Remove execdriver package

Signed-off-by: Tonis Tiigi <tonistiigi@gmail.com>
This commit is contained in:
Tonis Tiigi 2016-03-18 11:51:59 -07:00
parent 9c4570a958
commit 6eebe85290
32 changed files with 0 additions and 3035 deletions

View file

@ -1,9 +0,0 @@
This directory contains code pertaining to running containers and storing images
Code pertaining to running containers:
- execdriver
Code pertaining to storing images:
- graphdriver

View file

@ -1,133 +0,0 @@
package execdriver
import (
"errors"
"io"
"os/exec"
"time"
"github.com/opencontainers/runc/libcontainer"
)
// Context is a generic key value pair that allows
// arbitrary data to be sent
type Context map[string]string
// Define error messages
var (
ErrNotRunning = errors.New("Container is not running")
ErrWaitTimeoutReached = errors.New("Wait timeout reached")
ErrDriverAlreadyRegistered = errors.New("A driver already registered this docker init function")
ErrDriverNotFound = errors.New("The requested docker init has not been found")
)
// DriverCallback defines a callback function which is used in "Run" and "Exec".
// This allows work to be done in the parent process when the child is passing
// through PreStart, Start and PostStop events.
// Callbacks are provided a processConfig pointer and the pid of the child.
// The channel will be used to notify the OOM events.
type DriverCallback func(processConfig *ProcessConfig, pid int, chOOM <-chan struct{}) error
// Hooks is a struct containing function pointers to callbacks
// used by any execdriver implementation exploiting hooks capabilities
type Hooks struct {
// PreStart is called before container's CMD/ENTRYPOINT is executed
PreStart []DriverCallback
// Start is called after the container's process is full started
Start DriverCallback
// PostStop is called after the container process exits
PostStop []DriverCallback
}
// Terminal represents a pseudo TTY, it is for when
// using a container interactively.
type Terminal interface {
io.Closer
Resize(height, width int) error
}
// Driver is an interface for drivers to implement
// including all basic functions a driver should have
type Driver interface {
// Run executes the process, blocks until the process exits and returns
// the exit code. It's the last stage on Docker side for running a container.
Run(c *Command, pipes *Pipes, hooks Hooks) (ExitStatus, error)
// Exec executes the process in an existing container, blocks until the
// process exits and returns the exit code.
Exec(c *Command, processConfig *ProcessConfig, pipes *Pipes, hooks Hooks) (int, error)
// Kill sends signals to process in container.
Kill(c *Command, sig int) error
// Pause pauses a container.
Pause(c *Command) error
// Unpause unpauses a container.
Unpause(c *Command) error
// Name returns the name of the driver.
Name() string
// GetPidsForContainer returns a list of pid for the processes running in a container.
GetPidsForContainer(id string) ([]int, error)
// Terminate kills a container by sending signal SIGKILL.
Terminate(c *Command) error
// Clean removes all traces of container exec.
Clean(id string) error
// Stats returns resource stats for a running container
Stats(id string) (*ResourceStats, error)
// Update updates resource configs for a container
Update(c *Command) error
// SupportsHooks refers to the driver capability to exploit pre/post hook functionality
SupportsHooks() bool
}
// CommonResources contains the resource configs for a driver that are
// common across platforms.
type CommonResources struct {
Memory int64 `json:"memory"`
MemoryReservation int64 `json:"memory_reservation"`
CPUShares int64 `json:"cpu_shares"`
BlkioWeight uint16 `json:"blkio_weight"`
}
// ResourceStats contains information about resource usage by a container.
type ResourceStats struct {
*libcontainer.Stats
Read time.Time `json:"read"`
MemoryLimit int64 `json:"memory_limit"`
SystemUsage uint64 `json:"system_usage"`
}
// CommonProcessConfig is the common platform agnostic part of the ProcessConfig
// structure that describes a process that will be run inside a container.
type CommonProcessConfig struct {
exec.Cmd `json:"-"`
Tty bool `json:"tty"`
Entrypoint string `json:"entrypoint"`
Arguments []string `json:"arguments"`
Terminal Terminal `json:"-"` // standard or tty terminal
}
// CommonCommand is the common platform agnostic part of the Command structure
// which wraps an os/exec.Cmd to add more metadata
type CommonCommand struct {
ContainerPid int `json:"container_pid"` // the pid for the process inside a container
ID string `json:"id"`
MountLabel string `json:"mount_label"` // TODO Windows. More involved, but can be factored out
Mounts []Mount `json:"mounts"`
Network *Network `json:"network"`
ProcessConfig ProcessConfig `json:"process_config"` // Describes the init process of the container.
ProcessLabel string `json:"process_label"` // TODO Windows. More involved, but can be factored out
Resources *Resources `json:"resources"`
Rootfs string `json:"rootfs"` // root fs of the container
WorkingDir string `json:"working_dir"`
TmpDir string `json:"tmpdir"` // Directory used to store docker tmpdirs.
}

View file

@ -1,323 +0,0 @@
// +build !windows
package execdriver
import (
"encoding/json"
"io/ioutil"
"os"
"path/filepath"
"strconv"
"strings"
"time"
"github.com/docker/docker/daemon/execdriver/native/template"
"github.com/docker/docker/pkg/idtools"
"github.com/docker/docker/pkg/mount"
"github.com/docker/go-units"
"github.com/opencontainers/runc/libcontainer"
"github.com/opencontainers/runc/libcontainer/cgroups/fs"
"github.com/opencontainers/runc/libcontainer/configs"
blkiodev "github.com/opencontainers/runc/libcontainer/configs"
)
// Mount contains information for a mount operation.
type Mount struct {
Source string `json:"source"`
Destination string `json:"destination"`
Writable bool `json:"writable"`
Data string `json:"data"`
Propagation string `json:"mountpropagation"`
}
// Resources contains all resource configs for a driver.
// Currently these are all for cgroup configs.
type Resources struct {
CommonResources
// Fields below here are platform specific
BlkioWeightDevice []*blkiodev.WeightDevice `json:"blkio_weight_device"`
BlkioThrottleReadBpsDevice []*blkiodev.ThrottleDevice `json:"blkio_throttle_read_bps_device"`
BlkioThrottleWriteBpsDevice []*blkiodev.ThrottleDevice `json:"blkio_throttle_write_bps_device"`
BlkioThrottleReadIOpsDevice []*blkiodev.ThrottleDevice `json:"blkio_throttle_read_iops_device"`
BlkioThrottleWriteIOpsDevice []*blkiodev.ThrottleDevice `json:"blkio_throttle_write_iops_device"`
MemorySwap int64 `json:"memory_swap"`
KernelMemory int64 `json:"kernel_memory"`
CPUQuota int64 `json:"cpu_quota"`
CpusetCpus string `json:"cpuset_cpus"`
CpusetMems string `json:"cpuset_mems"`
CPUPeriod int64 `json:"cpu_period"`
Rlimits []*units.Rlimit `json:"rlimits"`
OomKillDisable bool `json:"oom_kill_disable"`
PidsLimit int64 `json:"pids_limit"`
MemorySwappiness int64 `json:"memory_swappiness"`
}
// ProcessConfig is the platform specific structure that describes a process
// that will be run inside a container.
type ProcessConfig struct {
CommonProcessConfig
// Fields below here are platform specific
Privileged bool `json:"privileged"`
User string `json:"user"`
Console string `json:"-"` // dev/console path
}
// Ipc settings of the container
// It is for IPC namespace setting. Usually different containers
// have their own IPC namespace, however this specifies to use
// an existing IPC namespace.
// You can join the host's or a container's IPC namespace.
type Ipc struct {
ContainerID string `json:"container_id"` // id of the container to join ipc.
HostIpc bool `json:"host_ipc"`
}
// Pid settings of the container
// It is for PID namespace setting. Usually different containers
// have their own PID namespace, however this specifies to use
// an existing PID namespace.
// Joining the host's PID namespace is currently the only supported
// option.
type Pid struct {
HostPid bool `json:"host_pid"`
}
// UTS settings of the container
// It is for UTS namespace setting. Usually different containers
// have their own UTS namespace, however this specifies to use
// an existing UTS namespace.
// Joining the host's UTS namespace is currently the only supported
// option.
type UTS struct {
HostUTS bool `json:"host_uts"`
}
// Network settings of the container
type Network struct {
Mtu int `json:"mtu"`
ContainerID string `json:"container_id"` // id of the container to join network.
NamespacePath string `json:"namespace_path"`
HostNetworking bool `json:"host_networking"`
}
// Command wraps an os/exec.Cmd to add more metadata
type Command struct {
CommonCommand
// Fields below here are platform specific
AllowedDevices []*configs.Device `json:"allowed_devices"`
AppArmorProfile string `json:"apparmor_profile"`
AutoCreatedDevices []*configs.Device `json:"autocreated_devices"`
CapAdd []string `json:"cap_add"`
CapDrop []string `json:"cap_drop"`
CgroupParent string `json:"cgroup_parent"` // The parent cgroup for this command.
GIDMapping []idtools.IDMap `json:"gidmapping"`
GroupAdd []string `json:"group_add"`
Ipc *Ipc `json:"ipc"`
OomScoreAdj int `json:"oom_score_adj"`
Pid *Pid `json:"pid"`
ReadonlyRootfs bool `json:"readonly_rootfs"`
RemappedRoot *User `json:"remap_root"`
SeccompProfile string `json:"seccomp_profile"`
UIDMapping []idtools.IDMap `json:"uidmapping"`
UTS *UTS `json:"uts"`
NoNewPrivileges bool `json:"no_new_privileges"`
}
// SetRootPropagation sets the root mount propagation mode.
func SetRootPropagation(config *configs.Config, propagation int) {
config.RootPropagation = propagation
}
// InitContainer is the initialization of a container config.
// It returns the initial configs for a container. It's mostly
// defined by the default template.
func InitContainer(c *Command) *configs.Config {
container := template.New()
container.Hostname = getEnv("HOSTNAME", c.ProcessConfig.Env)
container.Cgroups.Name = c.ID
container.Cgroups.Resources.AllowedDevices = c.AllowedDevices
container.Devices = filterDevices(c.AutoCreatedDevices, (c.RemappedRoot.UID != 0))
container.Rootfs = c.Rootfs
container.Readonlyfs = c.ReadonlyRootfs
// This can be overridden later by driver during mount setup based
// on volume options
SetRootPropagation(container, mount.RPRIVATE)
container.Cgroups.Parent = c.CgroupParent
// check to see if we are running in ramdisk to disable pivot root
container.NoPivotRoot = os.Getenv("DOCKER_RAMDISK") != ""
return container
}
func filterDevices(devices []*configs.Device, userNamespacesEnabled bool) []*configs.Device {
if !userNamespacesEnabled {
return devices
}
filtered := []*configs.Device{}
// if we have user namespaces enabled, these devices will not be created
// because of the mknod limitation in the kernel for an unprivileged process.
// Rather, they will be bind-mounted, which will only work if they exist;
// check for existence and remove non-existent entries from the list
for _, device := range devices {
if _, err := os.Stat(device.Path); err == nil {
filtered = append(filtered, device)
}
}
return filtered
}
func getEnv(key string, env []string) string {
for _, pair := range env {
parts := strings.SplitN(pair, "=", 2)
if parts[0] == key {
return parts[1]
}
}
return ""
}
// SetupCgroups setups cgroup resources for a container.
func SetupCgroups(container *configs.Config, c *Command) error {
if c.Resources != nil {
container.Cgroups.Resources.CpuShares = c.Resources.CPUShares
container.Cgroups.Resources.Memory = c.Resources.Memory
container.Cgroups.Resources.MemoryReservation = c.Resources.MemoryReservation
container.Cgroups.Resources.MemorySwap = c.Resources.MemorySwap
container.Cgroups.Resources.KernelMemory = c.Resources.KernelMemory
container.Cgroups.Resources.CpusetCpus = c.Resources.CpusetCpus
container.Cgroups.Resources.CpusetMems = c.Resources.CpusetMems
container.Cgroups.Resources.CpuPeriod = c.Resources.CPUPeriod
container.Cgroups.Resources.CpuQuota = c.Resources.CPUQuota
container.Cgroups.Resources.BlkioWeight = c.Resources.BlkioWeight
container.Cgroups.Resources.BlkioWeightDevice = c.Resources.BlkioWeightDevice
container.Cgroups.Resources.BlkioThrottleReadBpsDevice = c.Resources.BlkioThrottleReadBpsDevice
container.Cgroups.Resources.BlkioThrottleWriteBpsDevice = c.Resources.BlkioThrottleWriteBpsDevice
container.Cgroups.Resources.BlkioThrottleReadIOPSDevice = c.Resources.BlkioThrottleReadIOpsDevice
container.Cgroups.Resources.BlkioThrottleWriteIOPSDevice = c.Resources.BlkioThrottleWriteIOpsDevice
container.Cgroups.Resources.OomKillDisable = c.Resources.OomKillDisable
container.Cgroups.Resources.PidsLimit = c.Resources.PidsLimit
container.Cgroups.Resources.MemorySwappiness = c.Resources.MemorySwappiness
}
return nil
}
// Returns the network statistics for the network interfaces represented by the NetworkRuntimeInfo.
func getNetworkInterfaceStats(interfaceName string) (*libcontainer.NetworkInterface, error) {
out := &libcontainer.NetworkInterface{Name: interfaceName}
// This can happen if the network runtime information is missing - possible if the
// container was created by an old version of libcontainer.
if interfaceName == "" {
return out, nil
}
type netStatsPair struct {
// Where to write the output.
Out *uint64
// The network stats file to read.
File string
}
// Ingress for host veth is from the container. Hence tx_bytes stat on the host veth is actually number of bytes received by the container.
netStats := []netStatsPair{
{Out: &out.RxBytes, File: "tx_bytes"},
{Out: &out.RxPackets, File: "tx_packets"},
{Out: &out.RxErrors, File: "tx_errors"},
{Out: &out.RxDropped, File: "tx_dropped"},
{Out: &out.TxBytes, File: "rx_bytes"},
{Out: &out.TxPackets, File: "rx_packets"},
{Out: &out.TxErrors, File: "rx_errors"},
{Out: &out.TxDropped, File: "rx_dropped"},
}
for _, netStat := range netStats {
data, err := readSysfsNetworkStats(interfaceName, netStat.File)
if err != nil {
return nil, err
}
*(netStat.Out) = data
}
return out, nil
}
// Reads the specified statistics available under /sys/class/net/<EthInterface>/statistics
func readSysfsNetworkStats(ethInterface, statsFile string) (uint64, error) {
data, err := ioutil.ReadFile(filepath.Join("/sys/class/net", ethInterface, "statistics", statsFile))
if err != nil {
return 0, err
}
return strconv.ParseUint(strings.TrimSpace(string(data)), 10, 64)
}
// Stats collects all the resource usage information from a container.
func Stats(containerDir string, containerMemoryLimit int64, machineMemory int64) (*ResourceStats, error) {
f, err := os.Open(filepath.Join(containerDir, "state.json"))
if err != nil {
return nil, err
}
defer f.Close()
type network struct {
Type string
HostInterfaceName string
}
state := struct {
CgroupPaths map[string]string `json:"cgroup_paths"`
Networks []network
}{}
if err := json.NewDecoder(f).Decode(&state); err != nil {
return nil, err
}
now := time.Now()
mgr := fs.Manager{Paths: state.CgroupPaths}
cstats, err := mgr.GetStats()
if err != nil {
return nil, err
}
stats := &libcontainer.Stats{CgroupStats: cstats}
// if the container does not have any memory limit specified set the
// limit to the machines memory
memoryLimit := containerMemoryLimit
if memoryLimit == 0 {
memoryLimit = machineMemory
}
for _, iface := range state.Networks {
switch iface.Type {
case "veth":
istats, err := getNetworkInterfaceStats(iface.HostInterfaceName)
if err != nil {
return nil, err
}
stats.Interfaces = append(stats.Interfaces, istats)
}
}
return &ResourceStats{
Stats: stats,
Read: now,
MemoryLimit: memoryLimit,
}, nil
}
// User contains the uid and gid representing a Unix user
type User struct {
UID int `json:"root_uid"`
GID int `json:"root_gid"`
}
// ExitStatus provides exit reasons for a container.
type ExitStatus struct {
// The exit code with which the container exited.
ExitCode int
// Whether the container encountered an OOM.
OOMKilled bool
}

