func finalizeNamespace(args *InitArgs) error {
if err := utils.CloseExecFrom(3); err != nil {
return err
}
// We use the native drivers default template so that things like caps are consistent
// across both drivers
container := template.New()
if !args.Privileged {
// drop capabilities in bounding set before changing user
if err := capabilities.DropBoundingSet(container.Capabilities); err != nil {
return fmt.Errorf("drop bounding set %s", err)
}
// preserve existing capabilities while we change users
if err := system.SetKeepCaps(); err != nil {
return fmt.Errorf("set keep caps %s", err)
}
}
if err := namespaces.SetupUser(args.User); err != nil {
return fmt.Errorf("setup user %s", err)
}
if !args.Privileged {
if err := system.ClearKeepCaps(); err != nil {
return fmt.Errorf("clear keep caps %s", err)
}
var (
adds []string
drops []string
)
if args.CapAdd != "" {
adds = strings.Split(args.CapAdd, ":")
}
if args.CapDrop != "" {
drops = strings.Split(args.CapDrop, ":")
}
caps, err := execdriver.TweakCapabilities(container.Capabilities, adds, drops)
if err != nil {
return err
}
// drop all other capabilities
if err := capabilities.DropCapabilities(caps); err != nil {
return fmt.Errorf("drop capabilities %s", err)
}
}
if err := setupWorkingDirectory(args); err != nil {
return err
}
return nil
}
func keepCapabilities(adds []string, drops []string) []string {
container := nativeTemplate.New()
caps, err := execdriver.TweakCapabilities(container.Capabilities, adds, drops)
var newCaps []string
for _, cap := range caps {
newCaps = append(newCaps, strings.ToLower(cap))
}
if err != nil {
return []string{}
}
return newCaps
}
func keepCapabilities(adds []string, drops []string) ([]string, error) {
container := nativeTemplate.New()
logrus.Debugf("adds %s drops %s\n", adds, drops)
caps, err := execdriver.TweakCapabilities(container.Capabilities, adds, drops)
if err != nil {
return nil, err
}
var newCaps []string
for _, cap := range caps {
logrus.Debugf("cap %s\n", cap)
realCap := execdriver.GetCapability(cap)
numCap := fmt.Sprintf("%d", realCap.Value)
newCaps = append(newCaps, numCap)
}
return newCaps, 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
}
// Config takes ContainerJSON and Daemon Info and converts it into the opencontainers spec.
func Config(c types.ContainerJSON, info types.Info, capabilities []string) (config *specs.LinuxSpec, err error) {
config = &specs.LinuxSpec{
Spec: specs.Spec{
Version: SpecVersion,
Platform: specs.Platform{
OS: info.OSType,
Arch: info.Architecture,
},
Process: specs.Process{
Terminal: c.Config.Tty,
User: specs.User{
// TODO: user stuffs
},
Args: append([]string{c.Path}, c.Args...),
Env: c.Config.Env,
Cwd: c.Config.WorkingDir,
},
Root: specs.Root{
Path: "rootfs",
Readonly: c.HostConfig.ReadonlyRootfs,
},
Mounts: []specs.MountPoint{},
},
}
// make sure the current working directory is not blank
if config.Process.Cwd == "" {
config.Process.Cwd = DefaultCurrentWorkingDirectory
}
// get the user
if c.Config.User != "" {
u, err := user.LookupUser(c.Config.User)
if err != nil {
config.Spec.Process.User = specs.User{
UID: uint32(u.Uid),
GID: uint32(u.Gid),
}
} else {
//return nil, fmt.Errorf("Looking up user (%s) failed: %v", c.Config.User, err)
logrus.Warnf("Looking up user (%s) failed: %v", c.Config.User, err)
}
}
// add the additional groups
for _, group := range c.HostConfig.GroupAdd {
g, err := user.LookupGroup(group)
if err != nil {
return nil, fmt.Errorf("Looking up group (%s) failed: %v", group, err)
}
config.Spec.Process.User.AdditionalGids = append(config.Spec.Process.User.AdditionalGids, uint32(g.Gid))
}
// get the hostname, if the hostname is the name as the first 12 characters of the id,
// then set the hostname as the container name
if c.ID[:12] == c.Config.Hostname {
config.Hostname = strings.TrimPrefix(c.Name, "/")
}
// get mounts
mounts := map[string]bool{}
for _, mount := range c.Mounts {
mounts[mount.Destination] = true
config.Mounts = append(config.Mounts, specs.MountPoint{
Name: mount.Destination,
Path: mount.Destination,
})
}
// add /etc/hosts and /etc/resolv.conf if we should have networking
if c.HostConfig.NetworkMode != "none" && c.HostConfig.NetworkMode != "host" {
DefaultMounts = append(DefaultMounts, NetworkMounts...)
