fix(tcp): recover stalled TCP client connects

App/tcp_client.c

@@ -20,6 +20,7 @@ typedef struct {
     struct pbuf *hold_pbuf;
     uint16_t hold_offset;
     uint32_t next_retry_ms;
+    uint32_t connect_start_ms;
     uint8_t index;
     tcp_client_instance_config_t config;
     tcp_client_status_t status;
@@ -37,6 +38,11 @@ static uint16_t ring_used(uint16_t head, uint16_t tail, uint16_t size)
     return (head >= tail) ? (uint16_t)(head - tail) : (uint16_t)(size - tail + head);
 }
 
+static bool tick_reached(uint32_t now, uint32_t deadline)
+{
+    return ((int32_t)(now - deadline) >= 0);
+}
+
 static void tcp_client_reset_rx_state(tcp_client_ctx_t *ctx)
 {
     if (ctx == NULL) {
@@ -51,6 +57,31 @@ static void tcp_client_reset_rx_state(tcp_client_ctx_t *ctx)
     ctx->rx_tail = 0u;
 }
 
+static void tcp_client_abort_connect_timeout(tcp_client_ctx_t *ctx, uint32_t now)
+{
+    if (ctx == NULL) {
+        return;
+    }
+
+    if (ctx->pcb != NULL) {
+        tcp_arg(ctx->pcb, NULL);
+        tcp_recv(ctx->pcb, NULL);
+        tcp_sent(ctx->pcb, NULL);
+        tcp_err(ctx->pcb, NULL);
+        tcp_client_reset_rx_state(ctx);
+        tcp_abort(ctx->pcb);
+        ctx->pcb = NULL;
+    } else {
+        tcp_client_reset_rx_state(ctx);
+    }
+
+    ctx->connect_start_ms = 0u;
+    ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
+    ctx->status.errors++;
+    ctx->status.connect_timeout_count++;
+    ctx->next_retry_ms = now + ctx->config.reconnect_interval_ms;
+}
+
 static void tcp_client_fill_ring_from_pbuf(tcp_client_ctx_t *ctx)
 {
     struct pbuf *q;
@@ -111,6 +142,7 @@ static err_t tcp_client_on_recv(void *arg, struct tcp_pcb *pcb, struct pbuf *p,
         tcp_err(pcb, NULL);
         tcp_abort(pcb);
         ctx->pcb = NULL;
+        ctx->connect_start_ms = 0u;
         ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
         ctx->next_retry_ms = HAL_GetTick() + ctx->config.reconnect_interval_ms;
         return ERR_ABRT;
@@ -147,6 +179,7 @@ static void tcp_client_on_err(void *arg, err_t err)
     }
     tcp_client_reset_rx_state(ctx);
     ctx->pcb = NULL;
+    ctx->connect_start_ms = 0u;
     ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
     ctx->status.errors++;
     ctx->next_retry_ms = HAL_GetTick() + ctx->config.reconnect_interval_ms;
@@ -162,6 +195,7 @@ static err_t tcp_client_on_connected(void *arg, struct tcp_pcb *pcb, err_t err)
     }
     if (err != ERR_OK) {
         ctx->pcb = NULL;
+        ctx->connect_start_ms = 0u;
         ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
         ctx->status.errors++;
         ctx->next_retry_ms = HAL_GetTick() + ctx->config.reconnect_interval_ms;
@@ -169,6 +203,7 @@ static err_t tcp_client_on_connected(void *arg, struct tcp_pcb *pcb, err_t err)
     }
 
     ctx->pcb = pcb;
+    ctx->connect_start_ms = 0u;
     ctx->status.state = TCP_CLIENT_STATE_CONNECTED;
     tcp_nagle_disable(pcb);
     tcp_arg(pcb, ctx);
@@ -228,6 +263,7 @@ int tcp_client_connect(uint8_t instance)
         if (err != ERR_OK) {
             tcp_abort(pcb);
             ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
+            ctx->connect_start_ms = 0u;
             ctx->status.errors++;
             ctx->next_retry_ms = HAL_GetTick() + ctx->config.reconnect_interval_ms;
             return -1;
@@ -241,6 +277,7 @@ int tcp_client_connect(uint8_t instance)
              ctx->config.remote_ip[3]);
 
