vendor runc v1.0.0-rc91

Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>

vendor/github.com/opencontainers/runc/libcontainer/configs/cgroup_linux.go

@@ -1,5 +1,9 @@
 package configs
 
+import (
+	systemdDbus "github.com/coreos/go-systemd/v22/dbus"
+)
+
 type FreezerState string
 
 const (
@@ -29,18 +33,16 @@ type Cgroup struct {
 
 	// Resources contains various cgroups settings to apply
 	*Resources
+
+	// SystemdProps are any additional properties for systemd,
+	// derived from org.systemd.property.xxx annotations.
+	// Ignored unless systemd is used for managing cgroups.
+	SystemdProps []systemdDbus.Property `json:"-"`
 }
 
 type Resources struct {
-	// If this is true allow access to any kind of device within the container.  If false, allow access only to devices explicitly listed in the allowed_devices list.
-	// Deprecated
-	AllowAllDevices *bool `json:"allow_all_devices,omitempty"`
-	// Deprecated
-	AllowedDevices []*Device `json:"allowed_devices,omitempty"`
-	// Deprecated
-	DeniedDevices []*Device `json:"denied_devices,omitempty"`
-
-	Devices []*Device `json:"devices"`
+	// Devices is the set of access rules for devices in the container.
+	Devices []*DeviceRule `json:"devices"`
 
 	// Memory limit (in bytes)
 	Memory int64 `json:"memory"`
@@ -124,7 +126,4 @@ type Resources struct {
 
 	// CpuWeight sets a proportional bandwidth limit.
 	CpuWeight uint64 `json:"cpu_weight"`
-
-	// CpuMax sets she maximum bandwidth limit (format: max period).
-	CpuMax string `json:"cpu_max"`
 }

vendor/github.com/opencontainers/runc/libcontainer/configs/config.go

@@ -8,7 +8,7 @@ import (
 	"time"
 
 	"github.com/opencontainers/runtime-spec/specs-go"
-
+	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 )
 
@@ -70,9 +70,10 @@ type Arg struct {
 
 // Syscall is a rule to match a syscall in Seccomp
 type Syscall struct {
-	Name   string `json:"name"`
-	Action Action `json:"action"`
-	Args   []*Arg `json:"args"`
+	Name     string `json:"name"`
+	Action   Action `json:"action"`
+	ErrnoRet *uint  `json:"errnoRet"`
+	Args     []*Arg `json:"args"`
 }
 
 // TODO Windows. Many of these fields should be factored out into those parts
@@ -175,7 +176,7 @@ type Config struct {
 
 	// Hooks are a collection of actions to perform at various container lifecycle events.
 	// CommandHooks are serialized to JSON, but other hooks are not.
-	Hooks *Hooks
+	Hooks Hooks
 
 	// Version is the version of opencontainer specification that is supported.
 	Version string `json:"version"`
@@ -202,17 +203,50 @@ type Config struct {
 	RootlessCgroups bool `json:"rootless_cgroups,omitempty"`
 }
 
-type Hooks struct {
+type HookName string
+type HookList []Hook
+type Hooks map[HookName]HookList
+
+const (
 	// Prestart commands are executed after the container namespaces are created,
 	// but before the user supplied command is executed from init.
-	Prestart []Hook
+	// Note: This hook is now deprecated
+	// Prestart commands are called in the Runtime namespace.
+	Prestart HookName = "prestart"
+
+	// CreateRuntime commands MUST be called as part of the create operation after
+	// the runtime environment has been created but before the pivot_root has been executed.
+	// CreateRuntime is called immediately after the deprecated Prestart hook.
+	// CreateRuntime commands are called in the Runtime Namespace.
+	CreateRuntime = "createRuntime"
+
+	// CreateContainer commands MUST be called as part of the create operation after
+	// the runtime environment has been created but before the pivot_root has been executed.
+	// CreateContainer commands are called in the Container namespace.
+	CreateContainer = "createContainer"
+
+	// StartContainer commands MUST be called as part of the start operation and before
+	// the container process is started.
+	// StartContainer commands are called in the Container namespace.
+	StartContainer = "startContainer"
 