}
// if we aren't doing something crazy like mounting a default mount ourselves,
// the we can mount it the default way
for _, mount := range DefaultMounts {
if _, ok := mounts[mount.Path]; !ok {
config.Mounts = append(config.Mounts, mount)
}
}
// set privileged
if c.HostConfig.Privileged {
// allow all caps
capabilities = execdriver.GetAllCapabilities()
}
// get the capabilities
config.Linux.Capabilities, err = execdriver.TweakCapabilities(capabilities, c.HostConfig.CapAdd.Slice(), c.HostConfig.CapDrop.Slice())
if err != nil {
return nil, fmt.Errorf("setting capabilities failed: %v", err)
}
// add CAP_ prefix
// TODO: this is awful
for i, cap := range config.Linux.Capabilities {
if !strings.HasPrefix(cap, "CAP_") {
config.Linux.Capabilities[i] = fmt.Sprintf("CAP_%s", cap)
}
}
// if we have a container that needs a terminal but no env vars, then set
// default env vars for the terminal to function
if config.Spec.Process.Terminal && len(config.Spec.Process.Env) <= 0 {
config.Spec.Process.Env = DefaultTerminalEnv
}
if config.Spec.Process.Terminal {
// make sure we have TERM set
var termSet bool
for _, env := range config.Spec.Process.Env {
if strings.HasPrefix(env, "TERM=") {
termSet = true
break
}
}
if !termSet {
// set the term variable
config.Spec.Process.Env = append(config.Spec.Process.Env, fmt.Sprintf("TERM=%s", DefaultTerminal))
}
}
return config, nil
}
// Config takes ContainerJSON and converts it into the opencontainers spec.
func Config(c types.ContainerJSON, osType, architecture string, capabilities []string, idroot, idlen uint32) (config *specs.Spec, err error) {
// for user namespaces use defaults unless another range specified
if idroot == 0 {
idroot = DefaultUserNSHostID
}
if idlen == 0 {
idlen = DefaultUserNSMapSize
}
config = &specs.Spec{
Version: SpecVersion,
Platform: specs.Platform{
OS: osType,
Arch: architecture,
},
Process: specs.Process{
Terminal: c.Config.Tty,
User: specs.User{
// TODO: user stuffs
},
Args: append([]string{c.Path}, c.Args...),
Env: c.Config.Env,
Cwd: c.Config.WorkingDir,
// TODO: add parsing of Ulimits
Rlimits: []specs.Rlimit{
{
Type: "RLIMIT_NOFILE",
Hard: uint64(1024),
Soft: uint64(1024),
},
},
NoNewPrivileges: true,
ApparmorProfile: c.AppArmorProfile,
},
Root: specs.Root{
Path: "rootfs",
Readonly: c.HostConfig.ReadonlyRootfs,
},
Mounts: []specs.Mount{},
Linux: specs.Linux{
Namespaces: []specs.Namespace{
{
Type: "ipc",
},
{
Type: "uts",
},
{
Type: "mount",
},
},
UIDMappings: []specs.IDMapping{
{
ContainerID: 0,
HostID: idroot,
Size: idlen,
},
},
GIDMappings: []specs.IDMapping{
{
ContainerID: 0,
HostID: idroot,
Size: idlen,
},
},
Resources: &specs.Resources{
Devices: []specs.DeviceCgroup{
{
Allow: false,
Access: sPtr("rwm"),
},
},
DisableOOMKiller: c.HostConfig.Resources.OomKillDisable,
OOMScoreAdj: &c.HostConfig.OomScoreAdj,
Memory: &specs.Memory{
Limit: uint64ptr(c.HostConfig.Resources.Memory),
Reservation: uint64ptr(c.HostConfig.Resources.MemoryReservation),
Swap: uint64ptr(c.HostConfig.Resources.MemorySwap),
Swappiness: uint64ptr(*c.HostConfig.Resources.MemorySwappiness),
Kernel: uint64ptr(c.HostConfig.Resources.KernelMemory),
},
CPU: &specs.CPU{
Shares: uint64ptr(c.HostConfig.Resources.CPUShares),
Quota: uint64ptr(c.HostConfig.Resources.CPUQuota),
Period: uint64ptr(c.HostConfig.Resources.CPUPeriod),
Cpus: &c.HostConfig.Resources.CpusetCpus,
Mems: &c.HostConfig.Resources.CpusetMems,
},
Pids: &specs.Pids{
Limit: &c.HostConfig.Resources.PidsLimit,
},
BlockIO: &specs.BlockIO{
Weight: &c.HostConfig.Resources.BlkioWeight,
// TODO: add parsing for Throttle/Weight Devices
},
},
RootfsPropagation: "",
},
}
// make sure the current working directory is not blank
if config.Process.Cwd == "" {
config.Process.Cwd = DefaultCurrentWorkingDirectory
}
// get the user
if c.Config.User != "" {
u, err := user.LookupUser(c.Config.User)
if err != nil {
config.Process.User = specs.User{
UID: uint32(u.Uid),
GID: uint32(u.Gid),
}
} else {
//return nil, fmt.Errorf("Looking up user (%s) failed: %v", c.Config.User, err)
logrus.Warnf("Looking up user (%s) failed: %v", c.Config.User, err)