View file

@ -1,66 +0,0 @@
package execdriver
import "github.com/docker/go-connections/nat"
// Mount contains information for a mount operation.
type Mount struct {
Source string `json:"source"`
Destination string `json:"destination"`
Writable bool `json:"writable"`
}
// Resources contains all resource configs for a driver.
// Currently these are all for cgroup configs.
type Resources struct {
CommonResources
// Fields below here are platform specific
}
// ProcessConfig is the platform specific structure that describes a process
// that will be run inside a container.
type ProcessConfig struct {
CommonProcessConfig
// Fields below here are platform specific
ConsoleSize [2]int `json:"-"` // h,w of initial console size
}
// Network settings of the container
type Network struct {
Interface *NetworkInterface `json:"interface"`
ContainerID string `json:"container_id"` // id of the container to join network.
}
// NetworkInterface contains network configs for a driver
type NetworkInterface struct {
MacAddress string `json:"mac"`
Bridge string `json:"bridge"`
IPAddress string `json:"ip"`
// PortBindings is the port mapping between the exposed port in the
// container and the port on the host.
PortBindings nat.PortMap `json:"port_bindings"`
}
// Command wraps an os/exec.Cmd to add more metadata
type Command struct {
CommonCommand
// Fields below here are platform specific
FirstStart bool `json:"first_start"` // Optimization for first boot of Windows
Hostname string `json:"hostname"` // Windows sets the hostname in the execdriver
LayerFolder string `json:"layer_folder"` // Layer folder for a command
LayerPaths []string `json:"layer_paths"` // Layer paths for a command
Isolation string `json:"isolation"` // Isolation technology for the container
ArgsEscaped bool `json:"args_escaped"` // True if args are already escaped
HvPartition bool `json:"hv_partition"` // True if it's an hypervisor partition
EpList []string `json:"endpoints"` // List of network endpoints for HNS
}
// ExitStatus provides exit reasons for a container.
type ExitStatus struct {
// The exit code with which the container exited.
ExitCode int
}

View file

@ -1,15 +0,0 @@
// +build freebsd
package execdrivers
import (
"fmt"
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/pkg/sysinfo"
)
// NewDriver returns a new execdriver.Driver from the given name configured with the provided options.
func NewDriver(options []string, root, libPath string, sysInfo *sysinfo.SysInfo) (execdriver.Driver, error) {
return nil, fmt.Errorf("jail driver not yet supported on FreeBSD")
}

View file

@ -1,16 +0,0 @@
// +build linux
package execdrivers
import (
"path"
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/daemon/execdriver/native"
"github.com/docker/docker/pkg/sysinfo"
)
// NewDriver returns a new execdriver.Driver from the given name configured with the provided options.
func NewDriver(options []string, root, libPath string, sysInfo *sysinfo.SysInfo) (execdriver.Driver, error) {
return native.NewDriver(path.Join(root, "execdriver", "native"), options)
}

View file

@ -1,14 +0,0 @@
// +build windows
package execdrivers
import (
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/daemon/execdriver/windows"
"github.com/docker/docker/pkg/sysinfo"
)
// NewDriver returns a new execdriver.Driver from the given name configured with the provided options.
func NewDriver(options []string, root, libPath string, sysInfo *sysinfo.SysInfo) (execdriver.Driver, error) {
return windows.NewDriver(root, options)
}

