslt() failed\r\n",
+               (unsigned)waited);
+    } else {
         res = (struct tls_scan_bss_t *)buf;
+        printf("recovery: scan done after %u ms, %u AP(s)\r\n", (unsigned)waited,
+               (unsigned)res->count);
         /* struct tls_bss_info_t has no length field - res->bss is a fixed-size-element
            array (struct tls_scan_bss_t::bss[1] is a flexible-array-style head), so index
            it directly. Cap by what actually fits in buf in case res->count ever reports
@@ -88,11 +123,28 @@ void nl_recovery_try(void)
         printf("recovery: cannot start download\r\n");
         return;
     }
-    /* Give the download task the boot phase; on success the SDK resets the chip. */
+    /* Give the download task the boot phase; on success the SDK resets the chip. Past
+       NL_RECOVERY_DOWNLOAD_MAX_MS, stop feeding the watchdog instead of continuing to wait -
+       see the comment on that constant. */
+    {
+        uint32_t waited = 0;
+        bool feeding = true;
+
         while (nl_ota_busy()) {
+            if (feeding) {
+                if (waited >= NL_RECOVERY_DOWNLOAD_MAX_MS) {
+                    printf("recovery: download still running after %u ms, letting the watchdog reset\r\n",
+                           (unsigned)NL_RECOVERY_DOWNLOAD_MAX_MS);
+                    feeding = false;
+                } else {
                     tls_watchdog_clr();
-        tls_os_time_delay(HZ / 2);
                 }
+            }
+            tls_os_time_delay(HZ / 2);
+            waited += 500;
+        }
+        if (feeding)
             printf("recovery: update failed (%s), continuing normal boot\r\n",
                    nl_ota_last_error() ? nl_ota_last_error() : "unknown");
     }
+}

src/net/nl_wifi_w600.c

@@ -345,6 +345,13 @@ bool nl_wifi_join_blocking(const char *ssid, const char *pass, uint32_t timeout_
 {
     uint32_t waited = 0;
 
+    /* C2: if a stale SDK softap bit from a previous onboarding/fallback boot is still set
+     * in TLS_PARAM_ID_WPROTOCOL, sys_net_status_changed() (sdk/Platform/Sys/tls_sys.c)
+     * recreates the SDK's own default-params softap the moment this join reaches
+     * NETIF_IP_NET_UP - exactly the "lamp stuck in onboarding" case recovery exists for.
+     * Clear it once, up front, same as nl_wifi_start() does. */
+    clear_softap_param();
+
     s_recovery_join_done = false;
     s_recovery_join_ok = false;
     tls_netif_add_status_event(recovery_join_event);