}
}
// add the additional groups
for _, group := range c.HostConfig.GroupAdd {
g, err := user.LookupGroup(group)
if err != nil {
return nil, fmt.Errorf("Looking up group (%s) failed: %v", group, err)
}
config.Process.User.AdditionalGids = append(config.Process.User.AdditionalGids, uint32(g.Gid))
}
// get the hostname, if the hostname is the name as the first 12 characters of the id,
// then set the hostname as the container name
if c.ID[:12] == c.Config.Hostname {
config.Hostname = strings.TrimPrefix(c.Name, "/")
}
// set privileged
if c.HostConfig.Privileged {
// allow all caps
capabilities = execdriver.GetAllCapabilities()
}
// get the capabilities
config.Process.Capabilities, err = execdriver.TweakCapabilities(capabilities, c.HostConfig.CapAdd, c.HostConfig.CapDrop)
if err != nil {
return nil, fmt.Errorf("setting capabilities failed: %v", err)
}
// add CAP_ prefix
// TODO: this is awful
for i, cap := range config.Process.Capabilities {
if !strings.HasPrefix(cap, "CAP_") {
config.Process.Capabilities[i] = fmt.Sprintf("CAP_%s", cap)
}
}
// if we have a container that needs a terminal but no env vars, then set
// default env vars for the terminal to function
if config.Process.Terminal && len(config.Process.Env) <= 0 {
config.Process.Env = DefaultTerminalEnv
}
if config.Process.Terminal {
// make sure we have TERM set
var termSet bool
for _, env := range config.Process.Env {
if strings.HasPrefix(env, "TERM=") {
termSet = true
break
}
}
if !termSet {
// set the term variable
config.Process.Env = append(config.Process.Env, fmt.Sprintf("TERM=%s", DefaultTerminal))
}
}
// check namespaces
if !c.HostConfig.NetworkMode.IsHost() {
config.Linux.Namespaces = append(config.Linux.Namespaces, specs.Namespace{
Type: "network",
})
}
if !c.HostConfig.PidMode.IsHost() {
config.Linux.Namespaces = append(config.Linux.Namespaces, specs.Namespace{
Type: "pid",
})
}
if c.HostConfig.UsernsMode.Valid() && !c.HostConfig.NetworkMode.IsHost() && !c.HostConfig.PidMode.IsHost() && !c.HostConfig.Privileged {
config.Linux.Namespaces = append(config.Linux.Namespaces, specs.Namespace{
Type: "user",
})
} else {
// reset uid and gid mappings
config.Linux.UIDMappings = []specs.IDMapping{}
config.Linux.GIDMappings = []specs.IDMapping{}
}
// get mounts
mounts := map[string]bool{}
for _, mount := range c.Mounts {
mounts[mount.Destination] = true
var opt []string
if mount.RW {
opt = append(opt, "rw")
}
if mount.Mode != "" {
opt = append(opt, mount.Mode)
}
opt = append(opt, []string{"rbind", "rprivate"}...)
config.Mounts = append(config.Mounts, specs.Mount{
Destination: mount.Destination,
Type: "bind",
Source: mount.Source,
Options: opt,
})
}
// add /etc/hosts and /etc/resolv.conf if we should have networking
if c.HostConfig.NetworkMode != "none" && c.HostConfig.NetworkMode != "host" {
DefaultMounts = append(DefaultMounts, NetworkMounts...)
}
// if we aren't doing something crazy like mounting a default mount ourselves,
// the we can mount it the default way
for _, mount := range DefaultMounts {
if _, ok := mounts[mount.Destination]; !ok {
config.Mounts = append(config.Mounts, mount)
}
}
// fix default mounts for cgroups and devpts without user namespaces
// see: https://github.com/opencontainers/runc/issues/225#issuecomment-136519577
if len(config.Linux.UIDMappings) == 0 {
for k, mount := range config.Mounts {
switch mount.Destination {
case "/sys/fs/cgroup":
config.Mounts[k].Options = append(config.Mounts[k].Options, "ro")
case "/dev/pts":
config.Mounts[k].Options = append(config.Mounts[k].Options, "gid=5")
}
}
}
// parse additional groups and add them to gid mappings
if err := parseMappings(config, c.HostConfig); err != nil {
return nil, err
}
// parse devices
if err := parseDevices(config, c.HostConfig); err != nil {
return nil, err
}
// parse security opt
if err := parseSecurityOpt(config, c.HostConfig); err != nil {
return nil, err
}
// set privileged
if c.HostConfig.Privileged {
if !c.HostConfig.ReadonlyRootfs {
// clear readonly for cgroup
// config.Mounts["cgroup"] = DefaultMountpoints["cgroup"]
}
}
return config, nil
}