View file

@ -1,514 +0,0 @@
// +build linux,cgo
package native
import (
"fmt"
"path/filepath"
"strings"
"syscall"
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/pkg/mount"
"github.com/docker/docker/profiles/seccomp"
"github.com/docker/docker/volume"
"github.com/opencontainers/runc/libcontainer/apparmor"
"github.com/opencontainers/runc/libcontainer/configs"
"github.com/opencontainers/runc/libcontainer/devices"
)
// createContainer populates and configures the container type with the
// data provided by the execdriver.Command
func (d *Driver) createContainer(c *execdriver.Command, hooks execdriver.Hooks) (container *configs.Config, err error) {
container = execdriver.InitContainer(c)
if err := d.createIpc(container, c); err != nil {
return nil, err
}
if err := d.createPid(container, c); err != nil {
return nil, err
}
if err := d.createUTS(container, c); err != nil {
return nil, err
}
if err := d.setupRemappedRoot(container, c); err != nil {
return nil, err
}
if err := d.createNetwork(container, c, hooks); err != nil {
return nil, err
}
if c.ProcessConfig.Privileged {
if !container.Readonlyfs {
// clear readonly for /sys
for i := range container.Mounts {
if container.Mounts[i].Destination == "/sys" {
container.Mounts[i].Flags &= ^syscall.MS_RDONLY
}
}
container.ReadonlyPaths = nil
}
// clear readonly for cgroup
for i := range container.Mounts {
if container.Mounts[i].Device == "cgroup" {
container.Mounts[i].Flags &= ^syscall.MS_RDONLY
}
}
container.MaskPaths = nil
if err := d.setPrivileged(container); err != nil {
return nil, err
}
} else {
if err := d.setCapabilities(container, c); err != nil {
return nil, err
}
if c.SeccompProfile == "" {
container.Seccomp, err = seccomp.GetDefaultProfile()
if err != nil {
return nil, err
}
}
}
// add CAP_ prefix to all caps for new libcontainer update to match
// the spec format.
for i, s := range container.Capabilities {
if !strings.HasPrefix(s, "CAP_") {
container.Capabilities[i] = fmt.Sprintf("CAP_%s", s)
}
}
container.AdditionalGroups = c.GroupAdd
if c.AppArmorProfile != "" {
container.AppArmorProfile = c.AppArmorProfile
}
if c.SeccompProfile != "" && c.SeccompProfile != "unconfined" {
container.Seccomp, err = seccomp.LoadProfile(c.SeccompProfile)
if err != nil {
return nil, err
}
}
if err := execdriver.SetupCgroups(container, c); err != nil {
return nil, err
}
container.OomScoreAdj = c.OomScoreAdj
if container.Readonlyfs {
for i := range container.Mounts {
switch container.Mounts[i].Destination {
case "/proc", "/dev", "/dev/pts", "/dev/mqueue":
continue
}
container.Mounts[i].Flags |= syscall.MS_RDONLY
}
/* These paths must be remounted as r/o */
container.ReadonlyPaths = append(container.ReadonlyPaths, "/dev")
}
if err := d.setupMounts(container, c); err != nil {
return nil, err
}
d.setupLabels(container, c)
d.setupRlimits(container, c)
container.NoNewPrivileges = c.NoNewPrivileges
return container, nil
}
func (d *Driver) createNetwork(container *configs.Config, c *execdriver.Command, hooks execdriver.Hooks) error {
if c.Network == nil {
return nil
}
if c.Network.ContainerID != "" {
d.Lock()
active := d.activeContainers[c.Network.ContainerID]
d.Unlock()
if active == nil {
return fmt.Errorf("%s is not a valid running container to join", c.Network.ContainerID)
}
state, err := active.State()
if err != nil {
return err
}
container.Namespaces.Add(configs.NEWNET, state.NamespacePaths[configs.NEWNET])
return nil
}
if c.Network.NamespacePath != "" {
container.Namespaces.Add(configs.NEWNET, c.Network.NamespacePath)
return nil
}
// only set up prestart hook if the namespace path is not set (this should be
// all cases *except* for --net=host shared networking)
container.Hooks = &configs.Hooks{
Prestart: []configs.Hook{
configs.NewFunctionHook(func(s configs.HookState) error {
if len(hooks.PreStart) > 0 {
for _, fnHook := range hooks.PreStart {
// A closed channel for OOM is returned here as it will be
// non-blocking and return the correct result when read.
chOOM := make(chan struct{})
close(chOOM)
if err := fnHook(&c.ProcessConfig, s.Pid, chOOM); err != nil {
return err
}
}
}
return nil
}),
},
}
return nil
}
func (d *Driver) createIpc(container *configs.Config, c *execdriver.Command) error {
if c.Ipc.HostIpc {
container.Namespaces.Remove(configs.NEWIPC)
return nil
}
if c.Ipc.ContainerID != "" {
d.Lock()
active := d.activeContainers[c.Ipc.ContainerID]
d.Unlock()
if active == nil {
return fmt.Errorf("%s is not a valid running container to join", c.Ipc.ContainerID)
}
state, err := active.State()
if err != nil {
return err
}
container.Namespaces.Add(configs.NEWIPC, state.NamespacePaths[configs.NEWIPC])
}
return nil
}
func (d *Driver) createPid(container *configs.Config, c *execdriver.Command) error {
if c.Pid.HostPid {
container.Namespaces.Remove(configs.NEWPID)
return nil
}
return nil
}
func (d *Driver) createUTS(container *configs.Config, c *execdriver.Command) error {
if c.UTS.HostUTS {
container.Namespaces.Remove(configs.NEWUTS)
container.Hostname = ""
return nil
}
return nil
}
func (d *Driver) setupRemappedRoot(container *configs.Config, c *execdriver.Command) error {
if c.RemappedRoot.UID == 0 {
container.Namespaces.Remove(configs.NEWUSER)
return nil
}
// convert the Docker daemon id map to the libcontainer variant of the same struct
// this keeps us from having to import libcontainer code across Docker client + daemon packages
cuidMaps := []configs.IDMap{}
cgidMaps := []configs.IDMap{}
for _, idMap := range c.UIDMapping {
cuidMaps = append(cuidMaps, configs.IDMap(idMap))
}
for _, idMap := range c.GIDMapping {
cgidMaps = append(cgidMaps, configs.IDMap(idMap))
}
container.UidMappings = cuidMaps
container.GidMappings = cgidMaps
for _, node := range container.Devices {
node.Uid = uint32(c.RemappedRoot.UID)
node.Gid = uint32(c.RemappedRoot.GID)
}
// TODO: until a kernel/mount solution exists for handling remount in a user namespace,
// we must clear the readonly flag for the cgroups mount (@mrunalp concurs)
for i := range container.Mounts {
if container.Mounts[i].Device == "cgroup" {
container.Mounts[i].Flags &= ^syscall.MS_RDONLY
}
}
return nil
}
func (d *Driver) setPrivileged(container *configs.Config) (err error) {
container.Capabilities = execdriver.GetAllCapabilities()
container.Cgroups.Resources.AllowAllDevices = true
hostDevices, err := devices.HostDevices()
if err != nil {
return err
}
container.Devices = hostDevices
if apparmor.IsEnabled() {
container.AppArmorProfile = "unconfined"
}
return nil
}
func (d *Driver) setCapabilities(container *configs.Config, c *execdriver.Command) (err error) {
container.Capabilities, err = execdriver.TweakCapabilities(container.Capabilities, c.CapAdd, c.CapDrop)
return err
}
func (d *Driver) setupRlimits(container *configs.Config, c *execdriver.Command) {
if c.Resources == nil {
return
}
for _, rlimit := range c.Resources.Rlimits {
container.Rlimits = append(container.Rlimits, configs.Rlimit{
Type: rlimit.Type,
Hard: rlimit.Hard,
Soft: rlimit.Soft,
})
}
}
// If rootfs mount propagation is RPRIVATE, that means all the volumes are
// going to be private anyway. There is no need to apply per volume
// propagation on top. This is just an optimization so that cost of per volume
// propagation is paid only if user decides to make some volume non-private
// which will force rootfs mount propagation to be non RPRIVATE.
func checkResetVolumePropagation(container *configs.Config) {
if container.RootPropagation != mount.RPRIVATE {
return
}
for _, m := range container.Mounts {
m.PropagationFlags = nil
}
}
func getMountInfo(mountinfo []*mount.Info, dir string) *mount.Info {
for _, m := range mountinfo {
if m.Mountpoint == dir {
return m
}
}
return nil
}
// Get the source mount point of directory passed in as argument. Also return
// optional fields.
func getSourceMount(source string) (string, string, error) {
// Ensure any symlinks are resolved.
sourcePath, err := filepath.EvalSymlinks(source)
if err != nil {
return "", "", err
}
mountinfos, err := mount.GetMounts()
if err != nil {
return "", "", err
}
mountinfo := getMountInfo(mountinfos, sourcePath)
if mountinfo != nil {
return sourcePath, mountinfo.Optional, nil
}
path := sourcePath
for {
path = filepath.Dir(path)
mountinfo = getMountInfo(mountinfos, path)
if mountinfo != nil {
return path, mountinfo.Optional, nil
}
if path == "/" {
break
}
}
// If we are here, we did not find parent mount. Something is wrong.
return "", "", fmt.Errorf("Could not find source mount of %s", source)
}
// Ensure mount point on which path is mounted, is shared.
func ensureShared(path string) error {
sharedMount := false
sourceMount, optionalOpts, err := getSourceMount(path)
if err != nil {
return err
}
// Make sure source mount point is shared.
optsSplit := strings.Split(optionalOpts, " ")
for _, opt := range optsSplit {
if strings.HasPrefix(opt, "shared:") {
sharedMount = true
break
}
}
if !sharedMount {
return fmt.Errorf("Path %s is mounted on %s but it is not a shared mount.", path, sourceMount)
}
return nil
}
// Ensure mount point on which path is mounted, is either shared or slave.
func ensureSharedOrSlave(path string) error {
sharedMount := false
slaveMount := false
sourceMount, optionalOpts, err := getSourceMount(path)
if err != nil {
return err
}
// Make sure source mount point is shared.
optsSplit := strings.Split(optionalOpts, " ")
for _, opt := range optsSplit {
if strings.HasPrefix(opt, "shared:") {
sharedMount = true
break
} else if strings.HasPrefix(opt, "master:") {
slaveMount = true
break
}
}
if !sharedMount && !slaveMount {
return fmt.Errorf("Path %s is mounted on %s but it is not a shared or slave mount.", path, sourceMount)
}
return nil
}
func (d *Driver) setupMounts(container *configs.Config, c *execdriver.Command) error {
userMounts := make(map[string]struct{})
for _, m := range c.Mounts {
userMounts[m.Destination] = struct{}{}
}
// Filter out mounts that are overridden by user supplied mounts
var defaultMounts []*configs.Mount
_, mountDev := userMounts["/dev"]
for _, m := range container.Mounts {
if _, ok := userMounts[m.Destination]; !ok {
if mountDev && strings.HasPrefix(m.Destination, "/dev/") {
container.Devices = nil
continue
}
defaultMounts = append(defaultMounts, m)
}
}
container.Mounts = defaultMounts
mountPropagationMap := map[string]int{
"private": mount.PRIVATE,
"rprivate": mount.RPRIVATE,
"shared": mount.SHARED,
"rshared": mount.RSHARED,
"slave": mount.SLAVE,
"rslave": mount.RSLAVE,
}
for _, m := range c.Mounts {
for _, cm := range container.Mounts {
if cm.Destination == m.Destination {
return fmt.Errorf("Duplicate mount point '%s'", m.Destination)
}
}
if m.Source == "tmpfs" {
var (
data = "size=65536k"
flags = syscall.MS_NOEXEC | syscall.MS_NOSUID | syscall.MS_NODEV
err error
)
if m.Data != "" {
flags, data, err = mount.ParseTmpfsOptions(m.Data)
if err != nil {
return err
}
}
container.Mounts = append(container.Mounts, &configs.Mount{
Source: m.Source,
Destination: m.Destination,
Data: data,
Device: "tmpfs",
Flags: flags,
PropagationFlags: []int{mountPropagationMap[volume.DefaultPropagationMode]},
})
continue
}
flags := syscall.MS_BIND | syscall.MS_REC
var pFlag int
if !m.Writable {
flags |= syscall.MS_RDONLY
}
// Determine property of RootPropagation based on volume
// properties. If a volume is shared, then keep root propagation
// shared. This should work for slave and private volumes too.
//
// For slave volumes, it can be either [r]shared/[r]slave.
//
// For private volumes any root propagation value should work.
pFlag = mountPropagationMap[m.Propagation]
if pFlag == mount.SHARED || pFlag == mount.RSHARED {
if err := ensureShared(m.Source); err != nil {
return err
}
rootpg := container.RootPropagation
if rootpg != mount.SHARED && rootpg != mount.RSHARED {
execdriver.SetRootPropagation(container, mount.SHARED)
}
} else if pFlag == mount.SLAVE || pFlag == mount.RSLAVE {
if err := ensureSharedOrSlave(m.Source); err != nil {
return err
}
rootpg := container.RootPropagation
if rootpg != mount.SHARED && rootpg != mount.RSHARED && rootpg != mount.SLAVE && rootpg != mount.RSLAVE {
execdriver.SetRootPropagation(container, mount.RSLAVE)
}
}
mount := &configs.Mount{
Source: m.Source,
Destination: m.Destination,
Device: "bind",
Flags: flags,
}
if pFlag != 0 {
mount.PropagationFlags = []int{pFlag}
}
container.Mounts = append(container.Mounts, mount)
}
checkResetVolumePropagation(container)
return nil
}
func (d *Driver) setupLabels(container *configs.Config, c *execdriver.Command) {
container.ProcessLabel = c.ProcessLabel
container.MountLabel = c.MountLabel
}