 	// Poststart commands are executed after the container init process starts.
-	Poststart []Hook
+	// Poststart commands are called in the Runtime Namespace.
+	Poststart = "poststart"
 
 	// Poststop commands are executed after the container init process exits.
-	Poststop []Hook
-}
+	// Poststop commands are called in the Runtime Namespace.
+	Poststop = "poststop"
+)
+
+// TODO move this to runtime-spec
+// See: https://github.com/opencontainers/runtime-spec/pull/1046
+const (
+	Creating = "creating"
+	Created  = "created"
+	Running  = "running"
+	Stopped  = "stopped"
+)
 
 type Capabilities struct {
 	// Bounding is the set of capabilities checked by the kernel.
@@ -227,32 +261,39 @@ type Capabilities struct {
 	Ambient []string
 }
 
-func (hooks *Hooks) UnmarshalJSON(b []byte) error {
-	var state struct {
-		Prestart  []CommandHook
-		Poststart []CommandHook
-		Poststop  []CommandHook
+func (hooks HookList) RunHooks(state *specs.State) error {
+	for i, h := range hooks {
+		if err := h.Run(state); err != nil {
+			return errors.Wrapf(err, "Running hook #%d:", i)
+		}
 	}
 
+	return nil
+}
+
+func (hooks *Hooks) UnmarshalJSON(b []byte) error {
+	var state map[HookName][]CommandHook
+
 	if err := json.Unmarshal(b, &state); err != nil {
 		return err
 	}
 
-	deserialize := func(shooks []CommandHook) (hooks []Hook) {
-		for _, shook := range shooks {
-			hooks = append(hooks, shook)
+	*hooks = Hooks{}
+	for n, commandHooks := range state {
+		if len(commandHooks) == 0 {
+			continue
 		}
 
-		return hooks
+		(*hooks)[n] = HookList{}
+		for _, h := range commandHooks {
+			(*hooks)[n] = append((*hooks)[n], h)
+		}
 	}
 
-	hooks.Prestart = deserialize(state.Prestart)
-	hooks.Poststart = deserialize(state.Poststart)
-	hooks.Poststop = deserialize(state.Poststop)
 	return nil
 }
 
-func (hooks Hooks) MarshalJSON() ([]byte, error) {
+func (hooks *Hooks) MarshalJSON() ([]byte, error) {
 	serialize := func(hooks []Hook) (serializableHooks []CommandHook) {
 		for _, hook := range hooks {
 			switch chook := hook.(type) {
@@ -267,9 +308,12 @@ func (hooks Hooks) MarshalJSON() ([]byte, error) {
 	}
 
 	return json.Marshal(map[string]interface{}{
-		"prestart":  serialize(hooks.Prestart),
-		"poststart": serialize(hooks.Poststart),
-		"poststop":  serialize(hooks.Poststop),
+		"prestart":        serialize((*hooks)[Prestart]),
+		"createRuntime":   serialize((*hooks)[CreateRuntime]),
+		"createContainer": serialize((*hooks)[CreateContainer]),
+		"startContainer":  serialize((*hooks)[StartContainer]),
+		"poststart":       serialize((*hooks)[Poststart]),
+		"poststop":        serialize((*hooks)[Poststop]),
 	})
 }
 

vendor/github.com/opencontainers/runc/libcontainer/configs/device.go

@@ -1,8 +1,12 @@
 package configs
 
 import (
+	"errors"
 	"fmt"
 	"os"
+	"strconv"
+
+	"golang.org/x/sys/unix"
 )
 
 const (
@@ -12,21 +16,11 @@ const (
 // TODO Windows: This can be factored out in the future
 
 type Device struct {
-	// Device type, block, char, etc.
-	Type rune `json:"type"`
+	DeviceRule
 
 	// Path to the device.
 	Path string `json:"path"`
 
-	// Major is the device's major number.
-	Major int64 `json:"major"`
-
-	// Minor is the device's minor number.
-	Minor int64 `json:"minor"`
-
-	// Cgroup permissions format, rwm.
-	Permissions string `json:"permissions"`
-
 	// FileMode permission bits for the device.
 	FileMode os.FileMode `json:"file_mode"`
 