     ctx->status.state = TCP_CLIENT_STATE_CONNECTING;
+    ctx->connect_start_ms = HAL_GetTick();
     tcp_arg(pcb, ctx);
     tcp_err(pcb, tcp_client_on_err);
     err = tcp_connect(pcb, &remote_addr, ctx->config.remote_port, tcp_client_on_connected);
@@ -248,6 +285,7 @@ int tcp_client_connect(uint8_t instance)
         tcp_err(pcb, NULL);
         tcp_abort(pcb);
         ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
+        ctx->connect_start_ms = 0u;
         ctx->status.errors++;
         ctx->next_retry_ms = HAL_GetTick() + ctx->config.reconnect_interval_ms;
         return -1;
@@ -274,6 +312,7 @@ int tcp_client_disconnect(uint8_t instance)
         tcp_abort(ctx->pcb);
         ctx->pcb = NULL;
     }
+    ctx->connect_start_ms = 0u;
     ctx->status.state = TCP_CLIENT_STATE_DISCONNECTED;
     tcp_client_reset_rx_state(ctx);
     return 0;
@@ -368,21 +407,26 @@ uint16_t tcp_client_peek(uint8_t instance, uint8_t *data, uint16_t max_len)
 void tcp_client_drop(uint8_t instance, uint16_t len)
 {
     tcp_client_ctx_t *ctx;
-    uint16_t dropped = 0u;
+    uint16_t acked = 0u;
 
     if (instance >= TCP_CLIENT_INSTANCE_COUNT || len == 0u) {
         return;
     }
 
     ctx = &g_clients[instance];
-    while (dropped < len && ctx->rx_tail != ctx->rx_head) {
+    while (acked < len) {
-        ctx->rx_tail = (uint16_t)((ctx->rx_tail + 1u) % TCP_CLIENT_RX_BUFFER_SIZE);
-        dropped++;
-    }
-    if (dropped > 0u && ctx->pcb != NULL) {
-        tcp_recved(ctx->pcb, dropped);
-    }
         tcp_client_fill_ring_from_pbuf(ctx);
+        if (ctx->rx_tail == ctx->rx_head) {
+            break;
+        }
+        while (acked < len && ctx->rx_tail != ctx->rx_head) {
+            ctx->rx_tail = (uint16_t)((ctx->rx_tail + 1u) % TCP_CLIENT_RX_BUFFER_SIZE);
+            acked++;
+        }
+    }
+    if (acked > 0u && ctx->pcb != NULL) {
+        tcp_recved(ctx->pcb, acked);
+    }
 }
 
 bool tcp_client_is_connected(uint8_t instance)
@@ -410,9 +454,12 @@ void tcp_client_poll(void)
             continue;
         }
         if ((ctx->pcb != NULL) && (ctx->status.state == TCP_CLIENT_STATE_CONNECTING)) {
+            if ((uint32_t)(now - ctx->connect_start_ms) >= TCP_CLIENT_CONNECT_TIMEOUT_MS) {
+                tcp_client_abort_connect_timeout(ctx, now);
+            }
             continue;
         }
-        if (now >= ctx->next_retry_ms) {
+        if (tick_reached(now, ctx->next_retry_ms)) {
             ctx->status.reconnect_count++;
             ctx->next_retry_ms = now + ctx->config.reconnect_interval_ms;
             (void)tcp_client_connect(i);

App/tcp_client.h

@@ -16,6 +16,7 @@ extern "C" {
 #define TCP_CLIENT_INSTANCE_COUNT      2u
 #define TCP_CLIENT_RX_BUFFER_SIZE      480u
 #define TCP_CLIENT_RECONNECT_DELAY_MS  3000u
+#define TCP_CLIENT_CONNECT_TIMEOUT_MS  10000u
 
 typedef enum {
     TCP_CLIENT_STATE_IDLE = 0,
@@ -39,6 +40,7 @@ typedef struct {
     uint32_t rx_bytes;
     uint32_t tx_bytes;
     uint32_t reconnect_count;
+    uint32_t connect_timeout_count;
     uint32_t errors;
 } tcp_client_status_t;