View file

@ -1,606 +0,0 @@
// +build linux,cgo
package native
import (
"fmt"
"io"
"io/ioutil"
"os"
"os/exec"
"path/filepath"
"strings"
"sync"
"syscall"
"time"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/pkg/parsers"
"github.com/docker/docker/pkg/pools"
"github.com/docker/docker/pkg/reexec"
sysinfo "github.com/docker/docker/pkg/system"
"github.com/docker/docker/pkg/term"
aaprofile "github.com/docker/docker/profiles/apparmor"
"github.com/opencontainers/runc/libcontainer"
"github.com/opencontainers/runc/libcontainer/apparmor"
"github.com/opencontainers/runc/libcontainer/cgroups/systemd"
"github.com/opencontainers/runc/libcontainer/configs"
"github.com/opencontainers/runc/libcontainer/system"
"github.com/opencontainers/runc/libcontainer/utils"
)
// Define constants for native driver
const (
DriverName = "native"
Version = "0.2"
defaultApparmorProfile = "docker-default"
)
// Driver contains all information for native driver,
// it implements execdriver.Driver.
type Driver struct {
root string
activeContainers map[string]libcontainer.Container
machineMemory int64
factory libcontainer.Factory
sync.Mutex
}
// NewDriver returns a new native driver, called from NewDriver of execdriver.
func NewDriver(root string, options []string) (*Driver, error) {
meminfo, err := sysinfo.ReadMemInfo()
if err != nil {
return nil, err
}
if err := sysinfo.MkdirAll(root, 0700); err != nil {
return nil, err
}
if apparmor.IsEnabled() {
if err := aaprofile.InstallDefault(defaultApparmorProfile); err != nil {
apparmorProfiles := []string{defaultApparmorProfile}
// Allow daemon to run if loading failed, but are active
// (possibly through another run, manually, or via system startup)
for _, policy := range apparmorProfiles {
if err := aaprofile.IsLoaded(policy); err != nil {
return nil, fmt.Errorf("AppArmor enabled on system but the %s profile could not be loaded.", policy)
}
}
}
}
// choose cgroup manager
// this makes sure there are no breaking changes to people
// who upgrade from versions without native.cgroupdriver opt
cgm := libcontainer.Cgroupfs
// parse the options
for _, option := range options {
key, val, err := parsers.ParseKeyValueOpt(option)
if err != nil {
return nil, err
}
key = strings.ToLower(key)
switch key {
case "native.cgroupdriver":
// override the default if they set options
switch val {
case "systemd":
if systemd.UseSystemd() {
cgm = libcontainer.SystemdCgroups
} else {
// warn them that they chose the wrong driver
logrus.Warn("You cannot use systemd as native.cgroupdriver, using cgroupfs instead")
}
case "cgroupfs":
cgm = libcontainer.Cgroupfs
default:
return nil, fmt.Errorf("Unknown native.cgroupdriver given %q. try cgroupfs or systemd", val)
}
default:
return nil, fmt.Errorf("Unknown option %s\n", key)
}
}
f, err := libcontainer.New(
root,
cgm,
libcontainer.InitPath(reexec.Self(), DriverName),
)
if err != nil {
return nil, err
}
return &Driver{
root: root,
activeContainers: make(map[string]libcontainer.Container),
machineMemory: meminfo.MemTotal,
factory: f,
}, nil
}
// Run implements the exec driver Driver interface,
// it calls libcontainer APIs to run a container.
func (d *Driver) Run(c *execdriver.Command, pipes *execdriver.Pipes, hooks execdriver.Hooks) (execdriver.ExitStatus, error) {
destroyed := false
var err error
c.TmpDir, err = ioutil.TempDir("", c.ID)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
defer os.RemoveAll(c.TmpDir)
// take the Command and populate the libcontainer.Config from it
container, err := d.createContainer(c, hooks)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
p := &libcontainer.Process{
Args: append([]string{c.ProcessConfig.Entrypoint}, c.ProcessConfig.Arguments...),
Env: c.ProcessConfig.Env,
Cwd: c.WorkingDir,
User: c.ProcessConfig.User,
}
wg := sync.WaitGroup{}
writers, err := setupPipes(container, &c.ProcessConfig, p, pipes, &wg)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
cont, err := d.factory.Create(c.ID, container)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
if err := cont.Start(p); err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
d.Lock()
d.activeContainers[c.ID] = cont
d.Unlock()
defer func() {
if !destroyed {
cont.Destroy()
}
d.cleanContainer(c.ID)
}()
//close the write end of any opened pipes now that they are dup'ed into the container
for _, writer := range writers {
writer.Close()
}
// 'oom' is used to emit 'oom' events to the eventstream, 'oomKilled' is used
// to set the 'OOMKilled' flag in state
oom := notifyOnOOM(cont)
oomKilled := notifyOnOOM(cont)
if hooks.Start != nil {
pid, err := p.Pid()
if err != nil {
p.Signal(os.Kill)
p.Wait()
return execdriver.ExitStatus{ExitCode: -1}, err
}
hooks.Start(&c.ProcessConfig, pid, oom)
}
waitF := p.Wait
if nss := cont.Config().Namespaces; !nss.Contains(configs.NEWPID) {
// we need such hack for tracking processes with inherited fds,
// because cmd.Wait() waiting for all streams to be copied
waitF = waitInPIDHost(p, cont)
}
ps, err := waitF()
if err != nil {
execErr, ok := err.(*exec.ExitError)
if !ok {
return execdriver.ExitStatus{ExitCode: -1}, err
}
ps = execErr.ProcessState
}
// wait for all IO goroutine copiers to finish
wg.Wait()
cont.Destroy()
destroyed = true
// oomKilled will have an oom event if any process within the container was
// OOM killed at any time, not only if the init process OOMed.
//
// Perhaps we only want the OOMKilled flag to be set if the OOM
// resulted in a container death, but there isn't a good way to do this
// because the kernel's cgroup oom notification does not provide information
// such as the PID. This could be heuristically done by checking that the OOM
// happened within some very small time slice for the container dying (and
// optionally exit-code 137), but I don't think the cgroup oom notification
// can be used to reliably determine this
//
// Even if there were multiple OOMs, it's sufficient to read one value
// because libcontainer's oom notify will discard the channel after the
// cgroup is destroyed
_, oomKill := <-oomKilled
return execdriver.ExitStatus{ExitCode: utils.ExitStatus(ps.Sys().(syscall.WaitStatus)), OOMKilled: oomKill}, nil
}
// notifyOnOOM returns a channel that signals if the container received an OOM notification
// for any process. If it is unable to subscribe to OOM notifications then a closed
// channel is returned as it will be non-blocking and return the correct result when read.
func notifyOnOOM(container libcontainer.Container) <-chan struct{} {
oom, err := container.NotifyOOM()
if err != nil {
logrus.Warnf("Your kernel does not support OOM notifications: %s", err)
c := make(chan struct{})
close(c)
return c
}
return oom
}
func killCgroupProcs(c libcontainer.Container) {
var procs []*os.Process
if err := c.Pause(); err != nil {
logrus.Warn(err)
}
pids, err := c.Processes()
if err != nil {
// don't care about childs if we can't get them, this is mostly because cgroup already deleted
logrus.Warnf("Failed to get processes from container %s: %v", c.ID(), err)
}
for _, pid := range pids {
if p, err := os.FindProcess(pid); err == nil {
procs = append(procs, p)
if err := p.Kill(); err != nil {
logrus.Warn(err)
}
}
}
if err := c.Resume(); err != nil {
logrus.Warn(err)
}
for _, p := range procs {
if _, err := p.Wait(); err != nil {
logrus.Warn(err)
}
}
}
func waitInPIDHost(p *libcontainer.Process, c libcontainer.Container) func() (*os.ProcessState, error) {
return func() (*os.ProcessState, error) {
pid, err := p.Pid()
if err != nil {
return nil, err
}
process, err := os.FindProcess(pid)
s, err := process.Wait()
if err != nil {
execErr, ok := err.(*exec.ExitError)
if !ok {
return s, err
}
s = execErr.ProcessState
}
killCgroupProcs(c)
p.Wait()
return s, err
}
}
// Kill implements the exec driver Driver interface.
func (d *Driver) Kill(c *execdriver.Command, sig int) error {
d.Lock()
active := d.activeContainers[c.ID]
d.Unlock()
if active == nil {
return fmt.Errorf("active container for %s does not exist", c.ID)
}
state, err := active.State()
if err != nil {
return err
}
if state.InitProcessPid == -1 {
return fmt.Errorf("avoid sending signal %d to container %s with pid -1", sig, c.ID)
}
return syscall.Kill(state.InitProcessPid, syscall.Signal(sig))
}
// Pause implements the exec driver Driver interface,
// it calls libcontainer API to pause a container.
func (d *Driver) Pause(c *execdriver.Command) error {
d.Lock()
active := d.activeContainers[c.ID]
d.Unlock()
if active == nil {
return fmt.Errorf("active container for %s does not exist", c.ID)
}
return active.Pause()
}
// Unpause implements the exec driver Driver interface,
// it calls libcontainer API to unpause a container.
func (d *Driver) Unpause(c *execdriver.Command) error {
d.Lock()
active := d.activeContainers[c.ID]
d.Unlock()
if active == nil {
return fmt.Errorf("active container for %s does not exist", c.ID)
}
return active.Resume()
}
// Terminate implements the exec driver Driver interface.
func (d *Driver) Terminate(c *execdriver.Command) error {
defer d.cleanContainer(c.ID)
container, err := d.factory.Load(c.ID)
if err != nil {
return err
}
defer container.Destroy()
state, err := container.State()
if err != nil {
return err
}
pid := state.InitProcessPid
currentStartTime, err := system.GetProcessStartTime(pid)
if err != nil {
return err
}
if state.InitProcessStartTime == currentStartTime {
err = syscall.Kill(pid, 9)
syscall.Wait4(pid, nil, 0, nil)
}
return err
}
// Name implements the exec driver Driver interface.
func (d *Driver) Name() string {
return fmt.Sprintf("%s-%s", DriverName, Version)
}
// GetPidsForContainer implements the exec driver Driver interface.
func (d *Driver) GetPidsForContainer(id string) ([]int, error) {
d.Lock()
active := d.activeContainers[id]
d.Unlock()
if active == nil {
return nil, fmt.Errorf("active container for %s does not exist", id)
}
return active.Processes()
}
func (d *Driver) cleanContainer(id string) error {
d.Lock()
delete(d.activeContainers, id)
d.Unlock()
return os.RemoveAll(filepath.Join(d.root, id))
}
func (d *Driver) createContainerRoot(id string) error {
return os.MkdirAll(filepath.Join(d.root, id), 0655)
}
// Clean implements the exec driver Driver interface.
func (d *Driver) Clean(id string) error {
return os.RemoveAll(filepath.Join(d.root, id))
}
// Stats implements the exec driver Driver interface.
func (d *Driver) Stats(id string) (*execdriver.ResourceStats, error) {
d.Lock()
c := d.activeContainers[id]
d.Unlock()
if c == nil {
return nil, execdriver.ErrNotRunning
}
now := time.Now()
stats, err := c.Stats()
if err != nil {
return nil, err
}
memoryLimit := c.Config().Cgroups.Resources.Memory
// if the container does not have any memory limit specified set the
// limit to the machines memory
if memoryLimit == 0 {
memoryLimit = d.machineMemory
}
return &execdriver.ResourceStats{
Stats: stats,
Read: now,
MemoryLimit: memoryLimit,
}, nil
}
// Update updates configs for a container
func (d *Driver) Update(c *execdriver.Command) error {
d.Lock()
cont := d.activeContainers[c.ID]
d.Unlock()
if cont == nil {
return execdriver.ErrNotRunning
}
config := cont.Config()
if err := execdriver.SetupCgroups(&config, c); err != nil {
return err
}
if err := cont.Set(config); err != nil {
return err
}
return nil
}
// TtyConsole implements the exec driver Terminal interface.
type TtyConsole struct {
console libcontainer.Console
}
// NewTtyConsole returns a new TtyConsole struct.
func NewTtyConsole(console libcontainer.Console, pipes *execdriver.Pipes, wg *sync.WaitGroup) (*TtyConsole, error) {
tty := &TtyConsole{
console: console,
}
if err := tty.AttachPipes(pipes, wg); err != nil {
tty.Close()
return nil, err
}
return tty, nil
}
// Resize implements Resize method of Terminal interface
func (t *TtyConsole) Resize(h, w int) error {
return term.SetWinsize(t.console.Fd(), &term.Winsize{Height: uint16(h), Width: uint16(w)})
}
// AttachPipes attaches given pipes to TtyConsole
func (t *TtyConsole) AttachPipes(pipes *execdriver.Pipes, wg *sync.WaitGroup) error {
wg.Add(1)
go func() {
defer wg.Done()
if wb, ok := pipes.Stdout.(interface {
CloseWriters() error
}); ok {
defer wb.CloseWriters()
}
pools.Copy(pipes.Stdout, t.console)
}()
if pipes.Stdin != nil {
go func() {
pools.Copy(t.console, pipes.Stdin)
pipes.Stdin.Close()
}()
}
return nil
}
// Close implements Close method of Terminal interface
func (t *TtyConsole) Close() error {
return t.console.Close()
}
func setupPipes(container *configs.Config, processConfig *execdriver.ProcessConfig, p *libcontainer.Process, pipes *execdriver.Pipes, wg *sync.WaitGroup) ([]io.WriteCloser, error) {
writers := []io.WriteCloser{}
rootuid, err := container.HostUID()
if err != nil {
return writers, err
}
if processConfig.Tty {
cons, err := p.NewConsole(rootuid)
if err != nil {
return writers, err
}
term, err := NewTtyConsole(cons, pipes, wg)
if err != nil {
return writers, err
}
processConfig.Terminal = term
return writers, nil
}
// not a tty--set up stdio pipes
term := &execdriver.StdConsole{}
processConfig.Terminal = term
// if we are not in a user namespace, there is no reason to go through
// the hassle of setting up os-level pipes with proper (remapped) ownership
// so we will do the prior shortcut for non-userns containers
if rootuid == 0 {
p.Stdout = pipes.Stdout
p.Stderr = pipes.Stderr
r, w, err := os.Pipe()
if err != nil {
return writers, err
}
if pipes.Stdin != nil {
go func() {
io.Copy(w, pipes.Stdin)
w.Close()
}()
p.Stdin = r
}
return writers, nil
}
// if we have user namespaces enabled (rootuid != 0), we will set
// up os pipes for stderr, stdout, stdin so we can chown them to
// the proper ownership to allow for proper access to the underlying
// fds
var fds []uintptr
copyPipes := func(out io.Writer, in io.ReadCloser) {
defer wg.Done()
io.Copy(out, in)
in.Close()
}
//setup stdout
r, w, err := os.Pipe()
if err != nil {
w.Close()
return writers, err
}
writers = append(writers, w)
fds = append(fds, r.Fd(), w.Fd())
if pipes.Stdout != nil {
wg.Add(1)
go copyPipes(pipes.Stdout, r)
}
term.Closers = append(term.Closers, r)
p.Stdout = w
//setup stderr
r, w, err = os.Pipe()
if err != nil {
w.Close()
return writers, err
}
writers = append(writers, w)
fds = append(fds, r.Fd(), w.Fd())
if pipes.Stderr != nil {
wg.Add(1)
go copyPipes(pipes.Stderr, r)
}
term.Closers = append(term.Closers, r)
p.Stderr = w
//setup stdin
r, w, err = os.Pipe()
if err != nil {
r.Close()
return writers, err
}
fds = append(fds, r.Fd(), w.Fd())
if pipes.Stdin != nil {
go func() {
io.Copy(w, pipes.Stdin)
w.Close()
}()
p.Stdin = r
}
for _, fd := range fds {
if err := syscall.Fchown(int(fd), rootuid, rootuid); err != nil {
return writers, fmt.Errorf("Failed to chown pipes fd: %v", err)
}
}
return writers, nil
}
// SupportsHooks implements the execdriver Driver interface.
// The libcontainer/runC-based native execdriver does exploit the hook mechanism
func (d *Driver) SupportsHooks() bool {
return true
}

