fix: enhance bridge port handling and sniffer management with improved error logging and interface checks

backend/src/network_sniffer.py

@@ -4,6 +4,7 @@ import asyncpg
 import threading
 import logging
 import os
+import socket
 from typing import List, Dict
 
 logging.basicConfig(level=logging.INFO)
@@ -23,29 +24,55 @@ bridge_ports_cache = {}
 # -------------------------------------------------------------------
 async_loop = asyncio.new_event_loop()
 
+
 def start_async_loop(loop):
     asyncio.set_event_loop(loop)
     loop.run_forever()
 
+
 threading.Thread(target=start_async_loop, args=(async_loop,), daemon=True).start()
 
 
 # -------------------------------------------------------------------
 #  BRIDGE PORT HANDLING
 # -------------------------------------------------------------------
+def check_interface_exists(iface: str) -> bool:
+    return os.path.isdir(f"/sys/class/net/{iface}")
+
+
+def check_interface_up(iface: str) -> bool:
+    try:
+        with open(f"/sys/class/net/{iface}/operstate", "r") as f:
+            return f.read().strip() == "up"
+    except FileNotFoundError:
+        return False
+
+
 def get_bridge_ports(bridge: str) -> List[str]:
     if bridge in bridge_ports_cache:
         return bridge_ports_cache[bridge]
 
     base = f"/sys/class/net/{bridge}/brif/"
     if not os.path.isdir(base):
-        logger.error(f"Bridge {bridge} does not exist")
+        logger.error(f"[ERROR] Bridge '{bridge}' does not exist.")
         return []
 
+    try:
         ports = os.listdir(base)
-    bridge_ports_cache[bridge] = ports
-    logger.info(f"Bridge {bridge} ports: {ports}")
-    return ports
+    except PermissionError:
+        logger.error(f"[ERROR] No permissions to read bridge ports for '{bridge}'.")
+        return []
+
+    ok_ports = []
+    for p in ports:
+        if not check_interface_exists(p):
+            logger.warning(f"[WARN] Port '{p}' in bridge but does not exist in /sys/class/net")
+            continue
+        ok_ports.append(p)
+
+    bridge_ports_cache[bridge] = ok_ports
+    logger.info(f"[INFO] Bridge {bridge} ports: {ok_ports}")
+    return ok_ports
 
 
 def determine_direction(pkt_iface: str, bridge: str):
@@ -59,6 +86,7 @@ def determine_direction(pkt_iface: str, bridge: str):
 #  DATABASE INSERTION
 # -------------------------------------------------------------------
 async def db_insert_packet(pkt_info: dict):
+    conn = None
     try:
         conn = await asyncpg.connect(DB_DSN)
 
@@ -91,10 +119,11 @@ async def db_insert_packet(pkt_info: dict):
             pkt_info["raw"]
         )
 
-    except Exception as e:
+    except (asyncpg.PostgresError, ConnectionError, OSError) as e:
         logger.exception(f"DB insert failed: {e}")
+
     finally:
-        if 'conn' in locals():
+        if conn:
             await conn.close()
 
 
@@ -125,39 +154,71 @@ def handle_packet(pkt, bridge: str):
 
 
 # -------------------------------------------------------------------
-#  SNIFFING
+#  CLEAN STOPPING SNIFFERS
 # -------------------------------------------------------------------
+def sniffer_loop(ifname: str, stop_event: threading.Event, bridge: str):
+    logger.info(f"Sniffer STARTED on {ifname}")
+
+    if not check_interface_exists(ifname):
+        logger.error(f"[ERROR] Interface {ifname} does not exist. Stopping sniffer.")
+        return
+
+    if not check_interface_up(ifname):
+        logger.error(f"[ERROR] Interface {ifname} is DOWN. Stopping sniffer.")
+        return
+
+    while not stop_event.is_set():
+        try:
+            sniff(
+                iface=ifname,
+                prn=lambda pkt: handle_packet(pkt, bridge),
+                store=False,
+                timeout=1  # periodic return so we can check stop_event
+            )
+        except PermissionError:
+            logger.error(f"[ERROR] Permission denied sniffing on {ifname}. Run as root.")
+            break
+        except OSError as e:
+            logger.error(f"[ERROR] Sniffer error on {ifname}: {e}")
+            break
+        except Exception as e:
+            logger.exception(f"Unexpected sniffer error on {ifname}: {e}")
+            break
+
+    logger.info(f"Sniffer STOPPED on {ifname}")
+
+
 def start_sniffer_thread(bridge: str):
     ports = get_bridge_ports(bridge)
+    if not ports:
+        logger.error(f"[ERROR] Could not start sniffer: No valid ports found for bridge {bridge}")
+        return {}
 
     for iface in ports:
         if iface in sniffer_threads:
+            logger.info(f"[INFO] Sniffer on {iface} is already running")
             continue
 
         stop_event = threading.Event()
         thread_stop_flags[iface] = stop_event
 
-        def sniff_blocking(ifname=iface):
-            logger.info(f"Sniffer STARTED on {ifname}")
-
-            sniff(
-                iface=ifname,
-                prn=lambda pkt: handle_packet(pkt, bridge),
-                store=False,
-                stop_filter=lambda _: stop_event.is_set(),
-                timeout=1  # ensure periodic stop checks
+        thread = threading.Thread(
+            target=sniffer_loop,
+            args=(iface, stop_event, bridge),
+            daemon=True
         )
 
-            logger.info(f"Sniffer STOPPED on {ifname}")
-
-        thread = threading.Thread(target=sniff_blocking, daemon=True)
         sniffer_threads[iface] = thread
         thread.start()
 
     return sniffer_threads
 
 
+# -------------------------------------------------------------------
+#  PUBLIC ASYNC START/STOP METHODS
+# -------------------------------------------------------------------
 async def start_sniffing(bridge: str):
+    logger.info(f"Starting sniffing for bridge {bridge}")
     start_sniffer_thread(bridge)
     return True
 
@@ -179,16 +240,14 @@ async def stop_sniffing():
 
 
 # -------------------------------------------------------------------
-#  STATUS HELPER (for API endpoint)
+#  STATUS HELPER
 # -------------------------------------------------------------------
 def get_sniffer_status():
-    """
-    Return a dict of interface → running/stopped status.
-    Useful for an API /status endpoint.
-    """
-    status = {}
-
+    out = {}
     for iface, t in sniffer_threads.items():
-        status[iface] = "running" if t.is_alive() else "stopped"
-
-    return status
+        out[iface] = {
+            "running": t.is_alive(),
+            "exists": check_interface_exists(iface),
+            "up": check_interface_up(iface)
+        }
+    return out