package docker import ( "container/list" "fmt" "github.com/dotcloud/docker/auth" "io" "io/ioutil" "log" "os" "os/exec" "path" "sort" "strings" "time" ) type Capabilities struct { MemoryLimit bool SwapLimit bool } type Runtime struct { root string repository string containers *list.List networkManager *NetworkManager graph *Graph repositories *TagStore authConfig *auth.AuthConfig idIndex *TruncIndex capabilities *Capabilities kernelVersion *KernelVersionInfo } var sysInitPath string func init() { sysInitPath = SelfPath() } func (runtime *Runtime) List() []*Container { containers := new(History) for e := runtime.containers.Front(); e != nil; e = e.Next() { containers.Add(e.Value.(*Container)) } return *containers } func (runtime *Runtime) getContainerElement(id string) *list.Element { for e := runtime.containers.Front(); e != nil; e = e.Next() { container := e.Value.(*Container) if container.Id == id { return e } } return nil } func (runtime *Runtime) Get(name string) *Container { id, err := runtime.idIndex.Get(name) if err != nil { return nil } e := runtime.getContainerElement(id) if e == nil { return nil } return e.Value.(*Container) } func (runtime *Runtime) Exists(id string) bool { return runtime.Get(id) != nil } func (runtime *Runtime) containerRoot(id string) string { return path.Join(runtime.repository, id) } func (runtime *Runtime) Create(config *Config) (*Container, error) { // Lookup image img, err := runtime.repositories.LookupImage(config.Image) if err != nil { return nil, err } // Generate id id := GenerateId() // Generate default hostname // FIXME: the lxc template no longer needs to set a default hostname if config.Hostname == "" { config.Hostname = id[:12] } container := &Container{ // FIXME: we should generate the ID here instead of receiving it as an argument Id: id, Created: time.Now(), Path: config.Cmd[0], Args: config.Cmd[1:], //FIXME: de-duplicate from config Config: config, Image: img.Id, // Always use the resolved image id NetworkSettings: &NetworkSettings{}, // FIXME: do we need to store this in the container? SysInitPath: sysInitPath, } container.root = runtime.containerRoot(container.Id) // Step 1: create the container directory. // This doubles as a barrier to avoid race conditions. if err := os.Mkdir(container.root, 0700); err != nil { return nil, err } // If custom dns exists, then create a resolv.conf for the container if len(config.Dns) > 0 { container.ResolvConfPath = path.Join(container.root, "resolv.conf") f, err := os.Create(container.ResolvConfPath) if err != nil { return nil, err } defer f.Close() for _, dns := range config.Dns { if _, err := f.Write([]byte("nameserver " + dns + "\n")); err != nil { return nil, err } } } else { container.ResolvConfPath = "/etc/resolv.conf" } // Step 2: save the container json if err := container.ToDisk(); err != nil { return nil, err } // Step 3: register the container if err := runtime.Register(container); err != nil { return nil, err } return container, nil } func (runtime *Runtime) Load(id string) (*Container, error) { container := &Container{root: runtime.containerRoot(id)} if err := container.FromDisk(); err != nil { return nil, err } if container.Id != id { return container, fmt.Errorf("Container %s is stored at %s", container.Id, id) } if container.State.Running { container.State.Ghost = true } if err := runtime.Register(container); err != nil { return nil, err } return container, nil } // Register makes a container object usable by the runtime as func (runtime *Runtime) Register(container *Container) error { if container.runtime != nil || runtime.Exists(container.Id) { return fmt.Errorf("Container is already loaded") } if err := validateId(container.Id); err != nil { return err } // init the wait lock container.waitLock = make(chan struct{}) // FIXME: if the container is supposed to be running but is not, auto restart it? // if so, then we need to restart monitor and init a new lock // If the container is supposed to be running, make sure of it if container.State.Running { if output, err := exec.Command("lxc-info", "-n", container.Id).CombinedOutput(); err != nil { return err } else { if !strings.Contains(string(output), "RUNNING") { Debugf("Container %s was supposed to be running be is not.", container.Id) container.State.setStopped(-127) if err := container.ToDisk(); err != nil { return err } } } } // Even if not running, we init the lock (prevents races in start/stop/kill) container.State.initLock() container.runtime = runtime // Attach to stdout and stderr container.stderr = newWriteBroadcaster() container.stdout = newWriteBroadcaster() // Attach to stdin if container.Config.OpenStdin { container.stdin, container.stdinPipe = io.Pipe() } else { container.stdinPipe = NopWriteCloser(ioutil.Discard) // Silently drop stdin } // done runtime.containers.PushBack(container) runtime.idIndex.Add(container.Id) // If the container is not running or just has been flagged not running // then close the wait lock chan (will be reset upon start) if !container.State.Running { close(container.waitLock) } else { container.allocateNetwork() go container.monitor() } return nil } func (runtime *Runtime) LogToDisk(src *writeBroadcaster, dst string) error { log, err := os.OpenFile(dst, os.O_RDWR|os.O_APPEND|os.O_CREATE, 0600) if err != nil { return err } src.AddWriter(log) return nil } func (runtime *Runtime) Destroy(container *Container) error { element := runtime.getContainerElement(container.Id) if element == nil { return fmt.Errorf("Container %v not found - maybe it was already destroyed?", container.Id) } if err := container.Stop(10); err != nil { return err } if mounted, err := container.Mounted(); err != nil { return err } else if mounted { if err := container.Unmount(); err != nil { return fmt.Errorf("Unable to unmount container %v: %v", container.Id, err) } } // Deregister the container before removing its directory, to avoid race conditions runtime.idIndex.Delete(container.Id) runtime.containers.Remove(element) if err := os.RemoveAll(container.root); err != nil { return fmt.Errorf("Unable to remove filesystem for %v: %v", container.Id, err) } return nil } // Commit creates a new filesystem image from the current state of a container. // The image can optionally be tagged into a repository func (runtime *Runtime) Commit(id, repository, tag, comment, author string) (*Image, error) { container := runtime.Get(id) if container == nil { return nil, fmt.Errorf("No such container: %s", id) } // FIXME: freeze the container before copying it to avoid data corruption? // FIXME: this shouldn't be in commands. rwTar, err := container.ExportRw() if err != nil { return nil, err } // Create a new image from the container's base layers + a new layer from container changes img, err := runtime.graph.Create(rwTar, container, comment, author) if err != nil { return nil, err } // Register the image if needed if repository != "" { if err := runtime.repositories.Set(repository, tag, img.Id, true); err != nil { return img, err } } return img, nil } func (runtime *Runtime) restore() error { dir, err := ioutil.ReadDir(runtime.repository) if err != nil { return err } for _, v := range dir { id := v.Name() container, err := runtime.Load(id) if err != nil { Debugf("Failed to load container %v: %v", id, err) continue } Debugf("Loaded container %v", container.Id) } return nil } // FIXME: harmonize with NewGraph() func NewRuntime() (*Runtime, error) { runtime, err := NewRuntimeFromDirectory("/var/lib/docker") if err != nil { return nil, err } k, err := GetKernelVersion() if err != nil { return nil, err } runtime.kernelVersion = k if CompareKernelVersion(k, &KernelVersionInfo{Kernel: 3, Major: 8, Minor: 0}) < 0 { log.Printf("WARNING: You are running linux kernel version %s, which might be unstable running docker. Please upgrade your kernel to 3.8.0.", k.String()) } if cgroupMemoryMountpoint, err := FindCgroupMountpoint("memory"); err != nil { log.Printf("WARNING: %s\n", err) } else { _, err1 := ioutil.ReadFile(path.Join(cgroupMemoryMountpoint, "memory.limit_in_bytes")) _, err2 := ioutil.ReadFile(path.Join(cgroupMemoryMountpoint, "memory.soft_limit_in_bytes")) runtime.capabilities.MemoryLimit = err1 == nil && err2 == nil if !runtime.capabilities.MemoryLimit { log.Printf("WARNING: Your kernel does not support cgroup memory limit.") } _, err = ioutil.ReadFile(path.Join(cgroupMemoryMountpoint, "memory.memsw.limit_in_bytes")) runtime.capabilities.SwapLimit = err == nil if !runtime.capabilities.SwapLimit { log.Printf("WARNING: Your kernel does not support cgroup swap limit.") } } return runtime, nil } func NewRuntimeFromDirectory(root string) (*Runtime, error) { runtimeRepo := path.Join(root, "containers") if err := os.MkdirAll(runtimeRepo, 0700); err != nil && !os.IsExist(err) { return nil, err } g, err := NewGraph(path.Join(root, "graph")) if err != nil { return nil, err } repositories, err := NewTagStore(path.Join(root, "repositories"), g) if err != nil { return nil, fmt.Errorf("Couldn't create Tag store: %s", err) } if NetworkBridgeIface == "" { NetworkBridgeIface = DefaultNetworkBridge } netManager, err := newNetworkManager(NetworkBridgeIface) if err != nil { return nil, err } authConfig, err := auth.LoadConfig(root) if err != nil && authConfig == nil { // If the auth file does not exist, keep going return nil, err } runtime := &Runtime{ root: root, repository: runtimeRepo, containers: list.New(), networkManager: netManager, graph: g, repositories: repositories, authConfig: authConfig, idIndex: NewTruncIndex(), capabilities: &Capabilities{}, } if err := runtime.restore(); err != nil { return nil, err } return runtime, nil } type History []*Container func (history *History) Len() int { return len(*history) } func (history *History) Less(i, j int) bool { containers := *history return containers[j].When().Before(containers[i].When()) } func (history *History) Swap(i, j int) { containers := *history tmp := containers[i] containers[i] = containers[j] containers[j] = tmp } func (history *History) Add(container *Container) { *history = append(*history, container) sort.Sort(history) }