View file

@ -1,14 +0,0 @@
// +build !linux
package native
import (
"fmt"
"github.com/docker/docker/daemon/execdriver"
)
// NewDriver returns a new native driver, called from NewDriver of execdriver.
func NewDriver(root string, options []string) (execdriver.Driver, error) {
return nil, fmt.Errorf("native driver not supported on non-linux")
}

View file

@ -1,14 +0,0 @@
// +build linux,!cgo
package native
import (
"fmt"
"github.com/docker/docker/daemon/execdriver"
)
// NewDriver returns a new native driver, called from NewDriver of execdriver.
func NewDriver(root string, options []string) (execdriver.Driver, error) {
return nil, fmt.Errorf("native driver not supported on non-linux")
}

View file

@ -1,96 +0,0 @@
// +build linux
package native
import (
"fmt"
"os"
"os/exec"
"strings"
"sync"
"syscall"
"github.com/docker/docker/daemon/execdriver"
"github.com/opencontainers/runc/libcontainer"
// Blank import 'nsenter' so that init in that package will call c
// function 'nsexec()' to do 'setns' before Go runtime take over,
// it's used for join to exist ns like 'docker exec' command.
_ "github.com/opencontainers/runc/libcontainer/nsenter"
"github.com/opencontainers/runc/libcontainer/utils"
)
// Exec implements the exec driver Driver interface,
// it calls libcontainer APIs to execute a container.
func (d *Driver) Exec(c *execdriver.Command, processConfig *execdriver.ProcessConfig, pipes *execdriver.Pipes, hooks execdriver.Hooks) (int, error) {
active := d.activeContainers[c.ID]
if active == nil {
return -1, fmt.Errorf("No active container exists with ID %s", c.ID)
}
user := processConfig.User
if c.RemappedRoot.UID != 0 && user == "" {
//if user namespaces are enabled, set user explicitly so uid/gid is set to 0
//otherwise we end up with the overflow id and no permissions (65534)
user = "0"
}
p := &libcontainer.Process{
Args: append([]string{processConfig.Entrypoint}, processConfig.Arguments...),
Env: c.ProcessConfig.Env,
Cwd: c.WorkingDir,
User: user,
}
if processConfig.Privileged {
p.Capabilities = execdriver.GetAllCapabilities()
}
// add CAP_ prefix to all caps for new libcontainer update to match
// the spec format.
for i, s := range p.Capabilities {
if !strings.HasPrefix(s, "CAP_") {
p.Capabilities[i] = fmt.Sprintf("CAP_%s", s)
}
}
config := active.Config()
wg := sync.WaitGroup{}
writers, err := setupPipes(&config, processConfig, p, pipes, &wg)
if err != nil {
return -1, err
}
if err := active.Start(p); err != nil {
return -1, err
}
//close the write end of any opened pipes now that they are dup'ed into the container
for _, writer := range writers {
writer.Close()
}
if hooks.Start != nil {
pid, err := p.Pid()
if err != nil {
p.Signal(os.Kill)
p.Wait()
return -1, err
}
// A closed channel for OOM is returned here as it will be
// non-blocking and return the correct result when read.
chOOM := make(chan struct{})
close(chOOM)
hooks.Start(&c.ProcessConfig, pid, chOOM)
}
ps, err := p.Wait()
if err != nil {
exitErr, ok := err.(*exec.ExitError)
if !ok {
return -1, err
}
ps = exitErr.ProcessState
}
// wait for all IO goroutine copiers to finish
wg.Wait()
return utils.ExitStatus(ps.Sys().(syscall.WaitStatus)), nil
}

View file

@ -1,40 +0,0 @@
// +build linux
package native
import (
"fmt"
"os"
"runtime"
"github.com/docker/docker/pkg/reexec"
"github.com/opencontainers/runc/libcontainer"
)
func init() {
reexec.Register(DriverName, initializer)
}
func fatal(err error) {
if lerr, ok := err.(libcontainer.Error); ok {
lerr.Detail(os.Stderr)
os.Exit(1)
}
fmt.Fprintln(os.Stderr, err)
os.Exit(1)
}
func initializer() {
runtime.GOMAXPROCS(1)
runtime.LockOSThread()
factory, err := libcontainer.New("")
if err != nil {
fatal(err)
}
if err := factory.StartInitialization(); err != nil {
fatal(err)
}
panic("unreachable")
}

