Minor changes to error handling in main loop + changes to boot script and config

nimd.conf

@@ -7,7 +7,7 @@ restartDelay         = 10
 sigtermDelay         = 90
 
 [Logging]
-level   = info
+level   = debug
 logFile = /var/log/nimd
 
 [Filesystem]

scripts/boot.sh

@@ -112,4 +112,4 @@ if ! [[ $memory -eq "" ]]
    qemumem="-m ${memory}"
 fi
 
-qemu-system-x86_64 -net nic,model=virtio,netdev=user.0 -netdev user,id=user.0 -drive file=vm.qcow2,if=virtio -enable-kvm -kernel ${kernel} -initrd ${initrd} ${qemumem}  -append 'root=/dev/vda1 rw quiet modules=ext4 console=ttyS0,9600 console=ttyS0 init=/bin/nimd'
+qemu-system-x86_64 -net nic,model=virtio,netdev=user.0 -netdev user,id=user.0 -drive file=vm.qcow2,if=virtio -enable-kvm -kernel ${kernel} -initrd ${initrd} ${qemumem}  -append 'root=/dev/vda1 rw quiet modules=ext4 console=ttyS0 init=/bin/nimd'

src/core/mainloop.nim

@@ -40,44 +40,49 @@ proc mainLoop*(logger: Logger, config: NimDConfig, startServices: bool = true) =
         logger.switchToFile()
         logger.debug("Entering accept() loop")
         while true:
-            serverSocket.accept(clientSocket)
+            try:
-            logger.debug(&"Received connection on control socket")
+                serverSocket.accept(clientSocket)
-            if clientSocket.recv(opType, size=1) == 0:
+                logger.debug(&"Received connection on control socket")
-                logger.debug(&"Client has disconnected, waiting for new connections")
+                if clientSocket.recv(opType, size=1) == 0:
-                continue
+                    logger.debug(&"Client has disconnected, waiting for new connections")
-            logger.debug(&"Received operation type '{opType}' via control socket")
+                    continue
-            # The operation type is a single byte:
+                logger.debug(&"Received operation type '{opType}' via control socket")
-            # - 'p' -> poweroff
+                # The operation type is a single byte:
-            # - 'r' -> reboot
+                # - 'p' -> poweroff
-            # - 'h' -> halt
+                # - 'r' -> reboot
-            # - 's' -> Services-related operations (start, stop, get status, etc.)
+                # - 'h' -> halt
-            # - 'l' -> Reload in-memory configuration
+                # - 's' -> Services-related operations (start, stop, get status, etc.)
-            # - 'c' -> Check NimD status (returns "1" if up)
+                # - 'l' -> Reload in-memory configuration
-            case opType:
+                # - 'c' -> Check NimD status (returns "1" if up)
-                of "p":
+                case opType:
-                    logger.info("Received shutdown request")
+                    of "p":
-                    shutdown(logger)
+                        logger.info("Received shutdown request")
-                of "r":
+                        shutdown(logger)
-                    logger.info("Received reboot request")
+                    of "r":
-                    reboot(logger)
+                        logger.info("Received reboot request")
-                of "h":
+                        reboot(logger)
-                    logger.info("Received halt request")
+                    of "h":
-                    halt(logger)
+                        logger.info("Received halt request")
-                of "s":
+                        halt(logger)
-                    logger.info("Received service-related request")
+                    of "s":
-                    # TODO: Operate on services
+                        logger.info("Received service-related request")
-                of "l":
+                        # TODO: Operate on services
-                    logger.info("Received reload request")
+                    of "l":
-                    mainLoop(logger, parseConfig(logger, "/etc/nimd/nimd.conf"), startServices=false)
+                        logger.info("Received reload request")
-                of "c":
+                        mainLoop(logger, parseConfig(logger, "/etc/nimd/nimd.conf"), startServices=false)
-                    logger.info("Received check request, responding")
+                    of "c":
-                    clientSocket.send("1")
+                        logger.info("Received check request, responding")
-                else:
+                        clientSocket.send("1")
-                    logger.warning(&"Received unknown operation type '{opType}' via control socket, ignoring it")
+                    else:
-                    discard
+                        logger.warning(&"Received unknown operation type '{opType}' via control socket, ignoring it")
-            clientSocket.close()
+                        discard
+            except:
+                logger.error(&"An error occurred while idling on control socket: {getCurrentExceptionMsg()}")
+            finally:
+                clientSocket.close()
+                clientSocket = newSocket(AF_INET, SOCK_STREAM, IPPROTO_TCP)
     except:
-        logger.critical(&"A critical error has occurred while running, restarting the mainloop in {config.restartDelay} seconds! Error -> {getCurrentExceptionMsg()}")
+        logger.critical(&"A critical error has occurred while running, restarting the main loop in {config.restartDelay} seconds! Error -> {getCurrentExceptionMsg()}")
         sleepSeconds(config.restartDelay)
         # We *absolutely* cannot die
         mainLoop(logger, config, startServices=false)

src/main.nim

@@ -59,7 +59,7 @@ proc addStuff =
                           depends=(@[newDependency(Other, errorer)]), provides=(@[]),
                           stdin="/dev/null", stderr="", stdout="")
     var shell = newService(name="login", description="A simple login shell", kind=Simple,
-                           workDir=getCurrentDir(), runlevel=Boot, exec="/bin/login -f root",
+                           workDir=getCurrentDir(), runlevel=Default, exec="/bin/login -f root",
                            supervised=true, restart=Always, restartDelay=0, depends=(@[]), provides=(@[]),
                            useParentStreams=true, stdin="/dev/null", stderr="/proc/self/fd/2", stdout="/proc/self/fd/1"
                            )
@@ -74,8 +74,12 @@ proc checkControlSocket(logger: Logger, config: NimDConfig): bool =
     ## socket
     result = true
     var stat_result: Stat
-    if posix.stat(cstring(config.sock), stat_result) == -1:
+    if posix.stat(cstring(config.sock), stat_result) == -1 and posix.errno != 2:
         logger.warning(&"Could not stat() {config.sock}, assuming NimD instance isn't running")
+    elif posix.errno == 2:
+        logger.debug(&"Control socket path is clear, starting up")
+        posix.errno = 0
+        # 2 is ENOENT, which means the file does not exist
     # I stole this from /usr/lib/python3.10/stat.py
     elif (int(stat_result.st_mode) and 0o170000) != 0o140000:
         logger.fatal(&"{config.sock} exists and is not a socket")