@@ -35,23 +29,154 @@ type Device struct {
 
 	// Gid of the device.
 	Gid uint32 `json:"gid"`
+}
 
-	// Write the file to the allowed list
+// DevicePermissions is a cgroupv1-style string to represent device access. It
+// has to be a string for backward compatibility reasons, hence why it has
+// methods to do set operations.
+type DevicePermissions string
+
+const (
+	deviceRead uint = (1 << iota)
+	deviceWrite
+	deviceMknod
+)
+
+func (p DevicePermissions) toSet() uint {
+	var set uint
+	for _, perm := range p {
+		switch perm {
+		case 'r':
+			set |= deviceRead
+		case 'w':
+			set |= deviceWrite
+		case 'm':
+			set |= deviceMknod
+		}
+	}
+	return set
+}
+
+func fromSet(set uint) DevicePermissions {
+	var perm string
+	if set&deviceRead == deviceRead {
+		perm += "r"
+	}
+	if set&deviceWrite == deviceWrite {
+		perm += "w"
+	}
+	if set&deviceMknod == deviceMknod {
+		perm += "m"
+	}
+	return DevicePermissions(perm)
+}
+
+// Union returns the union of the two sets of DevicePermissions.
+func (p DevicePermissions) Union(o DevicePermissions) DevicePermissions {
+	lhs := p.toSet()
+	rhs := o.toSet()
+	return fromSet(lhs | rhs)
+}
+
+// Difference returns the set difference of the two sets of DevicePermissions.
+// In set notation, A.Difference(B) gives you A\B.
+func (p DevicePermissions) Difference(o DevicePermissions) DevicePermissions {
+	lhs := p.toSet()
+	rhs := o.toSet()
+	return fromSet(lhs &^ rhs)
+}
+
+// Intersection computes the intersection of the two sets of DevicePermissions.
+func (p DevicePermissions) Intersection(o DevicePermissions) DevicePermissions {
+	lhs := p.toSet()
+	rhs := o.toSet()
+	return fromSet(lhs & rhs)
+}
+
+// IsEmpty returns whether the set of permissions in a DevicePermissions is
+// empty.
+func (p DevicePermissions) IsEmpty() bool {
+	return p == DevicePermissions("")
+}
+
+// IsValid returns whether the set of permissions is a subset of valid
+// permissions (namely, {r,w,m}).
+func (p DevicePermissions) IsValid() bool {
+	return p == fromSet(p.toSet())
+}
+
+type DeviceType rune
+
+const (
+	WildcardDevice DeviceType = 'a'
+	BlockDevice    DeviceType = 'b'
+	CharDevice     DeviceType = 'c' // or 'u'
+	FifoDevice     DeviceType = 'p'
+)
+
+func (t DeviceType) IsValid() bool {
+	switch t {
+	case WildcardDevice, BlockDevice, CharDevice, FifoDevice:
+		return true
+	default:
+		return false
+	}
+}
+
+func (t DeviceType) CanMknod() bool {
+	switch t {
+	case BlockDevice, CharDevice, FifoDevice:
+		return true
+	default:
+		return false
+	}
+}
+
+func (t DeviceType) CanCgroup() bool {
+	switch t {
+	case WildcardDevice, BlockDevice, CharDevice:
+		return true
+	default:
+		return false
+	}
+}
+
+type DeviceRule struct {
+	// Type of device ('c' for char, 'b' for block). If set to 'a', this rule
+	// acts as a wildcard and all fields other than Allow are ignored.
+	Type DeviceType `json:"type"`
+
+	// Major is the device's major number.
+	Major int64 `json:"major"`
+
+	// Minor is the device's minor number.
+	Minor int64 `json:"minor"`
+
+	// Permissions is the set of permissions that this rule applies to (in the
+	// cgroupv1 format -- any combination of "rwm").
+	Permissions DevicePermissions `json:"permissions"`
+
+	// Allow specifies whether this rule is allowed.
 	Allow bool `json:"allow"`
 }
 