View file

@ -1,106 +0,0 @@
package template
import (
"syscall"
"github.com/opencontainers/runc/libcontainer/apparmor"
"github.com/opencontainers/runc/libcontainer/configs"
)
const defaultMountFlags = syscall.MS_NOEXEC | syscall.MS_NOSUID | syscall.MS_NODEV
// New returns the docker default configuration for libcontainer
func New() *configs.Config {
container := &configs.Config{
Capabilities: []string{
"CHOWN",
"DAC_OVERRIDE",
"FSETID",
"FOWNER",
"MKNOD",
"NET_RAW",
"SETGID",
"SETUID",
"SETFCAP",
"SETPCAP",
"NET_BIND_SERVICE",
"SYS_CHROOT",
"KILL",
"AUDIT_WRITE",
},
Namespaces: configs.Namespaces([]configs.Namespace{
{Type: "NEWNS"},
{Type: "NEWUTS"},
{Type: "NEWIPC"},
{Type: "NEWPID"},
{Type: "NEWNET"},
{Type: "NEWUSER"},
}),
Cgroups: &configs.Cgroup{
ScopePrefix: "docker", // systemd only
Resources: &configs.Resources{
AllowAllDevices: false,
MemorySwappiness: -1,
},
},
Mounts: []*configs.Mount{
{
Source: "proc",
Destination: "/proc",
Device: "proc",
Flags: defaultMountFlags,
},
{
Source: "tmpfs",
Destination: "/dev",
Device: "tmpfs",
Flags: syscall.MS_NOSUID | syscall.MS_STRICTATIME,
Data: "mode=755",
},
{
Source: "devpts",
Destination: "/dev/pts",
Device: "devpts",
Flags: syscall.MS_NOSUID | syscall.MS_NOEXEC,
Data: "newinstance,ptmxmode=0666,mode=0620,gid=5",
},
{
Source: "mqueue",
Destination: "/dev/mqueue",
Device: "mqueue",
Flags: defaultMountFlags,
},
{
Source: "sysfs",
Destination: "/sys",
Device: "sysfs",
Flags: defaultMountFlags | syscall.MS_RDONLY,
},
{
Source: "cgroup",
Destination: "/sys/fs/cgroup",
Device: "cgroup",
Flags: defaultMountFlags | syscall.MS_RDONLY,
},
},
MaskPaths: []string{
"/proc/kcore",
"/proc/latency_stats",
"/proc/timer_stats",
},
ReadonlyPaths: []string{
"/proc/asound",
"/proc/bus",
"/proc/fs",
"/proc/irq",
"/proc/sys",
"/proc/sysrq-trigger",
},
}
if apparmor.IsEnabled() {
container.AppArmorProfile = "docker-default"
}
return container
}

View file

@ -1,3 +0,0 @@
// +build !linux
package template

View file

@ -1,24 +0,0 @@
package execdriver
import (
"io"
)
// Pipes is a wrapper around a container's output for
// stdin, stdout, stderr
type Pipes struct {
Stdin io.ReadCloser
Stdout, Stderr io.Writer
}
// NewPipes returns a wrapper around a container's output
func NewPipes(stdin io.ReadCloser, stdout, stderr io.Writer, useStdin bool) *Pipes {
p := &Pipes{
Stdout: stdout,
Stderr: stderr,
}
if useStdin {
p.Stdin = stdin
}
return p
}

View file

@ -1,55 +0,0 @@
package execdriver
import (
"io"
"os/exec"
)
// StdConsole defines standard console operations for execdriver
type StdConsole struct {
// Closers holds io.Closer references for closing at terminal close time
Closers []io.Closer
}
// NewStdConsole returns a new StdConsole struct
func NewStdConsole(processConfig *ProcessConfig, pipes *Pipes) (*StdConsole, error) {
std := &StdConsole{}
if err := std.AttachPipes(&processConfig.Cmd, pipes); err != nil {
return nil, err
}
return std, nil
}
// AttachPipes attaches given pipes to exec.Cmd
func (s *StdConsole) AttachPipes(command *exec.Cmd, pipes *Pipes) error {
command.Stdout = pipes.Stdout
command.Stderr = pipes.Stderr
if pipes.Stdin != nil {
stdin, err := command.StdinPipe()
if err != nil {
return err
}
go func() {
defer stdin.Close()
io.Copy(stdin, pipes.Stdin)
}()
}
return nil
}
// Resize implements Resize method of Terminal interface
func (s *StdConsole) Resize(h, w int) error {
// we do not need to resize a non tty
return nil
}
// Close implements Close method of Terminal interface
func (s *StdConsole) Close() error {
for _, c := range s.Closers {
c.Close()
}
return nil
}

View file

@ -1,125 +0,0 @@
// +build !windows
package execdriver
import (
"fmt"
"strings"
"github.com/docker/docker/pkg/stringutils"
"github.com/syndtr/gocapability/capability"
)
var capabilityList Capabilities
func init() {
last := capability.CAP_LAST_CAP
// hack for RHEL6 which has no /proc/sys/kernel/cap_last_cap
if last == capability.Cap(63) {
last = capability.CAP_BLOCK_SUSPEND
}
for _, cap := range capability.List() {
if cap > last {
continue
}
capabilityList = append(capabilityList,
&CapabilityMapping{
Key: strings.ToUpper(cap.String()),
Value: cap,
},
)
}
}
type (
// CapabilityMapping maps linux capability name to its value of capability.Cap type
// Capabilities is one of the security systems in Linux Security Module (LSM)
// framework provided by the kernel.
// For more details on capabilities, see http://man7.org/linux/man-pages/man7/capabilities.7.html
CapabilityMapping struct {
Key string `json:"key,omitempty"`
Value capability.Cap `json:"value,omitempty"`
}
// Capabilities contains all CapabilityMapping
Capabilities []*CapabilityMapping
)
// String returns <key> of CapabilityMapping
func (c *CapabilityMapping) String() string {
return c.Key
}
// GetCapability returns CapabilityMapping which contains specific key
func GetCapability(key string) *CapabilityMapping {
for _, capp := range capabilityList {
if capp.Key == key {
cpy := *capp
return &cpy
}
}
return nil
}
// GetAllCapabilities returns all of the capabilities
func GetAllCapabilities() []string {
output := make([]string, len(capabilityList))
for i, capability := range capabilityList {
output[i] = capability.String()
}
return output
}
// TweakCapabilities can tweak capabilities by adding or dropping capabilities
// based on the basics capabilities.
func TweakCapabilities(basics, adds, drops []string) ([]string, error) {
var (
newCaps []string
allCaps = GetAllCapabilities()
)
// look for invalid cap in the drop list
for _, cap := range drops {
if strings.ToLower(cap) == "all" {
continue
}
if !stringutils.InSlice(allCaps, cap) {
return nil, fmt.Errorf("Unknown capability drop: %q", cap)
}
}
// handle --cap-add=all
if stringutils.InSlice(adds, "all") {
basics = allCaps
}
if !stringutils.InSlice(drops, "all") {
for _, cap := range basics {
// skip `all` already handled above
if strings.ToLower(cap) == "all" {
continue
}
// if we don't drop `all`, add back all the non-dropped caps
if !stringutils.InSlice(drops, cap) {
newCaps = append(newCaps, strings.ToUpper(cap))
}
}
}
for _, cap := range adds {
// skip `all` already handled above
if strings.ToLower(cap) == "all" {
continue
}
if !stringutils.InSlice(allCaps, cap) {
return nil, fmt.Errorf("Unknown capability to add: %q", cap)
}
// add cap if not already in the list
if !stringutils.InSlice(newCaps, cap) {
newCaps = append(newCaps, strings.ToUpper(cap))
}
}
return newCaps, nil
}

View file

@ -1,8 +0,0 @@
// +build windows
package windows
// Clean implements the exec driver Driver interface.
func (d *Driver) Clean(id string) error {
return nil
}

View file

@ -1,36 +0,0 @@
//+build windows
package windows
import (
"errors"
"syscall"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
)
// createCommandLine creates a command line from the Entrypoint and args
// of the ProcessConfig. It escapes the arguments if they are not already
// escaped
func createCommandLine(processConfig *execdriver.ProcessConfig, alreadyEscaped bool) (commandLine string, err error) {
// While this should get caught earlier, just in case, validate that we
// have something to run.
if processConfig.Entrypoint == "" {
return "", errors.New("No entrypoint specified")
}
// Build the command line of the process
commandLine = processConfig.Entrypoint
logrus.Debugf("Entrypoint: %s", processConfig.Entrypoint)
for _, arg := range processConfig.Arguments {
logrus.Debugf("appending %s", arg)
if !alreadyEscaped {
arg = syscall.EscapeArg(arg)
}
commandLine += " " + arg
}
logrus.Debugf("commandLine: %s", commandLine)
return commandLine, nil
}

View file

@ -1,89 +0,0 @@
// +build windows
package windows
import (
"fmt"
"syscall"
"github.com/Microsoft/hcsshim"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
)
// Exec implements the exec driver Driver interface.
func (d *Driver) Exec(c *execdriver.Command, processConfig *execdriver.ProcessConfig, pipes *execdriver.Pipes, hooks execdriver.Hooks) (int, error) {
var (
term execdriver.Terminal
err error
exitCode int32
)
active := d.activeContainers[c.ID]
if active == nil {
return -1, fmt.Errorf("Exec - No active container exists with ID %s", c.ID)
}
createProcessParms := hcsshim.CreateProcessParams{
EmulateConsole: processConfig.Tty, // Note NOT c.ProcessConfig.Tty
WorkingDirectory: c.WorkingDir,
}
// Configure the environment for the process // Note NOT c.ProcessConfig.Env
createProcessParms.Environment = setupEnvironmentVariables(processConfig.Env)
// Create the commandline for the process // Note NOT c.ProcessConfig
createProcessParms.CommandLine, err = createCommandLine(processConfig, false)
if err != nil {
return -1, err
}
// Start the command running in the container.
pid, stdin, stdout, stderr, err := hcsshim.CreateProcessInComputeSystem(c.ID, pipes.Stdin != nil, true, !processConfig.Tty, createProcessParms)
if err != nil {
// TODO Windows: TP4 Workaround. In Hyper-V containers, there is a limitation
// of one exec per container. This should be fixed post TP4. CreateProcessInComputeSystem
// will return a specific error which we handle here to give a good error message
// back to the user instead of an inactionable "An invalid argument was supplied"
if herr, ok := err.(*hcsshim.HcsError); ok && herr.Err == hcsshim.WSAEINVAL {
return -1, fmt.Errorf("The limit of docker execs per Hyper-V container has been exceeded")
}
logrus.Errorf("CreateProcessInComputeSystem() failed %s", err)
return -1, err
}
// Now that the process has been launched, begin copying data to and from
// the named pipes for the std handles.
setupPipes(stdin, stdout, stderr, pipes)
// Note NOT c.ProcessConfig.Tty
if processConfig.Tty {
term = NewTtyConsole(c.ID, pid)
} else {
term = NewStdConsole()
}
processConfig.Terminal = term
// Invoke the start callback
if hooks.Start != nil {
// A closed channel for OOM is returned here as it will be
// non-blocking and return the correct result when read.
chOOM := make(chan struct{})
close(chOOM)
hooks.Start(&c.ProcessConfig, int(pid), chOOM)
}
if exitCode, err = hcsshim.WaitForProcessInComputeSystem(c.ID, pid, hcsshim.TimeoutInfinite); err != nil {
if herr, ok := err.(*hcsshim.HcsError); ok && herr.Err == syscall.ERROR_BROKEN_PIPE {
logrus.Debugf("Exiting Run() after WaitForProcessInComputeSystem failed with recognised error %s", err)
return hcsshim.WaitErrExecFailed, nil
}
logrus.Warnf("WaitForProcessInComputeSystem failed (container may have been killed): %s", err)
return -1, err
}
logrus.Debugln("Exiting Run()", c.ID)
return int(exitCode), nil
}

View file

@ -1,11 +0,0 @@
// +build windows
package windows
import "fmt"
// GetPidsForContainer implements the exec driver Driver interface.
func (d *Driver) GetPidsForContainer(id string) ([]int, error) {
// TODO Windows: Implementation required.
return nil, fmt.Errorf("GetPidsForContainer: GetPidsForContainer() not implemented")
}

View file

@ -1,63 +0,0 @@
// +build windows
package windows
import (
"fmt"
"io"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
)
// General comment. Handling I/O for a container is very different to Linux.
// We use a named pipe to HCS to copy I/O both in and out of the container,
// very similar to how docker daemon communicates with a CLI.
// startStdinCopy asynchronously copies an io.Reader to the container's
// process's stdin pipe and closes the pipe when there is no more data to copy.
func startStdinCopy(dst io.WriteCloser, src io.Reader) {
// Anything that comes from the client stdin should be copied
// across to the stdin named pipe of the container.
go func() {
defer dst.Close()
bytes, err := io.Copy(dst, src)
log := fmt.Sprintf("Copied %d bytes from stdin.", bytes)
if err != nil {
log = log + " err=" + err.Error()
}
logrus.Debugf(log)
}()
}
// startStdouterrCopy asynchronously copies data from the container's process's
// stdout or stderr pipe to an io.Writer and closes the pipe when there is no
// more data to copy.
func startStdouterrCopy(dst io.Writer, src io.ReadCloser, name string) {
// Anything that comes from the container named pipe stdout/err should be copied
// across to the stdout/err of the client
go func() {
defer src.Close()
bytes, err := io.Copy(dst, src)
log := fmt.Sprintf("Copied %d bytes from %s.", bytes, name)
if err != nil {
log = log + " err=" + err.Error()
}
logrus.Debugf(log)
}()
}
// setupPipes starts the asynchronous copying of data to and from the named
// pipes used byt he HCS for the std handles.
func setupPipes(stdin io.WriteCloser, stdout, stderr io.ReadCloser, pipes *execdriver.Pipes) {
if stdin != nil {
startStdinCopy(stdin, pipes.Stdin)
}
if stdout != nil {
startStdouterrCopy(pipes.Stdout, stdout, "stdout")
}
if stderr != nil {
startStdouterrCopy(pipes.Stderr, stderr, "stderr")
}
}

View file

@ -1,19 +0,0 @@
// +build windows
package windows
import (
"fmt"
"github.com/docker/docker/daemon/execdriver"
)
// Pause implements the exec driver Driver interface.
func (d *Driver) Pause(c *execdriver.Command) error {
return fmt.Errorf("Windows: Containers cannot be paused")
}
// Unpause implements the exec driver Driver interface.
func (d *Driver) Unpause(c *execdriver.Command) error {
return fmt.Errorf("Windows: Containers cannot be paused")
}

View file

@ -1,366 +0,0 @@
// +build windows
package windows
import (
"encoding/json"
"fmt"
"os"
"path/filepath"
"strconv"
"strings"
"syscall"
"time"
"github.com/Microsoft/hcsshim"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
)
// defaultContainerNAT is the default name of the container NAT device that is
// preconfigured on the server.
const defaultContainerNAT = "ContainerNAT"
// Win32 error codes that are used for various workarounds
// These really should be ALL_CAPS to match golangs syscall library and standard
// Win32 error conventions, but golint insists on CamelCase.
const (
CoEClassstring = syscall.Errno(0x800401F3) // Invalid class string
ErrorNoNetwork = syscall.Errno(1222) // The network is not present or not started
ErrorBadPathname = syscall.Errno(161) // The specified path is invalid
ErrorInvalidObject = syscall.Errno(0x800710D8) // The object identifier does not represent a valid object
)
type layer struct {
ID string
Path string
}
type portBinding struct {
Protocol string
InternalPort int
ExternalPort int
}
type natSettings struct {
Name string
PortBindings []portBinding
}
type networkConnection struct {
NetworkName string
// TODO Windows: Add Ip4Address string to this structure when hooked up in
// docker CLI. This is present in the HCS JSON handler.
EnableNat bool
Nat natSettings
}
type networkSettings struct {
MacAddress string
}
type device struct {
DeviceType string
Connection interface{}
Settings interface{}
}
type mappedDir struct {
HostPath string
ContainerPath string
ReadOnly bool
}
type containerInit struct {
SystemType string // HCS requires this to be hard-coded to "Container"
Name string // Name of the container. We use the docker ID.
Owner string // The management platform that created this container
IsDummy bool // Used for development purposes.
VolumePath string // Windows volume path for scratch space
Devices []device // Devices used by the container
IgnoreFlushesDuringBoot bool // Optimization hint for container startup in Windows
LayerFolderPath string // Where the layer folders are located
Layers []layer // List of storage layers
ProcessorWeight int64 `json:",omitempty"` // CPU Shares 0..10000 on Windows; where 0 will be omitted and HCS will default.
HostName string // Hostname
MappedDirectories []mappedDir // List of mapped directories (volumes/mounts)
SandboxPath string // Location of unmounted sandbox (used for Hyper-V containers, not Windows Server containers)
HvPartition bool // True if it a Hyper-V Container
EndpointList []string // List of endpoints to be attached to container
}
// defaultOwner is a tag passed to HCS to allow it to differentiate between
// container creator management stacks. We hard code "docker" in the case
// of docker.
const defaultOwner = "docker"
// Run implements the exec driver Driver interface
func (d *Driver) Run(c *execdriver.Command, pipes *execdriver.Pipes, hooks execdriver.Hooks) (execdriver.ExitStatus, error) {
var (
term execdriver.Terminal
err error
)
// Allocate Network only if there is no network interface
cu := &containerInit{
SystemType: "Container",
Name: c.ID,
Owner: defaultOwner,
IsDummy: dummyMode,
VolumePath: c.Rootfs,
IgnoreFlushesDuringBoot: c.FirstStart,
LayerFolderPath: c.LayerFolder,
ProcessorWeight: c.Resources.CPUShares,
HostName: c.Hostname,
EndpointList: c.EpList,
}
cu.HvPartition = c.HvPartition
if cu.HvPartition {
cu.SandboxPath = filepath.Dir(c.LayerFolder)
} else {
cu.VolumePath = c.Rootfs
cu.LayerFolderPath = c.LayerFolder
}
for _, layerPath := range c.LayerPaths {
_, filename := filepath.Split(layerPath)
g, err := hcsshim.NameToGuid(filename)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
cu.Layers = append(cu.Layers, layer{
ID: g.ToString(),
Path: layerPath,
})
}
// Add the mounts (volumes, bind mounts etc) to the structure
mds := make([]mappedDir, len(c.Mounts))
for i, mount := range c.Mounts {
mds[i] = mappedDir{
HostPath: mount.Source,
ContainerPath: mount.Destination,
ReadOnly: !mount.Writable}
}
cu.MappedDirectories = mds
// TODO Windows. At some point, when there is CLI on docker run to
// enable the IP Address of the container to be passed into docker run,
// the IP Address needs to be wired through to HCS in the JSON. It
// would be present in c.Network.Interface.IPAddress. See matching
// TODO in daemon\container_windows.go, function populateCommand.
if c.Network.Interface != nil {
var pbs []portBinding
// Enumerate through the port bindings specified by the user and convert
// them into the internal structure matching the JSON blob that can be
// understood by the HCS.
for i, v := range c.Network.Interface.PortBindings {
proto := strings.ToUpper(i.Proto())
if proto != "TCP" && proto != "UDP" {
return execdriver.ExitStatus{ExitCode: -1}, fmt.Errorf("invalid protocol %s", i.Proto())
}
if len(v) > 1 {
return execdriver.ExitStatus{ExitCode: -1}, fmt.Errorf("Windows does not support more than one host port in NAT settings")
}
for _, v2 := range v {
var (
iPort, ePort int
err error
)
if len(v2.HostIP) != 0 {
return execdriver.ExitStatus{ExitCode: -1}, fmt.Errorf("Windows does not support host IP addresses in NAT settings")
}
if ePort, err = strconv.Atoi(v2.HostPort); err != nil {
return execdriver.ExitStatus{ExitCode: -1}, fmt.Errorf("invalid container port %s: %s", v2.HostPort, err)
}
if iPort, err = strconv.Atoi(i.Port()); err != nil {
return execdriver.ExitStatus{ExitCode: -1}, fmt.Errorf("invalid internal port %s: %s", i.Port(), err)
}
if iPort < 0 || iPort > 65535 || ePort < 0 || ePort > 65535 {
return execdriver.ExitStatus{ExitCode: -1}, fmt.Errorf("specified NAT port is not in allowed range")
}
pbs = append(pbs,
portBinding{ExternalPort: ePort,
InternalPort: iPort,
Protocol: proto})
}
}
// TODO Windows: TP3 workaround. Allow the user to override the name of
// the Container NAT device through an environment variable. This will
// ultimately be a global daemon parameter on Windows, similar to -b
// for the name of the virtual switch (aka bridge).
cn := os.Getenv("DOCKER_CONTAINER_NAT")
if len(cn) == 0 {
cn = defaultContainerNAT
}
dev := device{
DeviceType: "Network",
Connection: &networkConnection{
NetworkName: c.Network.Interface.Bridge,
// TODO Windows: Fixme, next line. Needs HCS fix.
EnableNat: false,
Nat: natSettings{
Name: cn,
PortBindings: pbs,
},
},
}
if c.Network.Interface.MacAddress != "" {
windowsStyleMAC := strings.Replace(
c.Network.Interface.MacAddress, ":", "-", -1)
dev.Settings = networkSettings{
MacAddress: windowsStyleMAC,
}
}
cu.Devices = append(cu.Devices, dev)
} else {
logrus.Debugln("No network interface")
}
configurationb, err := json.Marshal(cu)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
configuration := string(configurationb)
// TODO Windows TP5 timeframe. Remove when TP4 is no longer supported.
// The following a workaround for Windows TP4 which has a networking
// bug which fairly frequently returns an error. Back off and retry.
maxAttempts := 5
for i := 0; i < maxAttempts; i++ {
err = hcsshim.CreateComputeSystem(c.ID, configuration)
if err == nil {
break
}
if !TP4RetryHack {
return execdriver.ExitStatus{ExitCode: -1}, err
}
if herr, ok := err.(*hcsshim.HcsError); ok {
if herr.Err != syscall.ERROR_NOT_FOUND && // Element not found
herr.Err != syscall.ERROR_FILE_NOT_FOUND && // The system cannot find the file specified
herr.Err != ErrorNoNetwork && // The network is not present or not started
herr.Err != ErrorBadPathname && // The specified path is invalid
herr.Err != CoEClassstring && // Invalid class string
herr.Err != ErrorInvalidObject { // The object identifier does not represent a valid object
logrus.Debugln("Failed to create temporary container ", err)
return execdriver.ExitStatus{ExitCode: -1}, err
}
logrus.Warnf("Invoking Windows TP4 retry hack (%d of %d)", i, maxAttempts-1)
time.Sleep(50 * time.Millisecond)
}
}
// Start the container
logrus.Debugln("Starting container ", c.ID)
err = hcsshim.StartComputeSystem(c.ID)
if err != nil {
logrus.Errorf("Failed to start compute system: %s", err)
return execdriver.ExitStatus{ExitCode: -1}, err
}
defer func() {
// Stop the container
if forceKill {
logrus.Debugf("Forcibly terminating container %s", c.ID)
if err := hcsshim.TerminateComputeSystem(c.ID, hcsshim.TimeoutInfinite, "exec-run-defer"); err != nil {
logrus.Warnf("Ignoring error from TerminateComputeSystem %s", err)
}
} else {
logrus.Debugf("Shutting down container %s", c.ID)
if err := hcsshim.ShutdownComputeSystem(c.ID, hcsshim.TimeoutInfinite, "exec-run-defer"); err != nil {
if herr, ok := err.(*hcsshim.HcsError); !ok ||
(herr.Err != hcsshim.ERROR_SHUTDOWN_IN_PROGRESS &&
herr.Err != ErrorBadPathname &&
herr.Err != syscall.ERROR_PATH_NOT_FOUND) {
logrus.Warnf("Ignoring error from ShutdownComputeSystem %s", err)
}
}
}
}()
createProcessParms := hcsshim.CreateProcessParams{
EmulateConsole: c.ProcessConfig.Tty,
WorkingDirectory: c.WorkingDir,
ConsoleSize: c.ProcessConfig.ConsoleSize,
}
// Configure the environment for the process
createProcessParms.Environment = setupEnvironmentVariables(c.ProcessConfig.Env)
createProcessParms.CommandLine, err = createCommandLine(&c.ProcessConfig, c.ArgsEscaped)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
// Start the command running in the container.
pid, stdin, stdout, stderr, err := hcsshim.CreateProcessInComputeSystem(c.ID, pipes.Stdin != nil, true, !c.ProcessConfig.Tty, createProcessParms)
if err != nil {
logrus.Errorf("CreateProcessInComputeSystem() failed %s", err)
return execdriver.ExitStatus{ExitCode: -1}, err
}
// Now that the process has been launched, begin copying data to and from
// the named pipes for the std handles.
setupPipes(stdin, stdout, stderr, pipes)
//Save the PID as we'll need this in Kill()
logrus.Debugf("PID %d", pid)
c.ContainerPid = int(pid)
if c.ProcessConfig.Tty {
term = NewTtyConsole(c.ID, pid)
} else {
term = NewStdConsole()
}
c.ProcessConfig.Terminal = term
// Maintain our list of active containers. We'll need this later for exec
// and other commands.
d.Lock()
d.activeContainers[c.ID] = &activeContainer{
command: c,
}
d.Unlock()
if hooks.Start != nil {
// A closed channel for OOM is returned here as it will be
// non-blocking and return the correct result when read.
chOOM := make(chan struct{})
close(chOOM)
hooks.Start(&c.ProcessConfig, int(pid), chOOM)
}
exitCode, err := hcsshim.WaitForProcessInComputeSystem(c.ID, pid, hcsshim.TimeoutInfinite)
if err != nil {
if herr, ok := err.(*hcsshim.HcsError); ok && herr.Err != syscall.ERROR_BROKEN_PIPE {
logrus.Warnf("WaitForProcessInComputeSystem failed (container may have been killed): %s", err)
}
// Do NOT return err here as the container would have
// started, otherwise docker will deadlock. It's perfectly legitimate
// for WaitForProcessInComputeSystem to fail in situations such
// as the container being killed on another thread.
return execdriver.ExitStatus{ExitCode: hcsshim.WaitErrExecFailed}, nil
}
logrus.Debugf("Exiting Run() exitCode %d id=%s", exitCode, c.ID)
return execdriver.ExitStatus{ExitCode: int(exitCode)}, nil
}
// SupportsHooks implements the execdriver Driver interface.
// The windows driver does not support the hook mechanism
func (d *Driver) SupportsHooks() bool {
return false
}