-func (d *Device) CgroupString() string {
-	return fmt.Sprintf("%c %s:%s %s", d.Type, deviceNumberString(d.Major), deviceNumberString(d.Minor), d.Permissions)
-}
-
-func (d *Device) Mkdev() int {
-	return int((d.Major << 8) | (d.Minor & 0xff) | ((d.Minor & 0xfff00) << 12))
-}
-
-// deviceNumberString converts the device number to a string return result.
-func deviceNumberString(number int64) string {
-	if number == Wildcard {
-		return "*"
+func (d *DeviceRule) CgroupString() string {
+	var (
+		major = strconv.FormatInt(d.Major, 10)
+		minor = strconv.FormatInt(d.Minor, 10)
+	)
+	if d.Major == Wildcard {
+		major = "*"
 	}
-	return fmt.Sprint(number)
+	if d.Minor == Wildcard {
+		minor = "*"
+	}
+	return fmt.Sprintf("%c %s:%s %s", d.Type, major, minor, d.Permissions)
+}
+
+func (d *DeviceRule) Mkdev() (uint64, error) {
+	if d.Major == Wildcard || d.Minor == Wildcard {
+		return 0, errors.New("cannot mkdev() device with wildcards")
+	}
+	return unix.Mkdev(uint32(d.Major), uint32(d.Minor)), nil
 }

vendor/github.com/opencontainers/runc/libcontainer/configs/device_defaults.go

@@ -1,111 +0,0 @@
-// +build linux
-
-package configs
-
-var (
-	// DefaultSimpleDevices are devices that are to be both allowed and created.
-	DefaultSimpleDevices = []*Device{
-		// /dev/null and zero
-		{
-			Path:        "/dev/null",
-			Type:        'c',
-			Major:       1,
-			Minor:       3,
-			Permissions: "rwm",
-			FileMode:    0666,
-		},
-		{
-			Path:        "/dev/zero",
-			Type:        'c',
-			Major:       1,
-			Minor:       5,
-			Permissions: "rwm",
-			FileMode:    0666,
-		},
-
-		{
-			Path:        "/dev/full",
-			Type:        'c',
-			Major:       1,
-			Minor:       7,
-			Permissions: "rwm",
-			FileMode:    0666,
-		},
-
-		// consoles and ttys
-		{
-			Path:        "/dev/tty",
-			Type:        'c',
-			Major:       5,
-			Minor:       0,
-			Permissions: "rwm",
-			FileMode:    0666,
-		},
-
-		// /dev/urandom,/dev/random
-		{
-			Path:        "/dev/urandom",
-			Type:        'c',
-			Major:       1,
-			Minor:       9,
-			Permissions: "rwm",
-			FileMode:    0666,
-		},
-		{
-			Path:        "/dev/random",
-			Type:        'c',
-			Major:       1,
-			Minor:       8,
-			Permissions: "rwm",
-			FileMode:    0666,
-		},
-	}
-	DefaultAllowedDevices = append([]*Device{
-		// allow mknod for any device
-		{
-			Type:        'c',
-			Major:       Wildcard,
-			Minor:       Wildcard,
-			Permissions: "m",
-		},
-		{
-			Type:        'b',
-			Major:       Wildcard,
-			Minor:       Wildcard,
-			Permissions: "m",
-		},
-
-		{
-			Path:        "/dev/console",
-			Type:        'c',
-			Major:       5,
-			Minor:       1,
-			Permissions: "rwm",
-		},
-		// /dev/pts/ - pts namespaces are "coming soon"
-		{
-			Path:        "",
-			Type:        'c',
-			Major:       136,
-			Minor:       Wildcard,
-			Permissions: "rwm",
-		},
-		{
-			Path:        "",
-			Type:        'c',
-			Major:       5,
-			Minor:       2,
-			Permissions: "rwm",
-		},
-
-		// tuntap
-		{
-			Path:        "",
-			Type:        'c',
-			Major:       10,
-			Minor:       200,
-			Permissions: "rwm",
-		},
-	}, DefaultSimpleDevices...)
-	DefaultAutoCreatedDevices = append([]*Device{}, DefaultSimpleDevices...)
-)