View file

@ -1,14 +0,0 @@
// +build windows
package windows
import (
"fmt"
"github.com/docker/docker/daemon/execdriver"
)
// Stats implements the exec driver Driver interface.
func (d *Driver) Stats(id string) (*execdriver.ResourceStats, error) {
return nil, fmt.Errorf("Windows: Stats not implemented")
}

View file

@ -1,24 +0,0 @@
// +build windows
package windows
// StdConsole is for when using a container non-interactively
type StdConsole struct {
}
// NewStdConsole returns a new StdConsole struct.
func NewStdConsole() *StdConsole {
return &StdConsole{}
}
// Resize implements Resize method of Terminal interface.
func (s *StdConsole) Resize(h, w int) error {
// we do not need to resize a non tty
return nil
}
// Close implements Close method of Terminal interface.
func (s *StdConsole) Close() error {
// nothing to close here
return nil
}

View file

@ -1,49 +0,0 @@
// +build windows
package windows
import (
"fmt"
"syscall"
"github.com/Microsoft/hcsshim"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
)
// Terminate implements the exec driver Driver interface.
func (d *Driver) Terminate(p *execdriver.Command) error {
return kill(p.ID, p.ContainerPid, syscall.SIGTERM)
}
// Kill implements the exec driver Driver interface.
func (d *Driver) Kill(p *execdriver.Command, sig int) error {
return kill(p.ID, p.ContainerPid, syscall.Signal(sig))
}
func kill(id string, pid int, sig syscall.Signal) error {
logrus.Debugf("WindowsExec: kill() id=%s pid=%d sig=%d", id, pid, sig)
var err error
context := fmt.Sprintf("kill: sig=%d pid=%d", sig, pid)
if sig == syscall.SIGKILL || forceKill {
// Terminate the compute system
if err := hcsshim.TerminateComputeSystem(id, hcsshim.TimeoutInfinite, context); err != nil {
logrus.Errorf("Failed to terminate %s - %q", id, err)
}
} else {
// Terminate Process
if err = hcsshim.TerminateProcessInComputeSystem(id, uint32(pid)); err != nil {
logrus.Warnf("Failed to terminate pid %d in %s: %q", pid, id, err)
// Ignore errors
err = nil
}
// Shutdown the compute system
if err := hcsshim.ShutdownComputeSystem(id, hcsshim.TimeoutInfinite, context); err != nil {
logrus.Errorf("Failed to shutdown %s - %q", id, err)
}
}
return err
}

View file

@ -1,32 +0,0 @@
// +build windows
package windows
import (
"github.com/Microsoft/hcsshim"
)
// TtyConsole implements the exec driver Terminal interface.
type TtyConsole struct {
id string
processid uint32
}
// NewTtyConsole returns a new TtyConsole struct.
func NewTtyConsole(id string, processid uint32) *TtyConsole {
tty := &TtyConsole{
id: id,
processid: processid,
}
return tty
}
// Resize implements Resize method of Terminal interface.
func (t *TtyConsole) Resize(h, w int) error {
return hcsshim.ResizeConsoleInComputeSystem(t.id, t.processid, h, w)
}
// Close implements Close method of Terminal interface.
func (t *TtyConsole) Close() error {
return nil
}

View file

@ -1,14 +0,0 @@
// +build !windows
package windows
import (
"fmt"
"github.com/docker/docker/daemon/execdriver"
)
// NewDriver returns a new execdriver.Driver
func NewDriver(root, initPath string) (execdriver.Driver, error) {
return nil, fmt.Errorf("Windows driver not supported on non-Windows")
}

View file

@ -1,14 +0,0 @@
// +build windows
package windows
import (
"github.com/docker/docker/daemon/execdriver"
)
// Update updates resource configs for a container.
func (d *Driver) Update(c *execdriver.Command) error {
// Updating resource isn't supported on Windows
// but we should return nil for enabling updating container
return nil
}

View file

@ -1,123 +0,0 @@
// +build windows
package windows
import (
"fmt"
"strings"
"sync"
"github.com/Microsoft/hcsshim"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/dockerversion"
"github.com/docker/docker/pkg/parsers"
"github.com/docker/engine-api/types/container"
)
// TP4RetryHack is a hack to retry CreateComputeSystem if it fails with
// known return codes from Windows due to bugs in TP4.
var TP4RetryHack bool
// This is a daemon development variable only and should not be
// used for running production containers on Windows.
var dummyMode bool
// This allows the daemon to terminate containers rather than shutdown
// This allows the daemon to force kill (HCS terminate) rather than shutdown
var forceKill bool
// DefaultIsolation allows users to specify a default isolation technology for
// when running a container on Windows. For example docker daemon -D
// --exec-opt isolation=hyperv will cause Windows to always run containers
// as Hyper-V containers unless otherwise specified.
var DefaultIsolation container.Isolation = "process"
// Define name and version for windows
var (
DriverName = "Windows 1854"
Version = dockerversion.Version + " " + dockerversion.GitCommit
)
type activeContainer struct {
command *execdriver.Command
}
// Driver contains all information for windows driver,
// it implements execdriver.Driver
type Driver struct {
root string
activeContainers map[string]*activeContainer
sync.Mutex
}
// Name implements the exec driver Driver interface.
func (d *Driver) Name() string {
return fmt.Sprintf("\n Name: %s\n Build: %s \n Default Isolation: %s", DriverName, Version, DefaultIsolation)
}
// NewDriver returns a new windows driver, called from NewDriver of execdriver.
func NewDriver(root string, options []string) (*Driver, error) {
for _, option := range options {
key, val, err := parsers.ParseKeyValueOpt(option)
if err != nil {
return nil, err
}
key = strings.ToLower(key)
switch key {
case "dummy":
switch val {
case "1":
dummyMode = true
logrus.Warn("Using dummy mode in Windows exec driver. This is for development use only!")
}
case "forcekill":
switch val {
case "1":
forceKill = true
logrus.Warn("Using force kill mode in Windows exec driver. This is for testing purposes only.")
}
case "isolation":
if !container.Isolation(val).IsValid() {
return nil, fmt.Errorf("Unrecognised exec driver option 'isolation':'%s'", val)
}
if container.Isolation(val).IsHyperV() {
DefaultIsolation = "hyperv"
}
logrus.Infof("Windows default isolation: '%s'", val)
default:
return nil, fmt.Errorf("Unrecognised exec driver option %s\n", key)
}
}
// TODO Windows TP5 timeframe. Remove this next block of code once TP4
// is no longer supported. Also remove the workaround in run.go.
//
// Hack for TP4.
// This overcomes an issue on TP4 which causes CreateComputeSystem to
// intermittently fail. It's predominantly here to make Windows to Windows
// CI more reliable.
TP4RetryHack = hcsshim.IsTP4()
return &Driver{
root: root,
activeContainers: make(map[string]*activeContainer),
}, nil
}
// setupEnvironmentVariables convert a string array of environment variables
// into a map as required by the HCS. Source array is in format [v1=k1] [v2=k2] etc.
func setupEnvironmentVariables(a []string) map[string]string {
r := make(map[string]string)
for _, s := range a {
arr := strings.Split(s, "=")
if len(arr) == 2 {
r[arr[0]] = arr[1]
}
}
return r
}